阅读说明：本技术 车辆驾驶故障检测方法、装置、设备及存储介质 (Vehicle driving fault detection method, device, equipment and storage medium ) 是由 胡坚耀 李沐 钟晓文 林凡 于 2021-09-28 设计创作，主要内容包括：本发明公开了一种车辆驾驶故障检测方法、装置、设备及存储介质,其通过深度学习模型构建驾驶故障监控模型,结合核密度估计得到所述驾驶故障监控模型的故障监控阈值,并采用动态加权因子突出故障敏感特征的影响以提高故障检测的灵敏度,得到动态监控指标,并判断动态监控指标是否超过故障监控阈值,以实现车辆行驶过程中驾驶故障的实时智能监控,进而能在检测到车辆有危险故障发生时,可以及时向驾驶员发出警告,避免交通事故意外的发生。(The invention discloses a vehicle driving fault detection method, a device, equipment and a storage medium, wherein a driving fault monitoring model is constructed through a deep learning model, a fault monitoring threshold value of the driving fault monitoring model is obtained by combining nuclear density estimation, the influence of fault sensitivity characteristics is highlighted by adopting a dynamic weighting factor so as to improve the sensitivity of fault detection, a dynamic monitoring index is obtained, and whether the dynamic monitoring index exceeds the fault monitoring threshold value is judged, so that the real-time intelligent monitoring of driving faults in the driving process of a vehicle is realized, and further, when dangerous faults of the vehicle are detected, a warning can be timely sent to a driver, and the occurrence of traffic accidents is avoided.)

1. A vehicle driving failure detection method, characterized by comprising:

inputting the acquired driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model;

based on the driving fault monitoring model, combining a kernel density estimation method to obtain a control limit of the driving fault monitoring model, and taking the control limit as a fault monitoring threshold;

acquiring vehicle driving test data acquired in real time, and acquiring an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model;

considering the fault sensitivity difference of the characteristics, introducing a dynamic weighting factor to process the abnormal monitoring index to obtain a dynamic monitoring index;

and when the dynamic monitoring index is larger than the fault monitoring threshold value, judging that the vehicle has a driving fault, and sending a fault warning to a vehicle driver.

2. The vehicle driving malfunction detection method according to claim 1, characterized in that the method further includes:

introducing a static weighting factor to process the abnormal monitoring index to obtain a static monitoring index;

and when the static monitoring index is larger than the fault monitoring threshold value, sending driving safety reminding to a vehicle driver.

3. The vehicle driving fault detection method according to claim 1, wherein the step of inputting the acquired driving training data samples into a pre-constructed deep learning model to obtain a driving fault monitoring model specifically comprises:

constructing a depth support vector data model based on depth features;

and inputting the acquired driving training data sample into the deep support vector data model to obtain a driving fault monitoring model.

4. The vehicle driving fault detection method according to claim 1, wherein the obtaining of the control limit of the driving fault monitoring model based on the driving fault monitoring model by combining a kernel density estimation method, and using the control limit as a fault monitoring threshold specifically includes:

acquiring a training sample monitoring index of the driving training data sample in the driving fault monitoring model;

constructing a probability density function of the monitoring index of the training sample based on the driving training data sample;

and calculating an estimated value of the probability density function when the confidence level is preset to obtain a corresponding control limit, and taking the control limit as a fault monitoring threshold.

5. The vehicle driving fault detection method according to claim 1, wherein the obtaining of the vehicle driving test data collected in real time and the obtaining of the abnormality monitoring index of the vehicle driving test data based on the driving fault monitoring model specifically include:

inputting vehicle driving test data acquired in real time into the driving fault monitoring model, and performing multi-layer feature extraction to obtain a depth feature set of the test data;

and carrying out hypersphere modeling on the depth feature set, calculating the square of the distance from the depth feature set to the center of a hypersphere, and taking the calculated square of the distance as an abnormal monitoring index of the driving test data.

6. The vehicle driving fault detection method according to claim 1, wherein the step of processing the abnormal monitoring indicator by introducing a dynamic weighting factor in consideration of fault sensitivity differences of the features to obtain a dynamic monitoring indicator specifically comprises:

acquiring a depth characteristic set extracted from the driving test data in the driving fault monitoring model;

aiming at the depth feature set, establishing a history window { D with the length of D in a sliding window mode_t-d+1,i,D_t-d+2,i,…,D_t,iEstimating the abnormal monitoring index according to the dynamic data of the historical window to obtain a dynamic monitoring index:

wherein DD_tIn order to monitor the indicators dynamically,as dynamic weight factors, P_t-j+1，iProbability of failure at time (t-j +1) within a time window of length d for the ith depth feature, y_t-j+1，i ^(L)Is the ith depth characteristic of the L-th network at the (t-j +1) th time, O_iIs the center of the hyper-sphere corresponding to the ith depth characteristic, s is a new variable, s is more than or equal to 0 and less than or equal to d, and represents the s moment in a time window with the length of d,representing the static weight factor of the ith depth feature at time s,representing the dynamic weight factor, D, over a time window of length D after the introduction of a new sample point j_t-s+1,iRepresenting the monitoring index of the ith depth feature at time s in a time window of length d,representing the average value of the monitoring indexes of the ith depth feature in a time window with the length of D, n is the number of samples, D_lim,iThe maximum value of the monitoring index of the ith depth feature.

