An abnormal positioning method, device and equipment of a vehicle and a storage medium

Abstract

Embodiments of the present application provide an abnormal positioning method, device and equipment of a vehicle and a storage medium, the method comprising: determining target combination data corresponding to functional operation abnormality of the vehicle according to at least one type of operation data generated during functional operation of the vehicle, the target combination data comprising an operation problem category and feature data associated with each operation problem category; training a set machine learning model according to the target combination data to obtain a classification model; and positioning an operation abnormality object of the vehicle during functional operation abnormality according to the classification model. By using the method, the problem data of functional operation abnormality is determined through each operation data, and then the classification model is obtained by model training based on the problem data. According to the classification model, the operation abnormality object affecting the decision effect under different problems is determined, automatic positioning of the operation abnormality object is realized, the efficiency and accuracy of abnormal positioning are improved, and the labor cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of vehicle anomaly location technology, and in particular to a method, apparatus, device, and storage medium for vehicle anomaly location. Background Technology

[0002] The development of intelligent vehicles requires various operational processes, during which sensors and software algorithms collect, calculate, and generate a large amount of relevant data. The collected data is primarily used for software decision-making; after calculations are completed by different decision-making modules, decision results are generated. Both the raw data collected by sensors and the decision data generated by different software calculations play a crucial role in the overall intelligent driving decision-making process. Therefore, the ability to detect the quality of various data and determine the impact of different types of data on the overall intelligent driving system is a vital foundation for determining autonomous driving capabilities.

[0003] With the increasing scale and distributed nature of modern software, the incidence of defects in algorithm decision-making modules is rising. Current technologies primarily rely on manual detection and identification of relevant data based on historical experience. However, the large volume of data makes defect detection difficult, hindering rapid location and wasting human resources. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and storage medium for locating abnormal vehicle locations, enabling rapid and automatic location of abnormal objects, improving the efficiency and accuracy of abnormal location, and reducing labor costs.

[0005] Firstly, this embodiment provides a method for locating anomalies in a vehicle, the method comprising:

[0006] Based on at least one type of operational data generated during the functional operation of the vehicle, a target combination of data corresponding to the abnormal functional operation of the vehicle is determined. The target combination of data includes operational problem categories and feature data associated with each operational problem category.

[0007] The machine learning model is trained based on the target combination data to obtain a classification model;

[0008] Based on the classification model, locate the abnormal operation object of the vehicle when its functions are malfunctioning.

[0009] Secondly, this embodiment provides a vehicle anomaly location device, the device comprising:

[0010] The data determination module is used to determine the target combination data corresponding to the abnormal operation of the vehicle when the vehicle is in a functional operation, based on at least one type of operational data generated during the functional operation of the vehicle. The target combination data includes the operational problem category and the feature data associated with each operational problem category.

[0011] The model determination module is used to train a set machine learning model based on the target combination data to obtain a classification model;

[0012] An anomaly localization module is used to locate the abnormal operation object of the vehicle when its functions are abnormal, based on the classification model.

[0013] Thirdly, this embodiment provides an electronic device, the electronic device comprising:

[0014] At least one processor; and

[0015] A memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the vehicle anomaly localization method according to any embodiment of the present invention.

[0017] Fourthly, this embodiment provides a computer-readable storage medium, wherein the computer program is executed by the at least one processor to enable the at least one processor to perform the vehicle anomaly localization method according to any embodiment of the present invention.

[0018] This invention provides a method, apparatus, device, and storage medium for locating vehicle anomalies. The method includes: determining target combination data corresponding to a vehicle's functional malfunction based on at least one type of operational data generated during the vehicle's functional operation; the target combination data including operational problem categories and associated feature data for each operational problem category; training a set machine learning model based on the target combination data to obtain a classification model; and locating the abnormal operation object of the vehicle when its functional operation is abnormal based on the classification model. This technical solution determines the problem data of functional malfunction through various operational data, preprocesses the problem data, trains the model based on the processed data to obtain a classification model, and determines the abnormal operation object affecting decision-making under different problems based on the classification model. Compared to the prior art where anomaly location is performed manually, this technical solution achieves rapid and automatic location of abnormal operation objects based on operational data, improving the efficiency and accuracy of anomaly location and reducing labor costs.

[0019] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a method for locating anomalies in a vehicle according to Embodiment 1 of the present invention.

[0022] Figure 2 This is a flowchart illustrating a method for locating anomalies in a vehicle according to Embodiment 2 of the present invention.

[0023] Figure 3 This is a schematic diagram of the structure of a vehicle anomaly location device provided in Embodiment 3 of the present invention;

[0024] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "original," "target," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Example 1

[0028] Figure 1 This is a flowchart illustrating a method for abnormal vehicle location according to Embodiment 1 of the present invention. This method is applicable to abnormal vehicle location based on operational data. The method can be executed by an abnormal vehicle location device, which can be implemented in hardware and / or software and can be integrated into an electronic device.

[0029] like Figure 1 As shown, the vehicle anomaly localization method provided in this embodiment may specifically include the following steps:

[0030] S110. Based on at least one type of operational data generated during the functional operation of the vehicle, determine the target combination data corresponding to the abnormal functional operation of the vehicle.

