An AI-based method and device for end-side analysis of vehicle CAN signals

Abstract

An AI-based method and device for end-side analysis of vehicle CAN signals, relating to the field of vehicle-side data processing technology, is disclosed. In this method, the end-side AI automatic analysis device extracts temporal feature matrices and spatial feature vectors from the arbitration process and data domain of the target vehicle CAN signal. Based on these, a signal feature vector is constructed. This signal feature vector is then input into a preset signal recognition model to obtain the signal identifier of the target vehicle CAN signal. Based on the signal identifier, the target vehicle CAN signal is categorized to generate a structured vehicle CAN signal data table. This avoids the need for extensive pre-defined physical operations to categorize the target vehicle CAN signal, thereby improving the categorization efficiency of vehicle CAN signals.

Description

Technical Field

[0001] This application relates to the field of vehicle-side data processing technology, and in particular to an AI analysis method and device for vehicle CAN signal end-side based on artificial intelligence. Background Technology

[0002] With the rapid development of the automotive industry towards intelligence and connectivity, modern vehicles integrate a large number of electronic control units (ECUs). These units communicate and collaborate via the Controller Area Network (CAN) bus. As the vehicle's internal neural network, the CAN bus carries massive amounts of data reflecting the vehicle's operating status, covering various key information from engine operating conditions and transmission status to body control systems and battery management systems. Therefore, acquiring and analyzing CAN bus signals is of significant technical value for enabling rapid vehicle fault diagnosis, efficient fleet safety management, and performance monitoring and maintenance of new energy vehicles.

[0003] To meet these requirements, relevant technologies generally employ a stimulus-response correlation analysis method to identify unknown CAN signals. Specifically, analysts first connect a CAN analyzer to the vehicle's CAN bus to record all data messages broadcast on the bus. Then, the analysts perform a series of preset physical operations on the vehicle as stimuli, such as steadily pressing or releasing the accelerator pedal, repeatedly turning the steering wheel, and turning the headlights on or off one by one. During this process, the analysis software continuously monitors the data payload (usually 8 bytes) of all CAN messages, using algorithms or manual observation to find message data that changes synchronously with the specific physical operation in time. When the message data change pattern of a certain CAN identifier is highly correlated with a certain physical stimulus, that CAN identifier is categorized as belonging to the vehicle function corresponding to that stimulus.

[0004] However, when it is necessary to support large-scale, multi-brand mixed fleets, or when a repair shop encounters a new or uncommon vehicle model, the relevant technologies need to be stimulated through a large number of pre-set physical operations, resulting in low efficiency in classifying vehicle CAN signals. Summary of the Invention

[0005] This application provides an AI-based method and device for end-side analysis of vehicle CAN signals, which improves the classification efficiency of vehicle CAN signals.

[0006] Firstly, an AI-based method for end-side analysis of vehicle CAN signals is provided. This method includes: extracting a temporal feature matrix and a spatial feature vector from the arbitration process and data domain of the target vehicle CAN signal; constructing a signal feature vector based on the temporal feature matrix and the spatial feature vector; inputting the signal feature vector into a preset signal recognition model to obtain the signal identifier of the target vehicle CAN signal; and classifying the target vehicle CAN signal based on the signal identifier to generate a structured vehicle CAN signal data table.

[0007] By adopting the above technical solution, the edge AI automatic analysis device extracts the timing feature matrix and spatial feature vector from the arbitration process and data domain of the target vehicle's CAN signal. Based on the timing feature matrix and spatial feature vector, it constructs a signal feature vector, inputs the signal feature vector into a preset signal recognition model to obtain the signal identifier of the target vehicle's CAN signal, and classifies the target vehicle's CAN signal based on the signal identifier to generate a structured vehicle CAN signal data table. This avoids a large number of preset physical operations to excite and classify the target vehicle's CAN signal, thereby improving the classification efficiency of vehicle CAN signals.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the steps of extracting timing feature matrices and spatial feature vectors from the arbitration process and data domain of the target vehicle's CAN signal specifically include: extracting the binary sequence of the arbitration identifier and remote transmission request bit from the arbitration domain of the target vehicle's CAN signal; constructing a node contention timing diagram based on the priority mapping table of the arbitration identifier, the node contention timing diagram recording the switching timing of dominant and recessive bits of each CAN node during the arbitration process; determining arbitration conflict events based on the node contention timing diagram; calculating the arbitration contention strength coefficient based on the occurrence frequency and duration of the arbitration conflict events; associating and mapping the arbitration contention strength coefficient with the bit timing features of the target vehicle's CAN signal to construct an arbitration-aware timing feature matrix; and spatially encoding the data bytes in the data domain of the target vehicle's CAN signal according to byte position and bit weight to generate spatial feature vectors.

[0009] By employing the above technical solution, the arbitration contention strength coefficient, reflecting the bus contention state, is combined with the bit timing characteristics of the signal. The constructed arbitration-aware timing feature matrix carries environmental information about signal transmission, i.e., the contention pressure under which it was successfully transmitted. Simultaneously, spatial encoding of the data field preserves the inherent structure of the data bytes. This ensures that the final feature vector not only describes the signal itself but also depicts the network state at the time of its generation, providing deeper background information for the AI ​​model and thus improving the accuracy of signal recognition.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the step of constructing a node contention timing diagram based on the priority mapping table of arbitration identifiers specifically includes: collecting arbitration domain data within a preset time window from the CAN bus of the target vehicle; statistically analyzing the occurrence frequency and time distribution of each arbitration identifier within the preset time window; calculating the bus occupancy rate of each CAN node based on the occurrence frequency, and constructing a static priority sequence according to the priority rules in the CAN protocol specification; determining a dynamic priority adjustment mode based on the deviation between the actual arbitration result and the static priority sequence; combining the static priority sequence and the dynamic priority adjustment mode to generate a priority mapping table of arbitration identifiers; and constructing a node contention timing diagram based on the priority mapping table.