7. The vehicle driving fault detection method according to claim 2, wherein the introducing of the static weighting factor to process the abnormal monitoring index to obtain a static monitoring index specifically includes:

acquiring a depth feature set extracted from the driving test data in the driving fault monitoring model, and calculating the sub-components of each depth feature in the depth feature set in the abnormal monitoring index;

calculating the fault probability of each depth feature in the depth feature set;

calculating a static weight factor of each depth feature according to the fault probability of each depth feature;

and obtaining a static monitoring index according to the static weighting factor of each depth feature and the sub-component of each depth feature in the abnormal monitoring index.

8. A vehicle driving failure detection apparatus, characterized by comprising:

the driving fault monitoring model acquisition module is used for inputting the acquired driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model;

the fault monitoring threshold acquisition module is used for obtaining a control limit of the driving fault monitoring model by combining a kernel density estimation method based on the driving fault monitoring model and taking the control limit as a fault monitoring threshold;

the abnormal monitoring index acquisition module is used for acquiring vehicle driving test data acquired in real time and acquiring an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model;

the dynamic monitoring index acquisition module is used for taking the fault sensitivity difference of the characteristics into consideration and introducing a dynamic weighting factor to process the abnormal monitoring index to obtain a dynamic monitoring index;

and the fault warning module is used for judging that the vehicle has a driving fault when the dynamic monitoring index is larger than the fault monitoring threshold value, and sending a fault warning to a vehicle driver.

9. A vehicle driving failure detection apparatus comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the vehicle driving failure detection method according to any one of claims 1 to 7 when executing the computer program.

10. A storage medium characterized by comprising a stored computer program, wherein the apparatus on which the storage medium is stored is controlled to execute the vehicle driving failure detection method according to any one of claims 1 to 7 when the computer program is executed.

Technical Field

The present invention relates to the field of network technologies, and in particular, to a method, an apparatus, a device, and a storage medium for detecting a vehicle driving fault.

Background

With the continuous development of the economic strength of China, automobiles become popular transportation means for China from rare products at the end of the last century, the traffic data display is carried out by the end of 2020, the quantity of Chinese automobiles is up to 2.75 hundred million, and compared with the area of only 40 million of 1990, the change of turning over the sky and covering the land can be said to happen.

With the increasing number of people driving vehicles year by year, the number of vehicles driving on the road is increasing day by day, but until now, an effective method for intelligently detecting faults of the vehicles does not exist, and the accident faults possibly occurring in the driving process of the vehicles cannot be effectively and intelligently monitored and analyzed, so that the vehicle owners cannot timely process the faulty vehicles when dangerous faults occur on the road, the probability of traffic accidents is increased, personal and property safety of pedestrians and surrounding driving vehicles is seriously damaged, and the construction of a stable and ordered driving order is adversely affected.

Disclosure of Invention

Aspects of the present invention provide a vehicle driving fault detection method, apparatus, device, and storage medium, which can monitor a driving fault occurring during a vehicle driving process and timely issue a warning to a driver when the occurrence of the driving fault is detected.

The invention provides a vehicle driving fault detection method in a first aspect, which comprises the following steps:

inputting the acquired driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model;

based on the driving fault monitoring model, combining a kernel density estimation method to obtain a control limit of the driving fault monitoring model, and taking the control limit as a fault monitoring threshold;

acquiring vehicle driving test data acquired in real time, and acquiring an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model;

considering the fault sensitivity difference of the characteristics, introducing a dynamic weighting factor to process the abnormal monitoring index to obtain a dynamic monitoring index;

and when the dynamic monitoring index is larger than the fault monitoring threshold value, judging that the vehicle has a driving fault, and sending a fault warning to a vehicle driver.

As an improvement of the above, the method further comprises:

introducing a static weighting factor to process the abnormal monitoring index to obtain a static monitoring index;

and when the static monitoring index is larger than the fault monitoring threshold value, sending driving safety reminding to a vehicle driver.

As an improvement of the above scheme, the inputting the acquired driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model specifically includes:

constructing a depth support vector data model based on depth features;

and inputting the acquired driving training data sample into the deep support vector data model to obtain a driving fault monitoring model.

