Method, system, and storage medium for failure prediction based on vehicle historical data

Abstract

The application relates to vehicle fault prediction, in particular to a fault prediction method based on vehicle historical data, a computer system for vehicle fault prediction and a computer storage medium. The method comprises the following steps: A, extracting a plurality of sample data sets from vehicle historical data based on different sample selection strategies; B, respectively performing invalid feature elimination and availability screening on each of the plurality of sample data sets; and C, training a combined model by using each of the plurality of sample data sets, and obtaining a fault prediction result based on the trained combined model. The fault prediction scheme provided by the application combines a combined perspective with a combined model, thereby improving the accuracy of fault prediction while improving the accuracy and comprehensiveness of data mining.

Description

Technical Field

[0001] This invention relates to vehicle fault prediction, and more specifically, to a fault prediction method based on historical vehicle data, a computer system for vehicle fault prediction, and a computer storage medium. Background Technology

[0002] Currently, most vehicle malfunction handling methods remain at the reactive stage, relying on customer complaints for repairs. With the development of information technology and machine learning, some existing technologies attempt to utilize big data analysis and predictive models to provide early warnings of vehicle malfunctions. However, because some vehicle malfunctions occur suddenly, lacking corresponding warning signs beforehand, the collected vehicle operation data lacks information characterizing these signs, resulting in low fault recall and low fault prediction accuracy. Summary of the Invention

[0003] To address or at least alleviate one or more of the above problems, this invention proposes a fault prediction method based on historical vehicle data, a computer system for vehicle fault prediction, and a computer storage medium. The fault prediction scheme proposed in this invention employs a combination of perspectives and models, improving both the accuracy and comprehensiveness of data mining while simultaneously enhancing the accuracy of fault prediction.

[0004] According to a first aspect of the present invention, a fault prediction method based on vehicle historical data is provided, the method comprising: A) extracting multiple sample datasets from the vehicle historical data based on different sample selection strategies; B) performing invalid feature removal and availability screening for each of the multiple sample datasets; and C) training a combined model using each of the multiple sample datasets, and obtaining a fault prediction result based on the trained combined model.

[0005] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, step A includes: A1, receiving vehicle history data of a cluster of faulty vehicles and a cluster of non-faulty vehicles, wherein the vehicle history data includes historical data from at least one source within the vehicle during a first time period ending at the time of the fault; A2, extracting positive sample data from the vehicle history data based on a positive sample selection strategy; A3, extracting multiple sets of negative sample data from the vehicle history data based on multiple negative sample selection strategies; and A4, combining the positive sample data with each of the multiple sets of negative sample data to generate multiple sample datasets for fault prediction.

[0006] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, in step A2, the positive sample selection strategy includes: extracting historical data from the vehicle historical data of the faulty vehicle cluster during a second time period ending at the time of the fault, wherein the second time period is shorter than the first time period.

[0007] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, in step A3, the negative sample selection strategy includes at least two of the following: randomly selecting a subset of non-faulty vehicles from the non-faulty vehicle cluster, and randomly extracting a first set of negative sample data from the vehicle history data of the non-faulty vehicle subset; extracting historical data from the vehicle history data of the faulty vehicle cluster during a third time period ending at the start of the second time period as a second set of negative sample data, wherein the third time period is shorter than the first time period; and selecting a subset of faulty vehicles from the faulty vehicle cluster whose cumulative running time is less than the minimum fault duration, and extracting a third set of negative sample data from the vehicle history data of the faulty vehicle subset.

[0008] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, step B includes one of the following: using variance filtering to remove non-divergent features in each sample dataset; inputting the sample dataset into the algorithm model to calculate the feature importance of each feature, and removing features based on the magnitude of the feature importance.

[0009] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, step B includes: B1, for each sample dataset, calculating the variance of each feature in the sample dataset and removing features with zero variance from the sample dataset.

[0010] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, step B includes performing the following operations for each sample dataset: B2, inputting the original sample dataset into a first algorithm model to obtain the actual feature importance of each feature; B3, randomly shuffling the labels in the original sample dataset, and inputting the shuffled sample dataset into the first algorithm model to obtain the feature importance of each feature under the random labels; B4, repeating step B3 N times to obtain N feature importances of each feature under the random labels; and B5, comparing the actual feature importance with the N feature importances under the random labels, and performing feature removal based on the comparison results.

[0011] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, step B5 includes: for each feature, calculating a statistical feature value of the importance of the N features, the statistical feature value including the 75th percentile of the importance of the N features; calculating the difference between the actual feature importance and the statistical feature value; and if the difference is less than or equal to a first threshold, then removing the feature.