[0011] By adopting the above technical solution, the static priority sequence specified by the CAN protocol is combined with the dynamic priority adjustment mode to generate a priority mapping table that conforms to the protocol basis and reflects the actual operating characteristics of the network, making the subsequent node contention timing diagram more accurate.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of determining the dynamic priority adjustment mode based on the deviation between the actual arbitration result and the static priority sequence specifically includes: determining the response capability coefficient of each CAN node based on the dominant bit holding time and recessive bit switching delay of each CAN node during the actual arbitration process; performing cross-analysis of the response capability coefficient and the bus occupancy rate of the corresponding node to obtain the response degradation mode of high-priority nodes and the preemption success mode of low-priority nodes; constructing an arbitration competition situation diagram based on the response degradation mode and the preemption success mode, the arbitration competition situation diagram reflecting the actual competitive advantage of each CAN node under different network loads; obtaining the advantageous node cluster and the disadvantageous node cluster in the network through topological analysis of the arbitration competition situation diagram; predicting the probability distribution of arbitration results within a future preset time window based on the advantageous node cluster and the disadvantageous node cluster; and using the probability distribution of arbitration results as dynamic weights to correct the static priority sequence to obtain a dynamic priority adjustment mode that reflects the real network state.

[0013] By employing the aforementioned technical solution, the response capability coefficient is quantified by analyzing the dominant bit holding time and recessive bit switching delay of nodes, revealing the response degradation mode of high-priority nodes and the preemption success mode of low-priority nodes. Based on the arbitration competition situation diagram constructed from these modes, topology analysis of the diagram identifies the advantageous and disadvantageous node clusters in the target vehicle's CAN network. Based on the probability distribution of arbitration results within the predicted future time window of the node clusters, this probability distribution is used as a dynamic weight to correct the static priority sequence, thereby improving the matching degree between the dynamic priority adjustment mode and the network state.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the step of inputting the signal feature vector into a preset signal recognition model to obtain the signal identifier of the target vehicle's CAN signal specifically includes: determining a feature weight allocation strategy based on the dimensional distribution characteristics of the signal feature vector, wherein the feature weight allocation strategy is used to balance the contributions of temporal features and spatial features in signal recognition; normalizing the signal feature vector according to the feature weight allocation strategy to generate a standardized feature vector; inputting the standardized feature vector into a multi-layer neural network for feature mapping to obtain a signal category probability distribution; determining a candidate signal identifier set based on the signal category probability distribution and a preset confidence threshold; calculating the similarity score between each candidate signal identifier in the candidate signal identifier set and historically identified signals; and comprehensively evaluating and determining the final signal identifier based on the similarity score and the signal category probability distribution.

[0015] By adopting the above technical solution, the feature weight allocation strategy determined based on the dimensional distribution characteristics of the signal feature vector can adaptively balance the contributions of temporal and spatial features in signal recognition, reducing the recognition bias caused by feature imbalance. By comprehensively evaluating the signal category probability distribution and historical similarity scores to determine the final signal identifier, the accuracy of target vehicle CAN signal recognition is improved.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, the step of classifying target vehicle CAN signals based on signal identifiers to generate a structured vehicle CAN signal data table specifically includes: classifying signal identifiers with the same or similar functional attributes into the same functional domain; determining the functional domain priority sequence based on the differences in signal characteristics and real-time requirements of each functional domain; constructing a dependency graph between functional domains based on the vehicle system architecture knowledge base, the dependency graph reflecting the control logic and data flow between different functional domains; designing a hierarchical data storage structure based on the dependency graph and the functional domain priority sequence; assigning metadata tags containing timestamps, signal sources, and data integrity check codes to each of the classified CAN signals according to the hierarchical data storage structure; and establishing a signal traceability index based on the metadata tags to generate a structured vehicle CAN signal data table that supports fast querying and correlation analysis.

[0017] By adopting the above technical solution, signal identifiers with the same functional attributes are categorized into functional domains, facilitating systematic analysis and management. The hierarchical data storage structure is designed to optimize data organization and access efficiency based on the dependencies and priority differences between functional domains. The signal tracing index built based on metadata tags supports multi-dimensional fast querying and correlation analysis, improving the usability and query performance of the structured data table.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of establishing a signal tracing index based on metadata tags and generating a structured vehicle CAN signal data table that supports fast querying and correlation analysis, the method further includes: calculating the data activity index of each signal of the target vehicle based on the number of data changes and the total number of data transmission bytes within a preset observation period for each signal in each functional domain; assigning monitoring weight coefficients to each signal of the target vehicle according to the urgency requirements of vehicle fault diagnosis, wherein the monitoring weight coefficients are positively correlated with the safety criticality of the vehicle function corresponding to the signal; performing a weighted operation on the data activity index and the monitoring weight coefficients to obtain a comprehensive importance score for each signal; sorting and reorganizing the signal entries in the structured data table according to the comprehensive importance score from high to low; adjusting the physical layout of data storage according to the reorganized signal sequence, placing high-scoring signals in the high-speed access area of ​​the memory and low-scoring signals in the normal access area of ​​the memory.

[0019] By adopting the above technical solution, the logical structure of the data table is optimized through signal sorting and reorganization based on comprehensive importance scores, enabling important signals to be accessed and processed preferentially. The physical storage layout, adjusted according to the reorganized signal sequence, places high-scoring signals in a high-speed access area and low-scoring signals in a normal access area. This storage optimization strategy improves the access speed of critical signals and the overall system response performance.

[0020] Secondly, embodiments of this application provide an edge AI automatic analysis device, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the edge AI automatic analysis device to perform the method as described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on an edge AI automatic analysis device, cause the edge AI automatic analysis device to execute the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on an edge AI automatic analysis device, cause the edge AI automatic analysis device to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the edge AI automatic analysis device provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. Because the edge AI automatic analysis device extracts the timing feature matrix and spatial feature vector from the arbitration process and data domain of the target vehicle's CAN signal, constructs a signal feature vector based on the timing feature matrix and spatial feature vector, inputs the signal feature vector into a preset signal recognition model to obtain the signal identifier of the target vehicle's CAN signal, and classifies the target vehicle's CAN signal based on the signal identifier to generate a structured vehicle CAN signal data table, it avoids a large number of preset physical operations to excite and classify the target vehicle's CAN signal, thereby improving the classification efficiency of vehicle CAN signals.