As an improvement of the above scheme, the obtaining of the control limit of the driving fault monitoring model based on the driving fault monitoring model by combining a kernel density estimation method, and using the control limit as a fault monitoring threshold specifically includes:

acquiring a training sample monitoring index of the driving training data sample in the driving fault monitoring model;

constructing a probability density function of the monitoring index of the training sample based on the driving training data sample;

and calculating an estimated value of the probability density function when the confidence level is preset to obtain a corresponding control limit, and taking the control limit as a fault monitoring threshold.

As an improvement of the above scheme, the acquiring vehicle driving test data collected in real time and obtaining an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model specifically include:

inputting vehicle driving test data acquired in real time into the driving fault monitoring model, and performing multi-layer feature extraction to obtain a depth feature set of the test data;

and carrying out hypersphere modeling on the depth feature set, calculating the square of the distance from the depth feature set to the center of a hypersphere, and taking the calculated square of the distance as an abnormal monitoring index of the driving test data.

As an improvement of the above scheme, the step of taking into account the fault sensitivity difference of the features and introducing a dynamic weighting factor to process the abnormal monitoring index to obtain a dynamic monitoring index specifically includes:

acquiring a depth characteristic set extracted from the driving test data in the driving fault monitoring model;

aiming at the depth feature set, establishing a history window { D with the length of D in a sliding window mode_t-d+1,i,D_t-d+2,i,…,D_t,iEstimating the abnormal monitoring index according to the dynamic data of the historical window to obtain a dynamic monitoring index:

wherein DD_tIn order to monitor the indicators dynamically,as dynamic weight factors, P_t-j+1,iProbability of failure at time (t-j +1) within a time window of length d for the ith depth feature, y_t-j+1,i ^(L)Is the ith depth characteristic of the L-th network at the (t-j +1) th time, O_iIs the center of the hyper-sphere corresponding to the ith depth characteristic, s is a new variable, s is more than or equal to 0 and less than or equal to d, and represents the s moment in a time window with the length of d,representing the static weight factor of the ith depth feature at time s,representing the dynamic weight factor, D, over a time window of length D after the introduction of a new sample point j_t-s+1,iRepresenting the monitoring index of the ith depth feature at time s in a time window of length d,representing the average value of the monitoring indexes of the ith depth feature in a time window with the length of D, n is the number of samples, D_lim,iThe maximum value of the monitoring index of the ith depth feature.

As an improvement of the above scheme, the introducing a static weighting factor to process the abnormal monitoring index to obtain a static monitoring index specifically includes:

acquiring a depth feature set extracted from the driving test data in the driving fault monitoring model, and calculating the sub-components of each depth feature in the depth feature set in the abnormal monitoring index;

calculating the fault probability of each depth feature in the depth feature set;

calculating a static weight factor of each depth feature according to the fault probability of each depth feature;

and obtaining a static monitoring index according to the static weighting factor of each depth feature and the sub-component of each depth feature in the abnormal monitoring index.

A second aspect of the present invention provides a vehicle driving failure detection apparatus, including:

the driving fault monitoring model acquisition module is used for inputting the acquired driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model;

the fault monitoring threshold acquisition module is used for obtaining a control limit of the driving fault monitoring model by combining a kernel density estimation method based on the driving fault monitoring model and taking the control limit as a fault monitoring threshold;

the abnormal monitoring index acquisition module is used for acquiring vehicle driving test data acquired in real time and acquiring an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model;

the dynamic monitoring index acquisition module is used for taking the fault sensitivity difference of the characteristics into consideration and introducing a dynamic weighting factor to process the abnormal monitoring index to obtain a dynamic monitoring index;

and the fault warning module is used for judging that the vehicle has a driving fault when the dynamic monitoring index is larger than the fault monitoring threshold value, and sending a fault warning to a vehicle driver.

A third aspect of the present invention provides a vehicle driving failure detection apparatus comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the vehicle driving failure detection method as described above when executing the computer program.

A fourth aspect of the present invention provides a storage medium including a stored computer program, wherein the apparatus in which the storage medium is located is controlled to execute the vehicle driving failure detection method as described above when the computer program is run.

Compared with the prior art, the vehicle driving fault detection method, the vehicle driving fault detection device, the vehicle driving fault detection equipment and the storage medium have the following beneficial effects:

the vehicle driving fault detection method provided by the invention comprises the steps of inputting an obtained driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model; based on the driving fault monitoring model, combining a kernel density estimation method to obtain a control limit of the driving fault monitoring model, and taking the control limit as a fault monitoring threshold; acquiring vehicle driving test data acquired in real time, and acquiring an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model; considering the fault sensitivity difference of the characteristics, introducing a dynamic weighting factor to process the abnormal monitoring index to obtain a dynamic monitoring index; when the dynamic monitoring index is larger than the fault monitoring threshold, judging that the vehicle has a driving fault, sending a fault warning to a vehicle driver, constructing a driving fault monitoring model through a deep learning model, obtaining the fault monitoring threshold of the driving fault monitoring model by combining kernel density estimation, adopting a dynamic weighting factor to highlight the influence of fault sensitivity characteristics so as to improve the sensitivity of fault detection, obtaining the dynamic monitoring index, and judging whether the dynamic monitoring index exceeds the fault monitoring threshold, so as to realize real-time intelligent monitoring of the driving fault in the driving process of the vehicle, and further, when the dangerous fault of the vehicle is detected, timely sending a warning to the driver so as to avoid the occurrence of traffic accidents.