[0012] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, step B further includes: using a first algorithm model to perform usability evaluation on each of the plurality of sample datasets, and filtering out sample datasets with AUC values ​​less than or equal to a second threshold.

[0013] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, step C includes: C1, training a first algorithm model using each of the M sample datasets to obtain M trained first algorithm models; C2, training a second algorithm model using each of the M sample datasets to obtain M trained second algorithm models; C3, testing the M trained first algorithm models and the M trained second algorithm models using one of the M sample datasets to obtain the accuracy of each model; and C4, weighting the prediction results of the M trained first algorithm models and the M trained second algorithm models with the accuracy of each model as the weight to obtain the fault prediction result under the combined model.

[0014] According to a second aspect of the present invention, a computer system for predicting vehicle faults is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements any one of the methods described according to the first aspect of the present invention.

[0015] According to a third aspect of the present invention, a computer storage medium is provided, the computer storage medium including instructions that, when executed, perform any one of the methods described according to the first aspect of the present invention.

[0016] On the one hand, the fault prediction scheme according to one or more embodiments of the present invention adopts a combined perspective (e.g., extracting multiple sample datasets based on multiple sample selection strategies) to avoid the bias caused by defining positive and negative samples from a single perspective, thereby enabling a more comprehensive and accurate mining of fault symptom information in vehicle historical operation data.

[0017] On the other hand, the fault prediction scheme according to one or more embodiments of the present invention removes invalid features in the sample dataset based on, for example, variance filtering or feature importance, and further improves the performance of the model by using a combined model, thereby achieving a balance between the overall computational load and the fault prediction accuracy of the model. Attached Figure Description

[0018] The above and / or other aspects and advantages of the present invention will become clearer and more readily understood from the following description taken in conjunction with the accompanying drawings, in which the same or similar elements are denoted by the same reference numerals. In the drawings:

[0019] Figure 1 A flowchart of a fault prediction method 10 based on vehicle historical data according to one or more embodiments of the present invention; and

[0020] Figure 2 This is a block diagram of a computer system 20 for vehicle fault prediction according to one or more embodiments of the present invention. Detailed Implementation

[0021] The following detailed description is merely exemplary in nature and is not intended to limit the disclosed technology or its application and use. Furthermore, it is not intended to be bound by any express or implied theory presented in the foregoing technical fields, background art, or the following detailed description.

[0022] In the following detailed description of the embodiments, numerous specific details are set forth in order to provide a more thorough understanding of the disclosed technology. However, it will be apparent to those skilled in the art that the disclosed technology can be practiced without these specific details. In other instances, well-known features have not been described in detail to avoid unnecessarily complicating the description.

[0023] Terms such as "comprising" and "including" indicate that, in addition to the units and steps directly and explicitly described in the specification, the technical solution of the present invention does not exclude the presence of other units and steps not directly or explicitly described. Terms such as "first" and "second" do not indicate the order of the units in terms of time, space, size, etc., but are merely used to distinguish the units. The term "vehicle" or other similar terms used herein include general motor vehicles as well as hybrid electric vehicles, electric vehicles, plug-in hybrid electric vehicles, etc.

[0024] In the following, various exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0025] Figure 1 This is a flowchart of a fault prediction method 10 based on vehicle historical data according to one or more embodiments of the present invention.

[0026] like Figure 1 As shown, in step S110, multiple sample datasets are extracted from vehicle historical data based on different sample selection strategies. As described in the background section, because some vehicle malfunctions occur suddenly and lack corresponding warning signs beforehand, it is difficult to discover obvious malfunction signs from vehicle historical operating data solely based on expert experience analysis, resulting in low fault recall and low fault prediction accuracy. This invention adopts a combined perspective approach, that is, defining sample datasets from multiple angles based on different sample selection strategies (e.g., multiple negative sample selection strategies), thereby avoiding the bias caused by defining sample datasets from a single angle and more comprehensively and accurately mining fault sign information from vehicle historical operating data.

[0027] Optionally, in step S110, historical vehicle data from faulty vehicle clusters (e.g., a set of faulty vehicles) and non-faulty vehicle clusters (e.g., a set of non-faulty vehicles) are first received. Since actual vehicle faults are related to various physical quantities under specific conditions, there is an objective correlation between historical vehicle data and vehicle faults, and therefore it can be used for vehicle fault prediction. For example, historical vehicle data may include the steering gear deflection angle, which can be used to analyze steering faults.

