A vehicle fault data processing method and system based on artificial intelligence

Abstract

This application provides an artificial intelligence-based vehicle fault data processing method and system, belonging to the field of data processing technology. It can collect vehicle operating data and update feature values ​​in a real-time feature pool according to the definition of a feature registration area. When a feature value in the real-time feature pool is abnormal, an abnormal event corresponding to the abnormal feature value is generated and written to an abnormal queue. Temporal heterogeneous fusion feature vectors are extracted from the real-time feature pool and the feature registration area, and input into a vehicle performance prediction model to obtain vehicle performance degradation parameters. The vehicle performance degradation parameters, summary information of the abnormal queue, and environmental load stress degree are input into a vehicle fault diagnosis model to obtain vehicle fault diagnosis results. Based on the vehicle fault diagnosis results, a strategy mapping table is queried to obtain vehicle fault handling strategy information. The fault handling strategy information is filled into the strategy control area to obtain a strategy data package, which is then sent to the vehicle. This method can effectively handle vehicle faults.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a vehicle fault data processing method and system based on artificial intelligence. Background Technology

[0002] With the continuous development of Internet of Vehicles (IoT) and intelligent driving technologies, modern vehicles have become highly complex mobile intelligent terminals, with their internal sensors and controllers continuously generating massive amounts of multi-source heterogeneous time-series data. Deep analysis of this data based on artificial intelligence technology to achieve early prediction and intelligent handling of vehicle faults is of great significance for improving driving safety and reducing maintenance costs, and has become a research hotspot in the field of intelligent vehicles.

[0003] However, existing vehicle fault prediction technologies lack a unified and standardized management framework for multi-source data. Features are defined arbitrarily, and data quality varies widely, leading to inconsistent data at the input sources of the prediction model and resulting in inaccurate predictions. Secondly, most methods treat the prediction model as an isolated component, with its input consisting solely of instantaneous sensor data, which is relatively simplistic. This prevents the model from making reliable inferences under complex and ever-changing operating conditions, resulting in high false alarm and false negative rates. Consequently, they are unable to effectively handle vehicle faults. Summary of the Invention

[0004] The technical problem this application aims to solve is to provide a vehicle fault data processing method and system based on artificial intelligence, which can improve the accuracy of fault prediction results and effectively handle vehicle faults. The specific solution is as follows:

[0005] An artificial intelligence-based vehicle fault data processing method includes:

[0006] In response to a fault data processing instruction, a feature registration area, a real-time feature pool, an anomaly queue, and a policy control area for the vehicle corresponding to the fault data processing instruction are created.

[0007] According to the definition of the feature registration area, the vehicle's operating data is collected and the feature values ​​in the real-time feature pool are updated. When the feature values ​​in the real-time feature pool are abnormal, an abnormal event corresponding to the abnormal feature value is generated and the abnormal event is written into the abnormal queue.

[0008] Temporal heterogeneous fusion feature vectors are extracted from the real-time feature pool and the feature registration area, and input into the pre-trained vehicle performance prediction model to obtain the vehicle performance degradation parameter output by the prediction model. The vehicle performance degradation parameter represents the continuous value of the direction and rate of decline of the vehicle's health status in the future.

[0009] The vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree are input into a pre-trained vehicle fault diagnosis model to obtain the vehicle fault diagnosis result output by the vehicle fault diagnosis model; the environmental load stress degree represents the fault stress pressure of the environmental information of the vehicle.

[0010] Based on the vehicle fault diagnosis results, query the strategy mapping table to obtain the vehicle's fault handling strategy information;

[0011] The fault handling strategy information is filled into the strategy control area to obtain a strategy data packet;

[0012] The policy data packet is sent to the vehicle so that the vehicle's controller can execute the fault handling policy information in the policy data packet.

[0013] Optionally, the training process of the vehicle performance prediction model in the above method includes:

[0014] A training dataset is obtained, which contains multiple training samples. Each training sample includes data features of multi-source runtime sequence data of the vehicle within a historical time period and corresponding vehicle performance labels. The data features include time-series features, statistical features, and frequency domain features.

[0015] A bidirectional long short-term memory network model based on an attention mechanism is constructed as an initial vehicle performance prediction model. The initial vehicle performance prediction model includes an input layer, a bidirectional LSTM layer, an attention layer, and a fully connected output layer.

[0016] The initial vehicle performance prediction model is trained based on the training samples.

[0017] If the initial vehicle performance prediction model meets the first pre-set training completion condition, the initial vehicle performance prediction model that meets the first training completion condition will be used as the vehicle performance prediction model.

[0018] Optionally, in the above method, inputting the vehicle performance degradation parameters, the summary information of the anomaly queue, and the environmental load stress degree into the pre-trained vehicle fault diagnosis model includes:

[0019] The vehicle performance degradation parameters, summary information of the abnormal queue, and environmental load stress degree are standardized and feature-weightedly fused to generate a comprehensive diagnostic feature vector.

[0020] The comprehensive diagnostic feature vector is input into the vehicle fault diagnosis model, wherein the vehicle fault diagnosis model is a deep neural network model based on multi-task learning, which learns a deep representation of features through a shared feature extraction layer and generates diagnostic results including fault type classification results, fault severity level and root cause component location information through multiple parallel output branches.

[0021] Optionally, the above method involves standardizing and weighted fusing the vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree to generate a comprehensive diagnostic feature vector, including:

[0022] Obtain the length of the exception queue in the current period, the feature value identifiers of the exception events in the exception queue, and the average value of the feature values;

[0023] A summary of the anomaly queue is generated based on the length of the anomaly queue in the current period, the feature value identifiers of the anomaly events in the anomaly queue, and the average value of the feature values.