[0031] In this embodiment, it is necessary to locate vehicle anomalies based on operational data. Therefore, when executing the vehicle anomaly location method provided in this embodiment, it is necessary to first obtain at least one type of operational data generated during the vehicle's functional operation. Operational data is generated during the vehicle's functional operation, and the vehicle needs to be operated functionally to obtain this data.

[0032] Before operating the vehicle's functions, it is necessary to ensure that the designated sensors are installed on the vehicle and can operate normally and collect data during vehicle operation. This is to ensure that specific problems can be identified through the data later. For example, sensors mainly include cameras, Global Positioning System (GPS), LiDAR, millimeter-wave radar, and Controller Area Network (CAN).

[0033] Before the vehicle can operate its functions, the internally developed algorithm modules (i.e., the computational decision-making modules) need to be compiled and deployed to the vehicle's intelligent decision-making module. For example, these algorithm modules may include perception, localization, fusion, planning, or control modules. After ensuring successful deployment and operation, it is also necessary to ensure that the data collected by the sensors can be transmitted to each algorithm module in real-time via data stream as a data source. Different algorithm modules calculate using their own algorithm models, output the computational decision results, and then associate the raw sensor data with the algorithm module's computational decision results and save them to the vehicle's storage device.

[0034] The vehicle is driven to the actual operating section for operation, and environmental data, as well as decision-making process and result data of the algorithm module, are collected at every moment during the operation. Any abnormal situations and problem data are also recorded.

[0035] During the operation of autonomous driving algorithms, sensors need to collect data on the surrounding environment, interactions with targets, lane information, and other data. Furthermore, it is crucial to ensure the proper functioning of the vehicle-side problem recording system. Recorders are required to document any issues encountered during operation (e.g., not driving in the center, getting too close to targets, driving erratically), along with the time of the occurrence.

[0036] After the vehicle's functions are operational, the raw sensor data collected by the vehicle, the decision data from different algorithm modules, and the problem data recorded during operation are all stored in a unified storage device. For example, the data is stored in a data center, and files are organized into a unified directory with accompanying descriptions. The stored data is then managed and accessed via a server for subsequent processing and calculations.

[0037] Specifically, based on the operational data generated from the vehicle's functional operation, at least one type of operational data is obtained. Based on each type of operational data, target combination data corresponding to the vehicle's abnormal functional operation is determined. The target combination data includes the operational problem category and the feature data associated with each operational problem category. The vehicle is located abnormally based on the target combination data.

[0038] In this embodiment, the operational data may include raw data acquired by sensor devices, decision data obtained through calculations by the algorithm decision module, and problem data corresponding to various abnormal situations. It is understood that abnormal vehicle operation may be caused by malfunctions in the sensor or other acquisition devices leading to abnormal data acquisition, which in turn leads to abnormal decision data and thus abnormal vehicle operation, or by problems with the algorithm decision module leading to abnormal decision data and thus abnormal vehicle operation.

[0039] Decision data refers to the data calculated by the algorithm decision module based on environmental data. For example, if the perception algorithm detects a stationary obstacle 20 meters ahead, the perception module identifies the type of object and informs the control module. The control module then performs driving control calculations based on the identified information, the relative position of the obstacle, and lane line information, such as needing to perform a lateral displacement of 1 meter or stop within 5 seconds.

[0040] To determine the cause of vehicle malfunctions, this embodiment relies on various operational data. Different types of abnormal information and phenomena are statistically categorized, and the operational data is processed. Statistical analysis is performed on the problems encountered during vehicle operation and the data that may be related to those problems.

[0041] First, related data within the operational data are correlated, such as raw data collection, decision data, and problem data. Raw data collection is automatically collected by data acquisition devices installed on the vehicle, such as sensors. Therefore, the system can automatically record relevant information about the raw data collection, such as the collection time, data type, and data value. Decision data is automatically calculated by algorithm modules placed on the vehicle. Therefore, the vehicle-side system can automatically record relevant information about the decision data, such as the generation time. Considering time correlation, timestamps can be used as identifiers to correlate raw data collection and decision data corresponding to the same time. Since different sensors have different collection frequencies, algorithm calculation frequencies, and problem recording times, direct correlation using timestamps is not feasible. Instead, frequency compression can be used to align the frequencies of each data point.

[0042] The problem data mainly includes problem data automatically recorded by the system and problem data manually recorded during vehicle operation. For problem data automatically recorded by the system, the system can automatically record relevant information such as time. Therefore, for this type of data, the original collected data and decision data can be linked together using timestamps as identifiers, and then the system-recorded problem data can be linked together.

[0043] For manually recorded problem data, there is often a lag in the recording process. Therefore, correlation can be performed based on approximate timestamps. For example, the original collected data with the smallest time difference from the problem data can be correlated with the decision data, and then correlated with the manually recorded problem data.

[0044] It's important to understand that the problem data automatically recorded by the system is standardized and requires no further processing. Manually recorded problem data, on the other hand, is written in text and lacks regularity. In this embodiment, manually recorded problems are processed to obtain categorized problem data. For example, the categorization process could be based on word segmentation.

