阅读说明：本技术 一种盾构机姿态优化方法、系统及模型训练方法、系统 (Shield tunneling machine attitude optimization method and system and model training method and system ) 是由 张东明 吴惠明 常佳奇 李刚 黄宏伟 于 2021-08-10 设计创作，主要内容包括：本发明涉及一种盾构机姿态优化方法、系统及模型训练方法、系统。本发明能够使用聚类方法依据盾构施工历史数据对地层种类组合进行分类,从而作为后续各种机器学习方法的分段建模提供依据。依据分段结果,使用多种机器学习模型建模并根据准确率进行权重分配,就可以得到适用于各种地层种类组合的集成学习模型。所述集成学习模型可以针对指定地层,为具有更高准确率的机器学习模型分配更高的权重,因此具有更高的准确率,而且不同地层类型采用的机器学习模型不同,因此不会出现因地层突变导致的盾构机姿态预测不准现象,更加充分地利用了历史数据。(The invention relates to a method and a system for optimizing the posture of a shield tunneling machine and a method and a system for training a model. The invention can classify stratum type combinations by using a clustering method according to shield construction historical data, thereby providing a basis for segmented modeling of various subsequent machine learning methods. And according to the segmentation result, modeling by using various machine learning models and carrying out weight distribution according to the accuracy, so that the integrated learning model suitable for various stratum type combinations can be obtained. The integrated learning model can distribute higher weight for the machine learning model with higher accuracy aiming at the specified stratum, so that the integrated learning model has higher accuracy, and the machine learning models adopted by different stratum types are different, so that the phenomenon of inaccurate prediction of the posture of the shield machine caused by stratum mutation can not occur, and the historical data is more fully utilized.)

1. A method for training a shield tunneling machine attitude prediction model is characterized by comprising the following steps:

dividing shield construction historical data according to stratum type combinations appearing on the tunnel face at the same time by using a clustering algorithm to obtain n_k0A stratum combination sample set corresponding to stratum type combinations of each category; the shield construction historical data comprises shield construction parameter data, shield machine attitude parameter data and stratum data;

partitioning the stratigraphic combined sample set of each category into at least one interval sample set in time continuity; each sample in the inter-range sample set includes a t corresponding to each other_nShield construction parameter data and t at time_n+1The attitude parameter data of the shield machine at the moment, t_n+1The time is the t_nThe next moment in time;

selecting part of samples in each interval sample set to form a training set, and using the t_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The attitude parameter data of the shield machine at the moment is output parameters, and a machine learning model is trained; forming a machine learning model after test training of the test set by using another part of samples in the interval sample set, and calculating the accuracy rate of the machine learning model after training;

and determining a final trained machine learning model as a shield machine posture prediction model according to the accuracy corresponding to each interval sample set.

2. The method for training the shield tunneling machine attitude prediction model according to claim 1, wherein the machine learning model at least comprises 2 algorithm models, the training set respectively trains at least 2 algorithm models, and the test set respectively tests at least 2 trained algorithm models.

3. The method for training the attitude prediction model of the shield machine according to claim 2, wherein the step of determining the final trained machine learning model as the attitude prediction model of the shield machine according to the accuracy corresponding to each interval sample set specifically comprises:

calculating the average value of the accuracy of each trained algorithm model corresponding to all the interval sample sets of the same category, and determining the trained algorithm model corresponding to the interval sample set with the highest accuracy as a class target model;

calculating model weights in the same category according to the average value of the accuracy rates of the various trained algorithm models; and distributing the class target model according to the model weight to obtain a shield machine attitude prediction model corresponding to the class.

4. The method for training the attitude prediction model of the shield tunneling machine according to claim 1, wherein the shield construction historical data is divided according to the combination of stratum types appearing on the tunnel face at the same time by using a clustering algorithm to obtain n_k0The stratum combination sample set corresponding to the stratum type combination of each category specifically includes:

acquiring shield construction parameter data, shield machine attitude parameter data and stratum data, and recording as shield construction historical data; the shield construction parameter data comprises: cutter parameters, jack parameters, working condition parameters and grouting parameters; the shield tunneling machine attitude parameter data comprises: attitude data of the cutter head and the shield tail; the formation data includes: stratum type combination and stratum characteristic parameters simultaneously appearing on the tunnel face;

calculating the total number n of stratum category combinations simultaneously appearing on the face_s；

Randomly selecting n in the shield construction historical data_sUsing the sample as initial clustering centerSetting a maximum number of iterations n_maxWith minimum error change epsilon_min；

Calculating the distance from each sample to the clustering center, and taking the class with the minimum distance as the class of the sample;

recalculating new cluster centers for each category;

determine whether it isIf not, taking the new clustering center as the clustering center of the category, and returning to execute the step of calculating the distance between each sample and the clustering center; if yes, determining the obtained categories of all stratum category combinations and a stratum combination sample set corresponding to each category; the class is n_k0And (4) respectively.

5. The method for training the attitude prediction model of the shield tunneling machine according to claim 1, wherein a part of samples in each interval sample set is selected to form a training set, and the t is used as the t_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The method includes that the attitude parameter data of the shield machine at the moment are output parameters, and a machine learning model is trained, and the method specifically comprises the following steps:

establishing a machine learning model for each interval sample set;

training a corresponding machine learning model by utilizing a training set in the interval sample set;

and adjusting the hyper-parameters of the machine learning model by using a grid method in the training process to obtain the trained machine learning model.