[0028] For example, vehicle history data includes historical data from at least one source within the vehicle (e.g., onboard sensors, electronic control unit) during a first time period ending at the time of the failure. In some embodiments of this application, vehicle history data is generated based on vehicle sensor data. Continuing the example above, the steering gear deflection angle can be acquired, for example, by an angular deflection sensor. In other examples, vehicle history data can also be collected using, for example, position sensors, acceleration sensors, temperature sensors, etc. Of course, vehicle history data can also be obtained from other sources. For example, motor torque can be generated based on torque commands generated by the electronic control unit, and therefore vehicle history data can also be acquired from vehicle controllers such as the electronic control unit.

[0029] The sample data extraction strategy can be divided into a positive sample data extraction strategy and a negative sample data extraction strategy. Optionally, positive sample data can be extracted from vehicle historical data based on the same positive sample selection strategy. For example, historical data during a second time period (the second time period is shorter than the first time period) ending at the time of the failure can be extracted from the historical data of a cluster of faulty vehicles. Since the probability of data anomalies is highest in the period before the failure occurs, this period (i.e., the second time period) can be regarded as the data degradation period, and the data within the data degradation period can be used as positive samples. The specific duration of the second time period can be determined based on the failure type and business experience, and this invention does not impose a specific limitation on it.

[0030] Optionally, multiple sets of negative sample data can be extracted from vehicle history data based on various different negative sample selection strategies. For example, the negative sample selection strategies include at least two of strategies 1-3 detailed below.

[0031] Strategy 1: Randomly select a subset of non-faulty vehicles from the non-faulty vehicle cluster, and randomly extract the first set of negative sample data from the vehicle history data of the non-faulty vehicle subset.

[0032] Strategy 2 involves extracting historical data from the faulty vehicle cluster, specifically the third time period ending at the start of the second time period, as a second set of negative sample data. The third time period is shorter than the first time period. It is understood that comparing data from the same vehicle's data degradation period (e.g., the second time period) with data from its non-data degradation period (e.g., the third time period) makes it easier to identify the true fault information. Therefore, the historical data of the faulty vehicle cluster can also serve as a source of negative sample data.

[0033] Strategy 3 involves selecting a subset of faulty vehicles from the faulty vehicle cluster whose cumulative runtime is less than the minimum fault duration, and then extracting a third set of negative sample data from the historical data of these vehicles. The cumulative runtime refers to the total operating time of the vehicle since it rolled off the production line, and the minimum fault duration refers to the minimum time interval between the vehicle's production line completion and the occurrence of the fault within the faulty vehicle cluster.

[0034] Optionally, the positive sample data described above can be combined with each of the multiple sets of negative sample data (e.g., the first set of negative sample data, the second set of negative sample data, and the third set of negative sample data) to generate multiple sample datasets for fault prediction (e.g., the first sample dataset, the second sample dataset, and the third sample dataset).

[0035] In step S120, invalid feature removal and availability screening are performed for each of the multiple sample datasets. It is understood that invalid feature removal and availability screening for the sample datasets can bring at least the following benefits: reducing the size of the training data, reducing the overall computational load, and accelerating model training; reducing model complexity and avoiding overfitting; reducing feature input, which is beneficial for interpreting the model; and improving model accuracy.

[0036] Optionally, in the invalid feature removal stage, variance filtering can be used to remove non-divergent features from each sample dataset (e.g., the first sample dataset, the second sample dataset, and the third sample dataset). Non-divergent features refer to features on which samples show virtually no difference; that is, the feature does not contribute to the differentiation of samples. For example, the divergence of a feature can be determined based on variance. For instance, for each sample dataset, the variance of each feature in the sample dataset can be calculated, and features with zero variance can be removed from the sample dataset.

[0037] Alternatively, invalid feature removal can be performed based on algorithmic models. For example, each sample dataset can be input into an algorithmic model (e.g., LightGBM, Random Forest, XGBoost) to calculate the feature importance of each feature, and features can be removed based on their importance. Specifically, the following two algorithmic model-based invalid feature removal strategies are provided.

[0038] First, each sample dataset is input into a first algorithm model such as the LightGBM model, and the feature importance of each feature (e.g., including information gain and number of splits) is output, and features with zero information gain or zero feature importance are removed.

[0039] Second, each sample dataset is input into the first algorithm model to obtain the actual feature importance of each feature; the labels in the original sample dataset are randomly shuffled, and the shuffled sample dataset is input into the first algorithm model again to obtain the feature importance of each feature under the random labels; the above shuffled input operation is repeated N times (N is a positive integer) to obtain the N feature importances of each feature under the random labels; and the actual feature importance is compared with the N feature importances under the random labels, and feature removal is performed based on the comparison results.