[0045] Meanwhile, since the processed running data needs to be input into the model later, continuous running data needs to be discretized to obtain discrete data for input into the model.

[0046] For data that has undergone association and discretization processing, each data feature is input as a single-value feature into the model. When observing the final data, only the impact of a single data feature on the final classification result can be observed. However, during the algorithm's decision-making process, there may be combinations of two or more features that influence the result. Therefore, to explore and discover such phenomena, it is necessary to perform feature crossing on the independent features to construct cross features.

[0047] In this embodiment, cross-features, raw collected data, and decision data are all used as feature data associated with each operational problem category. Problem data automatically recorded by the system and manually recorded problem data after classification processing are used as operational problem categories. Operational problem categories and the feature data associated with each operational problem category are determined as target combination data.

[0048] S120. Train the set machine learning model based on the target combination data to obtain a classification model.

[0049] In this embodiment, after obtaining the target combination data through data association and preprocessing of all running data, a machine learning model is used to model the target combination data. Considering the possibility of outliers and missing values ​​in the data, a model that is insensitive to missing and outliers should be selected. This means the model should not be unable to compute due to missing values, nor should it suffer significant losses in stability and accuracy due to outlier values. Furthermore, this embodiment primarily focuses on the exploration and discovery of data features; therefore, the model needs to have high interpretability. Considering all factors, a random forest model is preferred in this embodiment. Specifically, the machine learning model is trained based on the target combination data to obtain a classification model.

[0050] For example, modeling can be performed based on the feature data associated with each operational problem category in the target combination data. The feature data associated with each operational problem category is input into the model, and the model is trained with the operational problem category as the target.

[0051] S130. Based on the classification model, locate the abnormal operation object when the vehicle's functions are malfunctioning.

[0052] The malfunctioning object could be a sensor data acquisition device or an algorithm decision-making module. In this embodiment, based on the trained model, the contribution of each feature data point can be determined, and the feature data related to the malfunction can be identified based on the contribution. The field names of the relevant feature data are used to determine which sensor device acquired the data or which algorithm module is involved. The involved sensor devices or algorithm modules are then identified as the malfunctioning objects when the vehicle malfunctions. The model is then used to calculate which specific data or algorithm decision-making modules affect the decision-making effect under different malfunction categories, ultimately accurately locating the problem through data analysis.

[0053] This invention provides a method for locating vehicle anomalies. The method includes: determining target combination data corresponding to a functional malfunction of the vehicle based on at least one type of operational data generated during the vehicle's functional operation; the target combination data including operational problem categories and associated feature data for each operational problem category; training a pre-defined machine learning model based on the target combination data to obtain a classification model; and locating the abnormal operational object of the vehicle when its functional operation is abnormal based on the classification model. This technical solution determines the problem data of functional malfunction through various operational data, preprocesses the problem data, trains the model based on the processed data to obtain a classification model, and determines the abnormal operational object affecting decision-making under different problems based on the classification model. Compared to the prior art where anomaly location is performed manually, this technical solution achieves rapid and automatic location of abnormal operational objects based on operational data, improving the efficiency and accuracy of anomaly location and reducing labor costs.

[0054] Example 2

[0055] Figure 2 This is a flowchart illustrating a method for locating vehicle anomalies according to Embodiment 2 of the present invention. This embodiment is a further optimization of the above embodiment. In this embodiment, the following are further optimized: "determining the target combination data corresponding to the vehicle when its function is abnormal based on at least one type of operation data generated during the vehicle's functional operation", "training a set machine learning model based on the target combination data to obtain a classification model", and "locating the abnormal operation object of the vehicle when its function is abnormal based on the target classification model".

[0056] like Figure 2 As shown in the figure, this embodiment 2 provides a method for locating abnormalities in a vehicle, which specifically includes the following steps:

[0057] S210. Based on the various operational data generated during the functional operation of the vehicle, determine the target associated data after associating the various operational data.

[0058] In this step, it is necessary to correlate the raw data collected by the sensors, the decision data calculated by the algorithm decision-making module, and the recorded question data. Since different sensors have different acquisition frequencies, algorithm calculation frequencies, and question recording times, direct correlation using timestamps is not feasible. Frequency alignment based on sensor acquisition and algorithm calculation frequencies is required, followed by correlation based on the recorded question time and the aligned data.

[0059] Preferably, the operational data includes: raw collected data, decision data, and problem data. Further, the step of determining the target associated data after associating the various operational data generated during the vehicle's functional operation can be optimized as follows:

[0060] a1. Compress and align the acquisition frequency of the original data and the calculation frequency of the decision data related to the original data in a way that aligns with the lowest frequency to obtain compressed original data and decision data.

[0061] In this embodiment, the raw collected data is first correlated with the decision data. Since the acquisition frequencies of different sensors and the calculation frequencies of algorithm modules vary, it is necessary to align the sensor acquisition frequencies with the algorithm calculation frequencies. Specifically, this is done by calculating the lowest frequency alignment for each sensor and its associated algorithm module. For example, if the lidar acquisition frequency is 20Hz and the related algorithm module calculation frequency is 10Hz, then the lidar acquisition frequency is reduced to 10Hz.

