Model training method, autonomous driving vehicle fault prediction method and system
By synchronizing the clock and training the model using the operation log data of autonomous vehicles, and combining gradient boosting trees and neural network models, a fault prediction model is established. This solves the problems of low efficiency and insufficient accuracy of traditional fault handling methods, and achieves efficient and accurate fault prediction and maintenance suggestion generation.
Patent Information
- Application Number
- CN202311775965.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-21
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2043-12-21
AI Technical Summary
Traditional methods for handling faults in autonomous vehicles are inefficient, inaccurate, and unreliable, relying on human experience, which can lead to misjudgments and long processing times.
By acquiring operational log data from multiple operating systems of autonomous vehicles, performing clock synchronization processing, establishing a training sample set, and training it using a gradient boosting tree model and a neural network model, the resulting fault prediction model is obtained. By combining the fault prediction results with the correlation between the maintenance plan, maintenance suggestion information is generated.
It improves the efficiency and accuracy of fault handling for autonomous vehicles, reduces fault location time, lowers economic losses, and enables real-time performance monitoring and maintenance of vehicle operating systems.
Smart Images

Figure CN117972416B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent driving technology, and in particular to a model training method, an automatic driving vehicle fault prediction method and a system. Background Art
[0002] Autonomous driving vehicles are usually composed of multiple working systems, which may include chassis systems, drive-by-wire systems, brake-by-wire systems, steering-by-wire systems, battery management systems, and autonomous driving systems. Failure in any working system will affect the normal driving of the vehicle.
[0003] In the related art, after the alarm information is reported on one or more working systems on the vehicle, the information associated with the alarm information is manually counted, and the cause of the fault is analyzed and the fault location is located based on the maintenance personnel's experience. This fault handling method is not only time-consuming and labor-intensive, but also prone to misjudgment, resulting in a long troubleshooting process.
[0004] Therefore, the traditional vehicle fault handling method has the problems of low processing efficiency, lack of accuracy and reliability. Summary of the invention
[0005] The present invention provides a model training method, an automatic driving vehicle fault prediction method and a system to solve the defects of low processing efficiency, inaccuracy and reliability of traditional vehicle fault handling methods.
[0006] In a first aspect, the present invention provides a model training method, comprising:
[0007] Obtaining operation log data samples of multiple working systems of the autonomous driving vehicle; wherein the multiple working systems have been pre-processed with clock synchronization;
[0008] Labeling the operation log data samples to establish a training sample set;
[0009] The pre-built gradient boosting tree model and the neural network model are trained respectively by the training sample set to obtain a first learning model and a second learning model;
[0010] The first learning model and the second learning model are combined to obtain a fault prediction model.
[0011] According to the model training method provided by the present invention, the operation log data samples are labeled to establish a training sample set, including:
[0012] Adding a unified time tag to the operation log data sample to obtain a first data sample;
[0013] Extracting an alarm data sample and a performance indicator sample from the first data sample;
[0014] Adding an alarm level label to the alarm data sample and adding a performance label to the performance indicator sample to obtain a second data sample;
[0015] A training sample set is established based on the second data sample.
[0016] According to the model training method provided by the present invention, a training sample set is established based on the second data sample, including:
[0017] Adding root cause features to the second data sample to obtain an initial sample set;
[0018] At least part of the data in the initial sample set is subjected to feature elimination processing and feature normalization processing to obtain a training sample set.
[0019] According to the model training method provided by the present invention, feature elimination processing is performed on at least part of the data in the initial sample set, including:
[0020] Determine the characteristic variance, missing value ratio and characteristic correlation coefficient of at least part of the data in the initial sample set;
[0021] Extracting target features that meet a rejection condition from at least part of the data; wherein the rejection condition includes at least one of the feature variance being lower than a preset variance threshold, the missing value ratio being higher than a preset ratio threshold, and the feature correlation coefficient being higher than a preset coefficient threshold;
[0022] The target feature is removed from at least part of the data.