[0040] Understandably, truly stable and important high-quality features become less important under random labels. Conversely, if a feature has low actual importance but its importance increases under random labels, then that feature is of poor quality and needs to be removed. In one example, invalid features can be removed based on statistical values ​​of feature importance. For instance, for each feature, calculate statistical values ​​of N feature importance, which can be the 75th percentile, mean, or other quantiles of the N feature importance; and if the difference between the actual feature importance and the statistical value is less than or equal to a first threshold (e.g., 0), then the feature is removed; otherwise, the feature is retained.

[0041] Furthermore, in the usability screening phase, a first algorithm model (e.g., LightGBM model, Random Forest model, XGBoost model) can be used to evaluate the usability of each of the multiple sample datasets, and the evaluation results can be used to determine whether the sample dataset can be used for subsequent modeling operations. For example, the training data of each sample dataset can be input into the first algorithm model for model training, and the test results can be obtained based on the trained model. The test results can be judged based on the AUC (Area Under Curve) metric. For example, if the AUC value is less than or equal to a second threshold (e.g., 0.5), the sample dataset can be used for subsequent modeling operations; otherwise, the sample dataset is deleted.

[0042] After invalid feature removal and availability screening, in order to more effectively obtain time-series information from vehicle historical data, time-series feature extraction should be performed on each sample dataset. Time-series feature extraction can be based on known time-series feature extraction methods in the art, including but not limited to steps such as time-series data integrity judgment, time-series information feature construction, and removal of irrelevant and redundant features. This invention does not specifically limit these steps.

[0043] Next, in step S130, the combined model is trained using each of the multiple sample datasets, and the fault prediction result is obtained based on the trained combined model. It should be understood that the sample datasets described in step S130 should include temporal feature data that has already undergone temporal feature extraction. The combined model includes at least two heterogeneous classification models, such as random forest models, LightGBM models, neural network (NN) models, K-nearest neighbor (KNN) models, etc.

[0044] In the model training phase of step S130, a first algorithm model (e.g., the LightGBM model) is trained using each of multiple (M, where M is a positive integer greater than 1) sample datasets to obtain M trained first algorithm models; and a second algorithm model (e.g., the Random Forest model) is trained using each of the M sample datasets to obtain M trained second algorithm models. Next, the M trained first algorithm models and M trained second algorithm models are tested using one of the M sample datasets (e.g., the first sample dataset, the second sample dataset, or the third sample dataset) to obtain the accuracy of each model. In the model ensemble phase, the prediction results of the M trained first algorithm models and the M trained second algorithm models are weighted by the accuracy of each model to obtain the fault prediction result under the combined model.

[0045] The method 10 of the present invention employs a combined perspective (e.g., extracting multiple sample datasets based on various sample selection strategies) to avoid bias caused by defining positive and negative samples from a single perspective, thereby enabling a more comprehensive and accurate extraction of fault symptom information from historical vehicle operation data. Furthermore, the method 10 of the present invention removes invalid features from the sample dataset based on, for example, variance filtering or feature importance, and further employs a combined model approach to improve model performance, thereby achieving a balance between overall computational cost and the model's fault prediction accuracy.

[0046] Figure 2 This is a block diagram of a computer system 20 for vehicle fault prediction according to an embodiment of the present invention. Figure 2 As shown, the computer system 20 includes a memory 210, a processor 220, and a computer program 230 stored in the memory 210 and executable on the processor 220. When the processor 220 executes the computer program 230, it can achieve the following: Figure 1 Method 10 is shown.

[0047] Furthermore, as described above, the present invention can also be implemented as a computer storage medium storing information for causing a computer to perform, such as Figure 1 The procedure of method 10 shown. Here, various types of computer storage media can be used as computer storage media, such as disks (e.g., magnetic disks, optical disks, etc.), cards (e.g., memory cards, optical cards, etc.), semiconductor memory (e.g., ROM, non-volatile memory, etc.), and tapes (e.g., magnetic tapes, cassette tapes, etc.).

[0048] Where applicable, the various embodiments provided by the present invention may be implemented using hardware, software, or a combination of hardware and software. Furthermore, where applicable, without departing from the scope of the invention, the various hardware and / or software components described herein may be combined into composite components comprising software, hardware, and / or both. Where applicable, without departing from the scope of the invention, the various hardware and / or software components described herein may be divided into sub-components comprising software, hardware, or both. Additionally, where applicable, it is contemplated that software components may be implemented as hardware components, and vice versa.