[0062] Specifically, frequency reduction can be achieved through data compression. The compressed data can be represented as follows:

[0063] HZ out =List(HZ) Zip ), Among them, HZ max This indicates the maximum frequency (Hz) between the sampling frequency and the calculation frequency. min This represents the minimum frequency between the sampling frequency and the calculation frequency, where n represents a positive integer.

[0064] For example: HZ lidar(a) =20Hz, Hz lidar(b) =5Hz, then the final output will be 5Hz, where the data HZ lidar(a)The data is compressed to 5Hz. The compression method involves summing the data from frames 1 to 4, dividing the sum by 4 to calculate the mean, and using this mean as the first output data point, which is then added to a list. This process is repeated for frames 5 to 8, and so on, until all results are calculated, resulting in a compressed dataset. This method is used to perform frequency compression on all relevant data, primarily including raw acquisition data and algorithm decision data.

[0065] b1. Associate the compressed raw collected data and decision data using timestamps to obtain the first associated data.

[0066] Specifically, the compressed raw collected data and decision data are correlated, and compressed data with the same timestamp are correlated. The correlated data is recorded as the first correlated data.

[0067] c1. Determine the target related data based on the first related data and the problem data.

[0068] In this step, the compressed raw collected data and the decision data are linked together to determine the first associated data and the problem data. The linked data is then identified as the target associated data. Here, the problem data includes both automatically registered abnormal data and manually recorded problem data. Similarly, the first associated data and the problem data can also be linked using timestamps.

[0069] Furthermore, based on the first set of related data and the problem data, the target related data is determined, including:

[0070] c11. Classify the problem data to obtain the running problem categories.

[0071] Specifically, the recording methods for acquiring problem data include system recording and manual recording. System-based problem data recording can be understood as the system automatically recording problems when preset judgment conditions are met, triggering the judgment. Manual problem data recording involves manually recording problems discovered during vehicle operation. For example, a manually recorded problem might be: "When a high-speed vehicle passes in front of this vehicle on its left, this vehicle experiences abnormal intermittent braking and sudden braking. This problem occurs frequently, affecting normal operation."

[0072] If the recording method is manual, the problem data is categorized to obtain the operational problem category. In this embodiment, if the problems discovered during operation are recorded manually, there will be issues with non-standard recording and time lag. Therefore, after collecting the problem data, all problem data needs to be uniformly categorized. The processing and categorization method here uses word segmentation matching in natural language to segment all recorded problems. The word segmentation method mainly uses publicly available tools, such as Jieba word segmentation and Han Language Processing (HanLP), etc., where the word segmentation modes include precise mode, full mode, search engine mode, etc. Preferably, precise mode is used, for example: "A high-speed vehicle passes the left front of this vehicle, and this vehicle abnormally brakes and brakes suddenly. This problem occurs frequently, affecting normal operation." This sentence is a specific recorded problem, and the word segmentation results are as follows: "High-speed / vehicle / passes / the / left front / of / this / vehicle / abnormal / brakes / , / sudden brakes / , / this / problem / frequently / occurs / , / affects / normal / operation". Regular expression matching was used to find the keywords " / point braking / " and " / emergency braking / ". These terms were then categorized as "abnormal braking". This method was used to classify all manually recorded problem data and obtain the operational problem categories.

[0073] If the recording method is system recording, then the problem data will be classified as an operational problem category. It is clear that system-recorded problem data can include relevant information such as the time the problem occurred and details of any anomalies discovered; these records are relatively standardized and therefore require minimal processing. Furthermore, the anomaly data automatically recorded by the vehicle system has already been automatically categorized and requires no further processing.

[0074] c12. Determine the target associated data based on the minimum time difference between the timestamp of the running problem category and the timestamp of the first associated data.

[0075] Specifically, the first set of associated data is linked to the running problem categories using timestamps. Because the data in the first set of associated data is compressed, some data cannot be linked using timestamps. Therefore, an alternative data association method, approximate timestamp association, will be used. The specific steps are as follows: Associate the record time of the problem data corresponding to the running problem category with the time of the compressed first set of associated data. If the data can be associated, the association is performed directly. If not, the time difference between the timestamp of the problem data and each timestamp of the compressed data is calculated. Then, the minimum calculated time difference is selected as the occurrence time of the associated problem data, and this data is marked. Simultaneously, a new column is added to map and match the marked data with the running problem categories; this matched data is determined as the target associated data.

[0076] S220. Based on the target association data and combined with the preset text-value mapping relationship, obtain numerical target association data.

[0077] If the categorical target data in the target association data is text-type data, it is necessary to perform text-to-numerical mapping on the text-type data, converting all text information into numerical values. For example, categorical target data, such as identifying lane line colors including white, yellow, blue, and orange, is mapped to the number 1 (white), 2 (yellow), 4 (blue), and 4 (orange) since the model cannot directly use text. All other text-type data is mapped to numerical values. The non-categorical target data used in this embodiment is already numerical in its data storage records, so no further conversion is needed. Ultimately, all target association data is converted into numerical target association data.

[0078] S230. Based on the numerical target association data and combined with the preset feature construction method, obtain the cross data after feature crossing and assign values ​​to the cross data.