6. The method for training the shield tunneling machine attitude prediction model according to claim 3, wherein the machine learning model comprises an MLP model, an SVM model and a GBR model, and the step of obtaining the shield tunneling machine attitude prediction model specifically comprises:

respectively calculating the average value m of the prediction accuracy rates of the MLP model, the SVM model and the GBR model on all interval sample sets of the same class_j，s_jAnd g_j；

According to the average value m_j，s_jAnd g_jCalculating model weightsAnd

and summing the MLP model, the SVM model and the GBR model according to the model weight to obtain a shield tunneling machine attitude prediction model.

7. The method for training the shield tunneling machine attitude prediction model according to claim 1, wherein the obtaining of the stratum combination sample set specifically comprises:

preprocessing the shield construction historical data, and removing abnormal values and values of a shutdown section to obtain shield construction reference data;

obtaining a shield machine attitude parameters according to the shield construction reference data, and recording the attitude parameters asb shield construction parameters, noteThe number of c + d formation parameters,wherein the content of the first and second substances,for the thickness of each combination of formation types on the face,each physical mechanical parameter combined for each stratum type; 1,2 … n;

merging the a shield machine attitude parameters, the b shield construction parameters and the c + d stratum parameters according to time, and simultaneously adding shield machine attitude parameter data at the next moment, wherein the shield machine attitude parameter data at the next moment are recorded asObtaining the stratum combination sample set, wherein the stratum combination sample set is as follows:

8. a shield tunneling machine attitude optimization method is characterized by comprising the following steps:

inputting the shield construction parameters at the current moment into a shield machine attitude prediction model to obtain shield machine attitude parameter data at the next moment; the shield machine attitude prediction model is obtained by the shield machine attitude prediction model training method of claim 1;

judging whether the attitude parameter data of the shield tunneling machine at the next moment exceeds a preset deviation limit value;

if yes, adjusting the shield construction parameters to serve as new shield construction parameters at the current moment, and returning to execute the step of inputting the shield construction parameters at the current moment into a shield machine attitude prediction model;

if not, controlling the shield machine to act according to the shield machine attitude parameter data at the next moment, and storing the shield machine attitude parameter data at the next moment to the shield construction parameters.

9. The utility model provides a shield constructs quick-witted attitude prediction model training system which characterized in that includes:

the stratum type combination division module is used for dividing the shield construction historical data according to stratum type combinations which simultaneously appear on the tunnel face by using a clustering algorithm to obtain n_k0A stratum combination sample set corresponding to stratum type combinations of each category; the shield construction historical data comprises shield construction parameter data, shield machine attitude parameter data and stratum data;

the interval sample set dividing module is used for dividing the stratum combination sample set of each category into at least one interval sample set according to time continuity; each sample in the inter-range sample set includes a t corresponding to each other_nShield construction parameter data and t at time_n+1The attitude parameter data of the shield machine at the moment, t_n+1The time is the t_nLower of the momentA moment;

a machine learning model training module for selecting part of samples in each interval sample set to form a training set, and using the t_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The attitude parameter data of the shield machine at the moment is output parameters, and a machine learning model is trained; forming a machine learning model after test training of the test set by using another part of samples in the interval sample set, and calculating the accuracy rate of the machine learning model after training;

and the model establishing module is used for determining a final trained machine learning model as the model for predicting the attitude of the shield machine according to the accuracy corresponding to each interval sample set.

10. A shield constructs quick-witted attitude optimization system, its characterized in that includes:

the shield machine attitude parameter data acquisition module is used for inputting the shield construction parameters at the current moment into the shield machine attitude prediction model to obtain the shield machine attitude parameter data at the next moment;

the judging module is used for judging whether the attitude parameter data of the shield machine at the next moment exceeds a preset deviation limit value;

if yes, adjusting the shield construction parameters to serve as new shield construction parameters at the current moment, and returning to execute the step of inputting the shield construction parameters at the current moment into a shield machine attitude prediction model;

if not, controlling the shield machine to act according to the shield machine attitude parameter data at the next moment, and storing the shield machine attitude parameter data at the next moment to the shield construction parameters.

Technical Field

The invention relates to the technical field of shield tunnel construction, in particular to a method and a system for optimizing the posture of a shield machine and a method and a system for training a model.

Background

With the rapid development of the Chinese social economy and production, a great deal of tunnel engineering emerges, and the construction standard of the tunnel is continuously improved. The shield method is widely applied to the construction of subway tunnels because of its high construction speed and small influence on the surrounding environment. However, the shield method has numerous construction parameters, and if the shield construction parameters cannot be reasonably set, the shield method cannot exert the construction advantages and even cause safety accidents. The posture of the shield machine is an important index required to be controlled in the shield method construction process, and if shield construction parameters are set improperly, the shield machine deviates from the design axis of the tunnel, so that the problems that the ground surface subsidence is too large, the segment assembly is difficult, the tunnel cannot be communicated and the like are caused. The reasonable setting of the shield construction parameters ensures that the posture of the shield machine meets the requirements, and has important significance for fully playing the advantages of the shield method, improving the construction efficiency and reducing the construction risk.