Drawings

FIG. 1 is a flow chart of a vehicle driving fault detection method provided by an embodiment of the invention;

fig. 2 is a block diagram of a vehicle driving failure detection apparatus according to an embodiment 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.

Referring to fig. 1, a flowchart of a method for detecting a driving fault of a vehicle according to an embodiment of the present invention is shown.

The vehicle driving fault detection method provided by the embodiment of the invention comprises the following steps of S11 to S15:

step S11, inputting the acquired driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model;

step S12, obtaining a control limit of the driving fault monitoring model by combining a kernel density estimation method based on the driving fault monitoring model, and taking the control limit as a fault monitoring threshold;

step S13, acquiring vehicle driving test data acquired in real time, and acquiring an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model;

step S14, taking the fault sensitivity difference of the characteristics into consideration, introducing a dynamic weighting factor to process the abnormal monitoring index to obtain a dynamic monitoring index;

and step S15, when the dynamic monitoring index is larger than the fault monitoring threshold value, judging that the vehicle has a driving fault, and sending a fault warning to a vehicle driver.

For example, in step S15, the failure warning may be a warning to warn the driver of a driving failure and the driver needs to stop at the side for checking. The fault warning may be broadcast in voice to alert the driver.

The vehicle driving fault detection method provided by the embodiment of the invention comprises the steps of inputting an obtained driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model; based on the driving fault monitoring model, combining a kernel density estimation method to obtain a control limit of the driving fault monitoring model, and taking the control limit as a fault monitoring threshold; acquiring vehicle driving test data acquired in real time, and acquiring an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model; considering the fault sensitivity difference of the characteristics, introducing a dynamic weighting factor to process the abnormal monitoring index to obtain a dynamic monitoring index; when the dynamic monitoring index is larger than the fault monitoring threshold, judging that the vehicle has a driving fault, sending a fault warning to a vehicle driver, constructing a driving fault monitoring model through a deep learning model, obtaining the fault monitoring threshold of the driving fault monitoring model by combining kernel density estimation, adopting a dynamic weighting factor to highlight the influence of fault sensitivity characteristics so as to improve the sensitivity of fault detection, obtaining the dynamic monitoring index, and judging whether the dynamic monitoring index exceeds the fault monitoring threshold, so as to realize real-time intelligent monitoring of the driving fault in the driving process of the vehicle, and further, when the dangerous fault of the vehicle is detected, timely sending a warning to the driver so as to avoid the occurrence of traffic accidents.

In one embodiment, the step S11 "inputting the acquired driving training data sample into a deep learning model constructed in advance to obtain a driving fault monitoring model", specifically includes:

constructing a depth support vector data model based on depth features;

and inputting the acquired driving training data sample into the deep support vector data model to obtain a driving fault monitoring model.

It can be understood that the support vector data model (SVDD) adopts a shallow learning framework, and it is difficult to effectively and intelligently monitor complex faults generated in the vehicle driving process, so that the embodiment of the invention improves and optimizes the support vector data model (SVDD), redefines an optimization objective function of the SVDD under the deep learning framework, and constructs a deep support vector data model (DSVDD) based on a deep feature. The SVDD is briefly introduced below, and a depth support vector data model based on depth features according to an embodiment of the present invention is introduced based on the SVDD.

SVDD is an important single classification algorithm, which maps complex data into a high-dimensional feature space, finds a hypersphere as small as possible to surround all training samples, and samples beyond the hypersphere boundary are regarded as abnormal samples. Given a dataset X ═ X₁，x₂…x_n]^T∈R^n×mWhere n and m are the number of samples and variables, respectively, SVDD referencesInto a non-linear mapping function, all samples x_nMapping into a feature space_n) And finding a minimum hyper-sphere in a feature space to realize the enclosure of data, wherein the corresponding optimization objective function is as follows:

wherein R is the radius of the hyper-sphere, O is the center of the hyper-sphere, sigma_iFor the relaxation variable, C is a penalty factor that balances the hypersphere volume with the out-of-boundary samples.