[0079] In this model, each feature data input is a single-value feature, and only the impact of a single feature data on the final classification result can be observed during the final data observation. However, during the decision-making process of the algorithm module, there may be situations where two or more features are combined, affecting the result. Therefore, in order to explore and discover such phenomena, it is necessary to perform feature crossing on the independent feature data to construct cross features.

[0080] Specifically, feature cross-processing is performed on numerical target-related data using a preset feature construction method to obtain cross-data, and values ​​are assigned to the cross-data.

[0081] Furthermore, the steps of obtaining the cross data after crossover and assigning values ​​to the cross data based on the numerical target association data and the preset feature construction method are optimized as follows:

[0082] a2. Extract a set number of related feature data from numerical target association data, and discretize the feature data to obtain a discrete feature dataset.

[0083] Considering the large number of basic feature data in numerical target association data, the feature cross-computation complexity and efficiency should be taken into account when constructing features. The feature cross-computation dimension should not exceed three-dimensional feature cross-combination. Otherwise, the computational load will be too large, leading to incomputability or excessive time consumption. Furthermore, generating too many features can cause overfitting during model calculation due to the excessive number of features. Preferably, the number is set to 3.

[0084] Because of the large amount of continuous feature data, if the feature data were continuous values, there would be an excessive number of numerical values, which would be detrimental to model training. Therefore, it is necessary to discretize the features, such as through binning. In this step, the binning method mainly divides the data into different types based on the data range. For example, if the vehicle speed range is 0–130, then the speed is divided into 6 bins: 0–20, 20–40, 40–60, 60–80, 80–100, and above 100. Then, data within different numerical ranges are categorized into these bins and assigned bin identifiers. For example, 0–20 is bin 1, 20–40 is bin 2, and so on, to facilitate feature cross-construction.

[0085] b2. Construct features from the feature dataset by performing a feature AND operation to obtain cross data.

[0086] In this step, the main feature construction method is the feature AND construction method. For example, if it is found that there are non-motorized vehicles in the surrounding area and at the same time it is found that the vehicle is blocking the intersection with traffic lights and the traffic lights are green, then the new cross data is that there are non-motorized vehicles in the surrounding area, the vehicle is at the intersection, and the traffic lights are green.

[0087] c2. Assign a value of 1 to the cross data that meets the preset parallel conditions, and assign a value of zero to the cross data that does not meet the preset parallel conditions.

[0088] For example, the new intersection feature at this time is 1, which means there are non-motorized vehicles around and the intersection is at a green light. Other features that do not meet the above conditions are set to 0. Similarly, all relevant feature data are calculated.

[0089] S240. Numerical target correlation data and cross data are used as feature data, and the determined operational problem categories and the feature data associated with each operational problem category are used as the target combination data corresponding to the vehicle when its function is abnormal.

[0090] Specifically, numerical target-related data and cross-data are both used as feature data. Clearly, target-related data includes the associated raw collected data and decision data. Furthermore, the determined operational problem categories and the associated feature data for each operational problem category are used as the target combination data corresponding to the vehicle's abnormal functional operation.

[0091] S250. Construct a model for the target combination data using a random forest model.

[0092] Specifically, based on the correlated data, a classification prediction model can be built. Due to business needs, there are many categories of operational problems during operation; therefore, a multi-classification model is required to meet the model's prediction requirements. Data preprocessing is necessary before building the model.

[0093] The preprocessing here mainly includes data cleaning. Since we don't know what phenomena occurred during the actual data collection, calculation, and decision-making process, or how the data would behave, and manual imputation later could negatively impact the original data representation due to improper methods, we cannot use data imputation to fill in or repair missing data in each field. Instead, we must delete the missing data or process it automatically using machine learning capabilities. Furthermore, considering that this embodiment focuses on discovering defects in the algorithmic decision-making software, we do not impute or delete missing data. Abnormal data only needs to be assessed, not processed; otherwise, the data would not accurately reflect the problematic data.

[0094] The selection of a classification model requires comprehensive consideration of the impact of missing values, outliers, and model interpretability. For example, the K-Nearest Neighbor (KNN) algorithm is characterized by the influence of missing values ​​and outliers, and its interpretability. Decision trees also exhibit this characteristic. Random forests, on the other hand, have relatively low impact from missing values ​​and outliers, and their model is interpretable. In this embodiment, random forests are preferred as the classification model. Specifically, a random forest model is used to construct the model from the associated data.

[0095] In this embodiment, considering the possibility of outliers and missing values ​​in the target combination data, a model that is insensitive to missing and outliers must be selected. This means the model should not be unable to compute due to missing values, nor should it suffer significant losses in stability and accuracy due to outlier values. Furthermore, this embodiment primarily focuses on feature data exploration and discovery; therefore, the model needs to have high interpretability. Taking all factors into consideration, a random forest model is the primary choice.

[0096] S260. The model is trained by taking the feature data in the target combination data as input and the running problem category associated with the feature data as the target, and a classification model is obtained.

[0097] Specifically, after building a model from the associated data using a random forest model, the feature data from the target combination data is input into the model, and the problem categories in the target combination data are used as the model training targets.