Therefore, how to design a method capable of accurately predicting the attitude of the shield tunneling machine becomes a technical problem to be solved in the field.

Disclosure of Invention

The invention aims to provide a method and a system for optimizing the posture of a shield tunneling machine and a method and a system for training a model. The invention can solve the problem of poor attitude control of the shield tunneling machine.

In order to achieve the purpose, the invention provides the following scheme:

a training method for a shield tunneling machine attitude prediction model comprises the following steps:

dividing shield construction historical data according to stratum type combinations appearing on the tunnel face at the same time by using a clustering algorithm to obtain n_k0A stratum combination sample set corresponding to stratum type combinations of each category; the shield construction historical data comprises shield construction parameter data, shield machine attitude parameter data and stratum data;

partitioning the stratigraphic combined sample set of each category into at least one interval sample set in time continuity; each sample in the inter-range sample set includes a t corresponding to each other_nShield construction parameter data and t at time_n+1The attitude parameter data of the shield machine at the moment, t_n+1The time is the t_nThe next moment in time;

selecting part of samples in each interval sample set to form a training set, and using the t_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The attitude parameter data of the shield machine at the moment is output parameters, and a machine learning model is trained; forming a machine learning model after test training of the test set by using another part of samples in the interval sample set, and calculating the accuracy rate of the machine learning model after training;

and determining a final trained machine learning model as a shield machine posture prediction model according to the accuracy corresponding to each interval sample set.

Optionally, the machine learning model at least includes 2 algorithm models, the training set trains at least 2 algorithm models, and the test set tests at least 2 trained algorithm models.

Optionally, the step of determining a final trained machine learning model according to the accuracy corresponding to each interval sample set is a shield machine posture prediction model, and specifically includes:

calculating the average value of the accuracy of each trained algorithm model corresponding to all the interval sample sets of the same category, and determining the trained algorithm model corresponding to the interval sample set with the highest accuracy as a class target model;

calculating model weights in the same category according to the average value of the accuracy rates of the various trained algorithm models; and distributing the class target model according to the model weight to obtain a shield machine attitude prediction model corresponding to the class.

Optionally, the shield construction historical data is divided according to the stratum type combination appearing on the tunnel face at the same time by using a clustering algorithm to obtain n_k0The stratum combination sample set corresponding to the stratum type combination of each category specifically includes:

acquiring shield construction parameter data, shield machine attitude parameter data and stratum data, and recording as shield construction historical data; the shield construction parameter data comprises: cutter parameters, jack parameters, working condition parameters and grouting parameters; the shield tunneling machine attitude parameter data comprises: attitude data of the cutter head and the shield tail; the formation data includes: stratum type combination and stratum characteristic parameters simultaneously appearing on the tunnel face;

calculating the total number n of stratum category combinations simultaneously appearing on the face_s；

Randomly selecting n in the shield construction historical data_sUsing the sample as initial clustering centerSetting a maximum number of iterations n_maxWith minimum error change epsilon_min；

Calculating the distance from each sample to the clustering center, and taking the class with the minimum distance as the class of the sample;

recalculating new cluster centers for each category;

judging whether an iteration termination condition is reached, if not, taking the new clustering center as the clustering center of the category, and returning to execute 'calculating the distance between each sample and the clustering center'; if yes, determining the obtained categories of all stratum category combinations and a stratum combination sample set corresponding to each category; said classIs n respectively_k0And (4) respectively.

Optionally, a part of samples in each interval sample set is selected to form a training set, and the t is used_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The method includes that the attitude parameter data of the shield machine at the moment are output parameters, and a machine learning model is trained, and the method specifically comprises the following steps:

establishing a machine learning model for each interval sample set;

training a corresponding machine learning model by utilizing a training set in the interval sample set;

and adjusting the hyper-parameters of the machine learning model by using a grid method in the training process to obtain the trained machine learning model.

Optionally, the machine learning model includes an MLP model, an SVM model and a GBR model, and the step of obtaining the shield machine attitude prediction model specifically includes:

respectively calculating the average value m of the prediction accuracy rates of the MLP model, the SVM model and the GBR model on all interval sample sets of the same class_j，s_jAnd g_j；

According to the average value m_j，s_jAnd g_jCalculating model weightsAnd

and summing the MLP model, the SVM model and the GBR model according to the model weight to obtain a shield tunneling machine attitude prediction model.

Optionally, the obtaining of the stratum combination sample set specifically includes:

preprocessing the shield construction historical data, and removing abnormal values and values of a shutdown section to obtain shield construction reference data;

obtaining a shield machine attitude parameters according to the shield construction reference data, and recording the attitude parameters asb shield construction parameters, noteThe number of c + d formation parameters,wherein the content of the first and second substances,for the thickness of each combination of formation types on the face,each physical mechanical parameter combined for each stratum type; 1,2 … n;

merging the a shield machine attitude parameters, the b shield construction parameters and the c + d stratum parameters according to time, and simultaneously adding shield machine attitude parameter data at the next moment, wherein the shield machine attitude parameter data at the next moment are recorded asObtaining the stratum combination sample set, wherein the stratum combination sample set is as follows:

the invention also provides a shield tunneling machine attitude optimization method, which comprises the following steps:

inputting the shield construction parameters at the current moment into a shield machine attitude prediction model to obtain shield machine attitude parameter data at the next moment; the shield machine attitude prediction model is obtained by the shield machine attitude prediction model training method;

judging whether the attitude parameter data of the shield tunneling machine at the next moment exceeds a preset deviation limit value;

if yes, adjusting the shield construction parameters to serve as new shield construction parameters at the current moment, and returning to execute the step of inputting the shield construction parameters at the current moment into a shield machine attitude prediction model;

if not, controlling the shield machine to act according to the shield machine attitude parameter data at the next moment, and storing the shield machine attitude parameter data at the next moment to the shield construction parameters.