Lagrangian functions can be built based on the above problem:

wherein alpha is_i>0 and beta_i>0 is the lagrange multiplier. Further analysis can lead to dual description of the original optimization problem:

wherein, K (x)_i，x_j)＝φ(x_n)^Tφ(x_n) The method is a kernel function operation, namely the contents of two vectors in the feature space can be calculated by using a kernel function in the original space, and a commonly used kernel function is a Gaussian kernel function.

Equation (3) describes a standard quadratic optimization problem that can be solved to obtain the center of the hyper-sphere:

the radius R can be obtained by calculating the distance of the support vector to the center:

R＝||φ(x^*)-O|| (5)

wherein x is^*Represents a correspondence α_i>Arbitrary vector x of 0_i。

For test sample x at time t_tDefining the square D of its distance to the center of the hyper-sphere_tAs a monitoring index, the expression is as follows:

by mixing D_tAnd R²Comparing to judge the condition of the process data; if D is_t≤R²Then the corresponding vector x_iClassifying the samples as normal samples; otherwise, the sample is regarded as an abnormal sample.

The basic framework of the SVDD model comprises two main steps, namely single-layer nonlinear mapping and hypersphere modeling. In the model structure, the nonlinear mapping is a space transformation, plays a role in feature extraction, and is completed by means of a kernel function in the actual calculation process. Because the basic SVDD only relates to one feature extraction layer, the complex data relation is difficult to effectively process, and the traditional SVDD algorithm cannot be directly applied to fault analysis in the automobile driving process.

Based on the limitation analysis of the traditional SVDD model structure, the embodiment of the invention provides a deep support vector data model for fault monitoring. According to the method, a depth feature extraction technology based on a multilayer neural network is introduced into an SVDD model to replace the traditional implicit nonlinear transformation, so that the mining and expression capacity of the model on the internal features of data is improved. Meanwhile, aiming at the difference problem of the depth features in fault information expression, a feature weighting unit is further designed, and the weight is calculated according to the fault sensitivity degree of the features, so that the monitoring capability of the complex faults is enhanced. The key to the construction of the depth support vector data model is two points: and (3) constructing a depth SVDD model and designing a feature weighting strategy.

Deep SVDD (DSVDD) is a new method developed on the basis of SVDD, and the basic idea is to perform feature extraction through a deep learning network and then construct an end-to-end anomaly detection network model on the basis of deep features. The method is different from the traditional SVDD method in that a plurality of feature extraction layers are used for obtaining data feature representation instead of simple single-layer kernel mapping.

For data { x_i∈R^mI ═ 1, 2, …, n }, the multi-layer feature extraction process can be described as a functional relationship:

wherein phi is^(l)(1 ≦ L ≦ L) represents the layer-by-layer nonlinear mapping function relationship in the deep network, W⁽⁾L is more than or equal to 1 and less than or equal to L, the weight parameters of the first-layer network are expressed, and finally the extracted characteristics are as follows:

y_i ^(L)＝Φ(x_i；W)＝φ^(L)(φ^(L-1)(…φ⁽¹⁾(x_i；W⁽¹⁾))…；W^(L-1))；W^(L)) (8)

wherein W ═ { W ═ W⁽¹⁾，W⁽²⁾，…，W^(L)Represents a depth network collective parameter set.

Due to the application of the depth feature extraction technology, the traditional SVDD optimization objective function is not applicable any more. The optimization goal of the depth SVDD model is to make the output depth feature y by training the debugging parameter set W_i ^(L)Distributed as densely as possible in a hypersphere with radius R and center O, and the corresponding optimization objective function:

wherein, ν is a balance parameter for adjusting the influence of abnormal data points outside the hyper-sphere, λ is a penalty coefficient for the network weight, and O is a sphere center specified in a priori. In the single classification task, it is generally assumed that the training data sets are all normal samples, and at this time, the objective function can be further simplified as follows:

the weight parameter W is optimized through a deep network optimization algorithm, training samples can be gathered to the position near the hypersphere center O in a deep feature space, a hypersphere for describing normal training data is formed, and the network weight parameter obtained through training is W^*。

For sample x_tThe abnormal degree monitoring index is defined as the depth characteristic y of the point_t ^(L)＝Φ(x_t；W^*) The square of the distance to the hypersphere center O, i.e.:

D_t＝||y_t ^(L)-O||²＝||φ(x_t；W^*)-O||² (11)

different from the SVDD monitoring index in the formula (6), the depth SVDD monitoring index described by the above formula can be directly calculated without using complex kernel function mapping.

In one embodiment, the step S12 "obtaining a control limit of the driving fault monitoring model based on the driving fault monitoring model by combining a kernel density estimation method, and using the control limit as a fault monitoring threshold", specifically includes:

acquiring a training sample monitoring index of the driving training data sample in the driving fault monitoring model;

constructing a probability density function of the monitoring index of the training sample based on the driving training data sample;

and calculating an estimated value of the probability density function when the confidence level is preset to obtain a corresponding control limit, and taking the control limit as a fault monitoring threshold.