[0098] S270. Based on the classification model, determine the contribution of each feature data to the classification model.

[0099] Specifically, based on the classification model, the contribution of each feature trained on the model is calculated.

[0100] S280. Based on the contribution level, locate the abnormal operation object when the vehicle's function is abnormal.

[0101] Specifically, based on the contribution of each feature data to the classification model, the abnormal objects in the vehicle's operation when the function is abnormal are located.

[0102] Furthermore, based on contribution level, the abnormal operation objects of the vehicle when its functions malfunction, including:

[0103] a3. Select the feature data whose sum of the number of consecutive numbers with the largest contribution is greater than a set threshold as the target feature data.

[0104] Specifically, after obtaining the contribution of each feature data, they are sorted from largest to smallest according to their contribution. For easier observation, features with a calculated contribution of 0 are removed.

[0105] Preferably, the threshold value can be set to 95%. Specifically, by accumulating the contribution values, the feature data with a total contribution value of the corresponding Top N exceeding 95% are taken as the target feature data.

[0106] b3. Determine the data acquisition device and algorithm decision module corresponding to the field names of the target feature data.

[0107] Specifically, determine the field names and corresponding values ​​of the target feature data, and determine the data acquisition devices and corresponding calculation and decision-making software corresponding to the field names.

[0108] c3. Data acquisition equipment and computing decision-making software are considered as objects of operational abnormality when the vehicle's functions malfunction.

[0109] Specifically, the identified data acquisition devices and computational decision-making software are considered as the objects of operational anomalies when the vehicle's functions malfunction. For example, the corresponding sensors used to acquire this data and the algorithmic decision-making modules, such as the control module, used to perform calculations on this data are determined based on the target feature data field names.

[0110] Furthermore, after determining that the sum of the contributions of the preceding consecutive data features is greater than a set threshold as the target feature data, the process also includes: constructing an interpreter for the classification model based on a set algorithm; and generating an interpretation information table and effect diagram through the interpreter to explain the contribution of each target data feature to each classification problem.

[0111] Considering that some of the identified features may be overlapping features, it is necessary to decompose the combined features to facilitate pinpointing the most fundamental problem in the data. Furthermore, some problems may not be clearly represented before combining individual features, but become more apparent after feature cross-combinations. Since the features calculated by random forests are all importance scores, it is impossible to know whether the feature's impact on the result is positive or negative, yet this result is relatively important for problem localization. In this case, the feature interpretation tool Shap is needed.

[0112] A random forest classification interpreter is built using Shap, primarily employing the Top N features selected through feature filtering, with the target problem as the training object. After training, the Shap model interpretation function generates a model interpretation information table and corresponding performance graphs. The positive and negative impacts of the performance graphs, as well as the degree of influence across different ranges, are used to interpret the impact of each feature on the target. A negative calculated feature value indicates that a smaller feature value makes problems or anomalies more likely to occur. A positive calculated feature value indicates that a larger feature value makes problems or anomalies more likely to occur.

[0113] This embodiment analyzes data collected by various sensors during operation and, based on data detection and judgment capabilities, accurately identifies abnormal sensor information and anomalous data collected by the sensors. It detects and identifies anomalous data based on the decision-making results calculated by various algorithm decision-making modules during autonomous driving operation and the effective range of data generated by different modules. It records anomalous information triggered during the autonomous driving operation verification process and anomalous phenomena discovered by operators. It statistically classifies various anomalous information and phenomena, models this data using machine learning models, and calculates which specific data or algorithm decision-making modules affect the decision-making effect under different problems, ultimately accurately locating the problem through data analysis.

[0114] Example 3

[0115] Figure 3 This is a schematic diagram of a vehicle anomaly location device provided in Embodiment 3 of the present invention. It is applicable to situations requiring anomaly location of a vehicle based on operational data. This device can be implemented in hardware and / or software and is generally integrated into an electronic device. For example... Figure 3 As shown, the device includes: a data determination module 31, a model determination module 32, and an anomaly location module 33, wherein...

[0116] Data determination module 31 is used to determine the target combination data corresponding to the abnormal operation of the vehicle when the vehicle is in a functional operation based on at least one type of operation data generated during the functional operation of the vehicle. The target combination data includes the operation problem category and the feature data associated with each operation problem category.

[0117] The model determination module 32 is used to train a set machine learning model based on the target combination data to obtain a classification model;

[0118] The anomaly localization module 33 is used to locate the abnormal object of the vehicle when the vehicle's function is abnormal, based on the classification model.

[0119] Optionally, the data determination module 32 may include:

[0120] The associated data determination unit is used to determine the target associated data after associating each set of operational data based on the operational data generated during the functional operation of the vehicle.

[0121] The numerical data determination unit is used to obtain numerical target-related data based on the target-related data and a preset text-numerical mapping relationship.

[0122] The cross-data acquisition unit is used to obtain cross-data after feature cross-data based on numerical target association data and a preset feature construction method, and to assign values ​​to the cross-data.

[0123] The combination determination unit is used to use numerical target correlation data and cross data as feature data, and to use the determined operation problem categories and the feature data associated with each operation problem category as the target combination data corresponding to the vehicle when the function is abnormal.

