A new energy vehicle electronic control fault prediction system

Abstract

This invention relates to the field of electronic control technology for new energy vehicles, and discloses a fault prediction system for electronic control systems in new energy vehicles. The system includes a real-time data acquisition module, a dynamic fault prediction model construction module, a fault difference calculation module, a multi-dimensional anomaly analysis module, a fault probability localization module, and an adaptive maintenance strategy module. The real-time data acquisition module collects sensor data from the electronic control system in real time; the dynamic fault prediction model construction module constructs a dynamic fault prediction model based on historical fault data; the fault difference calculation module inputs real-time data into the model and outputs theoretical fault indicators; the multi-dimensional anomaly analysis module compares theoretical and measured fault indicators to generate an anomaly difference matrix; the fault probability localization module inputs the matrix into a spatial correlation network to generate a fault probability distribution map; and the adaptive maintenance strategy module configures maintenance parameters according to the distribution map. This system can improve the accuracy and real-time performance of fault prediction, ensuring the stable operation of the electronic control system in new energy vehicles.

Description

Technical Field

[0001] This invention relates to the field of new energy vehicle electronic control technology, specifically a new energy vehicle electronic control fault prediction system. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the electronic control system, as the core control unit of the vehicle, directly affects vehicle driving safety and user experience. Currently, the structure of new energy vehicle electronic control systems is becoming increasingly complex, encompassing multiple subsystems such as battery management, motor control, and vehicle control. These subsystems interact and monitor their status through numerous sensors. However, in actual operation, the electronic control system is susceptible to complex operating conditions, environmental factors, and component aging, leading to various potential faults. If these faults are not detected and addressed in a timely manner, they may cause power interruption, a sudden drop in range, or even safety accidents.

[0003] Current methods for handling electronic control system faults primarily rely on reactive repairs or scheduled maintenance. Reactive repairs only detect faults after they occur, leading to vehicle downtime, increased repair costs, and the potential for more serious safety hazards due to the escalation of the fault. While scheduled maintenance can detect faults to some extent in advance, its fixed cycle cannot be flexibly adjusted according to the actual operating conditions of the vehicle. It struggles to effectively identify some intermittent or early-stage faults, potentially resulting in over-maintenance or under-maintenance.

[0004] Existing fault prediction technologies largely rely on single sensor data or static models for analysis. Single sensor data cannot comprehensively reflect the overall operating status of the electronic control system, easily leading to misjudgments or omissions. Static models, on the other hand, struggle to adapt to the dynamic changes of the electronic control system under different operating conditions, and their prediction accuracy gradually decreases over time, failing to meet the real-time and accurate fault prediction requirements of new energy vehicles. Furthermore, in terms of fault location and maintenance strategy formulation, existing technologies lack precise distribution analysis of fault probabilities, making it difficult to quickly pinpoint the root cause of faults. Maintenance parameter configuration also largely depends on experience, failing to achieve personalized and adaptive maintenance solutions, further impacting fault handling efficiency and vehicle operational reliability. Summary of the Invention

[0005] The purpose of this invention is to provide a new energy vehicle electronic control fault prediction system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a new energy vehicle electronic control fault prediction system, the system comprising:

[0007] The real-time data acquisition module is used to acquire sensor data from the electronic control system of new energy vehicles in real time.

[0008] The dynamic fault prediction model building module constructs a dynamic fault prediction model based on historical fault data.

[0009] The fault difference calculation module inputs the real-time collected sensor data of the electronic control system into the dynamic fault prediction model and outputs theoretical fault indicators.

[0010] The multidimensional anomaly analysis module compares the theoretical fault indicators with the measured fault indicators in multiple dimensions to generate an anomaly difference matrix.

[0011] The fault probability localization module inputs the anomaly difference matrix into a spatial correlation network to generate a fault probability distribution map.

[0012] The adaptive maintenance strategy module configures maintenance parameters based on the fault probability distribution map.

[0013] Preferably, the dynamic fault prediction model construction module specifically includes:

[0014] The historical data feature mining unit performs multi-dimensional decomposition processing on historical fault data, extracts the energy ratio of steady-state and transient components of fault features, establishes a correlation matrix between sensor data and fault modes, and aligns fault sequence patterns under different operating conditions.

[0015] The dynamic fault prediction model generation unit inputs the processed historical fault data into the hybrid prediction network. The hybrid prediction network includes a time series prediction unit based on the equipment aging curve to generate basic fault prediction values, a feature correction network with embedded frequency domain attention mechanism to correct prediction deviations caused by sensor distortion, and a feature compensator to dynamically adjust prediction weights based on real-time sensor data.