Specifically, the training sample monitoring index of the driving training data sample is obtained through formulas (7) to (11), and since the hypersphere radius is not defined in the depth SVDD model, another method must be found to set the detection threshold. In the embodiment of the invention, the probability distribution of the monitoring index of the training sample is obtained by adopting the kernel density estimation, and the control limit of the probability distribution is further calculated to be used as the fault monitoring threshold. When the monitoring index corresponding to the test sample exceeds the fault monitoring threshold value, the test sample is regarded as a fault data point.

In particular, Kernel Density Estimation (KDE) is a non-parametric probability density estimation technique that can estimate the probability density function of a random variable from given training data. For training data samples { x in the deep SVDD modeling process₁，x₂，…，x_nAnd calculating to obtain a corresponding training sample monitoring index D₁，D₂，…D_nConstruction of D based on training data set_tProbability density function f (D)_t) The expression is as follows:

wherein, g (x) represents the kernel function adopted by KDE, and a represents the width parameter of the kernel function.

On the premise of giving confidence coefficient delta, the corresponding control limit D can be solved through an integral formula of a probability density function_limI.e. by

The conceptual meaning of the confidence δ is: the probability that the monitoring index value corresponding to the normal sample has delta% is not more than D_lim。

In an embodiment, the step S13 "acquiring vehicle driving test data collected in real time, and obtaining an abnormality monitoring index of the vehicle driving test data based on the driving fault monitoring model" specifically includes:

inputting vehicle driving test data acquired in real time into the driving fault monitoring model, and performing multi-layer feature extraction to obtain a depth feature set of the test data;

and carrying out hypersphere modeling on the depth feature set, calculating the square of the distance from the depth feature set to the center of the hypersphere, and taking the calculated square of the distance as an abnormal monitoring index of the test data.

In the embodiment of the invention, vehicle driving test data is input into a driving fault monitoring model, multi-layer feature extraction is carried out to obtain a corresponding depth feature set, and an abnormal monitoring index of the test data is calculated according to a formula (11).

In an embodiment, the step S14 "considers the fault sensitivity difference of the features, and introduces a dynamic weighting factor to process the abnormal monitoring index, so as to obtain a dynamic monitoring index", specifically includes:

acquiring a depth characteristic set extracted from the driving test data in the driving fault monitoring model;

aiming at the depth feature set, establishing a history window { D with the length of D in a sliding window mode_t-d+1,i,D_t-d+2,i,…,D_t,iEstimating the abnormal monitoring index according to the dynamic data of the historical window to obtain a dynamic monitoring index:

wherein DD_tIn order to monitor the indicators dynamically,as dynamic weight factors, P_t-j+1,iProbability of failure at time (t-j +1) within a time window of length d for the ith depth feature, y_t-j+1,i ^(L)Is the ith depth characteristic of the L-th network at the (t-j +1) th time, O_iIs the center of the hyper-sphere corresponding to the ith depth characteristic, s is a new variable, s is more than or equal to 0 and less than or equal to d, and represents the s moment in a time window with the length of d,representing the static weight factor of the ith depth feature at time s,representing the dynamic weight factor, D, over a time window of length D after the introduction of a new sample point j_t-s+1,iRepresenting the monitoring index of the ith depth feature at time s in a time window of length d,representing the average value of the monitoring indexes of the ith depth feature in a time window with the length of D, n is the number of samples, D_lim,iThe maximum value of the monitoring index of the ith depth feature.

It should be noted that the values of s in the above equations (14) to (16) are randomly generated except that s is increased from 1 to d according to a specific specification.

Specifically, the monitoring index formula of the formula (11) is analyzed, and D can be seen_tEssence describes the euclidean distance in the depth feature space. Specific expression y listing depth features and center of hyper-sphere_t ^(L)＝[y_t,1 ^(L),y_t,2 ^(L),…y_t,k ^(L)]，O＝[O₁，O₂，…O_k]Then equation (11) can be expressed as:

D_t＝(y_t,1 ^(L)-O₁)²+(y_t,2 ^(L)-O₂)²+…+(y_t,k ^(L)-O_k)² (17)

as can be seen from equation (17), the monitoring index is treated equally for each depth feature. In practice, some faults may affect only a few features (which may be referred to as fault-sensitive features), and most of the features remain in a normal state. At this time, a minute amount of failure information may be buried by normal noise information, thereby making it difficult to effectively detect a failure. In essence, this is a phenomenon resulting from the difference in fault sensitivity of one different feature. In view of the problem, the embodiment of the present invention proposes to adopt a weighting strategy to highlight the influence of the fault sensitivity characteristics, thereby improving the sensitivity of complex fault detection.