[0124] Optionally, the operational data includes: raw collected data, decision data, and problem data;

[0125] The associated data determination unit may include:

[0126] The first acquisition subunit is used to compress and align the acquisition frequency of the original acquisition data and the calculation frequency of the decision data related to the original acquisition data in a way that aligns with the lowest frequency, so as to obtain the compressed original acquisition data and decision data.

[0127] The second acquisition subunit is used to associate the compressed original collected data and decision data with timestamps as identifiers to obtain the first associated data;

[0128] The association determination subunit is used to determine the target association data based on the first association data and the problem data.

[0129] Optionally, the associated sub-unit is determined, specifically for:

[0130] The problem data is categorized to obtain the types of operational problems;

[0131] The target associated data is determined based on the minimum time difference between the timestamp of the running problem category and the timestamp of the first associated data.

[0132] Optionally, the cross-acquisition unit is specifically used for:

[0133] Extract a set number of related feature data from numerical target association data, and discretize the feature data to obtain a discrete feature dataset;

[0134] The feature dataset is constructed by performing a feature AND operation to obtain cross data;

[0135] Cross data that meets the preset parallel conditions will be assigned a value of 1, and cross data that does not meet the preset parallel conditions will be assigned a value of 0.

[0136] Optionally, the model determination module 32 can be used for:

[0137] The model building unit is used to build a model from the target combination data using a random forest model.

[0138] The classification model determination unit is used to train the model with the feature data in the target combination data as input and the running question category associated with the feature data as the target, so as to obtain the classification model.

[0139] Optionally, the anomaly location module 33 may include:

[0140] The contribution determination unit is used to determine the contribution of each feature data to the classification model based on the classification model.

[0141] The anomaly localization unit is used to locate the abnormal objects when the vehicle's functions malfunction, based on the contribution level.

[0142] Furthermore, the anomaly localization unit is specifically used for:

[0143] The feature data whose sum of the contributions of the first consecutive numbers of consecutive numbers with the largest contribution is greater than a set threshold are identified as target feature data.

[0144] Determine the data acquisition devices and algorithm decision modules corresponding to the field names of the target feature data;

[0145] The data acquisition equipment and algorithm decision-making module are considered as abnormal operation objects when the vehicle's functions malfunction.

[0146] Optionally, the device further includes an interpretation module, specifically used for:

[0147] An interpreter for a classification model built based on a defined algorithm;

[0148] The interpreter generates an explanation information table and a result diagram to explain the contribution of each target data feature to each classification problem.

[0149] The vehicle anomaly location device provided in this embodiment of the invention can execute the vehicle anomaly location method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0150] Example 4

[0151] Figure 4 This is a schematic diagram of an electronic device according to Embodiment 4 of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0152] like Figure 4 As shown, the electronic device 40 includes at least one processor 41 and a memory, such as a read-only memory (ROM) 42 or a random access memory (RAM) 43, communicatively connected to the at least one processor 41. The memory stores computer programs executable by the at least one processor. The processor 41 can perform various appropriate actions and processes based on the computer program stored in the ROM 42 or loaded into the RAM 43 from storage unit 48. The RAM 43 may also store various programs and data required for the operation of the electronic device 40. The processor 41, ROM 42, and RAM 43 are interconnected via a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0153] Multiple components in electronic device 40 are connected to I / O interface 45, including: input unit 46, such as keyboard, mouse, etc.; output unit 47, such as various types of monitors, speakers, etc.; storage unit 48, such as disk, optical disk, etc.; and communication unit 49, such as network card, modem, wireless transceiver, etc. Communication unit 49 allows electronic device 40 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0154] Processor 41 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 41 performs the various methods and processes described above, such as the anomaly localization method for vehicles.

[0155] In some embodiments, the vehicle anomaly localization method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 48. In some embodiments, part or all of the computer program may be loaded into and / or installed on electronic device 40 via ROM 42 and / or communication unit 49. When the computer program is loaded into RAM 43 and executed by processor 41, one or more steps of the vehicle anomaly localization method described above may be performed. Alternatively, in other embodiments, processor 41 may be configured to perform the vehicle anomaly localization method by any other suitable means (e.g., by means of firmware).