[0024] Obtain the vehicle's environmental information and current load;

[0025] The environmental stress degree is calculated based on the environmental information, and the dynamic load weighting factor is calculated based on the current load.

[0026] The environmental load stress degree of the vehicle is determined based on the environmental stress degree and the dynamic load weighting factor.

[0027] The vehicle performance degradation parameters, anomaly queue summary information, and environmental load stress degree were normalized using the min-max normalization method.

[0028] The normalized features are weighted and fused to generate the comprehensive diagnostic feature vector.

[0029] Optionally, in the above method, filling the fault handling strategy information into the strategy control area to obtain the strategy data packet includes:

[0030] The fault handling strategy information is associated and mapped with the preset control fields in the strategy control area;

[0031] Generate a serial number corresponding to the fault handling strategy information and bind it to the fault handling strategy information;

[0032] Based on the severity level of the vehicle fault diagnosis results, a priority flag for the vehicle fault diagnosis results is generated.

[0033] Generate an instruction effective timestamp and a transmission status flag. The instruction effective timestamp is used to control the vehicle-mounted actuator to activate the strategy within a specified time window. The transmission status flag is used to indicate the transmission status.

[0034] The control field, serial number, priority flag, instruction effective timestamp, transmission status flag, and vehicle fault diagnosis result associated with the mapping are serialized and encapsulated according to a predefined data packet format to obtain a data packet;

[0035] Add a checksum to the data packet to generate a policy data packet.

[0036] An artificial intelligence-based vehicle fault data processing system includes:

[0037] A creation unit is used to create a feature registration area, a real-time feature pool, an anomaly queue, and a policy control area for the vehicle corresponding to the fault data processing instruction in response to the fault data processing instruction.

[0038] The first execution unit is used to collect the vehicle's operating data and update the feature values ​​in the real-time feature pool according to the definition of the feature registration area. When the feature values ​​in the real-time feature pool are abnormal, an abnormal event corresponding to the abnormal feature value is generated and the abnormal event is written into the abnormal queue.

[0039] The second execution unit is used to extract temporal heterogeneous fusion feature vectors from the real-time feature pool and the feature registration area, input them into the pre-trained vehicle performance prediction model, and obtain the vehicle performance decay parameter output by the prediction model. The vehicle performance decay parameter represents the continuous value of the direction and rate of decline of the vehicle's health status in the future.

[0040] The third execution unit is used to input the vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree into the pre-trained vehicle fault diagnosis model to obtain the vehicle fault diagnosis result output by the vehicle fault diagnosis model; the environmental load stress degree represents the fault stress pressure of the environmental information of the vehicle.

[0041] The query unit is used to query the strategy mapping table based on the vehicle fault diagnosis results to obtain the vehicle fault handling strategy information.

[0042] A generation unit is used to fill the fault handling strategy information into the strategy control area to obtain a strategy data packet;

[0043] The sending unit is configured to send the policy data packet to the vehicle so that the vehicle's controller can execute the fault handling policy information in the policy data packet.

[0044] Optionally, in the above-described system, the second execution unit includes:

[0045] The acquisition sub-unit is used to acquire the training dataset, which contains multiple training samples. Each training sample includes data features of the multi-source runtime sequence data of the vehicle within a historical time period and the corresponding vehicle performance label. The data features include time-series features, statistical features, and frequency domain features.

[0046] A sub-unit is constructed to build a bidirectional long short-term memory network model based on an attention mechanism as an initial vehicle performance prediction model. The initial vehicle performance prediction model includes an input layer, a bidirectional LSTM layer, an attention layer, and a fully connected output layer.

[0047] A training subunit is used to train the initial vehicle performance prediction model based on the training samples.

[0048] The first execution subunit is configured to, when the initial vehicle performance prediction model satisfies the first training completion condition, use the initial vehicle performance prediction model that satisfies the first training completion condition as the vehicle performance prediction model.

[0049] Optionally, in the aforementioned system, the third execution unit includes:

[0050] The first generation subunit is used to standardize and perform feature weighted fusion on the vehicle performance degradation parameters, summary information of the abnormal queue, and environmental load stress degree to generate a comprehensive diagnostic feature vector.

[0051] An execution subunit is used to input the comprehensive diagnostic feature vector into the vehicle fault diagnosis model, wherein the vehicle fault diagnosis model is a deep neural network model based on multi-task learning, which learns a deep representation of features through a shared feature extraction layer and generates diagnostic results including fault type classification results, fault severity level and root cause component location information through multiple parallel output branches.

[0052] Optionally, in the above system, the first generation subunit includes:

[0053] The first acquisition module is used to acquire the length of the abnormal queue in the current period, the feature value identifiers of the abnormal events in the abnormal queue, and the average value of the feature values;

[0054] The first generation module is used to generate summary information of the abnormal queue based on the length of the abnormal queue in the current period, the feature value identifiers of the abnormal events in the abnormal queue, and the average value of the feature values.

[0055] The second acquisition module is used to acquire the environmental information and current load of the vehicle.

[0056] The calculation module is used to calculate the environmental stress degree based on the environmental information and to calculate the dynamic load weighting factor based on the current load.

[0057] A determination module is used to determine the environmental load stress degree of the vehicle based on the environmental stress degree and the dynamic load weighting factor.

[0058] The normalization module is used to normalize the vehicle performance degradation parameters, anomaly queue summary information and environmental load stress degree respectively using the min-max normalization method;

[0059] The second generation module is used to perform weighted fusion of the normalized features to generate the comprehensive diagnostic feature vector.