[0026] 2. Because the edge-side AI automatic analysis device combines the arbitration contention strength coefficient, which reflects the bus contention state, with the bit timing characteristics of the signal, the constructed arbitration-aware timing feature matrix carries environmental information about the signal transmission, i.e., the contention pressure under which it was successfully transmitted. Simultaneously, spatial encoding of the data domain preserves the inherent structure of the data bytes. This ensures that the final feature vector not only describes the signal itself but also depicts the network state at the time of its generation, providing the AI ​​model with deeper background information and thus improving the accuracy of signal recognition.

[0027] 3. Because the edge AI automatic analysis device combines the static priority sequence specified by the CAN protocol with the dynamic priority adjustment mode, it generates a priority mapping table that conforms to the protocol basis and reflects the actual operating characteristics of the network, making the subsequent node contention timing diagram more accurate. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating an AI analysis method for vehicle CAN signal end-side based on artificial intelligence, as described in an embodiment of this application.

[0029] Figure 2 This is another flowchart illustrating an AI analysis method for vehicle CAN signals based on artificial intelligence, as described in this application.

[0030] Figure 3 This is a schematic diagram of the physical device structure of an edge AI automatic analysis device in the embodiments of this application. Detailed Implementation

[0031] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0033] This application provides an AI-based method and device for end-side analysis of vehicle CAN signals, which improves the classification efficiency of vehicle CAN signals.

[0034] The following describes an AI-based vehicle CAN signal end-side analysis method according to an embodiment of this application:

[0035] Please see Figure 1 This is a flowchart illustrating an AI analysis method for vehicle CAN signal terminals based on artificial intelligence, as described in this application.

[0036] S101. Extract the timing feature matrix and spatial feature vector from the arbitration process and data domain of the target vehicle's CAN signal.

[0037] The target vehicle CAN signal refers to the specific vehicle controller area network communication signal that needs to be analyzed and identified. The arbitration process refers to the competition mechanism for determining bus access rights by comparing the priorities of arbitration identifiers when multiple nodes on the CAN bus transmit data simultaneously. The data field represents the 8-byte data segment carrying the actual information content in the CAN frame structure. The timing feature matrix is ​​a two-dimensional data structure with time step as rows and feature type as columns, used to reflect the variation pattern of the CAN signal in the time dimension. The spatial feature vector represents a one-dimensional array describing the position, bit weight, and spatial distribution characteristics of each byte in the CAN signal data field.

[0038] Specifically, this step is performed after the edge-side AI automatic analysis device acquires the target vehicle's CAN bus data. It should be noted that the target vehicle CAN bus data required by this invention is acquired using a dedicated CAN analyzer or data logger with high-precision physical layer sampling capabilities. This device can record the level state and duration of each bit on the bus with a resolution of at least nanoseconds. This dedicated analyzer can be connected to the CAN bus via the vehicle's OBD interface, but its data acquisition depth and accuracy far exceed the scope of standard OBD diagnostic protocols. After the edge AI automatic analysis device acquires the CAN bus data of the target vehicle, it constructs a timing feature matrix by segmenting the time window and extracting multiple features in parallel. First, the continuous CAN signal of the target vehicle is segmented according to the preset time window length. Within each time window, the edge AI automatic analysis device performs bit-level parsing on the arbitration field of the CAN frame of the target vehicle, extracting the 11-bit or 29-bit arbitration identifier and the binary data of the RTR bit. By monitoring the timing of the level changes of multiple CAN nodes during the bus arbitration process, it calculates multiple timing feature parameters such as arbitration contention strength, node response delay, priority conflict frequency, and bus occupancy rate within the time window. Then, these feature parameters are used as column vectors and the time window sequence is used as row vectors to organize a two-dimensional timing feature matrix with the time step as the row and the feature type as the column. Meanwhile, the edge AI automatic analysis device extracts spatial features from the data field, parses the 8 bytes of the CAN frame data field byte by byte, assigns different position weights to each byte according to its absolute position in the frame (bytes 0 to 7), and considers the weight distribution of each bit within the byte (from the most significant bit to the least significant bit). Combined with the encoding characteristics of data types (such as integer, floating-point, and Boolean), the byte position and bit weight are encoded into a multi-dimensional vector to form a spatial feature vector that can reflect the spatial distribution pattern and semantic information of the data.

[0039] S102. Construct a signal feature vector based on the time-series feature matrix and the spatial feature vector.

[0040] Specifically, this step is executed immediately after the edge AI automatic analysis device extracts the temporal feature matrix and spatial feature vector. The edge AI automatic analysis device first performs a dimensionality transformation on the temporal feature matrix. Since the temporal feature matrix is ​​a two-dimensional structure (time step × number of features), while the spatial feature vector is a one-dimensional structure, the edge AI automatic analysis device uses methods such as matrix flattening, global average pooling, or temporal convolution to convert the temporal feature matrix into a one-dimensional temporal feature vector, thus achieving dimensional compatibility with the spatial feature vector. Subsequently, based on the difference in importance between temporal and spatial features in the CAN signal recognition task, the edge AI automatic analysis device determines the feature weights by calculating the information gain, variance contribution, or correlation coefficient of each feature component. The transformed temporal and spatial feature vectors are then weighted separately to highlight important features and suppress redundant information. Finally, the edge AI automatic analysis device uses strategies such as vector splicing, weighted summation, or attention fusion to combine the weighted temporal feature vector and spatial feature vector into a unified signal feature vector. This vector not only retains the temporal dynamic characteristics of the CAN signal, but also contains the spatial structure information of the data domain.

[0041] S103. Input the signal feature vector into the preset signal recognition model to obtain the signal identifier of the target vehicle's CAN signal.

[0042] Among them, the signal identifier is a unique identifier used to represent the specific type, functional classification, or abnormal state of the CAN signal.