For convenience, the sub-components of the ith depth feature in the fault monitoring index are defined as:

D_t,i＝(y_t,i ^(L)-O_i)² (18)

if the feature y_t,i ^(L)Is a fault sensitive feature, then the corresponding fault monitoring index sub-component D_t,iSignificant changes must occur. To measure the sensitivity of the depth feature to the fault, the fault probability of the depth feature is further defined as:

wherein D is_lim,iIs D_t,iThe corresponding delta% control limit can be calculated from the kernel density estimation method. For normal data changes, the probability of failure P_t,iApproaching 0. Otherwise, the probability of the fault-sensitive samples is close to 1. Gamma is an adjusting factor, and the distribution condition of the probability curve can be influenced by changing the value of gamma.

Specifically, the dynamic weight factors can more effectively mine fault information contained in the depth features, so that the fault detection performance is improved.

Further, in one embodiment, the vehicle driving failure detection method further includes:

introducing a static weighting factor to process the abnormal monitoring index to obtain a static monitoring index;

and when the static monitoring index is larger than the fault monitoring threshold value, sending driving safety reminding to a vehicle driver.

For example, the driving safety reminder can remind a driver of observing the driving condition of the vehicle with more attention during driving.

In the embodiment of the invention, a static weighting strategy is also designed, and the influence of fault sensitivity characteristics is highlighted by using a static weighting factor. In the concrete implementation, by introducing a static weighting strategy and a dynamic weighting strategy, the accuracy of fault detection can be effectively improved on the basis of keeping a low false alarm rate, and the intelligent detection of unexpected faults in the automobile driving process is effectively realized.

Specifically, the introducing a static weighting factor to process the abnormal monitoring index to obtain a static monitoring index specifically includes:

acquiring a depth feature set extracted from the driving test data in the driving fault monitoring model, and calculating the sub-components of each depth feature in the depth feature set in the abnormal monitoring index;

calculating the fault probability of each depth feature in the depth feature set;

calculating a static weight factor of each depth feature according to the fault probability of each depth feature;

and obtaining a static monitoring index according to the static weighting factor of each depth feature and the sub-component of each depth feature in the abnormal monitoring index.

In specific implementation, the sub-components of the ith depth feature in the fault monitoring index are calculated through formula (18), the fault probability of each depth feature is calculated through formula (19), and the static weight factor of the depth feature is calculated through formula (20):

the significance of the above formula is that D is satisfied_t,i>D_lim,iThe condition features are defined as fault-sensitive features, with corresponding weights greater than 1, the specific weight being dependent on the fault probability. Otherwise, the weight of the non-sensitive feature is set to 1, i.e. the original influence degree is maintained. Based on the weight factor, constructing a statically weighted static monitoring index SD_tThe following were used:

accordingly, referring to fig. 2, fig. 2 is a block diagram of a vehicle driving failure detection apparatus according to an embodiment of the present invention. The vehicle driving failure detection device 10 provided by the embodiment of the invention comprises:

the driving fault monitoring module obtaining module 11 is configured to input the obtained driving training data sample into a pre-constructed deep learning model to obtain a driving fault monitoring model;

a fault monitoring threshold obtaining module 12, configured to obtain a control limit of the driving fault monitoring model based on the driving fault monitoring model by using a kernel density estimation method, and use the control limit as a fault monitoring threshold;

the abnormal monitoring index acquisition module 13 is used for acquiring vehicle driving test data acquired in real time and acquiring an abnormal monitoring index of the vehicle driving test data based on the driving fault monitoring model;

a dynamic monitoring index obtaining module 14, configured to take into account the fault sensitivity difference of the features, introduce a dynamic weighting factor to process the abnormal monitoring index, so as to obtain a dynamic monitoring index;

and the fault warning module 15 is configured to determine that a vehicle has a driving fault when the dynamic monitoring index is greater than the fault monitoring threshold, and send a fault warning to a vehicle driver.

In an embodiment, the driving fault monitoring model obtaining module 11 is specifically configured to:

constructing a depth support vector data model based on depth features;

and inputting the acquired driving training data sample into the deep support vector data model to obtain a driving fault monitoring model.

In an embodiment, the failure monitoring threshold obtaining module 12 is specifically configured to:

acquiring a training sample monitoring index of the driving training data sample in the driving fault monitoring model;

constructing a probability density function of the monitoring index of the training sample based on the driving training data sample;

and calculating an estimated value of the probability density function when the confidence level is preset to obtain a corresponding control limit, and taking the control limit as a fault monitoring threshold.