[0060] Optionally, in the above-described system, the generation unit includes:

[0061] The mapping subunit is used to associate and map the fault handling strategy information with the preset control fields in the strategy control area;

[0062] The second generation subunit is used to generate a serial number corresponding to the fault handling strategy information and bind it to the fault handling strategy information.

[0063] The third generation subunit is used to generate a priority flag for the vehicle fault diagnosis result based on the fault severity level of the vehicle fault diagnosis result.

[0064] The fourth generation subunit is used to generate an instruction effective timestamp and a transmission status flag. The instruction effective timestamp is used to control the vehicle-end actuator to activate the strategy within a specified time window, and the transmission status flag is used to indicate the transmission status.

[0065] The encapsulation subunit is used to serialize and encapsulate the associated mapping control field, the serial number, the priority flag, the instruction effective timestamp, the transmission status flag, and the vehicle fault diagnosis result according to a predefined data packet format to obtain a data packet;

[0066] The fifth generation subunit is used to add a checksum to the data packet and generate a policy data packet.

[0067] Based on the above, this application provides an artificial intelligence-based vehicle fault data processing method and system. In response to a fault data processing instruction, a feature registration area, a real-time feature pool, an anomaly queue, and a policy control area for the vehicle corresponding to the fault data processing instruction are created. According to the definition of the feature registration area, the vehicle's operating data is collected and the feature values ​​in the real-time feature pool are updated. When a feature value in the real-time feature pool is abnormal, an anomaly event corresponding to the abnormal feature value is generated and written to the anomaly queue. Temporal heterogeneous fusion feature vectors are extracted from the real-time feature pool and the feature registration area, and input into a pre-trained vehicle performance prediction model to obtain vehicle performance degradation parameters output by the prediction model. The vehicle performance degradation parameter represents the continuous value of the direction and rate of decline in the vehicle's health status over future time. The vehicle performance degradation parameter, the summary information of the abnormal queue, and the environmental load stress degree are input into a pre-trained vehicle fault diagnosis model to obtain the vehicle fault diagnosis result output by the model. The environmental load stress degree represents the fault stress pressure exerted on the vehicle by the environmental information it is in. The vehicle fault diagnosis result is used to query a strategy mapping table to obtain the vehicle's fault handling strategy information. The fault handling strategy information is filled into the strategy control area to obtain a strategy data packet. The strategy data packet is sent to the vehicle so that the vehicle's controller executes the fault handling strategy information in the strategy data packet. Applying the method provided in this application can improve the accuracy of fault prediction results and effectively handle vehicle faults. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0069] Figure 1 A flowchart of a vehicle fault data processing method based on artificial intelligence provided in this application;

[0070] Figure 2 A flowchart of a process for generating a comprehensive diagnostic feature vector is provided in this application;

[0071] Figure 3 A flowchart of a process for obtaining a policy data packet is provided in this application;

[0072] Figure 4 This application provides a schematic diagram of the structure of a vehicle fault data processing system based on artificial intelligence. Detailed Implementation

[0073] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0074] In this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0075] This invention provides an artificial intelligence-based vehicle fault data processing method, applied to electronic devices, which can be edge servers or cloud-based. The method flowchart is shown below. Figure 1 As shown, it specifically includes:

[0076] S101: In response to the fault data processing instruction, create the feature registration area, real-time feature pool, anomaly queue and policy control area of ​​the vehicle corresponding to the fault data processing instruction.

[0077] In this embodiment, the feature registration area is configured to store the metadata definition of vehicle monitoring features, including parameters such as feature identifier, data source address, sampling frequency and compliance threshold range, to standardize data collection and anomaly judgment criteria.

[0078] Optionally, the real-time feature pool serves as a dynamic storage area, maintaining the current values ​​and status flags of features acquired and parsed in real time from the vehicle bus and sensors in key-value format.

[0079] Optionally, the exception queue adopts a first-in-first-out buffer structure to sequentially record exception events triggered when the feature value exceeds the threshold. The event record includes the feature ID, timestamp, and exception value.

[0080] Optionally, the policy control area maintains the communication control fields required for issuing instructions, including serial number, priority flag, effective timestamp, and retransmission counter.

[0081] S102: According to the definition of the feature registration area, collect the vehicle's operating data and update the feature values ​​in the real-time feature pool. When the feature values ​​in the real-time feature pool are abnormal, generate an abnormal event corresponding to the abnormal feature value and write the abnormal event into the abnormal queue.

[0082] Specifically, vehicle operation data can be obtained from a specified data source according to the collection rules defined in the feature registration area. After parsing, the corresponding feature values ​​in the real-time feature pool are updated. Abnormal states are determined by comparing the feature values ​​with the threshold range of the registration area: if the feature value exceeds the threshold of the registration area, an abnormal event object containing feature ID, timestamp, and deviation value is constructed and inserted into the tail of the abnormal queue.

[0083] S103: Extract the temporal heterogeneous fusion feature vector from the real-time feature pool and the feature registration area, input it into the pre-trained vehicle performance prediction model, and obtain the vehicle performance degradation parameter output by the prediction model. The vehicle performance degradation parameter represents the continuous value of the direction and rate of decline of the vehicle's health status in the future.

[0084] In this embodiment, a time-series fusion processor extracts multi-source heterogeneous time-series data from the real-time feature pool and combines it with signal metadata in the feature registration area. Through cross-modal alignment, deep convolution feature extraction, and attention-weighted fusion, a time-series heterogeneous fusion feature vector is generated. This time-series heterogeneous fusion feature vector is input into a pre-trained vehicle performance prediction model, which outputs a vehicle performance degradation parameter. This vehicle performance degradation parameter is a continuous value that quantitatively represents the rate and direction of degradation of the system's health status within a specific future time period.