[0023] According to the model training method provided by the present invention, the pre-built gradient boosting tree model is trained by the training sample set to obtain a first learning model, including:
[0024] Dividing the training sample set into a plurality of first sample subsets;
[0025] Based on the multiple first sample subsets, establish multiple first training and testing sample groups; wherein each of the first training and testing sample groups includes a first test set and a first training set, the first test set includes at least one of the first sample subsets, and the first training set includes the remaining first sample subsets except the first test set;
[0026] The gradient boosting tree model is trained and tested respectively by the plurality of first training and testing sample groups to obtain a plurality of first learning sub-models;
[0027] The multiple first learning sub-models are weightedly fused to obtain a first learning model.
[0028] According to the model training method provided by the present invention, the pre-built neural network model is trained by the training sample set to obtain a second learning model, including:
[0029] Dividing the training sample set into a plurality of second sample subsets;
[0030] Based on the multiple second sample subsets, establish multiple second training and testing sample groups; wherein each of the second training and testing sample groups includes a second test set and a second training set, the second test set includes at least one of the second sample subsets, and the second training set includes the remaining second sample subsets except the second test set;
[0031] The gradient boosting tree model is trained and tested respectively by the plurality of second training and testing sample groups to obtain a plurality of second learning sub-models;
[0032] The multiple second learning sub-models are weightedly fused to obtain a second learning model.
[0033] In a second aspect, the present invention further provides a method for predicting faults of an autonomous driving vehicle, comprising:
[0034] Receive operation log data uploaded by multiple working systems of the autonomous driving vehicle;
[0035] Inputting the operation log data into a fault prediction model to obtain a fault prediction result output by the fault prediction model;
[0036] Wherein, the fault prediction model is trained based on any of the model training methods described above.
[0037] According to the automatic driving vehicle fault prediction method provided by the present invention, after obtaining the fault prediction result output by the fault prediction model, the method further includes:
[0038] Establishing a correlation between each of the fault prediction results and the maintenance plan to obtain a relational database;
[0039] Retrieving the target maintenance plan corresponding to the fault prediction result in the relational database;
[0040] Based on the target maintenance plan, maintenance suggestion information corresponding to the fault prediction result is generated.
[0041] According to the automatic driving vehicle fault prediction method provided by the present invention, after generating the maintenance suggestion information corresponding to the fault prediction result, the method further includes:
[0042] Obtaining a score corresponding to the maintenance suggestion information;
[0043] The maintenance suggestion information with a score value greater than a preset score threshold, the fault prediction result corresponding to the maintenance suggestion information, and the operation log data are used as update data samples for updating the fault prediction model.
[0044] In a third aspect, the present invention further provides an automatic driving vehicle fault prediction system, comprising:
[0045] Multiple working systems of autonomous vehicles;
[0046] An electronic device is connected to the multiple working systems, the electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, it implements any one of the above methods for predicting faults of an autonomous driving vehicle.
[0047] The model training method, the method and system for predicting faults of autonomous vehicles provided by the present invention obtain operation log data samples and establish a training sample set, use the training sample set to train the gradient boosting tree model and the neural network model respectively, and finally merge the first learning model and the second learning model obtained by training to obtain a fault prediction model. Since the fault prediction model is obtained by training the operation log data samples of multiple working systems of the autonomous vehicle after adding labels, and integrates the training results of the gradient boosting tree model and the neural network model, the obtained fault prediction model is more accurate. The model is used in the vehicle fault handling process to improve the processing efficiency and accuracy of the vehicle fault handling link. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0049] Figure 1 It is a flowchart of a model training method provided by an embodiment of the present invention;
[0050] Figure 2 This is a schematic diagram of the principle of the gradient boosting tree model training process;
[0051] Figure 3 is a flow chart of a method for predicting faults of an autonomous driving vehicle provided by an embodiment of the present invention;
[0052] Figure 4 It is a schematic diagram of the principle of the training process and application process of the fault prediction model;
[0053] Figure 5 is a schematic diagram of the structure of an automatic driving vehicle fault prediction system provided by an embodiment of the present invention;
[0054] Figure 6 It is a structural diagram of multiple working systems. DETAILED DESCRIPTION
[0055] Embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.