The invention also provides a system for training the attitude prediction model of the shield tunneling machine, which comprises the following components:

the stratum type combination division module is used for dividing the shield construction historical data according to stratum type combinations which simultaneously appear on the tunnel face by using a clustering algorithm to obtain n_k0A stratum combination sample set corresponding to stratum type combinations of each category; the shield construction historical data comprises shield construction parameter data, shield machine attitude parameter data and stratum data;

the interval sample set dividing module is used for dividing the stratum combination sample set of each category into at least one interval sample set according to time continuity; each sample in the inter-range sample set includes a t corresponding to each other_nShield construction parameter data and t at time_n+1The attitude parameter data of the shield machine at the moment, t_n+1The time is the t_nThe next moment in time;

a machine learning model training module for selecting part of samples in each interval sample set to form a training set, and using the t_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The attitude parameter data of the shield machine at the moment is output parameters, and a machine learning model is trained; forming a machine learning model after test training of the test set by using another part of samples in the interval sample set, and calculating the accuracy rate of the machine learning model after training;

and the model establishing module is used for determining a final trained machine learning model as the model for predicting the attitude of the shield machine according to the accuracy corresponding to each interval sample set.

The invention also provides a shield tunneling machine attitude optimization system, which comprises:

the shield machine attitude parameter data acquisition module is used for inputting the shield construction parameters at the current moment into the shield machine attitude prediction model to obtain the shield machine attitude parameter data at the next moment;

the judging module is used for judging whether the attitude parameter data of the shield machine at the next moment exceeds a preset deviation limit value;

if yes, adjusting the shield construction parameters to serve as new shield construction parameters at the current moment, and returning to execute the step of inputting the shield construction parameters at the current moment into a shield machine attitude prediction model;

if not, controlling the shield machine to act according to the shield machine attitude parameter data at the next moment, and storing the shield machine attitude parameter data at the next moment to the shield construction parameters.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the invention provides a method and a system for optimizing the posture of a shield machine, and a method and a system for training a model. And according to the segmentation result, modeling by using various machine learning models and carrying out weight distribution according to the accuracy, so that the integrated learning model suitable for various stratum type combinations can be obtained. The integrated learning model can distribute higher weight for the machine learning model with higher accuracy aiming at the specified stratum, so that the integrated learning model has higher accuracy, and the machine learning models adopted by different stratum types are different, so that the phenomenon of inaccurate prediction of the posture of the shield machine caused by stratum mutation can not occur, and the historical data is more fully utilized.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.

Fig. 1 is a flowchart of a method for training a model for predicting a shield machine attitude according to embodiment 1 of the present invention;

FIG. 2 is a flow chart of the acquisition of a formation composition sample set;

FIG. 3 is a flow chart of stratum category combination classification;

FIG. 4 is a flow chart of a shield tunneling machine attitude prediction model construction;

fig. 5 is a flowchart of a method for optimizing the attitude of a shield tunneling machine according to embodiment 2 of the present invention;

fig. 6 is a structural block diagram of a training system of a shield tunneling machine attitude prediction model according to embodiment 3 of the present invention;

fig. 7 is a structural block diagram of a shield tunneling machine attitude optimization system provided in embodiment 4 of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention aims to develop a training method of a shield machine attitude prediction model and a shield machine attitude optimization method, which can distribute weights to different training models according to the test results of the existing machine learning models aiming at different stratums, and establish an integrated learning model using various machine learning models, thereby ensuring that the integrated learning model can adapt to various stratum conditions and greatly improving the accuracy and reliability of the integrated learning model.

The definitions of the terms of art in connection with the present invention are as follows:

formation: the term "rock layer" is used to refer to either consolidated rock or unconsolidated sediment, i.e., soil, as a layer or group of rock (soil) layers having certain uniform characteristics and properties and distinct from the upper and lower layers.

A shield method: a tunnel construction method is characterized in that a shield machine is used for excavating stratums and splicing tunnel segments.

The shield machine: a construction machine comprises a shell, a cutter head, pushing equipment, assembling equipment and other matched equipment, wherein the shell is a cylinder and plays a role in protection, and the other equipment is arranged inside the shell.

Cutter head: the shield machine is used for cutting the equipment of stratum, is located the shield machine front end, cuts off soil through rotatory extrusion.

Profiling cutters: the cutter extending out along the radial direction of the cutter head can enlarge the cutting range of the cutter head so as to help the shield machine to turn.

Pushing the jack: the jack for propelling the shield tunneling machine to advance is divided into an upper area, a right area, a lower area and a left area, and the posture of the shield tunneling machine can be adjusted by adjusting pressure values of the four areas.