[0043] Specifically, this step is executed after the edge AI automatic analysis device completes the construction of the signal feature vector. The edge AI automatic analysis device first performs feature weight analysis on the signal feature vector to establish a feature weight allocation strategy. Assuming the signal feature vector is 256-dimensional, with the first 128 dimensions representing temporal features and the last 128 dimensions representing spatial features, the edge AI automatic analysis device evaluates feature importance by calculating the variance of each dimension. For example, if the variance of the temporal feature is 0.8 and the variance of the spatial feature is 0.6, then the weight of the temporal feature is assigned as 0.57 (0.8 / 1.4), and the weight of the spatial feature is assigned as 0.43 (0.6 / 1.4). Subsequently, the edge AI automatic analysis device multiplies the first 128 dimensions of the signal feature vector by 0.57 and the last 128 dimensions by 0.43, and then performs L2 normalization on the entire vector to make the vector magnitude 1, generating a standardized feature vector. Next, the edge AI automatic analysis device inputs a 256-dimensional standardized feature vector into a fully connected neural network containing an input layer (256 nodes), a first hidden layer (128 nodes), a second hidden layer (64 nodes), a third hidden layer (32 nodes), and an output layer (10 signal category nodes). Using the ReLU activation function and forward propagation, the output layer uses the softmax function to generate probability values ​​for 10 signal categories, such as [0.05, 0.12, 0.78, 0.02, 0.01, 0.008, 0.006, 0.004, 0.002, 0.0]. Then, the device sets a confidence threshold of 0.1 and filters out candidate signal identifiers with probabilities greater than 0.1, namely category 2 (engine speed signal probability 0.78) and category 1 (vehicle speed signal probability 0.12), forming a candidate signal identifier set. The edge AI automatic analysis device retrieves the identified signal features of the same category from the historical database and calculates the cosine similarity between the current feature vector and the historical feature vector. For example, the similarity of category 2 is 0.92 and the similarity of category 1 is 0.76. Finally, the edge AI automatic analysis device comprehensively evaluates the candidate signal identifiers using a weighted average method. The specific calculation formula is: Comprehensive Score = Probability Distribution Score × Probability Weight + Similarity Score × Similarity Weight. The edge AI automatic analysis device presets the probability weight to be 0.6 and the similarity weight to be 0.4. For candidate category 2, the edge AI automatic analysis device calculates the comprehensive score as 0.78 × 0.6 + 0.92 × 0.4 = 0.468 + 0.368 = 0.836. For candidate category 1, the edge AI automatic analysis device calculates the comprehensive score as 0.12 × 0.6 + 0.76 × 0.4 = 0.072 + 0.304 = 0.376. The edge AI automatic analysis device compares the two comprehensive scores (0.836 > 0.376) and selects category 2 with the highest score as the final signal identifier.

[0044] S104. Classify the target vehicle's CAN signals based on signal identifiers and generate a structured vehicle CAN signal data table.

[0045] Among them, the structured vehicle CAN signal data table represents a collection of vehicle signal data stored in a hierarchical organizational structure, supporting fast query and correlation analysis.

[0046] Specifically, this step is performed after the edge AI automatic analysis device completes signal identifier recognition. The edge AI automatic analysis device first divides the identified signal identifiers into functional domains based on their functional attributes. Signal identifiers with the same or similar functional attributes are grouped into the same functional domain. For example, engine speed signals, engine temperature signals, and throttle position signals are grouped into the powertrain functional domain; vehicle speed signals, brake pressure signals, and ABS status signals are grouped into the braking system functional domain; steering angle signals and steering torque signals are grouped into the steering system functional domain; and battery voltage signals and charging status signals are grouped into the power management functional domain. Subsequently, the edge AI automatic analysis device determines the functional domain priority sequence based on the differences in signal characteristics and real-time requirements of each functional domain. By analyzing the signal sampling frequency, data change rate, and safety criticality level, the powertrain functional domain is set to priority 1 (highest), the braking system functional domain to priority 2, the steering system functional domain to priority 3, and the power management functional domain to priority 4. Next, the edge AI automatic analysis device constructs a dependency graph between functional domains based on the vehicle system architecture knowledge base. It analyzes the control logic and data flow relationships, such as the powertrain providing energy support to the braking system, the braking system and steering system working together, and power management providing power to all systems. A directed graph structure is then established to represent the dependencies between functional domains. Then, based on the dependency graph and the functional domain priority sequence, the edge AI automatic analysis device designs a hierarchical data storage structure. A tree structure is used, with high-priority functional domains as parent nodes and low-priority functional domains as child nodes. Dependencies are considered, and dependent functional domains are set as adjacent levels, forming a four-layer storage architecture: the first layer stores powertrain system signals, the second layer stores braking system signals, the third layer stores steering system signals, and the fourth layer stores power management signals. The edge AI automatic analysis device assigns metadata tags to each of the categorized CAN signals according to a hierarchical data storage structure. For example, the engine speed signal is assigned the tag {timestamp: 2025-09-23 14:30:25.123, signal source: ECU_Engine_01, data integrity check code: CRC32_0x8F2A1B3C, functional domain: power system, priority: 1, level: first layer}, and the brake pressure signal is assigned the tag {timestamp: 2025-09-23 14:30:25.124, signal source: ECU_Brake_02, data integrity check code: CRC32_0x7E9D4C5A, functional domain: braking system, priority: 2, level: second layer}.Finally, the edge AI automatic analysis device establishes a signal traceability index based on metadata tags. It supports time-series queries through timestamp indexes, source tracing through signal source indexes, category retrieval through functional domain indexes, and hierarchical access through hierarchical indexes. This generates a structured vehicle CAN signal data table that supports fast querying and correlation analysis. The data table adopts a key-value pair storage format, with the primary key being the signal identifier and the value being a composite structure containing the original signal data and complete metadata tags.