[0056] The terms used in the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the" and "the" used in the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0057] It should be understood that although the terms "first", "second", "third", etc. may be used to describe various information in the present invention, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.
[0058] Combine the following Figures 1 to 6 Describe the detailed scheme of the model training method, automatic driving vehicle fault prediction method and system provided by the embodiments of the present invention.
[0059] See also Figure 1 The model training method provided by the embodiment of the present invention specifically includes:
[0060] Step 110: Obtain operation log data samples of multiple working systems of the autonomous driving vehicle; wherein the multiple working systems have been pre-processed with clock synchronization.
[0061] In this embodiment, before the model training and model application phases begin, it is necessary to perform clock synchronization processing on multiple working systems of the autonomous driving vehicle to ensure the consistency of the clocks between the various working systems.
[0062] In some embodiments, a unified clock source can be used to synchronize the clocks of all working systems of the autonomous driving vehicle. Specifically, the clocks of the operating data, alarm data and other operating log data of each working system can be synchronized. In practical applications, a unified clock source can be set for devices that store operating log data, such as CAN (Controller Area Network) recorders, NAS (Network Attached Storage) storage devices, etc. The clock source can use the signals of positioning satellites such as GPS, Beidou, Galileo, etc. in the positioning system, or an IP clock source can be added to the vehicle system, and the clock synchronization protocol such as IEEE1588 can be used for clock synchronization.
[0063] In addition, the networks connecting the various working systems on the autonomous driving vehicle also need to undergo clock synchronization, including the clock of the vehicle's underlying CAN network, the clock of the on-board Ethernet, and the clock of the on-board industrial switch network. The cloud control center's clocks, including the computing and storage models, can also undergo clock synchronization.
[0064] In actual applications, on the one hand, the log and alarm information of other working systems can be collected through the autonomous driving system. The working systems in the autonomous driving system include the perception system, planning control system, positioning system, communication system, etc. On the other hand, the communication data between the various wire control systems of the vehicle chassis and the vehicle chassis VCU (Vehicle Control Unit) and the communication data between the vehicle chassis VCU and the autonomous driving system can be recorded through the CAN recorder. The above communication data is reported to the autonomous driving system according to the preset cycle. The working systems of the vehicle chassis, including the power system, braking system, steering system, load-carrying system, tire pressure system, etc., the alarm and log information of each working system are reported to the vehicle chassis VCU, and then reported to the autonomous driving system through the VCU.
[0065] Therefore, an operation log data sample including logs and alarm information of each working system and communication data can be obtained from the autonomous driving system.
[0066] Step 120: label the operation log data samples and establish a training sample set.
[0067] It is understandable that in order to ensure the data accuracy and completeness of the training sample set, the operation log data samples can be pre-processed by adding labels, thereby providing a reliable data basis for subsequent model training.
[0068] Step 130: Train the pre-built gradient boosting tree model and neural network model respectively through the training sample set to obtain a first learning model and a second learning model.
[0069] In this embodiment, two types of data models are trained respectively using the established training sample sets. The first type of data model adopts the gradient boosting tree model. The gradient boosting tree model uses the gradient boosting decision tree (GBDT) as the main model architecture. The gradient boosting tree is an algorithm that uses the negative gradient of the loss function in the current model as an approximate value of the residual to fit the regression tree.
[0070] For example, the gradient boosting tree model can specifically adopt the XGBoost model, which is a type of gradient boosting tree model. It also generates models serially and takes the sum of all models as output. The XGBoost model can perform a second-order Taylor expansion on the loss function, optimize the loss function using the second-order derivative information of the loss function, and choose whether to split the node based on whether the loss function is reduced. At the same time, the XGBoost model adds regularization, learning rate, column sampling, approximate optimal split point and other means to prevent overfitting, and the model reliability is higher.
[0071] In this embodiment, the second type of data model adopts a neural network model, which is composed of a widely interconnected neural network composed of multiple neurons, and can simulate the interactive reactions between biological nervous systems and objects in the real world. In the neural network model, each neuron is both an input and an output. Each neuron in the network transmits its output to all other neurons through connection weights, and at the same time receives information transmitted by all other neurons.
[0072] Step 140: Merge the first learning model and the second learning model to obtain a fault prediction model.