[0085] Optionally, signal metadata may include signal type and physical unit, etc.

[0086] In this embodiment, the vehicle performance prediction model may include a regression model based on a spatiotemporal Transformer architecture.

[0087] S104: Input the vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree into the pre-trained vehicle fault diagnosis model to obtain the vehicle fault diagnosis result output by the vehicle fault diagnosis model; the environmental load stress degree represents the fault stress pressure of the environmental information of the vehicle.

[0088] In this embodiment, a multi-dimensional diagnostic input vector is constructed, including: performance degradation parameters, summary information generated from anomaly queue statistics, and environmental load stress degree calculated by fusing ambient temperature, humidity, altitude, and road smoothness index and weighting by dynamic load. This input vector is input into a pre-trained multi-task fault diagnosis model, which outputs a structured diagnostic result that combines fault type classification, severity level assessment, and root cause component location.

[0089] Optionally, the summary information may include at least one of the following: anomaly frequency, latest anomaly feature type, and historical average anomaly value.

[0090] In this embodiment, the multi-task fault diagnosis model can be any type of machine learning model, such as a model that integrates graph neural networks and classifiers.

[0091] S105: Query the strategy mapping table based on the vehicle fault diagnosis results to obtain the vehicle fault handling strategy information.

[0092] In this embodiment, the diagnostic results are mapped to specific executable fault handling strategy instructions by querying a pre-configured strategy mapping table.

[0093] S106: Fill the fault handling strategy information into the strategy control area to obtain a strategy data packet;

[0094] In this embodiment, policy instructions and control fields in the policy control area are assembled according to a predetermined data packet format, and a checksum is added to generate a policy data packet.

[0095] S107: The policy data packet is sent to the vehicle so that the vehicle's controller can execute the fault handling policy information in the policy data packet.

[0096] In this embodiment, data packets are sent to the target vehicle via the vehicle-to-everything (V2X) communication protocol. The onboard controller then parses and executes the policy instructions, completing a closed-loop control process from cloud-based decision-making to vehicle-side execution.

[0097] The method provided in this application can improve the accuracy of fault prediction results and effectively handle vehicle faults.

[0098] In one embodiment provided in this application, based on the above implementation process, optionally, the training process of the vehicle performance prediction model includes:

[0099] A training dataset is obtained, which contains multiple training samples. Each training sample includes data features of multi-source runtime sequence data of the vehicle within a historical time period and corresponding vehicle performance labels. The data features include time-series features, statistical features, and frequency domain features.

[0100] A bidirectional long short-term memory network model based on an attention mechanism is constructed as an initial vehicle performance prediction model. The initial vehicle performance prediction model includes an input layer, a bidirectional LSTM layer, an attention layer, and a fully connected output layer.

[0101] The initial vehicle performance prediction model is trained based on the training samples.

[0102] If the initial vehicle performance prediction model meets the first pre-set training completion condition, the initial vehicle performance prediction model that meets the first training completion condition will be used as the vehicle performance prediction model.

[0103] In this embodiment, a training dataset can be obtained first. This dataset contains multi-source time-series data samples generated during vehicle operation in multiple historical time periods and their corresponding vehicle performance labels.

[0104] Among them, data features cover time-series features such as time series change patterns of sensor data, statistical features, and frequency domain features. Statistical features may include mean, variance, and extreme values, while frequency domain features may include frequency domain feature parameters obtained by Fourier transform.

[0105] Optionally, a bidirectional long short-term memory network based on an attention mechanism can be constructed as the initial prediction model. This initial prediction model includes: an input layer for receiving multidimensional feature data; a bidirectional LSTM layer for capturing long-term dependencies in the time series in both forward and backward directions; an attention layer for automatically weighting and highlighting the importance of features at key time steps; and a fully connected output layer for finally generating the predicted value.

[0106] During training, the initial model is iteratively optimized using training samples. Forward propagation is used to calculate prediction results, the mean squared error loss function is used to measure the deviation between the predicted values ​​and the true labels, and the backpropagation algorithm is used to adjust the model parameters. Optionally, the adaptive moment estimation (Adam) optimizer is used for parameter updates, and gradient clipping is introduced to prevent training divergence.

[0107] In this embodiment, when the model's performance on the validation set reaches a preset first training completion condition, such as when the loss function no longer decreases for several consecutive training epochs, or when the preset maximum number of training epochs is reached, the training process is terminated. The model parameters obtained at this point are then solidified into the final deployable vehicle performance prediction model. The specific threshold for this training completion condition can be determined through cross-validation or empirical values.

[0108] The model trained using the method provided in this application can effectively capture the spatiotemporal dependencies in vehicle operation and give greater attention to key fault symptom features, thereby improving the accuracy and reliability of predicting future performance degradation trends. The entire training process can be automated end-to-end without manual intervention in feature engineering, significantly improving model development efficiency.

[0109] In one embodiment provided in this application, based on the above implementation process, optionally, the step of inputting the vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree into the pre-trained vehicle fault diagnosis model includes:

[0110] The vehicle performance degradation parameters, summary information of the abnormal queue, and environmental load stress degree are standardized and feature-weightedly fused to generate a comprehensive diagnostic feature vector.

[0111] The comprehensive diagnostic feature vector is input into the vehicle fault diagnosis model, wherein the vehicle fault diagnosis model is a deep neural network model based on multi-task learning, which learns a deep representation of features through a shared feature extraction layer and generates diagnostic results including fault type classification results, fault severity level and root cause component location information through multiple parallel output branches.