The hinged jack is as follows: the jack for adjusting the shape of the shield machine is divided into an upper area, a right area, a lower area and a left area, and the shape of the shield machine can be adjusted by adjusting the stroke difference of the four areas.

Spiral unearthing machine: the soil body cut by the cutter head is conveyed to a machine on the belt conveyor through the rotation of the spiral member, and the rotational speed of the spiral member determines the unearthing speed.

Belt conveyor: and a belt device for conveying the soil from the spiral soil discharging machine to the residue soil vehicle.

Grouting the pipe after the wall: the slurry is transported to a transport pipeline of a gap between a duct piece and a stratum, a certain gap exists between the duct piece and the stratum after the duct piece is assembled and is separated from a shield tail, cement slurry and the like are required to be injected to fill the gap so as to avoid stratum deformation, and the pressure and the volume of grouting can be adjusted so as to achieve the purposes of minimizing stratum settlement and avoiding ground surface uplift.

A shield tail: the rear part of the shield machine is used for protecting a shell of equipment inside the shield machine.

Designing an axis of the tunnel: the directional line in the extending direction of the tunnel in the design drawing may be a straight line or a curved line.

The cross section of the tunnel is as follows: the tunnel is designed to be excavated into a section vertical to the axis of the tunnel.

Construction mileage: the distance traveled by the shield machine from the originating location to the designated location.

Shield construction parameters: various parameters required to be set in the construction process of the shield machine, such as the rotating speed of a cutter head and the like, and whether the parameter setting is reasonable or not determines the safety and the quality of the shield construction.

The shield machine posture: the shield machine axis is relative to the position relation of the design axis of the tunnel, and comprises cutter horizontal deviation, cutter vertical deviation, shield tail horizontal deviation and shield tail vertical deviation.

Horizontal deviation of the cutter head: the distance between the center of the cutter head and the design axis of the tunnel in the horizontal direction at the same mileage faces the advancing direction of the shield tunneling machine, the center of the cutter head on the left side of the design axis of the tunnel is positive, and the center of the cutter head on the right side of the design axis of the tunnel is negative.

Vertical deviation of the cutter head: the distance between the center of the cutter head and the design axis of the tunnel in the horizontal direction at the same mileage faces the advancing direction of the shield tunneling machine, the center of the cutter head above the design axis of the tunnel is positive, and the center of the cutter head below the design axis of the tunnel is negative.

Horizontal deviation of the shield tail: the distance between the center of the shield tail and the design axis of the tunnel in the horizontal direction at the same mileage faces the advancing direction of the shield machine, and the center of the shield tail on the left side of the design axis of the tunnel is positive, and the center of the shield tail on the right side of the design axis of the tunnel is negative.

Vertical deviation of shield tail: the distance between the center of the shield tail and the design axis of the tunnel in the horizontal direction at the same mileage faces the advancing direction of the shield machine, the center of the shield tail above the design axis of the tunnel is positive, and the center of the shield tail below the design axis of the tunnel is negative.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

Example 1:

referring to fig. 1, the present invention provides a method for training a model for predicting a shield machine attitude, which includes the following steps:

s1: dividing shield construction historical data according to stratum type combinations appearing on the tunnel face at the same time by using a clustering algorithm to obtain n_k0A stratum combination sample set corresponding to stratum type combinations of each category; the shield construction historical data comprises shield construction parameter data, shield machine attitude parameter data and stratum data;

s2: partitioning the stratigraphic combined sample set of each category into at least one interval sample set in time continuity; each sample in the inter-range sample set includes a t corresponding to each other_nShield construction parameter data and t at time_n+1The attitude parameter data of the shield machine at the moment, t_n+1The time is the t_nThe next moment in time;

s3: selecting part of samples in each interval sample set to form a training set, and using the t_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The attitude parameter data of the shield machine at the moment is output parameters, and a machine learning model is trained; forming a machine learning model after test training of the test set by using another part of samples in the interval sample set, and calculating the accuracy rate of the machine learning model after training;

in the present embodiment, the top p of each section is_train% of samples as training set, p_test% of samples as test set, where p_train:p_testBetween 3:1 and 4: 1.

S4: and determining a final trained machine learning model as a shield machine posture prediction model according to the accuracy corresponding to each interval sample set.

As shown in fig. 2 and 3, in step S1, the shield construction history data is divided according to the stratum type combination appearing on the tunnel face at the same time by using a clustering algorithm to obtain n_k0The stratum combination sample set corresponding to the stratum type combination of each category specifically comprises the following steps:

s11: calling an xlwt function packet to obtain shield construction parameter data, shield machine attitude parameter data and stratum data, and recording as shield construction historical data; and the shield construction parameter data and the shield machine attitude parameter data have corresponding data recording systems, such as a Robotec system. The shield construction parameter data comprises: cutter parameters, jack parameters, working condition parameters and grouting parameters; the shield tunneling machine attitude parameter data comprises: attitude data of the cutter head and the shield tail; the formation data includes: stratum type combination and stratum characteristic parameters simultaneously appearing on the tunnel face;

specifically, the tool parameters include: cutter head torque, cutter head rotating speed and profiling cutter stroke;