[0073] In this embodiment, by merging the two types of trained data models, namely the first learning model and the second learning model, the possibility of model overfitting is reduced, so that the final fault prediction model can have better fault analysis and prediction capabilities, thereby improving the model accuracy.
[0074] In one embodiment, labeling is performed on the operation log data samples to establish a training sample set, which specifically includes:
[0075] The first step is to add a unified time tag to the operation log data sample to obtain the first data sample.
[0076] In this embodiment, the operation log data samples obtained for each working system are generally in text format. The log information in the operation log data samples can be added to the database one by one, and a unified time tag can be loaded to ensure the format consistency of the time identification information. At the same time, the operation log data samples can also be preliminarily classified to improve the processing efficiency and data reliability of the sample preprocessing link.
[0077] The second step is to extract the alarm data samples and performance indicator samples from the first data samples.
[0078] In this embodiment, after adding the time tag, specific tags can be further added to the alarm data samples and the performance indicator samples. Therefore, the alarm data samples and the performance indicator samples in the first data sample need to be extracted first.
[0079] The third step is to add an alarm level label to the alarm data sample and add a performance label to the performance indicator sample to obtain a second data sample.
[0080] In some embodiments, alarm level labels can be added according to the degree of impact of the alarm information in the alarm data sample. For example, alarm data samples that affect the operation of the entire vehicle can be added with a "serious alarm" label, alarm data samples that affect the operation of the system can be added with a "general alarm" label, and alarm data samples that affect the performance of the system can be added with a "prompt alarm" label.
[0081] The performance label is mainly used to distinguish information related to the performance of the working system. In some embodiments, the performance indicator samples can be distinguished by adding a "performance related" label.
[0082] The fourth step is to establish a training sample set based on the second data sample.
[0083] For the second data sample obtained after labeling, a training sample set can be established by sample aggregation, thereby providing accurate and reliable data basis for model training.
[0084] In one embodiment, establishing a training sample set based on the second data sample specifically includes:
[0085] First, the root cause feature is added to the second data sample to obtain an initial sample set.
[0086] Before training begins, you can add root cause features to the second data sample, which is the root cause of the problem after the fault alarm. For example, the failure of the vehicle to operate due to high voltage power failure of the vehicle's electric brake is mainly related to the three working systems of ECU (Electronic Control Unit), BMS (Battery Management System) and autonomous driving system.
[0087] In this step, common fault causes can be determined and associated with operation log data samples of related working systems, including log information and alarm information of related working systems, and characteristic data of related working systems in a preset period before the fault occurs can be extracted, such as: {log information and alarm information of ECU system, log information and alarm information of BMS system, log information and alarm information of autonomous driving system, …}, and corresponding root cause tags can be added, such as: {low battery system power, battery system failure, motor system failure, …}.
[0088] In addition, each labeled second data sample contains a time series, and each time series can be feature processed to construct new features. Taking the BMS system failure cause of abnormal high voltage power failure of the vehicle as an example, the following new features can be constructed: {battery high voltage power-on time, battery high voltage output maximum current, battery high voltage output minimum current, battery high voltage output average current, battery pack 1 power, battery pack 2 power, battery pack 3 power, battery pack 4 power, ...}, etc. The missing values in each new feature are filled with the average value to obtain the initial sample set.
[0089] Then, feature elimination processing and feature normalization processing are performed on at least part of the data in the initial sample set to obtain a training sample set.
[0090] It can be understood that the accuracy and reliability of sample features in the training sample set can be improved through feature elimination and feature normalization.
[0091] In one embodiment, feature elimination is performed on at least part of the data in the initial sample set, specifically including:
[0092] Determine the characteristic variance, missing value ratio, and characteristic correlation coefficient of at least part of the data in the initial sample set;
[0093] Extracting target features that meet the elimination conditions from at least part of the data; wherein the elimination conditions include at least one of a feature variance lower than a preset variance threshold, a missing value ratio higher than a preset ratio threshold, and a feature correlation coefficient higher than a preset coefficient threshold;
[0094] Remove target features from at least part of the data.