[0112] In this embodiment, the vehicle performance degradation parameter can be set first. Summary information of the exception queue and environmental load stress Three types of heterogeneous input data underwent standardized preprocessing. Specifically, the performance degradation parameter was standardized using Z-score to eliminate dimensional differences; the anomaly queue summary information was normalized using Min-Max; and the environmental load stress degree, being a dimensionless indicator, was directly involved in subsequent fusion processing.

[0113] Optionally, in the feature weighting fusion stage, the three types of standardized features are dynamically weighted using a learnable attention weight matrix. Specifically, the calculation formula can be expressed as:

[0114]

[0115] Where W is the weight matrix and b is the bias term. This represents a vector concatenation operation. This indicates the vehicle performance degradation parameter after preprocessing. This represents the summary information of the preprocessed exception queue. This represents the preprocessed environmental load stress degree. The weighted eigenvector. After dimensionality reduction by a fully connected layer, a comprehensive diagnostic feature vector with uniform dimensions is generated.

[0116] In one embodiment provided in this application, based on the above implementation process, optionally, the process of standardizing and weighted fusing the vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree to generate a comprehensive diagnostic feature vector, such as... Figure 2 As shown, it includes:

[0117] S201: Obtain the length of the exception queue in the current period, the feature value identifiers of the exception events in the exception queue, and the average value of the feature values.

[0118] In this embodiment, the abnormal queue is traversed from the head to the tail, and the number of abnormal events accumulated in the current sampling period is counted as the queue length. At the same time, the set of feature identifiers in all abnormal records is extracted, and the arithmetic mean of the abnormal values ​​of each feature is calculated.

[0119] For example, the set of feature identifiers may include {"battery temperature", "engine vibration", "oil pressure"}, etc.

[0120] S202: Generate summary information of the abnormal queue based on the length of the abnormal queue in the current period, the feature value identifiers of the abnormal events in the abnormal queue, and the average value of the feature values.

[0121] In this embodiment, statistical information can be combined to generate structured summary information. Specifically, the generated JSON format data packet contains three fields: the queue_length field records the queue length, the feature_ids field stores the list of feature identifiers, and the average_values ​​field stores the average values ​​of anomalies corresponding to each feature.

[0122] S203: Obtain the environmental information and current load of the vehicle.

[0123] Optionally, environmental information, including temperature, humidity, altitude, and road vibration data collected by the IMU, can be acquired through an onboard sensor cluster; and real-time load data, including cargo weight, can be acquired through the vehicle body controller's CAN bus.

[0124] S204: Calculate the environmental stress degree based on the environmental information, and calculate the dynamic load weighting factor based on the current load.

[0125] In this embodiment, the environmental stress degree is calculated through four levels of quantification: first, the temperature, humidity, altitude, and vibration data are normalized to the [0,1] interval, then the preliminary value is obtained by weighted summation according to empirical weights (such as 0.3, 0.2, 0.2, 0.3), and finally multiplied by the road condition coefficient (1 can be set for smooth roads and 1.5 for bumpy roads).

[0126] Optionally, the dynamic load weighting factor is calculated using a piecewise function: 1.0 for no load, 1.3 for half load, 1.8 for heavy load, and 2.5 for overload.

[0127] S205: Determine the environmental load stress degree of the vehicle based on the environmental stress degree and the dynamic load weighting factor.

[0128] In this embodiment, the environmental stress degree can be multiplied by the load weighting factor to obtain the final environmental load stress degree. For example: environmental stress degree 0.7 × heavy load factor 1.8 = 1.26.

[0129] S206: The vehicle performance degradation parameters, abnormal queue summary information and environmental load stress degree are normalized by using the min-max normalization method.

[0130] In this embodiment, min-max normalization can be used to linearly map the vehicle performance degradation parameter to the [0,10] interval, normalize each numerical field in the anomaly queue summary information, and keep the environmental load stress degree at its original value.

[0131] S207: Weighted fusion of the normalized features to generate the comprehensive diagnostic feature vector.

[0132] In this embodiment, an attention weight matrix can be used to weight and concatenate the normalized three types of features to form a diagnostic feature vector. Optionally, the weight values ​​in the attention weight matrix are obtained through training with historical fault data, such as a performance parameter weight of 0.5, an anomaly queue weight of 0.3, and an environmental load weight of 0.2.

[0133] In one embodiment provided in this application, based on the above implementation process, optionally, the process of filling the fault handling strategy information into the strategy control area to obtain the strategy data packet is as follows: Figure 3 As shown, it includes:

[0134] S301: Associate and map the fault handling strategy information with the preset control fields in the strategy control area.

[0135] In this embodiment, a mapping relationship is established between the fault handling strategy information and the preset control fields in the strategy control area.

[0136] Optionally, the operation type in the policy instruction is extracted and mapped to the `command_type` field of the control field. The operation type can include speed limit, power reduction, or emergency stop, etc. Policy parameters are parsed and filled into the `command_parameters` field. Policy parameters can include specific parameters for the operation type, such as a speed limit of 80 km / h. Next, the target vehicle's VIN code is obtained and filled into the `target_vehicle_id` field.

[0137] S302: Generate a serial number corresponding to the fault handling strategy information and bind it to the fault handling strategy information.

[0138] In this embodiment, a distributed serial number generator is used to generate a unique instruction identifier. Specifically, the current millisecond-level timestamp is taken as the lower 32 bits, combined with the server node number as the middle 16 bits, and a sequence counter is appended as the higher 16 bits. These are then concatenated to generate a 64-bit serial number, which is then bound and stored with the fault handling strategy information.

[0139] S303: Generate a priority flag for the vehicle fault diagnosis results based on the fault severity level of the vehicle fault diagnosis results.