the jack parameters include: the sectional jacking force and stroke of the jacking jack and the sectional stroke of the hinged jack;

the working condition parameters comprise: the rotating speed of the spiral unearthing machine and the advancing speed of the shield machine;

the grouting parameters comprise: grouting pressure and grouting quantity after the wall;

the attitude data of the cutter head and the shield tail comprise: the cutter horizontal deviation, the cutter vertical deviation, the shield tail horizontal deviation and the shield tail vertical deviation;

the stratum characteristic parameters comprise the thickness of each stratum type combination and each physical and mechanical parameter of each stratum type combination; the physical mechanical parameters comprise cohesive force, internal friction angle, horizontal side pressure coefficient, compression modulus, liquid limit and saturation. The stratum type combination and the corresponding physical and mechanical parameters are obtained from a geological survey report, and the thickness of each stratum type combination is obtained from the drilling data according to an interpolation method.

Acquiring a stratum combination sample set according to the shield construction historical data, wherein the acquisition of the stratum combination sample set specifically comprises the following steps:

s111: preprocessing the shield construction historical data, and removing abnormal values and values of a shutdown section to obtain shield construction reference data;

s112: obtaining a shield machine attitude parameters according to the shield construction reference data, and recording the attitude parameters asb shield construction parameters, noteThe number of c + d formation parameters,wherein the content of the first and second substances,for the thickness of each combination of formation types on the face,each physical mechanical parameter combined for each stratum type; 1,2 … n;

s113: merging the a shield machine attitude parameters, the b shield construction parameters and the c + d stratum parameters according to time, and simultaneously adding shield machine attitude parameter data at the next moment, wherein the shield machine attitude parameter data at the next moment are recorded asObtaining the stratum combination sample set, wherein the stratum combination sample set is as follows:

the pretreatment comprises the following steps: noise reduction and structuring;

the noise reduction comprises the elimination of abnormal values and the elimination of shutdown segments;

the removing of the abnormal value specifically comprises the following steps:

a1: judging whether the data belongs to the value range of the corresponding parameter, if not, deleting the data;

a2: calculating the mean value mu and the variance sigma of each parameter;

a3: judging whether the data of each time of each parameter belongs to a confidence interval [ mu-3 sigma, mu +3 sigma ], and if not, deleting the data;

the elimination of the shutdown segment specifically comprises the following steps:

b1: judging whether the jack stroke changes or not, and if not, rejecting all data in the time period;

the structuralization comprises the extraction and mapping of the shield construction parameter data and the extraction, mapping and normalization of the stratum data;

the extraction and mapping of the shield construction parameter data specifically comprise the following steps:

c1: acquiring monitoring time interval delta t of attitude parameter data of shield tunneling machine_pEach monitoring point corresponds to the moment t_iAnd the attitude of the shield machine at the momentWherein a is the number of the attitude parameters of the shield machine, i is 1,2 … n;

c2: selectingCalculating the average value of the n shield construction parameters within the time range, wherein the average value of the n shield construction parameters is as follows:

wherein, b is the number of shield construction parameters;

c3: taking the average value of the n shield construction parameters as the delta t_pWithin a time interval, t_iObtaining shield construction parameter data consistent with the shield machine attitude parameter data according to the representative value of the shield construction parameter data at the moment;

the extraction and mapping of the formation data specifically comprises the following steps:

d1: various formation thickness distributions at each drilling location obtained by drilling;

d2: obtaining a stratum thickness function h (h) corresponding to the construction mileage x one by one through an interpolation method₁,h₂,h₃,...,h_n)＝f(x)，

D3: calculating the shield machine mileage at each shield machine attitude recording moment;

d4: obtaining the thickness of each stratum type combination on the tunnel face of the shield tunneling machine and each physical and mechanical parameter of each stratum type combination at the moment through the mileage of the shield tunneling machineWherein c is the number of each physical and mechanical parameter of each stratum type combination, d is the number of the thickness of each stratum type combination on the tunnel face, and i is 1,2 … n;

the normalization specifically comprises the following steps:

e1: according to the shield machine attitude, the shield construction parameters, the thickness on the tunnel face of the shield machine, various physical and mechanical parameters of each stratum and the shield machine attitude at the next momentObtaining a sample structure, the sample structure being:

wherein i is 1,2 … n;

e2: according to the sample structure, taking the parameter name of the sample structure as a column name, and taking the parameter value of the sample structure as a cell value to form a two-dimensional table;

e3: and according to the two-dimensional table, carrying out normalization operation on each column of values of the two-dimensional table by using a maximum-minimum normalization method to obtain a stratum combination sample set stored in the table.

S12: calculating the total number n of stratum category combinations simultaneously appearing on the face_s；

S13: randomly selecting n in the shield construction historical data_sUsing the sample as initial clustering centerSetting a maximum number of iterations n_maxWith minimum error change epsilon_min；

S14: calculating the distance from each sample to the clustering center, and taking the class with the minimum distance as the class of the sample;

s15: for each categoryn is the total number of samples in the ith class, and a new clustering center of each class is recalculated, wherein the calculation method of the new clustering center comprises the following steps:

s16: judging whether an iteration termination condition is reached, wherein the iteration termination condition is as follows: whether the maximum number of iterations n has been reached_maxOr the cluster center change positions of all the categories between two iterations are less than the minimum error change epsilon_min(ii) a If not, taking the new clustering center as the clustering center of the category, and returning to execute the step S14; if yes, determining the obtained categories of all stratum category combinations and a stratum combination sample set corresponding to each category; the class is n_k0And (4) respectively.