[0095] In this embodiment, the target features whose feature variance does not meet the requirements can be eliminated, that is, some features with smaller feature differences can be eliminated; the target features whose missing value ratio does not meet the requirements can also be eliminated, that is, features with more missing values can be eliminated; the target features whose feature correlation coefficient does not meet the requirements can also be eliminated, that is, features with higher correlation can be eliminated.
[0096] Through the feature elimination operation, some features that affect the accuracy of model training can be eliminated to improve the accuracy of the data in the entire training sample set.
[0097] In one embodiment, a pre-built gradient boosting tree model is trained using a training sample set to obtain a first learning model, specifically including:
[0098] Dividing the training sample set into a plurality of first sample subsets;
[0099] Based on the multiple first sample subsets, multiple first training and testing sample groups are established; wherein each first training and testing sample group includes a first test set and a first training set, the first test set includes at least one first sample subset, and the first training set includes the remaining first sample subsets except the first test set;
[0100] The gradient boosting tree model is trained and tested respectively through a plurality of first training and testing sample groups to obtain a plurality of first learning sub-models;
[0101] The multiple first learning sub-models are weightedly fused to obtain a first learning model.
[0102] In this embodiment, the gradient boosting tree model can be cross-validated using a training sample set, for example, by using a five-fold cross-validation.
[0103] Specifically, the training sample set can be divided into five first sample subsets A1, A2, A3, A4, and A5, wherein five first training test sample groups can be established, namely:
[0104] Use subset A1 as the test set, and subsets A2, A3, A4, and A5 as the training sets;
[0105] Use subset A2 as the test set, and subsets A1, A3, A4, and A5 as the training sets;
[0106] Use subset A3 as the test set, and subsets A1, A2, A4, and A5 as the training sets;
[0107] Use subset A4 as the test set, and subsets A1, A2, A3, and A5 as the training sets;
[0108] Subset A5 is used as the test set, and subsets A1, A2, A3, and A4 are used as training sets.
[0109] By training and testing the gradient boosting tree model through the five first training and testing sample groups, five first learning sub-models can be obtained, and finally the five first learning sub-models are weighted and fused to obtain a first learning model, that is, a machine learning model. In some embodiments, the weight values used in the weighted fusion process can be determined according to the training scores corresponding to each of the five first learning sub-models.
[0110] Figure 2 The principle of the gradient boosting tree model training process is shown in Figure 2 After the data X1 of a certain training set is input into the decision tree 1 of the gradient boosting tree model, the preliminary fault diagnosis results can be output, such as BMS system failure, ECU system failure, automatic driving system emergency stop, and CAN communication interruption emergency stop.
[0111] On the one hand, the preliminary fault diagnosis result of the BMS system failure will be input into the decision tree 2 together with the relevant training set data X2. After being processed by the decision tree 2, further fault diagnosis results can be output, such as low battery or no abnormality. At the same time, nodes trained according to the BMS log information within time T1 and nodes trained according to the BMS alarm information within time T1 can also be obtained.
[0112] On the other hand, the ECU system fault will be input into the decision tree 3 together with the relevant training set data X3. After being processed by the decision tree 3, further fault diagnosis results can be output, such as abnormal system parameter configuration, abnormal power system performance or no abnormality. At the same time, nodes trained according to the power system log information within time T2 and nodes trained according to the power system warning information within time T2 can also be obtained.
[0113] In one embodiment, the pre-built neural network model is trained by the training sample set to obtain the second learning model, which specifically includes:
[0114] Dividing the training sample set into a plurality of second sample subsets;
[0115] Based on the multiple second sample subsets, establish multiple second training and testing sample groups; wherein each second training and testing sample group includes a second test set and a second training set, the second test set includes at least one second sample subset, and the second training set includes the remaining second sample subsets except the second test set;
[0116] The gradient boosting tree model is trained and tested respectively through a plurality of second training and testing sample groups to obtain a plurality of second learning sub-models;
[0117] The multiple second learning sub-models are weightedly fused to obtain a second learning model.
