阅读说明：本技术 一种基于桩锤激震和决策树模型的地下溶洞探测方法 (A kind of underground karst cavity detection method to be impulsed based on pile monkey with decision-tree model ) 是由 陈敬松 李浩祖 *** 周立成 刘方刚 刘泽佳 汤立群 周玉锋 刘逸平 蒋震宇 于 2019-07-19 设计创作，主要内容包括：本发明公开了一种基于桩锤激震和决策树模型的地下溶洞探测方法,首先建立包含不同溶洞位置与直径的模型库；其次,对模型库进行有限元计算,得到在桩锤激震下,不同溶洞位置与直径情况下地表特定位置的加速度响应；随后,通过对加速度响应数据进行分析,提取加速度响应的特征量；接着将所提取的特征值与标签组合成数据库,将其随机分成训练集与测试集,然后将测试集数据输入决策树模型进行机器学习,初步得到溶洞探测模型；最后根据测试准确率对模型进行参数调整,从而得到最佳的基于决策树模型与有限元模拟计算的溶洞探测模型。本发明方法可有效实现对地下溶洞位置和几何尺寸的准确识别,且较现有的溶洞探测技术,成本更低,效率更高。(The invention discloses a kind of underground karst cavity detection methods to be impulsed based on pile monkey with decision-tree model, initially set up the model library comprising different solution cavity positions and diameter；Secondly, carrying out FEM calculation to model library, obtain in the case where pile monkey impulses, the acceleration responsive of earth's surface specific position in the case of different solution cavity positions and diameter；Then, by analyzing acceleration responsive data, the characteristic quantity of acceleration responsive is extracted；Then extracted characteristic value and tag combination are randomly divided into training set and test set at database, test set data input decision-tree model is then subjected to machine learning, tentatively obtains CAVE DETECTION model；Parameter adjustment is finally carried out to model according to test accuracy rate, to obtain the optimal CAVE DETECTION model based on decision-tree model and finite element simulation calculation.The method of the present invention can be realized effectively and be accurately identified to underground karst cavity position and geometric dimension, and more existing CAVE DETECTION technology, and cost is lower, more efficient.)

1. a kind of underground karst cavity detection method to be impulsed based on pile monkey with decision-tree model, it is characterised in that: firstly, utilizing numerical value The means of emulation establish the model library comprising different solution cavity positions and diameter；Secondly, carrying out FEM calculation to model library, obtain To in the case where pile monkey impulses, the acceleration responsive of earth's surface specific position in the case of different solution cavity positions and diameter；Then, by adding Speed responsive data are analyzed, and the characteristic quantity of acceleration responsive is extracted；Then by extracted characteristic value and tag combination at Database is randomly divided into training set and test set, and test set data input decision-tree model is then carried out machine learning, Tentatively obtain CAVE DETECTION model；Parameter adjustment is finally carried out to model according to test accuracy rate, to obtain optimal be based on The CAVE DETECTION model of decision-tree model and finite element simulation calculation；Itself the following steps are included:

1) by determining predictable solution cavity whether there is or not the response message under the solution cavity various working of solution cavity and different location and size Position and size；Then the range based on prediction establishes the model library comprising different solution cavity sizes and position, and model library impulses Source is set as pile monkey freely falling body under specified altitude and pounds to earth's surface, i.e., is impulsed using piling as the source of impulsing；

2) FEM calculation is carried out after the model library application of foundation impulses, impulsed earth's surface specific position nearby to obtain pile monkey Acceleration responsive；

3) by analyzing acceleration responsive data, corresponding characteristic value is extracted, and by characteristic value and set of tags composite number Training set is randomly divided into after test set according to library, then by database, and training set data is put into decision-tree model and carries out engineering In habit, the CAVE DETECTION model based on test set data is tentatively obtained；

4) the CAVE DETECTION Parameters in Mathematical Model tentatively obtained is adjusted, obtain it is optimal based on decision-tree model with it is limited The CAVE DETECTION model that member simulation calculates.

2. a kind of underground karst cavity detection method to be impulsed based on pile monkey with decision-tree model according to claim 1, special Sign is: in step 1), the model library of the solution cavity comprising different sizes and position is established, concrete condition is as follows:

When establishing model library, since the soil body belongs to semi-infinite body, and in order to simulate seismic wave in the propagation of infinite soil, Need to build on model the rebound enough to which wave can be prevented greatly, and in this model library, infinite element unit will be used as side Boundary's unit simulates semi-infinite body, in this way while reducing moulded dimension, ensure that not only to reduce and calculates the time but also will not be on side Boundary generates reflection；Wherein, Cin3D8 is set as to the unit of model outer layer, is infinite element unit, and remaining unit is set as C3D8r is finite element unit.