In an embodiment, the abnormality monitoring index obtaining module 13 is specifically configured to:

inputting vehicle driving test data acquired in real time into the driving fault monitoring model, and performing multi-layer feature extraction to obtain a depth feature set of the test data;

and carrying out hypersphere modeling on the depth feature set, calculating the square of the distance from the depth feature set to the center of a hypersphere, and taking the calculated square of the distance as an abnormal monitoring index of the driving test data.

In an embodiment, the dynamic monitoring index obtaining module 14 is specifically configured to:

acquiring a depth characteristic set extracted from the driving test data in the driving fault monitoring model;

aiming at the depth feature set, establishing a history window { D with the length of D in a sliding window mode_t-d+1,i,D_t-d+2,i,…,D_t,iEstimating the abnormal monitoring index according to the dynamic data of the historical window to obtain a dynamic monitoring index:

wherein DD_tIn order to monitor the indicators dynamically,is a dynamic weight factor;

in one embodiment, the vehicle driving failure detection apparatus 10 further includes:

the static monitoring index acquisition module is used for introducing a static weighting factor to process the abnormal monitoring index to obtain a static monitoring index;

and the driving safety reminding module is used for sending driving safety reminding to a vehicle driver when the static monitoring index is larger than the fault monitoring threshold value.

In an embodiment, the static monitoring index obtaining module is specifically configured to:

acquiring a depth feature set extracted from the driving test data in the driving fault monitoring model, and calculating the sub-components of each depth feature in the depth feature set in the abnormal monitoring index;

calculating the fault probability of each depth feature in the depth feature set;

calculating a static weight factor of each depth feature according to the fault probability of each depth feature;

and obtaining a static monitoring index according to the static weighting factor of each depth feature and the sub-component of each depth feature in the abnormal monitoring index.

It should be noted that the vehicle driving failure detection apparatus 10 provided in the embodiment of the present invention is used for executing all the steps and processes of the vehicle driving failure detection method provided in the above embodiment, and the working principles and the effects of the two correspond to each other, which is not described in detail herein.

Furthermore, the above-described device embodiments are merely illustrative, wherein the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. In addition, in the drawings of the embodiment of the apparatus provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, and may be specifically implemented as one or more communication buses or signal lines. One of ordinary skill in the art can understand and implement it without inventive effort. In addition, the vehicle driving fault detection device provided by the embodiment and the vehicle driving fault detection method provided by the embodiment of the invention belong to the same concept, and the specific implementation process and the specific technical scheme are detailed in the method embodiment and are not described again.

Accordingly, embodiments of the present invention also provide a vehicle driving failure detection apparatus, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements steps S11 to S15 of the vehicle driving failure detection method when executing the computer program. Alternatively, the processor implements the functions of the modules in the above-described device embodiments when executing the computer program, such as the driving failure monitoring model obtaining module 11, the failure monitoring threshold obtaining module 12, the abnormality monitoring index obtaining module 13, the dynamic monitoring index obtaining module 14, and the failure warning module 15.

Illustratively, the computer program may be partitioned into one or more modules that are stored in the memory and executed by the processor to implement the invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which are used for describing the execution process of the computer program in the vehicle driving failure detection apparatus/device.

The vehicle driving fault detection apparatus 10/device may be a desktop computer, a notebook computer, a palm computer, a cloud server, or other computing devices. The vehicle driving fault detection apparatus 10/device may include, but is not limited to, a processor, a memory. It will be understood by those skilled in the art that the schematic diagram is merely an example of the vehicle driving failure detection apparatus 10/device, and does not constitute a limitation of the vehicle driving failure detection apparatus 10/device, and may include more or less components than those shown, or combine some components, or different components, for example, the vehicle driving failure detection apparatus 10/device may further include an input-output device, a network access device, a bus, etc.

The Processor may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. The general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, the processor being the control center for the fault warning module apparatus 10/device, with various interfaces and lines connecting the various parts of the overall vehicle driving fault detection apparatus 10/device.

The memory may be used to store the computer programs and/or modules, and the processor may implement the various functions of the vehicle driving failure detection apparatus 10/device by running or executing the computer programs and/or modules stored in the memory, as well as invoking data stored in the memory. The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, a phonebook, etc.) created according to the use of the cellular phone, and the like. In addition, the memory may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), at least one magnetic disk storage device, a Flash memory device, or other volatile solid state storage device.

Wherein the vehicle driving failure detection apparatus 10/device integrated module/unit may be stored in a computer readable storage medium if it is implemented in the form of a software functional unit and sold or used as a separate product. Based on such understanding, all or part of the flow of the method according to the embodiments of the present invention may also be implemented by a computer program, which may be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method embodiments may be implemented. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. Correspondingly, the embodiment of the invention also provides a storage medium, which comprises a stored computer program, wherein when the computer program runs, the equipment where the storage medium is located is controlled to execute the steps S11 to S15 of the vehicle driving fault detection method.

The storage medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, and the like.

While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.