[0118] In this embodiment, the number of neurons in the first layer of the neural network model is equal to the number of features of the data samples, and the number of neurons in the last layer is equal to the number of label types. A four-layer neural network is set, and the number of neurons in the middle two layers is determined by the results of cross-validation of the neural network model. The second learning model finally obtained by training is the deep learning model.
[0119] It should be noted that the training process of the neural network model is basically the same as the training process of the above-mentioned gradient boosting tree model, and will not be repeated here.
[0120] In some embodiments, the process of merging the first learning model and the second learning model can be implemented using a stacking algorithm. The stacking algorithm can be regarded as a process of learning a model to fuse several existing models. In other words, the stacking algorithm can be regarded as a meta-learning, in which several base models (first layer) are inner-loops, and the meta-classifier (second layer) can be regarded as an outer-loop.
[0121] Based on the same general inventive concept, the present invention also protects an autonomous driving vehicle fault prediction method and system. The autonomous driving vehicle fault prediction method and system provided by the present invention are described below. The autonomous driving vehicle fault prediction method and system described below can be referenced to each other with the model training method described above.
[0122] Figure 3 It is a flow chart of a method for predicting faults of an autonomous driving vehicle provided in an embodiment of the present invention.
[0123] See also Figure 3 The automatic driving vehicle fault prediction method provided by the embodiment of the present invention specifically includes:
[0124] Step 210: Receive operation log data uploaded by multiple working systems of the autonomous driving vehicle.
[0125] Step 220: Input the operation log data into the fault prediction model to obtain the fault prediction result output by the fault prediction model.
[0126] Among them, the fault prediction model is trained based on the model training method provided in the above embodiments.
[0127] The solution provided in this embodiment can directly obtain accurate fault prediction results through the fault prediction model obtained through pre-training, thereby improving the efficiency and reliability of the vehicle fault prediction process.
[0128] In one embodiment, after obtaining the fault prediction result output by the fault prediction model, the method may further include:
[0129] The first step is to establish the association between each fault prediction result and the maintenance plan to obtain a relational database.
[0130] In this embodiment, the root cause of the problem in the fault prediction result, the system to which the problem belongs, and the corresponding maintenance plan can be presented in the form of an association table. Table 1 below exemplarily shows the structure of an association table.
[0131] Table 1 Correlation table between fault prediction results and maintenance plans
[0132] Serial number Root cause of the problem Problem attribution system Maintenance plan score 1 5G network failure causes RTK to not converge Positioning system Check 5G CPE and 5G network signal 95 2 Insufficient power causes the power system to shut down Battery System Vehicle battery replacement or charging 85 3 The lack of rear-facing LiDAR signal caused the autonomous driving system to make an emergency stop Perception System LiDAR to domain controller vehicle Ethernet line failure 90 4 The vehicle is parked on the lane line, causing the vehicle to be unable to move Autonomous Driving Systems The autonomous driving system autonomously moves the car to the AD road 80
[0133] In some embodiments, the scoring values for each maintenance plan may also be associated in the association table, and the scoring values may be manually set by the user based on the matching consistency between the maintenance result and the maintenance plan.
[0134] The second step is to retrieve the target maintenance plan corresponding to the fault prediction result in the relational database.
[0135] After the relational database is established, the corresponding target maintenance plan can be directly matched according to the fault prediction results, and the target maintenance plan and the fault prediction results can be displayed to the user together.
[0136] The third step is to generate maintenance suggestion information corresponding to the fault prediction results based on the target maintenance plan.
[0137] In some embodiments, the maintenance suggestion information may directly adopt the content corresponding to the target maintenance plan, or may be given after adaptively adjusting the content based on the target maintenance plan according to the current vehicle status.
[0138] In one embodiment, after generating maintenance suggestion information corresponding to the fault prediction result, the method may further include:
[0139] Obtain the score value corresponding to the maintenance suggestion information;
[0140] The maintenance suggestion information with a score value greater than a preset score threshold, the fault prediction result corresponding to the maintenance suggestion information, and the operation log data are used as update data samples for updating the fault prediction model.
[0141] In this embodiment, according to the maintenance suggestion information given for the actual faults in the application stage, the relevant data with higher scoring values can be used as new data samples to update the fault prediction model, so that the fault prediction model can be adaptively adjusted according to the actual application situation in the application stage, thereby improving the accuracy and reliability of the fault prediction process.