Dividing the stratum combination sample set of each category into at least one interval sample set according to time continuity to finally obtain n_kAnd (4) each interval.

In step S3, a part of the samples in each of the interval sample sets is selected to form a training set, and the t is used as the reference_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The method includes that the attitude parameter data of the shield machine at the moment are output parameters, and a machine learning model is trained, and the method specifically comprises the following steps:

s31: establishing a machine learning model for each interval sample set;

s32: training a corresponding machine learning model by utilizing a training set in the interval sample set;

s33: and adjusting the hyper-parameters of the machine learning model by using a grid method in the training process to obtain the trained machine learning model.

The machine learning model at least comprises 2 algorithm models, the training set respectively trains at least 2 algorithm models, and the test set respectively tests at least 2 trained algorithm models.

As shown in fig. 4, the machine learning model in this embodiment is 3 algorithm models, which are an MLP model, an SVM model, and a GBR model. The machine learning algorithm modeling of the invention is realized by python language.

The specific process of training the MLP model is as follows:

1) calling an xlwt function packet to read a stratum combination sample set stored in a table;

2) calling a PCA algorithm in the sklern function packet, instantiating the stored characteristics, and reducing the dimension of the input parameters of the stratum combination sample set to more than 90; the input parameters include: a shield machine attitude parameters, b shield construction parameters and c + d stratum parameters;

3) calling MLPRegressor algorithm in sklern, corresponding to n_kIndividual interval establishment n_kAn MLP model;

4) training the corresponding MLP model by using the training set of each interval in sequence, calculating the accuracy of the MLP model by using the corresponding test set, and selecting the hyper-parameters by using a grid method, wherein the main adjusted hyper-parameters are "hidden _ layer _ sizes", "activation", "solvent", "learning _ rate" and "alpla", recommended values are "hidden _ layer _ sizes ═(50,100), activation ═ localization ', solvent ═ sgd ', learning _ rate ═ adaptive ', alpha ═ 0.01", and the MLP model with the highest accuracy is obtained, and the accuracy in each interval is obtained.

The specific process of training the SVM model is as follows:

1) calling an xlwt function packet to read a stratum combination sample set stored in a table;

2) calling a PCA algorithm in the sklern function packet, instantiating the stored characteristics, and reducing the dimension of the input parameters of the stratum combination sample set to more than 90; the input parameters include: a shield machine attitude parameters, b shield construction parameters and c + d stratum parameters;

3) calling SVM algorithm in sklern, corresponding to n_kIndividual interval establishment n_kAn individual SVM model;

4) training the corresponding SVM models by using the training sets of each interval in sequence, calculating the accuracy of the SVM models by using the corresponding test sets, selecting the hyper-parameters by using a grid method, wherein the main adjusted hyper-parameters are ' gamma ', ' kernel ' and ' C ', the recommended values are ' gamma ' -auto ', kernel ' -linear ', and C ' -1 ', and obtaining the SVM model with the highest accuracy and the accuracy in each interval.

The specific process of GBR model training is as follows:

1) calling an xlwt function packet to read a stratum combination sample set stored in a table;

2) calling a PCA algorithm in the sklern function packet, instantiating the stored characteristics, and reducing the dimension of the input parameters of the stratum combination sample set to more than 90; the input parameters include: a shield machine attitude parameters, b shield construction parameters and c + d stratum parameters;

3) calling the GBR algorithm in sklern, corresponding to n_kIndividual interval establishment n_kA GBR model;

4) the corresponding GBR model is trained by using the training set of each interval in turn, the accuracy of the GBR model is calculated by using the corresponding test set, the hyper-parameters are selected by using a grid method, the main adjusted hyper-parameters are 'min _ samples _ split', 'min _ samples _ leaf', 'sub', 'n _ estimators', 'loss', 'leaving _ rate' and 'max _ depth', the recommended values are 'min _ samples _ split' 4, min _ samples _ leaf '2, sub _ sample' 0.5, n _ estimators '500, loss' lag ', leaving _ rate' 0.04, max _ depth '6', the GBR model with the highest accuracy is obtained, and the accuracy on each interval is obtained.

As shown in fig. 4, in step S4, determining the final trained machine learning model as the shield machine posture prediction model according to the accuracy corresponding to each section sample set, specifically includes the following steps:

s41: calculating the average value of the accuracy of each trained algorithm model corresponding to all the interval sample sets of the same category, and determining the trained algorithm model corresponding to the interval sample set with the highest accuracy as a class target model;

s42: calculating model weights in the same category according to the average value of the accuracy rates of the various trained algorithm models; and distributing the class target model according to the model weight to obtain a shield machine attitude prediction model corresponding to the class.

In the same category, the model weight is calculated by the average value of the accuracy of the various trained algorithm models, so that the training result is more accurate, certain relatively absolute factors can be objectively eliminated, and a foundation is laid for obtaining a shield machine attitude prediction model subsequently.