[0140] In this embodiment, a priority flag can be generated based on the severity level of the fault, for example:

[0141] Level 1 fault (critical): Priority flag is set to 0xFF (highest).

[0142] Level 2 fault (critical): Priority flag set to 0xCF.

[0143] Level 3 fault (general): Priority flag set to 0x7F.

[0144] Preventative maintenance: Priority flag set to 0x3F.

[0145] S304: Generate an instruction effective timestamp and a transmission status flag. The instruction effective timestamp is used to control the vehicle-side actuator to activate the strategy within a specified time window, and the transmission status flag is used to indicate the transmission status.

[0146] In this embodiment, the instruction take-off timestamp adopts a preset time format, and the calculation rule is: current time + level delay. Optionally, a level 1 fault takes effect immediately, a level 2 fault is delayed by a first preset duration, and a level 3 fault is delayed by a second preset duration; the second preset duration is longer than the first preset duration.

[0147] In this embodiment, the transmission status flag is initialized to 0x01, indicating that it is to be sent. It is subsequently changed to 0x02 (sent) / 0x04 (acknowledged) / 0x08 (failed) depending on the transmission status.

[0148] S305: The associated mapping control field, serial number, priority flag, instruction effective timestamp, transmission status flag, and vehicle fault diagnosis result are serialized and encapsulated according to a predefined data packet format to obtain a data packet.

[0149] In this embodiment, the encapsulation format includes a header identifier field, a protocol version field, a control segment field, and a data segment field, wherein the data segment uses a type-length-value (TLV) encoding format to organize the data.

[0150] S306: Add a checksum to the data packet to generate a policy data packet.

[0151] In this embodiment, a checksum is added to the serialized data packet to generate the final policy data packet. Specifically, a cyclic redundancy check algorithm can be used to calculate the data packet checksum using a predefined polynomial, and the checksum is appended to the end of the data packet to complete the data packet construction.

[0152] See Figure 4 The diagram below illustrates the structure of an artificial intelligence-based vehicle fault data processing system according to an embodiment of this application. The system includes:

[0153] The creation unit 401 is used to create a feature registration area, a real-time feature pool, an anomaly queue, and a policy control area for the vehicle corresponding to the fault data processing instruction in response to the fault data processing instruction.

[0154] The first execution unit 402 is used to collect the vehicle's operating data and update the feature values ​​in the real-time feature pool according to the definition of the feature registration area. When the feature values ​​in the real-time feature pool are abnormal, an abnormal event corresponding to the abnormal feature value is generated and the abnormal event is written into the abnormal queue.

[0155] The second execution unit 403 is used to extract temporal heterogeneous fusion feature vectors from the real-time feature pool and the feature registration area, input them into the pre-trained vehicle performance prediction model, and obtain the vehicle performance decay parameter output by the prediction model. The vehicle performance decay parameter represents the continuous value of the direction and rate of decline of the vehicle's health status in the future.

[0156] The third execution unit 404 is used to input the vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree into a pre-trained vehicle fault diagnosis model to obtain the vehicle fault diagnosis result output by the vehicle fault diagnosis model; the environmental load stress degree represents the fault stress pressure of the environmental information of the vehicle.

[0157] The query unit 405 is used to query the strategy mapping table based on the vehicle fault diagnosis results to obtain the vehicle fault handling strategy information.

[0158] The generation unit 406 is used to fill the fault handling strategy information into the strategy control area to obtain a strategy data packet;

[0159] The sending unit 407 is used to send the policy data packet to the vehicle so that the vehicle's controller can execute the fault handling policy information in the policy data packet.

[0160] In one embodiment provided in this application, based on the above-described solution, optionally, the second execution unit 403 includes:

[0161] The acquisition sub-unit is used to acquire the training dataset, which contains multiple training samples. Each training sample includes data features of the multi-source runtime sequence data of the vehicle within a historical time period and the corresponding vehicle performance label. The data features include time-series features, statistical features, and frequency domain features.

[0162] A sub-unit is constructed to build a bidirectional long short-term memory network model based on an attention mechanism as an initial vehicle performance prediction model. The initial vehicle performance prediction model includes an input layer, a bidirectional LSTM layer, an attention layer, and a fully connected output layer.

[0163] A training subunit is used to train the initial vehicle performance prediction model based on the training samples.

[0164] The first execution subunit is configured to, when the initial vehicle performance prediction model satisfies the first training completion condition, use the initial vehicle performance prediction model that satisfies the first training completion condition as the vehicle performance prediction model.

[0165] In one embodiment provided in this application, based on the above-described solution, optionally, the third execution unit 404 includes:

[0166] The first generation subunit is used to standardize and perform feature weighted fusion on the vehicle performance degradation parameters, summary information of the abnormal queue, and environmental load stress degree to generate a comprehensive diagnostic feature vector.

[0167] An execution subunit is used to input the comprehensive diagnostic feature vector into the vehicle fault diagnosis model, wherein the vehicle fault diagnosis model is a deep neural network model based on multi-task learning, which learns a deep representation of features through a shared feature extraction layer and generates diagnostic results including fault type classification results, fault severity level and root cause component location information through multiple parallel output branches.

[0168] In one embodiment provided in this application, based on the above-described solution, optionally, the first generation subunit includes:

[0169] The first acquisition module is used to acquire the length of the abnormal queue in the current period, the feature value identifiers of the abnormal events in the abnormal queue, and the average value of the feature values;

[0170] The first generation module is used to generate summary information of the abnormal queue based on the length of the abnormal queue in the current period, the feature value identifiers of the abnormal events in the abnormal queue, and the average value of the feature values.