3. a kind of underground karst cavity detection method to be impulsed based on pile monkey with decision-tree model according to claim 1, special Sign is: in step 2), since seismic wave will appear decaying in communication process, choose suitable landscape position come into The acquisition of row acceleration information just becomes extremely important, and the pile monkey based on step 1) impulses source, finds after analyzing data, Seismic wave data after radius 5m can be apparent from, and data just die down after radius 10m, bad to be divided Analysis, therefore when extracting acceleration surface response, it selects to impulse a radius in the circle for 5m-10m in distance piling No less than four sensors are evenly arranged on week to carry out data acquisition.

4. a kind of underground karst cavity detection method to be impulsed based on pile monkey with decision-tree model according to claim 1, special Sign is: in step 3), finding after carrying out tracing analysis to collected acceleration responsive data, seismic wave is on surface Propagation is broadly divided into three kinds: longitudinal wave, shear wave and R wave；And it carries out underground karst cavity detection and is primarily upon solution cavity distribution for earth's surface The influence of longitudinal wave, longitudinal wave calculation formula are as follows:

Wherein: v_pFor longitudinal wave, λ is Lame constants, and G is shear modulus, and ρ is density, and ν is Poisson's ratio, and E is elasticity modulus；

The time for inferring that longitudinal wave passes to earth's surface after solution cavity reflects using above-mentioned formula is the time corresponding to the second peak value；Cause This just chooses the second peak value and its corresponding time as characteristic quantity, then finds after analyzing data, when solution cavity buries When deep not deep enough, the longitudinal wave time through solution cavity reflection can approach with the longitudinal wave time propagated through earth's surface, and then influence the first trough The time of peak value；Therefore it based on above-mentioned analysis, determines to extract time t corresponding to the first trough_minWith acceleration value a_minWith And time t corresponding to secondary peak_maxWith acceleration value a_maxAs characteristic value；

After obtaining features described above value, since every group of operating condition corresponds to a group of labels, the distance H including solution cavity top apart from ground, The solution cavity centre of sphere is apart from place the vertical line distance D, solution cavity radius R, solution cavity orientation θ of impulsing that drive piles；And every group of operating condition corresponds to one simultaneously Group characteristic value, including time t corresponding to the first trough_minWith acceleration value a_minAnd time t corresponding to secondary peak_max With acceleration value a_max；Therefore, characteristic value and label one-to-one correspondence are combined, just can obtains required data set；With Data set is randomly divided into training set and test set according to 8:2 afterwards, wherein the effect of training set is to provide number for decision-tree model According to being learnt, thus machine learning prediction model required for obtaining, and test set is then the prediction mould for judge generation The accuracy of type.

5. a kind of underground karst cavity detection method to be impulsed based on pile monkey with decision-tree model according to claim 1, special Sign is: in step 4), being adjusted to the CAVE DETECTION Parameters in Mathematical Model tentatively obtained, specifically to decision-tree model In parameter, including decision tree depth capacity, split smallest sample number needed for smallest sample number, leaf node needed for internal node with Maximum eigenvalue is adjusted, to improve model accuracy rate, to obtain optimal based on decision-tree model and finite element modelling The CAVE DETECTION model of calculating.

Technical field

The present invention relates to the technical fields of CAVE DETECTION, refer in particular to a kind of ground to impulse based on pile monkey with decision-tree model Lower CAVE DETECTION method.

Background technique

A kind of form of the solution cavity as karsts developing area, generate development rely primarily on solubility and the rock fracture of rock with And the movement of underground water.Its presence is very fatal for pile foundation construction.Its meeting so that pile foundation in the construction process There is spillage, collapse hole, bit freezing, tiltedly bore and even cave in.This will injure the life and property safety of pile foundation construction personnel.Cause This, reasonable CAVE DETECTION method just becomes extremely important.It can predict the specifying information of solution cavity in advance, so as to close Reason is evaded solution cavity or is handled solution cavity, to reduce unnecessary loss.And traditional hole detection method includes highly dense Electrical method, elastic wave CT method, tube wave method of testing and shallow earthquake method etc. are spent, principle is using external electric field or elastic wave etc. not With under stratum distribution or propagation law to predict the position of underground karst cavity.Although these methods are answered extensively in engineering With, but still have some limitations and disadvantage.For example, the place meeting amplification data that high-density electric is poor in grounding requirement, To make the error of result increase；Elastic wave CT rule needs drill to be put into sensor to soil layer in testing location, To increase cost and workload；Although tube wave method of testing does not need to drill, but this method can not accurately obtain the position of solution cavity It sets and size, is simply possible to use in the buried depth of prediction solution cavity；Classical shallow earthquake rule need using the man-made explosions such as explosive as Driving source generates seismic wave, had both increased cost and has also resulted in certain risk.Therefore, need to develop it is a kind of it is safe and reliable, Inexpensive, efficient underground karst cavity detection method.In addition, the different location obtained in construction site by sensor, it is different when The response data at quarter not only substantial amounts, and since external environmental interference data show feature complicated and changeable.Therefore, it develops A kind of highly effective algorithm carries out analysis and data mining to these huge data sets, is to realize low cost, efficient Underground Dissolved The necessary means of exploratory tunnel excavating survey technology.