[0047] In some embodiments, after generating a structured vehicle CAN signal data table, the edge AI automatic analysis device further performs data optimization and monitoring tasks. The edge AI automatic analysis device calculates the data activity index of each signal of the target vehicle based on the number of data changes and the total number of bytes transmitted within a preset observation period in each functional domain. Assuming the observation period is 60 seconds, the edge AI automatic analysis device counts that the engine speed signal changes 1200 times and transmits 9600 bytes within 60 seconds. The edge AI automatic analysis device uses a weighted average formula to calculate the data activity index, where the number of data changes has a weight of 0.6 and the total number of bytes transmitted has a weight of 0.4. The specific calculation process is as follows: change frequency = 1200 times / 60 seconds = 20 times / second, transmission rate = 9600 bytes / 60 seconds = 160 bytes / second, data activity index = 20 times / second × 0.6 + 160 bytes / second × 0.4 = 12 times / second + 64 bytes / second = 76 activity points. Similarly, the brake pressure signal activity is calculated to be 36 activity points and the steering angle signal activity is calculated to be 28 activity points. The edge-side AI-powered automatic analysis equipment assigns monitoring weight coefficients to various signals of the target vehicle based on the urgency of vehicle fault diagnosis. For example, the monitoring weight for braking system-related signals is set to 0.9 (highest safety criticality), the monitoring weight for powertrain system signals is set to 0.8, the monitoring weight for steering system signals is set to 0.7, and the monitoring weight for power management signals is set to 0.6. The edge-side AI-powered automatic analysis equipment then performs a weighted calculation based on the data activity index and the monitoring weight coefficients to obtain a comprehensive importance score for each signal. The comprehensive score for the engine speed signal is 76 × 0.8 = 60.8 points, the comprehensive score for the brake pressure signal is 36 × 0.9 = 32.4 points, and the comprehensive score for the steering angle signal is 28 × 0.7 = 19.6 points. Based on the comprehensive importance score, the edge-side AI-powered automatic analysis equipment sorts and reorganizes the signal entries in the structured data table according to their scores from highest to lowest, placing the engine speed signal first, the brake pressure signal second, and the steering angle signal third, thus reorganizing the data table structure. The edge AI automatic analysis device adjusts the physical layout of data storage based on the recombined signal sequence, placing high-scoring signals with a comprehensive score greater than 50 in the high-speed access area of ​​the memory (such as SSD solid-state drives or memory cache), and placing low-scoring signals with a comprehensive score less than 50 in the normal access area of ​​the memory (such as mechanical hard drives), thereby improving the fast access performance of key signals.

[0048] In the above embodiments, the edge AI automatic analysis device extracts the timing feature matrix and spatial feature vector from the arbitration process and data domain of the target vehicle's CAN signal, constructs a signal feature vector based on the timing feature matrix and spatial feature vector, inputs the signal feature vector into a preset signal recognition model to obtain the signal identifier of the target vehicle's CAN signal, and classifies the target vehicle's CAN signal based on the signal identifier to generate a structured vehicle CAN signal data table. This avoids a large number of preset physical operations to excite and classify the target vehicle's CAN signal, thereby improving the classification efficiency of vehicle CAN signals.

[0049] However, due to the differences in hardware responsiveness among different ECU nodes, dynamic changes in network load, and the complex factors of concurrent arbitration competition among multiple nodes, the arbitration process exhibits a high degree of dynamism and uncertainty. The above embodiment provides a relatively simplified analysis of the arbitration competition situation, which may reduce the accuracy of signal recognition in complex network environments. To further improve the accuracy and robustness of signal recognition, especially to maintain stable recognition performance in high-load, multi-collision CAN bus environments, a more refined modeling and analysis of the arbitration process is needed.

[0050] Please see Figure 2 This is another flowchart illustrating an AI analysis method for vehicle CAN signals based on artificial intelligence, as described in this application.

[0051] S201. Extract the binary sequence of the arbitration identifier and remote transmission request bit from the arbitration field of the target vehicle's CAN signal.

[0052] The arbitration field in the CAN frame structure represents the data field used for bus access arbitration, containing an arbitration identifier and control bit information. The arbitration identifier is used to uniquely identify the CAN frame; it is 11 bits in a standard CAN frame and 29 bits in an extended CAN frame, with smaller values ​​indicating higher priority. The Remote Transmission Request bit refers to the RTR bit in the CAN frame, used to distinguish between data frames (RTR=0) and remote frames (RTR=1).

[0053] Specifically, this step is executed immediately after the edge-side AI automatic analysis device completes CAN bus data acquisition. The edge-side AI automatic analysis device first performs frame synchronization and boundary detection on the received raw CAN frame. It determines the start position of the CAN frame by identifying the Start of Frame (SOF) bit, and then locates the start and end boundaries of the arbitration field according to the CAN protocol standard. For standard CAN frames, the edge-side AI automatic analysis device extracts an 11-bit arbitration identifier starting from the first bit after the SOF bit, followed by the 12th bit, the RTR bit. For extended CAN frames, the device extracts the first 11 basic identifier bits, 1 SRR bit, 1 IDE bit, 18 extended identifier bits, and 1 RTR bit. The edge-side AI automatic analysis device uses bit synchronization clock recovery technology to ensure accurate bit boundary identification, and uses differential Manchester encoding parsing to convert level changes on the CAN bus into logic 0s and 1s, forming a complete binary sequence. For example, when a standard CAN frame is detected, the device extracts the binary sequence of the arbitration identifier as 10110010011 (corresponding to hexadecimal 0x5A3), with the RTR bit set to 0, and finally generates the complete arbitration field binary sequence 101100100110.

[0054] S202. Construct a node contention sequence diagram based on the priority mapping table of the arbitration identifier.

[0055] The arbitration identifier is a unique numerical identifier in the CAN frame used to identify the message type and the sending node; it is 11 bits in the standard frame and 29 bits in the extended frame. The priority mapping table represents a comprehensive priority lookup table generated by combining static priority sequences with dynamic priority adjustment modes, conforming to the CAN protocol specification while reflecting the actual network state. The node contention timing diagram is used to represent a two-dimensional graph with time on the horizontal axis and CAN nodes on the vertical axis, showing the contention behavior and timing relationships of each CAN node during the arbitration process.