[0156] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0157] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0158] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0159] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0160] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0161] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0162] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0163] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for locating anomalies in a vehicle, characterized in that, include: Based on the operational data of at least one type of functional operation abnormality generated during the functional operation of the vehicle, the target combination data corresponding to the functional operation abnormality of the vehicle is determined. The target combination data includes the operational problem category and the feature data associated with each operational problem category. The operational data includes: raw collected data, decision data and problem data. The machine learning model is trained based on the target combination data corresponding to the vehicle's abnormal operation to obtain a classification model. The classification model is trained with the feature data in the target combination data as input and the operation problem category associated with the feature data as the target. Based on the classification model, the abnormal operation object of the vehicle when its functions are abnormal is located. The abnormal operation object includes data acquisition equipment and algorithm decision module. The step of determining the target combination data corresponding to the vehicle's abnormal operation when it is operating abnormally, based on at least one type of operational data generated during the vehicle's functional operation, includes: Based on the operational data generated during the functional operation of the vehicle, the target associated data after associating the operational data is determined. Based on the target association data and combined with the preset text-value mapping relationship, numerical target association data is obtained; Based on the numerical target association data, and combined with a preset feature construction method, cross data after feature crossing is obtained and the cross data is assigned a value. The preset feature construction method is a feature AND construction method. The numerical target association data and the cross data are used as feature data, and the determined operation problem categories and the feature data associated with each operation problem category are used as the target combination data corresponding to the vehicle when the function is abnormal. The step of determining the target associated data after associating the various operational data generated during the vehicle's functional operation includes: The acquisition frequency of the original acquisition data and the calculation frequency of the decision data related to the original acquisition data are compressed and aligned in a manner that aligns with the lowest frequency to obtain compressed original acquisition data and decision data. The compressed raw collected data and decision data are associated with each other using timestamps to obtain the first associated data; Based on the first associated data and the problem data, determine the target associated data; The step of determining the target associated data based on the first associated data and the problem data includes: The problem data is categorized to obtain the running problem categories; The target associated data is determined based on the minimum time difference between the timestamp of the running problem category and the timestamp of the first associated data.

2. The method according to claim 1, characterized in that, The step of obtaining cross data after feature crossing based on the numerical target association data and a preset feature construction method, and assigning values ​​to the cross data, includes: Extract a set number of related feature data from the numerical target association data, and discretize the feature data to obtain a discrete feature dataset; The feature dataset is used to construct features by performing a feature AND operation to obtain cross data; Cross data that meets the preset parallel conditions will be assigned a value of 1, and cross data that does not meet the preset parallel conditions will be assigned a value of 0.

3. The method according to claim 1, characterized in that, The step of training a set machine learning model based on the target combination data to obtain a classification model includes: The target combination data is modeled using a random forest model; The model is trained by taking the feature data in the target combination data as input and the running question category associated with the feature data as the target, and thus obtaining the classification model.

4. The method according to claim 1, characterized in that, The step of locating the abnormal operation object of the vehicle when its functions malfunction according to the classification model includes: Based on the classification model, determine the contribution of each feature data point to the classification model; Based on the contribution level, locate the abnormal operation object of the vehicle when its functions are malfunctioning.

5. The method according to claim 4, characterized in that, The step of locating the abnormal operation object of the vehicle when its functions malfunction, based on the contribution level, includes: The feature data whose sum of the contributions of the first consecutive numbers of consecutive numbers with the largest contribution is greater than a set threshold are identified as target feature data. Determine the data acquisition device and algorithm decision module corresponding to the field names of the target feature data; The data acquisition device and the algorithm decision module are used as objects of abnormal operation when the vehicle's functions are malfunctioning.

6. The method according to claim 5, characterized in that, After determining the data features whose sum of contributions is greater than a set threshold from largest to smallest as target feature data, the following is also included: An interpreter for the classification model is constructed based on the defined algorithm; The interpreter generates an explanation information table and an effect diagram to explain the contribution of each target data feature to each classification problem.

7. An abnormal vehicle location device, characterized in that, include: The data determination module is used to determine the target combination data corresponding to the vehicle when it has a functional operation abnormality, based on the operation data of at least one type of functional operation abnormality generated during the functional operation of the vehicle. The target combination data includes the operation problem category and the feature data associated with each operation problem category. The operation data includes: raw collected data, decision data and problem data. The model determination module is used to train a set machine learning model based on the target combination data corresponding to the vehicle's abnormal operation, and obtain a classification model. The classification model is obtained by training with the feature data in the target combination data as input and the operation problem category associated with the feature data as the target. An anomaly localization module is used to locate the abnormal operation object of the vehicle when its functions are abnormal, based on the classification model. The abnormal operation object includes a data acquisition device and an algorithm decision module. The data determination module includes: The associated data determination unit is used to determine the target associated data after associating each of the operation data generated during the functional operation of the vehicle. The numerical data determination unit is used to obtain numerical target-related data based on the target-related data and a preset text-value mapping relationship. The cross-data acquisition unit is used to obtain cross-data after feature cross-data based on the numerical target association data and in combination with a preset feature construction method, and to assign values ​​to the cross-data. The preset feature construction method is a feature AND construction method. The combination determination unit is used to take the numerical target association data and the cross data as feature data, and take the determined operation problem category and the feature data associated with each operation problem category as the target combination data corresponding to the vehicle when the function is abnormal. The associated data determination unit includes: The first obtaining subunit is used to compress and align the acquisition frequency of the original acquisition data and the calculation frequency of the decision data related to the original acquisition data in a way that aligns with the lowest frequency, so as to obtain the compressed original acquisition data and decision data. The second acquisition subunit is used to associate the compressed original collected data and decision data with timestamps as identifiers to obtain the first associated data; The association determination subunit is used to determine the target association data based on the first association data and the problem data; Specifically, the association determination subunit is used for: The problem data is categorized to obtain the running problem categories; The target associated data is determined based on the minimum time difference between the timestamp of the running problem category and the timestamp of the first associated data.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the abnormal location method for the vehicle according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the abnormal location method for the vehicle according to any one of claims 1-6.

Images