[0171] The second acquisition module is used to acquire the environmental information and current load of the vehicle.

[0172] The calculation module is used to calculate the environmental stress degree based on the environmental information and to calculate the dynamic load weighting factor based on the current load.

[0173] A determination module is used to determine the environmental load stress degree of the vehicle based on the environmental stress degree and the dynamic load weighting factor.

[0174] The normalization module is used to normalize the vehicle performance degradation parameters, anomaly queue summary information and environmental load stress degree respectively using the min-max normalization method;

[0175] The second generation module is used to perform weighted fusion of the normalized features to generate the comprehensive diagnostic feature vector.

[0176] In one embodiment provided in this application, based on the above-described solution, optionally, the generation unit 406 includes:

[0177] The mapping subunit is used to associate and map the fault handling strategy information with the preset control fields in the strategy control area;

[0178] The second generation subunit is used to generate a serial number corresponding to the fault handling strategy information and bind it to the fault handling strategy information.

[0179] The third generation subunit is used to generate a priority flag for the vehicle fault diagnosis result based on the fault severity level of the vehicle fault diagnosis result.

[0180] The fourth generation subunit is used to generate an instruction effective timestamp and a transmission status flag. The instruction effective timestamp is used to control the vehicle-end actuator to activate the strategy within a specified time window, and the transmission status flag is used to indicate the transmission status.

[0181] The encapsulation subunit is used to serialize and encapsulate the associated mapping control field, the serial number, the priority flag, the instruction effective timestamp, the transmission status flag, and the vehicle fault diagnosis result according to a predefined data packet format to obtain a data packet;

[0182] The fifth generation subunit is used to add a checksum to the data packet and generate a policy data packet.

[0183] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0184] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0185] For ease of description, the above system is described by dividing it into various functional units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0186] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0187] The above provides a detailed description of an artificial intelligence-based vehicle fault data processing method provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A vehicle fault data processing method based on artificial intelligence, characterized in that, include: In response to a fault data processing instruction, a feature registration area, a real-time feature pool, an anomaly queue, and a policy control area for the vehicle corresponding to the fault data processing instruction are created. According to the definition of the feature registration area, the vehicle's operating data is collected and the feature values ​​in the real-time feature pool are updated. When the feature values ​​in the real-time feature pool are abnormal, an abnormal event corresponding to the abnormal feature value is generated and the abnormal event is written into the abnormal queue. Temporal heterogeneous fusion feature vectors are extracted from the real-time feature pool and the feature registration area, and input into the pre-trained vehicle performance prediction model to obtain the vehicle performance degradation parameter output by the prediction model. The vehicle performance degradation parameter represents the continuous value of the direction and rate of decline of the vehicle's health status in the future. The vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree are input into a pre-trained vehicle fault diagnosis model to obtain the vehicle fault diagnosis result output by the vehicle fault diagnosis model; the environmental load stress degree represents the fault stress pressure of the environmental information of the vehicle. Based on the vehicle fault diagnosis results, query the strategy mapping table to obtain the vehicle's fault handling strategy information; The fault handling strategy information is filled into the strategy control area to obtain a strategy data packet; The policy data packet is sent to the vehicle so that the vehicle's controller can execute the fault handling policy information in the policy data packet; The step of filling the fault handling strategy information into the strategy control area to obtain a strategy data packet includes: The fault handling strategy information is associated and mapped with the preset control fields in the strategy control area; Generate a serial number corresponding to the fault handling strategy information and bind it to the fault handling strategy information; Based on the severity level of the vehicle fault diagnosis results, a priority flag for the vehicle fault diagnosis results is generated. Generate an instruction effective timestamp and a transmission status flag. The instruction effective timestamp is used to control the vehicle-mounted actuator to activate the strategy within a specified time window. The transmission status flag is used to indicate the transmission status. The control field, serial number, priority flag, instruction effective timestamp, transmission status flag, and vehicle fault diagnosis result associated with the mapping are serialized and encapsulated according to a predefined data packet format to obtain a data packet; Add a checksum to the data packet to generate a policy data packet.

2. The method according to claim 1, characterized in that, The training process of the vehicle performance prediction model includes: A training dataset is obtained, which contains multiple training samples. Each training sample includes data features of multi-source runtime sequence data of the vehicle within a historical time period and corresponding vehicle performance labels. The data features include time-series features, statistical features, and frequency domain features. A bidirectional long short-term memory network model based on an attention mechanism is constructed as an initial vehicle performance prediction model. The initial vehicle performance prediction model includes an input layer, a bidirectional LSTM layer, an attention layer, and a fully connected output layer. The initial vehicle performance prediction model is trained based on the training samples. If the initial vehicle performance prediction model meets the first pre-set training completion condition, the initial vehicle performance prediction model that meets the first training completion condition will be used as the vehicle performance prediction model.

3. The method according to claim 1, characterized in that, The step of inputting the vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree into the pre-trained vehicle fault diagnosis model includes: The vehicle performance degradation parameters, summary information of the abnormal queue, and environmental load stress degree are standardized and feature-weightedly fused to generate a comprehensive diagnostic feature vector. The comprehensive diagnostic feature vector is input into the vehicle fault diagnosis model, wherein the vehicle fault diagnosis model is a deep neural network model based on multi-task learning, which learns a deep representation of features through a shared feature extraction layer and generates diagnostic results including fault type classification results, fault severity level and root cause component location information through multiple parallel output branches.