[0056] Specifically, this step is executed after the edge-side AI automatic analysis device completes the extraction of the arbitration domain binary sequence. The edge-side AI automatic analysis device collects arbitration domain data from the target vehicle's CAN bus within a preset time window. The time window is set to 500 milliseconds. Within this time window, the edge-side AI automatic analysis device continuously monitors the CAN bus and captures the arbitration domain information of all CAN frames. Assuming 150 CAN frames are collected, containing different types of arbitration identifiers such as 0x123, 0x456, and 0x789, the edge-side AI automatic analysis device statistically analyzes the frequency and time distribution of each arbitration identifier within the preset time window. For example, identifier 0x123 appears 45 times with an average interval of 11.1 milliseconds, identifier 0x456 appears 30 times with an average interval of 16.7 milliseconds, and identifier 0x789 appears 75 times with an average interval of 6.7 milliseconds. A frequency statistics table and a time distribution histogram are then established. The edge-side AI automatic analysis device calculates the bus occupancy rate of each CAN node based on its frequency of occurrence. The bus occupancy rate of the node corresponding to identifier 0x123 is 45 / 150 = 30%, the bus occupancy rate of the node corresponding to identifier 0x456 is 20%, and the bus occupancy rate of the node corresponding to identifier 0x789 is 50%. A static priority sequence is constructed according to the rule in the CAN protocol specification that lower values ​​indicate higher priority: 0x123 (highest priority), 0x456 (medium priority), and 0x789 (lowest priority). The edge-side AI automatic analysis device determines the response capability coefficient of each CAN node based on the dominant bit holding time and recessive bit switching delay during the actual arbitration process. The dominant bit holding time of the node corresponding to identifier 0x123 is measured to be 2.5 microseconds, and the recessive bit switching delay is 1.2 microseconds, resulting in a response capability coefficient of 1 / (2.5+1.2) = 0.27. Similarly, the response capability coefficients for other nodes are calculated to be 0.31 and 0.22, respectively. The edge-side AI-powered automatic analysis device cross-analyzes the response capability coefficient with the bus occupancy rate of the corresponding node. It finds that while high-priority node 0x123 has the highest static priority, its response capability coefficient is low, indicating a response degradation mode. Conversely, low-priority node 0x789, despite having the lowest static priority, has the highest bus occupancy rate, indicating a successful preemption mode. Based on these response degradation and successful preemption modes, the device constructs an arbitration competition landscape diagram, distributing nodes according to their response capability and bus occupancy rate in a two-dimensional coordinate system. This identifies advantageous node clusters (high response capability + moderate occupancy rate) and disadvantageous node clusters (low response capability + high occupancy rate). Through topology analysis of the arbitration competition landscape diagram, the device determines that node 0x456 belongs to an advantageous node cluster, and node 0x789 belongs to a disadvantageous node cluster. Based on these clusters, it predicts the probability distribution of the arbitration outcome within the next 500 milliseconds: node 0x123 has a 35% chance of winning, node 0x456 has a 40% chance, and node 0x789 has a 25% chance.The edge-side AI-powered automatic analysis device uses the probability distribution of arbitration results as dynamic weights to correct the static priority sequence, resulting in a dynamic priority adjustment mode: node 0x456 is dynamically prioritized to the highest level, node 0x123 maintains a medium priority, and node 0x789 maintains the lowest priority. The static priority sequence and the dynamic priority adjustment mode are combined to generate a priority mapping table. Finally, the edge-side AI-powered automatic analysis device constructs a node contention timing diagram based on the priority mapping table. Using time as the horizontal axis and node identifier as the vertical axis, it plots the bus access timing of each node within a time window, marking arbitration conflict points and priority switching events, thus forming a node contention timing diagram reflecting the network contention state.

[0057] S203. Determine arbitration conflict events based on node contention sequence diagram.

[0058] Specifically, this step is executed after the edge-side AI automatic analysis device completes the construction of the node contention timing diagram. The edge-side AI automatic analysis device performs a time window scan on the node contention timing diagram, setting the scan granularity to 1 microsecond, checking at each time point whether multiple nodes simultaneously initiate bus access requests. By analyzing the start timestamps of the signal lines of each node in the timing diagram, the edge-side AI automatic analysis device identifies that at time point t1=125.5ms, nodes 0x123 and 0x456 simultaneously begin sending the start-of-frame bit, and at time point t2=267.8ms, nodes 0x456, 0x789, and 0x1AB simultaneously attempt to access the bus. The edge-side AI automatic analysis device marks these times of concurrent access by multiple nodes as arbitration conflict events. The edge-side AI automatic analysis device organizes the identified arbitration conflict events in chronological order, recording basic information such as the occurrence time, involved nodes, and duration of each event.

[0059] S204. Calculate the arbitration competition intensity coefficient based on the frequency and duration of arbitration conflict events.

[0060] Specifically, this step is executed after the edge-side AI automatic analysis device identifies arbitration conflict events and arbitration delay events. The edge-side AI automatic analysis device first counts the frequency of arbitration conflict events within a preset observation time window. Assuming 15 arbitration conflict events are detected within a 500-millisecond observation window, the conflict frequency is calculated to be 15 / 500ms = 30 times / second. The edge-side AI automatic analysis device measures the duration of each arbitration conflict event, recording the complete time from the start of contention among multiple nodes to one node gaining bus access. For example, the first conflict lasts 3.2 microseconds, the second lasts 5.7 microseconds, and the third lasts 2.8 microseconds. The edge-side AI automatic analysis device calculates the average duration of all conflict events; assuming the average duration of 15 conflicts is 4.1 microseconds. The edge-side AI automatic analysis device calculates the arbitration contention intensity coefficient using a weighted calculation formula, comprehensively evaluating both conflict frequency and average duration. The formula is: Contention Intensity Coefficient = Conflict Frequency × Average Duration × Weighting Factor. The weighting factor is determined based on the CAN network's bus speed and number of nodes. For example, a high-speed CAN network (500kbps) has a weighting factor of 0.1, while a low-speed CAN network (125kbps) has a weighting factor of 0.2. Assuming the current network is high-speed, the contention intensity coefficient is 30 × 4.1 × 0.1 = 12.3. The edge-side AI automatic analysis device uses the calculated arbitration contention intensity coefficient as a quantitative indicator of the network contention status; a higher value indicates more intense network contention.