[0049] The software (such as program code and / or data) according to the invention can be stored on one or more computer storage media. It is also contemplated that the software identified herein can be implemented using one or more networked and / or otherwise general-purpose or special-purpose computers and / or computer systems. Where applicable, the order of the various steps described herein can be changed, combined into compound steps, and / or divided into sub-steps to provide the features described herein.

[0050] The embodiments and examples presented herein are provided to best illustrate embodiments of the invention and its particular applications, thereby enabling those skilled in the art to practice and use the invention. However, those skilled in the art will understand that the above description and examples are provided merely for ease of illustration and example. The descriptions presented are not intended to cover all aspects of the invention or to limit the invention to the precise forms disclosed.

Claims

1. A fault prediction method based on historical vehicle data, characterized in that, include: A1. Receive vehicle history data of faulty vehicle clusters and non-faulty vehicle clusters, wherein the vehicle history data includes historical data from at least one source within the vehicle during a first time period ending at the time of the fault. A. Multiple sample datasets are extracted from the vehicle historical data based on different sample selection strategies, including positive sample selection strategies and negative sample selection strategies. The positive sample selection strategy includes: extracting historical data from the vehicle history data of the faulty vehicle cluster during a second time period ending at the time of the fault, wherein the second time period is shorter than the first time period; The negative sample selection strategy includes: A subset of non-faulty vehicles is randomly selected from the cluster of non-faulty vehicles, and a first set of negative sample data is randomly extracted from the vehicle history data of the subset of non-faulty vehicles. Historical data from a third time period, ending at the start of the second time period, is extracted from the vehicle history data of the faulty vehicle cluster to serve as a second set of negative sample data, wherein the third time period is shorter than the first time period; and Select a subset of faulty vehicles from the faulty vehicle cluster whose cumulative running time is less than the minimum fault duration, and extract a third set of negative sample data from the vehicle history data of the faulty vehicle subset. B. Perform invalid feature removal and availability filtering for each of the multiple sample datasets; and C. Train the combined model using each of the multiple sample datasets, and obtain the fault prediction result based on the trained combined model.

2. The method according to claim 1, wherein, Step A includes: A2. Extract positive sample data from the vehicle historical data based on the positive sample selection strategy; A3. Extracting multiple sets of negative sample data from the vehicle historical data based on various negative sample selection strategies; and A4. Combine the positive sample data with each of the multiple sets of negative sample data to generate multiple sample datasets for fault prediction.

3. The method according to claim 1, wherein, Step B includes one of the following: Using variance filtering, non-divergent features in each sample dataset are removed. The sample dataset is input into the algorithm model to calculate the feature importance of each feature, and features are removed based on the magnitude of the feature importance.

4. The method according to claim 1, wherein, Step B includes: B1. For each sample dataset, calculate the variance of each feature in the sample dataset and remove features with zero variance from the sample dataset.

5. The method according to claim 1, wherein, Step B includes performing the following operations for each sample dataset: B2. Input the original sample dataset into the first algorithm model to obtain the actual feature importance of each feature; B3. Randomly shuffle the labels in the original sample dataset and input the shuffled sample dataset into the first algorithm model to obtain the feature importance of each feature under the random labels; B4. Repeat step B3 N times to obtain the N feature importance values ​​of each feature under random labels; as well as B5. Compare the actual feature importance with the N feature importances under the random label, and remove features based on the comparison results.

6. The method according to claim 5, wherein, Step B5 includes: For each feature, a statistical feature value of the importance of the N features is calculated, wherein the statistical feature value includes the 75th percentile of the importance of the N features; Calculate the difference between the actual feature importance and the statistical feature value; and If the difference is less than or equal to the first threshold, then the feature is removed.

7. The method according to claim 1, wherein, Step B further includes: The availability of each of the plurality of sample datasets is evaluated using the first algorithm model, and sample datasets with AUC values ​​less than or equal to the second threshold are filtered out.

8. The method according to claim 1, wherein, Step C includes: C1. Train the first algorithm model using each of the M sample datasets to obtain M trained first algorithm models. C2. Train the second algorithm model using each of the M sample datasets to obtain M trained second algorithm models. C3. Using one of the M sample datasets, test the M trained first algorithm models and the M trained second algorithm models to obtain the accuracy of each model; and C4. Using the accuracy of each model as a weight, perform a weighted average of the prediction results of the M trained first algorithm models and the M trained second algorithm models to obtain the fault prediction result under the combined model.

9. A computer system for predicting vehicle faults, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method of any one of claims 1 to 8.

10. A computer storage medium, characterized in that, The computer storage medium includes instructions that, when executed, perform the method according to any one of claims 1 to 8.

Images