[0142] In some embodiments, the executor of the above-mentioned autonomous driving vehicle fault prediction method can be an autonomous driving domain controller, and the fault prediction model training process can be completed in the cloud control center. In practical applications, the fault prediction function can be used as a low-priority task of the autonomous driving domain controller and queued in the autonomous driving domain controller to save the computing resources of the autonomous driving domain controller.
[0143] In a specific implementation, the training process and application process of the fault prediction model can be found in Figure 4 ,like Figure 4 As shown, on the one hand, in the cloud control center 310, the clocks of the working systems of the entire vehicle can be synchronized first, and then the operation log data samples of each working system can be obtained. After preprocessing the operation log data samples, the obtained training sample set can be used for model training to obtain a fault prediction model.
[0144] On the other hand, in the autonomous driving domain controller 320, the corresponding solution can be associated with the root cause of the fault obtained by fault analysis based on the input alarm information. Subsequently, after receiving the operation log data of each working system, the fault is predicted and the fault is handled according to the solution corresponding to the fault prediction result.
[0145] The solution provided in this embodiment can automatically provide the root cause and solution of the vehicle failure before or after the vehicle failure occurs, thereby saving the time of locating the vehicle failure, and can solve the vehicle failure in time, reducing the economic losses caused by the failure; at the same time, it can also detect the performance of each working system of the vehicle in real time, and automatically determine whether each working system needs to be repaired and parts replaced, saving the maintenance time of the vehicle.
[0146] Figure 5 It is a structural schematic diagram of an automatic driving vehicle fault prediction system provided by an embodiment of the present invention.
[0147] See also Figure 5 The automatic driving vehicle fault prediction system provided by the embodiment of the present invention specifically includes:
[0148] Multiple operating systems 400 of an autonomous vehicle;
[0149] The electronic device 500 is connected to the multiple working systems 400, and the electronic device 500 may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the automatic driving vehicle fault prediction method provided in the above embodiments.
[0150] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0151] In actual applications, the multiple working systems 400 of an autonomous driving vehicle generally involve a perception system, a battery management system, a power system, a braking system, a steering system, an autonomous driving system, etc. Figure 6 shows multiple operating systems of an autonomous vehicle, see Figure 6The autonomous driving system specifically includes multiple parts such as the perception system, the autonomous driving domain controller, and the RTK navigation system. The perception system specifically includes perception devices such as laser radar, high-definition camera, and millimeter-wave radar. The autonomous driving domain controller can be connected to the vehicle controller and the battery management system through the CAN bus in the communication system. The data of the autonomous driving domain controller will be stored in the NAS storage system, and the data of the CAN bus will be stored in the CAN recorder. The NAS storage system will be connected to the cloud control center through the 5G router. The data of the vehicle controller will be stored in the VCU log storage system, and the data of the battery management system will be stored in the BMS log storage system. The vehicle controller can monitor the tire pressure and load of the vehicle. At the same time, the vehicle controller is also connected to the power system, the braking system, and the steering system. The power system specifically includes the motor system and the transmission system. The braking system specifically includes three parts: EPB braking, electric braking, and parking brake. The steering system specifically includes steering sensors and steering hydraulic systems.
[0152] By connecting the above-mentioned working system with electronic equipment, the fault prediction function of the autonomous driving vehicle can be realized. It can be understood that the electronic device can be a computer or server deployed in the cloud control center.
[0153] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the autonomous driving vehicle fault prediction method provided in the above-mentioned embodiments.
[0154] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the automatic driving vehicle fault prediction method provided in the above-mentioned embodiments.