[0061] S205. Associate and map the arbitration competition intensity coefficient with the bit timing characteristics of the target vehicle's CAN signal to construct an arbitration-aware timing feature matrix.

[0062] Specifically, this step is performed after the edge-side AI automatic analysis device completes the arbitration contention strength coefficient calculation. The edge-side AI automatic analysis device extracts the bit timing features of the target vehicle's CAN signal, analyzes the 64 bits of data in each CAN frame bit by bit, and records the duration of each bit's level, transition frequency, and bit interval. For example, the duration of the first bit is 2 microseconds, the duration of the second bit is 4 microseconds, and the third bit has a level transition. The edge-side AI automatic analysis device organizes these bit-level timing information into a 64-dimensional timing feature vector. The edge AI automatic analysis device uses the calculated arbitration competition strength coefficient as a global weight factor and maps it to the bit-time features. Through matrix multiplication, the competition strength coefficient (12.3) from step S204 (assuming a high-speed network) is multiplied by each of the 64 bit-time feature values ​​to obtain a weighted time-time feature vector. For example, the original first-bit feature value of 0.5 becomes 0.5 × 12.3 = 6.15 after weighting, and the original second-bit feature value of 0.8 becomes 0.8 × 12.3 = 9.84 after weighting. The edge AI automatic analysis device constructs an arbitration-aware time-time feature matrix, with time windows as rows and weighted bit-time features as columns. Assuming it contains 100 time windows and 64 bit features, a 100 × 64 feature matrix is ​​formed. Each element in the matrix represents the arbitration-aware time-time feature value at a specific position within a specific time window. The edge AI automatic analysis device normalizes the constructed time-series feature matrix, using the max-min normalization method to scale all feature values ​​to the range of 0-1, so that features of different magnitudes have equal importance in subsequent analysis, and finally obtains a standardized arbitration-aware time-series feature matrix as a comprehensive feature representation of the CAN signal.

[0063] S206. Spatial encoding is performed on the data bytes in the data field of the target vehicle's CAN signal according to the byte position and bit weight to generate a spatial feature vector.

[0064] Specifically, this step can be performed after step S205 or in parallel with step S205. The edge AI automatic analysis device parses the data field structure of the target vehicle's CAN signal. Assuming the data field contains 8 bytes, located at positions 0 to 7 respectively, and each byte contains 8 binary bits, the edge AI automatic analysis device assigns a position code to each byte position and uses one-hot encoding to convert the byte position information into an 8-dimensional binary vector. For example, position 0 is encoded as [1, 0, 0, 0, 0, 0, 0, 0], and position 1 is encoded as [0, 1, 0, 0, 0, 0, 0, 0]. The edge AI automatic analysis device assigns bit weights to the binary bits in each data byte, setting the weight values ​​according to powers of 2. For example, the weight of the least significant bit is 2^0=1, the weight of the second least significant bit is 2^1=2, and the weight of the most significant bit is 2^7=128. The edge AI automatic analysis device multiplies the byte content by the bit weight to obtain a weighted value. For example, the byte value 01101001 is calculated to be 1×1+0×2+0×4+1×8+0×16+1×32+1×64+0×128=105. The edge AI automatic analysis device concatenates and fuses the position encoding and weighted values ​​to form a spatial feature representation for each byte. For example, the byte feature for position 0 is [1, 0, 0, 0, 0, 0, 0, 105], and the byte feature for position 1 is [0, 1, 0, 0, 0, 0, 0, 73]. The edge AI automatic analysis device combines the spatial features of all bytes sequentially to generate a 72-dimensional spatial feature vector (8 bytes × 9-dimensional features). The edge AI automatic analysis device standardizes the generated spatial feature vector, using the Z-score standardization method to eliminate the magnitude differences between different feature dimensions, ensuring the balance between positional and numerical information in the feature vector. The final standardized spatial feature vector serves as the spatial structure representation of the CAN signal data domain.

[0065] S207. Construct signal feature vectors based on temporal feature matrix and spatial feature vector.

[0066] S208. Input the signal feature vector into the preset signal recognition model to obtain the signal identifier of the target vehicle's CAN signal.

[0067] S209. Classify the target vehicle's CAN signals based on signal identifiers and generate a structured vehicle CAN signal data table.

[0068] Step S207 is similar to step S102, step S208 is similar to step S103, and step S209 is similar to step S104, so they will not be described again here.

[0069] In the above embodiments, the edge AI automatic analysis device transforms the abstract network competition state into quantifiable numerical features through the arbitration competition strength coefficient, and establishes a correlation mapping relationship with the bit timing features of the signal. This signal recognition method based on arbitration competition situation awareness can more accurately capture the essential features of signals in complex bus environments, thereby improving the accuracy of signal recognition in complex network environments.

[0070] The above describes an AI-based vehicle CAN signal end-side AI analysis method in the embodiments of this application. The following describes an exemplary end-side AI automatic analysis device 300 provided in the embodiments of this application.

[0071] Figure 3 This is an exemplary hardware structure diagram of the edge AI automatic analysis device 300 provided in this application embodiment. In some embodiments, the edge AI automatic analysis device 300 is a computer device, which includes a processor, a memory, and a network interface connected via an edge AI automatic analysis device bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores operating edge AI automatic analysis devices, computer programs, and a database. The internal memory provides an environment for the operation of the operating edge AI automatic analysis devices and computer programs in the non-volatile storage medium. The database of the computer device stores data. The network interface of the computer device is used to communicate with other external terminals or edge AI automatic analysis devices via a network connection. In some embodiments, the network interface can be a wired network interface; in some embodiments, the network interface can also be a wireless network interface. When the computer program is executed by the processor, it implements an artificial intelligence-based edge AI analysis method for vehicle CAN signals according to an embodiment of this application.

[0072] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0073] In some embodiments of this application, a computer-readable storage medium is also provided, including instructions that, when executed on the end-side AI automatic analysis device 300, cause the end-side AI automatic analysis device 300 to execute an artificial intelligence-based vehicle CAN signal end-side AI analysis method according to an embodiment of this application.