Summary of the invention

The purpose of the present invention is to overcome the shortcomings of the existing technology with it is insufficient, propose a kind of impulse based on pile monkey and decision The underground karst cavity detection method of tree-model, this method are impulsed using the pile monkey in pile foundation construction as the source of impulsing, and are swashed using this Acceleration surface response under focus effect carries out the position of underground karst cavity and geometric dimension as the input of decision-tree model Identification.

To achieve the above object, technical solution provided by the present invention are as follows: one kind is impulsed based on pile monkey and decision-tree model Underground karst cavity detection method, firstly, using the means of numerical simulation, establish include different solution cavity positions and diameter model Library；Secondly, carrying out FEM calculation to model library, obtain in the case where pile monkey impulses, earth's surface in the case of different solution cavity positions and diameter The acceleration responsive of specific position；Then, by analyzing acceleration responsive data, the feature of acceleration responsive is extracted Amount；Then extracted characteristic value and tag combination are randomly divided into training set and test set at database, then will be surveyed Examination collection data input decision-tree model carries out machine learning, tentatively obtains CAVE DETECTION model；Finally according to test accuracy rate pair Model carries out parameter adjustment, to obtain the optimal CAVE DETECTION model based on decision-tree model and finite element simulation calculation； Itself the following steps are included:

1) by determining predictable whether there is or not the response message under the solution cavity various working of solution cavity and different location and size Solution cavity position and size；Then the range based on prediction establishes the model library comprising different solution cavity sizes and position, model library The source of impulsing is set as pile monkey freely falling body under specified altitude and pounds to earth's surface, i.e., is impulsed using piling as the source of impulsing；

2) foundation model library application impulse after carry out FEM calculation, thus obtain pile monkey impulse nearby earth's surface it is specific The acceleration responsive of position；

3) by analyzing acceleration responsive data, corresponding characteristic value is extracted, and by characteristic value and tag combination Training set is randomly divided into after test set at database, then by database, and training set data is put into decision-tree model and carries out machine In device study, the CAVE DETECTION model based on test set data is tentatively obtained；

4) the CAVE DETECTION Parameters in Mathematical Model tentatively obtained is adjusted, obtain it is optimal based on decision-tree model with The CAVE DETECTION model of finite element simulation calculation.

In step 1), the model library of the solution cavity comprising different sizes and position is established, concrete condition is as follows:

When establishing model library, since the soil body belongs to semi-infinite body, and in order to simulate seismic wave in infinite soil It propagates, needs to build on model the rebound enough to which wave can be prevented greatly, and in this model library, infinite element unit will be used to make Semi-infinite body is simulated for boundary element, in this way while reducing moulded dimension, ensure that not only to reduce and calculate the time but also will not Reflection is generated on boundary；Wherein, Cin3D8 is set as to the unit of model outer layer, is infinite element unit, and remaining unit is set It is set to C3D8r, is finite element unit.

In step 2), since seismic wave will appear decaying in communication process, chooses suitable landscape position and come Carrying out acceleration information acquisition just becomes extremely important, and the pile monkey based on step 1) impulses source, sends out after analyzing data Existing, seismic wave data after radius 5m can be apparent from, and data just die down after radius 10m, it is bad into Row analysis, therefore when extracting acceleration surface response, select to drive piles in distance impulses a radius for 5m-10m Circumference on be evenly arranged no less than four sensors to carry out data acquisition.