[0155] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0156] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A model training method, characterized in that: include: Obtaining operation log data samples of multiple working systems of the autonomous driving vehicle; wherein the multiple working systems have been pre-processed with clock synchronization; Add a unified time label to the operation log data sample to obtain a first data sample; extract the alarm data sample and the performance indicator sample in the first data sample; add an alarm level label to the alarm data sample, and add a performance label to the performance indicator sample to obtain a second data sample; add root cause features to the second data sample, determine the common cause of the fault, and associate the operation log data samples of the relevant working system, specifically including the log information and alarm information of the relevant working system, extract the feature data of the relevant working system within a preset period before the fault occurs, and add the corresponding root cause label, each labeled second data sample includes a time series, perform feature processing on each time series, construct new features, and fill the missing values in each new feature with the average value to obtain an initial sample set; perform feature elimination processing and feature normalization processing on at least part of the data in the initial sample set to obtain a training sample set; The pre-constructed gradient boosting tree model and the neural network model are trained respectively by the training sample set to obtain a first learning model and a second learning model; wherein the number of neurons in the first layer of the neural network model is equal to the number of features of the data sample, the number of neurons in the last layer is equal to the number of types of labels, and the number of neurons in the middle two layers is determined by the result obtained by cross-validation of the neural network model; The first learning model and the second learning model are combined to obtain a fault prediction model.
2. The model training method according to claim 1, characterized in that: Performing feature elimination processing on at least part of the data in the initial sample set includes: Determine the characteristic variance, missing value ratio and characteristic correlation coefficient of at least part of the data in the initial sample set; Extracting target features that meet a rejection condition from at least part of the data; wherein the rejection condition includes at least one of the feature variance being lower than a preset variance threshold, the missing value ratio being higher than a preset ratio threshold, and the feature correlation coefficient being higher than a preset coefficient threshold; The target feature is removed from at least part of the data.
3. The model training method according to claim 1, characterized in that: The pre-built gradient boosting tree model is trained by the training sample set to obtain a first learning model, including: Dividing the training sample set into a plurality of first sample subsets; Based on the multiple first sample subsets, establish multiple first training and testing sample groups; wherein each of the first training and testing sample groups includes a first test set and a first training set, the first test set includes at least one of the first sample subsets, and the first training set includes the remaining first sample subsets except the first test set; The gradient boosting tree model is trained and tested respectively by the plurality of first training and testing sample groups to obtain a plurality of first learning sub-models; The multiple first learning sub-models are weightedly fused to obtain a first learning model.
4. The model training method according to claim 1, characterized in that: The pre-built neural network model is trained by the training sample set to obtain a second learning model, including: Dividing the training sample set into a plurality of second sample subsets; Based on the multiple second sample subsets, establish multiple second training and testing sample groups; wherein each of the second training and testing sample groups includes a second test set and a second training set, the second test set includes at least one of the second sample subsets, and the second training set includes the remaining second sample subsets except the second test set; The gradient boosting tree model is trained and tested respectively by the plurality of second training and testing sample groups to obtain a plurality of second learning sub-models; The multiple second learning sub-models are weightedly fused to obtain a second learning model.
5. A method for predicting faults of an autonomous driving vehicle, characterized in that: include: Receive operation log data uploaded by multiple working systems of the autonomous driving vehicle; Inputting the operation log data into a fault prediction model to obtain a fault prediction result output by the fault prediction model; The fault prediction model is trained based on the model training method according to any one of claims 1 to 4.
6. The automatic driving vehicle fault prediction method according to claim 5, characterized in that: After obtaining the fault prediction result output by the fault prediction model, the method further includes: Establishing a correlation between each of the fault prediction results and the maintenance plan to obtain a relational database; Retrieving the target maintenance plan corresponding to the fault prediction result in the relational database; Based on the target maintenance plan, maintenance suggestion information corresponding to the fault prediction result is generated.
7. The automatic driving vehicle fault prediction method according to claim 6, characterized in that: After generating the maintenance suggestion information corresponding to the fault prediction result, the method further includes: Obtaining a score corresponding to the maintenance suggestion information; The maintenance suggestion information with a score value greater than a preset score threshold, the fault prediction result corresponding to the maintenance suggestion information, and the operation log data are used as update data samples for updating the fault prediction model.
8. An automatic driving vehicle fault prediction system, characterized in that: include: Multiple working systems of autonomous vehicles; An electronic device connected to the multiple working systems, the electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the automatic driving vehicle fault prediction method as described in any one of claims 5 to 7 when executing the program.