[0074] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0075] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0076] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0077] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. An AI-based method for analyzing vehicle CAN signals at the terminal side, characterized in that, The method includes: Extract the timing feature matrix and spatial feature vector from the arbitration process and data domain of the target vehicle's CAN signal; The steps of extracting timing feature matrix and spatial feature vector from the arbitration process and data domain of the target vehicle's CAN signal specifically include: extracting the binary sequence of arbitration identifier and remote transmission request bit from the arbitration domain of the target vehicle's CAN signal; constructing a node contention timing diagram based on the priority mapping table of the arbitration identifier, wherein the node contention timing diagram records the switching timing of dominant and recessive bits of each CAN node during the arbitration process; determining arbitration conflict events based on the node contention timing diagram; calculating the arbitration contention strength coefficient based on the occurrence frequency and duration of the arbitration conflict events; associating and mapping the arbitration contention strength coefficient with the bit timing features of the target vehicle's CAN signal to construct an arbitration-aware timing feature matrix; and spatially encoding the data bytes in the data domain of the target vehicle's CAN signal according to byte position and bit weight to generate a spatial feature vector. Construct a signal feature vector based on the temporal feature matrix and the spatial feature vector; The signal feature vector is input into a preset signal recognition model to obtain the signal identifier of the target vehicle's CAN signal; Based on the signal identifiers, the target vehicle's CAN signals are categorized to generate a structured vehicle CAN signal data table.

2. The method according to claim 1, characterized in that, The step of constructing the node contention sequence graph based on the priority mapping table of the arbitration identifier specifically includes: Arbitration domain data is collected from the target vehicle's CAN bus within a preset time window; The frequency and time distribution of each arbitration identifier within the preset time window are statistically analyzed. The bus occupancy rate of each CAN node is calculated based on the frequency of occurrence, and a static priority sequence is constructed according to the priority rules in the CAN protocol specification. The dynamic priority adjustment mode is determined based on the deviation between the actual arbitration result and the static priority sequence. The static priority sequence and the dynamic priority adjustment mode are combined to generate the priority mapping table of the arbitration identifier; A node contention sequence diagram is constructed based on the priority mapping table.

3. The method according to claim 2, characterized in that, The step of determining the dynamic priority adjustment mode based on the deviation between the actual arbitration result and the static priority sequence specifically includes: The response capability coefficient of each CAN node is determined based on the dominant bit holding time and recessive bit switching delay of each CAN node during the actual arbitration process. By cross-analyzing the response capability coefficient with the bus occupancy rate of the corresponding node, the response degradation mode of high-priority nodes and the preemption success mode of low-priority nodes are obtained. An arbitration competition situation diagram is constructed based on the response degradation mode and the preemption success mode. The arbitration competition situation diagram reflects the actual competitive advantages of each CAN node under different network loads. By analyzing the topology of the arbitration competition situation diagram, the dominant node clusters and the disadvantaged node clusters in the network are obtained. Based on the dominant node cluster and the disadvantaged node cluster, predict the probability distribution of arbitration results within a future preset time window; Using the probability distribution of the arbitration results as dynamic weights, the static priority sequence is modified to obtain a dynamic priority adjustment mode that reflects the actual network state.

4. The method according to claim 1, characterized in that, The step of inputting the signal feature vector into a preset signal recognition model to obtain the signal identifier of the target vehicle's CAN signal specifically includes: A feature weight allocation strategy is determined based on the dimensional distribution characteristics of the signal feature vector. The feature weight allocation strategy is used to balance the contributions of temporal features and spatial features in signal recognition. The signal feature vector is normalized according to the feature weight allocation strategy to generate a standardized feature vector; The standardized feature vector is input into a multi-layer neural network for feature mapping to obtain the signal category probability distribution; A set of candidate signal identifiers is determined based on the probability distribution of the signal categories and a preset confidence threshold. Calculate the similarity score between each candidate signal identifier in the candidate signal identifier set and historically identified signals; Based on the similarity score and the signal category probability distribution, the final signal identifier is determined through comprehensive evaluation.

5. The method according to claim 1, characterized in that, The step of classifying the target vehicle's CAN signals based on the signal identifier and generating a structured vehicle CAN signal data table specifically includes: The signal identifiers that have the same or similar functional attributes are classified into the same functional domain; The functional domain priority sequence is determined based on the differences in signal characteristics and real-time requirements of each functional domain. A dependency graph between functional domains is constructed based on a vehicle system architecture knowledge base. The dependency graph reflects the control logic and data flow between different functional domains. Design a hierarchical data storage structure based on the dependency graph and the functional domain priority sequence; The categorized CAN signals are assigned metadata tags containing timestamps, signal sources, and data integrity check codes according to the hierarchical data storage structure. A signal tracing index is established based on the metadata tags, generating a structured vehicle CAN signal data table that supports fast querying and correlation analysis.

6. The method according to claim 5, characterized in that, After the step of establishing a signal tracing index based on the metadata tags and generating a structured vehicle CAN signal data table that supports fast querying and correlation analysis, the method further includes: The data activity index of each signal of the target vehicle is calculated based on the number of data changes and the total number of bytes transmitted within the preset observation period in each functional domain. Based on the urgency requirements of vehicle fault diagnosis, a monitoring weight coefficient is assigned to each signal of the target vehicle. The monitoring weight coefficient is positively correlated with the safety criticality of the vehicle function corresponding to the signal. The data activity index is weighted by the monitoring weight coefficient to obtain the comprehensive importance score of each signal; Based on the comprehensive importance score, the signal entries in the structured data table are sorted and reorganized according to the score from high to low; The physical layout of data storage is adjusted according to the recombined signal sequence, placing high-scoring signals in the high-speed access area of ​​the memory and low-scoring signals in the normal access area of ​​the memory.

7. An edge-side AI automatic analysis device, characterized in that, The edge AI automatic analysis device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the edge AI automatic analysis device to perform the method as described in any one of claims 1-6.

8. A computer program product containing instructions, characterized in that, When the computer program product is run on the edge AI automatic analysis device, the edge AI automatic analysis device performs the method as described in any one of claims 1-6.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the edge AI automatic analysis device, the edge AI automatic analysis device performs the method as described in any one of claims 1-6.

Images