4. The method according to claim 3, characterized in that, The standardization and feature-weighted fusion of the vehicle performance degradation parameters, anomaly queue summary information, and environmental load stress degree generate a comprehensive diagnostic feature vector, including: Obtain the length of the exception queue in the current period, the feature value identifiers of the exception events in the exception queue, and the average value of the feature values; A summary of the anomaly queue is generated based on the length of the anomaly queue in the current period, the feature value identifiers of the anomaly events in the anomaly queue, and the average value of the feature values. Obtain the vehicle's environmental information and current load; The environmental stress degree is calculated based on the environmental information, and the dynamic load weighting factor is calculated based on the current load. The environmental load stress degree of the vehicle is determined based on the environmental stress degree and the dynamic load weighting factor. The vehicle performance degradation parameters, anomaly queue summary information, and environmental load stress degree were normalized using the min-max normalization method. The normalized features are weighted and fused to generate the comprehensive diagnostic feature vector.

5. A vehicle fault data processing system based on artificial intelligence, characterized in that, include: A creation unit is used to create a feature registration area, a real-time feature pool, an anomaly queue, and a policy control area for the vehicle corresponding to the fault data processing instruction in response to the fault data processing instruction. The first execution unit is used to collect the vehicle's operating data and update the feature values ​​in the real-time feature pool according to the definition of the feature registration area. When the feature values ​​in the real-time feature pool are abnormal, an abnormal event corresponding to the abnormal feature value is generated and the abnormal event is written into the abnormal queue. The second execution unit is used to extract temporal heterogeneous fusion feature vectors from the real-time feature pool and the feature registration area, input them into the pre-trained vehicle performance prediction model, and obtain the vehicle performance decay parameter output by the prediction model. The vehicle performance decay parameter represents the continuous value of the direction and rate of decline of the vehicle's health status in the future. The third execution unit is used to input the vehicle performance degradation parameters, the summary information of the abnormal queue, and the environmental load stress degree into the pre-trained vehicle fault diagnosis model to obtain the vehicle fault diagnosis result output by the vehicle fault diagnosis model; the environmental load stress degree represents the fault stress pressure of the environmental information of the vehicle. The query unit is used to query the strategy mapping table based on the vehicle fault diagnosis results to obtain the vehicle fault handling strategy information. A generation unit is used to fill the fault handling strategy information into the strategy control area to obtain a strategy data packet; A sending unit is configured to send the policy data packet to the vehicle so that the vehicle's controller can execute the fault handling policy information in the policy data packet. The generation unit includes: The mapping subunit is used to associate and map the fault handling strategy information with the preset control fields in the strategy control area; The second generation subunit is used to generate a serial number corresponding to the fault handling strategy information and bind it to the fault handling strategy information. The third generation subunit is used to generate a priority flag for the vehicle fault diagnosis result based on the fault severity level of the vehicle fault diagnosis result. The fourth generation subunit is used to generate an instruction effective timestamp and a transmission status flag. The instruction effective timestamp is used to control the vehicle-end actuator to activate the strategy within a specified time window, and the transmission status flag is used to indicate the transmission status. The encapsulation subunit is used to serialize and encapsulate the associated mapping control field, the serial number, the priority flag, the instruction effective timestamp, the transmission status flag, and the vehicle fault diagnosis result according to a predefined data packet format to obtain a data packet; The fifth generation subunit is used to add a checksum to the data packet and generate a policy data packet.

6. The system according to claim 5, characterized in that, The second execution unit includes: The acquisition sub-unit is used to acquire the training dataset, which contains multiple training samples. Each training sample includes data features of the multi-source runtime sequence data of the vehicle within a historical time period and the corresponding vehicle performance label. The data features include time-series features, statistical features, and frequency domain features. A sub-unit is constructed to build a bidirectional long short-term memory network model based on an attention mechanism as an initial vehicle performance prediction model. The initial vehicle performance prediction model includes an input layer, a bidirectional LSTM layer, an attention layer, and a fully connected output layer. A training subunit is used to train the initial vehicle performance prediction model based on the training samples. The first execution subunit is configured to, when the initial vehicle performance prediction model satisfies the first training completion condition, use the initial vehicle performance prediction model that satisfies the first training completion condition as the vehicle performance prediction model.

7. The system according to claim 5, characterized in that, The third execution unit includes: The first generation subunit is used to standardize and perform feature weighted fusion on the vehicle performance degradation parameters, summary information of the abnormal queue, and environmental load stress degree to generate a comprehensive diagnostic feature vector. An execution subunit is used to input the comprehensive diagnostic feature vector into the vehicle fault diagnosis model, wherein the vehicle fault diagnosis model is a deep neural network model based on multi-task learning, which learns a deep representation of features through a shared feature extraction layer and generates diagnostic results including fault type classification results, fault severity level and root cause component location information through multiple parallel output branches.

8. The system according to claim 7, characterized in that, The first generating subunit includes: The first acquisition module is used to acquire the length of the abnormal queue in the current period, the feature value identifiers of the abnormal events in the abnormal queue, and the average value of the feature values; The first generation module is used to generate summary information of the abnormal queue based on the length of the abnormal queue in the current period, the feature value identifiers of the abnormal events in the abnormal queue, and the average value of the feature values. The second acquisition module is used to acquire the environmental information and current load of the vehicle. The calculation module is used to calculate the environmental stress degree based on the environmental information and to calculate the dynamic load weighting factor based on the current load. A determination module is used to determine the environmental load stress degree of the vehicle based on the environmental stress degree and the dynamic load weighting factor. The normalization module is used to normalize the vehicle performance degradation parameters, anomaly queue summary information and environmental load stress degree respectively using the min-max normalization method; The second generation module is used to perform weighted fusion of the normalized features to generate the comprehensive diagnostic feature vector.

Images