In step S42, the step of obtaining the shield machine attitude prediction model specifically includes:

s421: respectively calculating the average value m of the prediction accuracy rates of the MLP model, the SVM model and the GBR model on all interval sample sets of the same class_j，s_jAnd g_j；

S422: according to the average value m_j，s_jAnd g_jCalculating model weightsAnd

s423: and summing the MLP model, the SVM model and the GBR model according to the model weights to form a result weight distribution matrix of the three models, wherein the row names of the matrix are stratum type combinations, the column names of the matrix are model algorithms, and the shield machine attitude prediction model is obtained according to the weight distribution matrix.

So far, the shield attitude prediction model based on the ensemble learning is already built, and it should be noted that in the embodiment, only three algorithms of the MLP model, the SVM model and the GBR model are introduced, machine learning methods are various, and other methods can be added for the ensemble learning.

Example 2:

referring to fig. 5, the present invention provides a method for optimizing the attitude of a shield machine, including:

1) inputting the shield construction parameters at the current moment into a shield machine attitude prediction model to obtain shield machine attitude parameter data at the next moment; the shield machine attitude prediction model is obtained by the shield machine attitude prediction model training method in embodiment 1, and is not described herein again;

2) judging whether the attitude parameter data of the shield tunneling machine at the next moment exceeds a preset deviation limit value;

if yes, adjusting the shield construction parameters to serve as new shield construction parameters at the current moment, and returning to execute the step of inputting the shield construction parameters at the current moment into a shield machine attitude prediction model;

if not, controlling the shield machine to act according to the shield machine attitude parameter data at the next moment, and storing the shield machine attitude parameter data at the next moment to the shield construction parameters.

In practical application, firstly setting a limit value of the attitude deviation of the shield tunneling machine, wherein the standard is specified to be +/-50 mm, and can be usually +/-30 mm; secondly, select the construction parameter that the driver often adjusted in the shield structure work progress as treating the adjustment parameter, include: jacking force and stroke of a jacking jack, rotating speed and torque of a cutter head, stroke of a hinged jack, grouting amount and propelling speed. Thirdly, taking the construction parameters at the current moment as initial values of construction parameter optimization, and inputting the initial values into the integrated learning prediction model to obtain a predicted value of the shield machine posture at the next moment; and finally, judging whether the predicted value of the posture of the shield machine at the next moment exceeds a deviation limit value, if not, taking the construction parameter as the construction parameter value at the current moment to control the shield machine to act, if so, sequentially adjusting the construction parameters frequently adjusted by a shield driver, and inputting the construction parameters into a model for judging once every time, until the posture of the shield machine at the next moment meets the requirements, or the construction parameters exceed the value range.

Example 3:

the invention provides a training system of a shield machine attitude prediction model, which comprises:

a stratum category combination division module 1 for dividing the shield construction historical data according to stratum category combinations appearing on the tunnel face simultaneously by using a clustering algorithm to obtain n_k0A stratum combination sample set corresponding to stratum type combinations of each category; the shield construction historical data comprises shield construction parameter data, shield machine attitude parameter data and stratum data;

the interval sample set dividing module 2 is used for dividing the stratum combination sample set of each category into at least one interval sample set according to time continuity; each sample in the inter-range sample set includes a t corresponding to each other_nShield construction parameter data and t at time_n+1The attitude parameter data of the shield machine at the moment, t_n+1The time is the t_nThe next moment in time;

a machine learning model training module 3, configured to select a part of samples in each interval sample set to form a training set, and use the t_nThe shield construction parameter data at the moment is taken as an input parameter, and t is taken as_n+1The attitude parameter data of the shield machine at the moment is output parameters, and a machine learning model is trained; forming a machine learning model after test training of the test set by using another part of samples in the interval sample set, and calculating the accuracy rate of the machine learning model after training;

and the model establishing module 4 is used for determining a final trained machine learning model as the model for predicting the attitude of the shield machine according to the accuracy corresponding to each interval sample set.

Example 4:

the invention provides a shield machine attitude optimization system, which comprises:

the shield machine attitude parameter data acquisition module 5 is used for inputting the shield construction parameters at the current moment into the shield machine attitude prediction model to obtain the shield machine attitude parameter data at the next moment;

the judging module 6 is used for judging whether the attitude parameter data of the shield machine at the next moment exceeds a preset deviation limit value;

if yes, adjusting the shield construction parameters to serve as new shield construction parameters at the current moment, and returning to execute the step of inputting the shield construction parameters at the current moment into a shield machine attitude prediction model;

if not, controlling the shield machine to act according to the shield machine attitude parameter data at the next moment, and storing the shield machine attitude parameter data at the next moment to the shield construction parameters.

In conclusion, the invention can classify stratum type combinations by using a clustering method according to shield construction historical data, thereby providing a basis for the segmented modeling of various subsequent machine learning methods. And according to the segmentation result, modeling by using various machine learning models and carrying out weight distribution according to the accuracy, so that the integrated learning model suitable for various stratum type combinations can be obtained. The integrated learning model can distribute higher weight for the machine learning model with higher accuracy aiming at the specified stratum, so that the integrated learning model has higher accuracy, and the machine learning models adopted by different stratum types are different, so that the phenomenon of inaccurate prediction of the posture of the shield machine caused by stratum mutation can not occur, and the historical data is more fully utilized.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.