In step 3), found after carrying out tracing analysis to collected acceleration responsive data, seismic wave is on surface Propagation be broadly divided into three kinds: longitudinal wave, shear wave and R wave；And it carries out underground karst cavity detection and is primarily upon solution cavity distribution for ground The influence of table longitudinal wave, longitudinal wave calculation formula are as follows:

Wherein: v_pFor longitudinal wave, λ is Lame constants, and G is shear modulus, and ρ is density, and ν is Poisson's ratio, and E is springform Amount；

When inferring that longitudinal wave passes to the time of earth's surface as corresponding to the second peak value after solution cavity reflects using above-mentioned formula Between；Therefore the second peak value and its corresponding time are just chosen as characteristic quantity, is then found after analyzing data, when molten When hole buried depth is not deep enough, the longitudinal wave time through solution cavity reflection can approach with the longitudinal wave time propagated through earth's surface, and then influence first The time of trough peak value；Therefore it based on above-mentioned analysis, determines to extract time t corresponding to the first trough_minAnd acceleration value a_minAnd time t corresponding to secondary peak_maxWith acceleration value a_maxAs characteristic value；

After obtaining features described above value, since every group of operating condition corresponds to a group of labels, including solution cavity top apart from ground away from From H, the solution cavity centre of sphere is apart from place the vertical line distance D, solution cavity radius R, solution cavity orientation θ of impulsing that drive piles；And every group of operating condition is right simultaneously One group of characteristic value is answered, including time t corresponding to the first trough_minWith acceleration value a_minAnd the time corresponding to secondary peak t_maxWith acceleration value a_max；Therefore, characteristic value and label one-to-one correspondence are combined, just can obtains required data Collection；Data set is then randomly divided into training set and test set according to 8:2, wherein the effect of training set is for decision-tree model Data are provided to be learnt, thus machine learning prediction model required for obtaining, and test set is then for judging to generate The accuracy of prediction model.

In step 4), the CAVE DETECTION Parameters in Mathematical Model tentatively obtained is adjusted, specifically to decision tree mould Smallest sample number needed for smallest sample number, leaf node needed for parameter in type, including decision tree depth capacity, fractionation internal node It is adjusted with maximum eigenvalue, to improve model accuracy rate, to obtain optimal based on decision-tree model and finite element mould The quasi- CAVE DETECTION model calculated.

Compared with prior art, the present invention have the following advantages that with the utility model has the advantages that

1, more existing CAVE DETECTION technology, the method for the present invention do not need additionally to drill, it is only necessary to acquire acceleration surface Information can carry out CAVE DETECTION.

2, more existing CAVE DETECTION technology, the method for the present invention do not need to arrange multisensor, it is only necessary to neighbouring soil Layer is sampled, more convenient compared to traditional CAVE DETECTION method operation.

3, in terms of construction safety, the method for the present invention does not need that generation can be reduced in this way as the source that impulses using explosive The risk of accident, and it is also what scene can directly use that used pile monkey, which impulses, reduces measurement cost.

4, the method for the present invention combines machine learning and FEM calculation, is impulsed by the method for numerical simulation to pile monkey It is simulated, considerably reduces the time of model library foundation, improve efficiency.

5, the cost of the method for the present invention can be lower than traditional CAVE DETECTION method, due to mainly in combination with finite element software meter It calculates, and only needs to arrange a small amount of sensor in terms of DATA REASONING at the scene, so cost can be much lower.

Detailed description of the invention

Fig. 1 is the method for the present invention flow diagram.

Fig. 2 is earth's surface vertical acceleration response curve.

Fig. 3 is model library solution cavity distribution map.

Fig. 4 is acceleration surface sensor distribution map.

Specific embodiment

The present invention is further explained in the light of specific embodiments.

As shown in Figure 1, the underground karst cavity detection method to be impulsed based on pile monkey with decision-tree model provided by the present embodiment, The following steps are included:

1) the soil layer model library containing solution cavity is established.The size of model library is the cube of 30m*30m*30m, wherein soil layer sheet Structure model is classical Mohr-Coulomb (mole-coulomb) constitutive model, and it is infinite element that boundary mesh, which is then arranged, (Cin3D8), to simulate the feature of soil layer semi-infinite body, and remaining unit is set as C3D8r, is finite element unit.In this case In, it is assumed that solution cavity shape is sphere, and distance of the solution cavity top apart from ground is H, and the solution cavity centre of sphere is apart from the place that impulses of driving piles Vertical line distance is D, and solution cavity radius is R, and solution cavity orientation is θ, as shown in Figure 3: where model of the solution cavity top apart from ground distance H It encloses for 3m -10m, sampling interval 1m；A range of vertical line distance D is 0m -10m where the distance piling of the solution cavity centre of sphere impulses, Sampling interval is 1m；Solution cavity radius R takes 2m, 3m, 4m respectively；The range of solution cavity orientation θ is 0 ° -360 °, and sampling range is 5 °. Model library is established according to above different solution cavity size and position, is prepared for FEM calculation.

2) by calculating model library, the acceleration responsive of earth's surface specific position is obtained, but since seismic wave is passing It will appear decaying during broadcasting, therefore choose suitable landscape position just to become extremely important to carry out acceleration information acquisition. It being impulsed source based on pile monkey, is found after analyzing data, seismic wave data after radius 5m can be apparent from, and Data just die down after radius 10m, bad to be analyzed.Therefore when extracting acceleration surface response, selection It is evenly arranged no less than four sensors on the circumference for 5m--10m in the distance piling radius that impulses and carries out data Acquisition.In the present embodiment, FEM calculation is carried out to above model library, surface distance piling is extracted and impulses a radius Acceleration responsive at 10m annulus takes the response of wherein four points as primitive curve, 90 ° is spaced between four points, such as Fig. 4 It is shown.

It is found after carrying out tracing analysis to collected acceleration information, propagation of the seismic wave on surface is broadly divided into three Kind: longitudinal wave, shear wave and R wave.And our rule is primarily upon influence of the solution cavity distribution for earth's surface longitudinal wave.Longitudinal wave calculates public Formula is as follows:

Wherein: v_pFor longitudinal wave, λ is Lame constants, and G is shear modulus, and ρ is density, and ν is Poisson's ratio, and E is springform Amount.

When inferring that longitudinal wave passes to the time of earth's surface as corresponding to the second peak value after solution cavity reflects using above-mentioned formula Between.Therefore the second peak value and its corresponding time are just chosen as characteristic quantity.It is then found after analyzing data, when molten When hole buried depth is excessively shallow, the longitudinal wave time reflected through solution cavity can be closer to the longitudinal wave time propagated through earth's surface, and then influence first The time of trough peak value.Therefore based on above-mentioned analysis, this method determines to extract time (t corresponding to the first trough_min) and add Velocity amplitude (a_min) and secondary peak corresponding to time (t_max) and acceleration value (a_max) it is used as characteristic value, as shown in Figure 2. It is four measuring points at 10m annulus to carry out solution cavity prediction that radius in earth's surface is extracted under one operating condition, and measuring point includes upper every time Four characteristic values stated, therefore an operating condition includes 16 characteristic values.Then by each operating condition corresponding solution cavity top distance ground The distance H in face, solution cavity centre of sphere distance piling impulse a place vertical line distance D, solution cavity radius R, solution cavity orientation θ as label, mark Label are corresponded with characteristic value, and by Data Integration at data set, and data set is then randomly divided into training set according to 8:2 and is surveyed Examination collection, prepares for machine learning.

4) above-mentioned training set is subjected to machine learning using decision-tree model, tentatively obtains CAVE DETECTION model, then It is predicted using forecast set, then prediction result and test set label is compared, to obtain precision of prediction.Due in reality Do not need to the prediction in solution cavity orientation excessively accurate in the operating condition of border, therefore present case is analyzed permissible error is introduced.Its Middle permissible error is obtained after the result that decision-tree model the is predicted label comparison original with test set to test set 's.Error 1m refers to that it is accurate to be considered as predicting when the result of prediction is within the scope of the original label ± 1m of test set.It can must similarly miss Difference ± 2m.It is same that a vertical line distance D where 16 characteristic values impulse to the distance piling of solution cavity buried depth H and the solution cavity centre of sphere is stated in use When Shi Jinhang is predicted, a vertical line distance D where solution cavity buried depth H and the distance piling of the solution cavity centre of sphere impulse reaches license simultaneously to be missed When difference is 1m, accuracy rate is 62.82%；A vertical line distance D where the distance piling of solution cavity buried depth H and the solution cavity centre of sphere impulses reaches simultaneously To permissible error be 2m when, accuracy rate 81.82%.And when predicting solution cavity orientation, azimuth averaging is divided For four sections, at every a 90 °, judge the solution cavity centre of sphere whether in the section, accuracy rate 78.52%；Similarly it is divided into When eight sections, accuracy rate 60.98%；When being divided into 12 sections, accuracy rate 51.01%.

5) the CAVE DETECTION Parameters in Mathematical Model tentatively obtained is adjusted, specifically to the ginseng in decision-tree model Smallest sample number needed for smallest sample number, leaf node needed for number, including decision tree depth capacity, fractionation internal node and maximum are special Value indicative etc. is adjusted, to improve model accuracy rate, to obtain optimal based on decision-tree model and finite element simulation calculation CAVE DETECTION model.

Embodiment described above is only the preferred embodiments of the invention, and but not intended to limit the scope of the present invention, therefore All shapes according to the present invention change made by principle, should all be included within the scope of protection of the present invention.