[0016] The theoretical index calculation unit inputs the real-time collected sensor data of the electronic control system into the dynamic fault prediction model to obtain theoretical fault indices.

[0017] Preferably, the theoretical index calculation unit includes the following functions:

[0018] Based on adaptive filtering during the operation of the electronic control system, measurement noise caused by environmental interference is eliminated;

[0019] The correlation features of voltage, current, and temperature are integrated through a spatiotemporal feature fusion algorithm;

[0020] The output includes theoretical fault indicators within the normal operating condition fluctuation range, which are dynamically updated as the equipment ages.

[0021] Preferably, the multidimensional anomaly analysis module specifically includes:

[0022] The time-domain cumulative deviation calculation unit performs a sliding comparison between theoretical and measured fault indicators within a preset time window, uses a time series alignment algorithm to align asynchronously sampled fault sequences, calculates the cumulative deviation within each window, and generates a time-domain deviation vector.

[0023] The frequency domain energy shift detection unit decomposes the fault harmonic components of theoretical and measured indicators using a frequency domain decomposition algorithm, calculates the energy spectral density ratio, extracts the energy shift index of each harmonic, and constructs a frequency domain shift vector.

[0024] The sequence similarity evaluation unit matches the distance distribution between theoretical indices and measured sequences based on the structure matching algorithm, calculates the phase synchronization error at amplitude abrupt change points, quantifies the divergence measure of the interval distribution, and generates a sequence similarity vector.

[0025] The anomaly difference matrix generation unit concatenates the time-domain deviation vector, frequency-domain offset vector, and sequence similarity vector into tensors, and eliminates dimensional differences through feature importance-weighted normalization processing. The output is a third-order anomaly difference matrix with dimensions of device number multiplied by timestamp multiplied by difference type.

[0026] Preferably, the frequency domain energy shift detection unit specifically includes:

[0027] In the frequency domain analysis stage, the fault harmonic components of theoretical and measured indicators are first extracted. Multi-scale frequency band analysis is used to decompose each group of harmonic signals and extract the energy distribution characteristics within the preset sensitive frequency band interval. The sensitive frequency band interval covers the operating frequency range of typical electrical control equipment.

[0028] The energy density quantization unit quantifies the energy density of theoretical and measured indicators within the sensitive frequency band, and extracts the offset indicators of each order harmonic based on the relative offset between the two.

[0029] The frequency domain offset vector construction unit summarizes the energy offset results of all orders and constructs the frequency domain offset vector.

[0030] Preferably, the fault probability location module specifically includes:

[0031] The electronic control topology modeling unit constructs a topology diagram of the connection relationship between electronic control nodes based on vehicle location information, marks the line impedance parameters between each node, and superimposes the reverse current constraint condition of the power access point on the topology diagram to generate an electrified topology model including the impedance matrix and the node admittance matrix.

[0032] The anomaly propagation simulation unit maps the anomaly difference matrix to the corresponding node of the electronic control topology model, performs anomaly propagation inference based on graph neural network, and the calculation of the anomaly propagation inference includes calculating the attenuation factor of the abnormal current based on the node impedance parameters, capturing cross-regional anomaly correlation features through multi-head attention mechanism, and simulating the diffusion path of the abnormal current in the topology network using Monte Carlo method.

[0033] The probability distribution generation unit counts the frequency of abnormal current occurrences in simulated propagation for each line, calculates the abnormal current residence probability value in combination with the line impedance parameters, generates a fault probability distribution map covering the entire network, and marks the set of suspicious lines whose probability values ​​exceed the preset abnormal residence probability threshold.

[0034] The physical region positioning unit performs spatial clustering analysis on the fault probability distribution map, identifies abnormal probability clusters, and delineates the physical boundaries of the fault based on the vehicle location and the connection relationship between the electronic control topology.

[0035] Preferably, the fault probability localization module further includes outputs including suspicious device identifiers and abnormal propagation main paths;

[0036] The suspicious device identifier is based on the electronic control node connected to the suspicious line set, which binds the electronic control node to the actual equipment of the vehicle to form a suspicious device identifier set;

[0037] The main path of anomaly propagation is obtained by recording the node paths and their order in each round of propagation during the anomaly diffusion process in Monte Carlo simulation. Among all simulated paths, the frequency of occurrence of each path is counted, and the path sequence with the highest cumulative frequency is selected as the main path of anomaly propagation. The output sequence of the main path of anomaly propagation is a structured and ordered list of nodes.

[0038] Preferably, the adaptive maintenance strategy module specifically includes:

[0039] When the abnormal probability value of a certain area in the fault probability distribution map exceeds the preset abnormal residence probability threshold, a mode switching command is sent to the monitoring terminal to which the area belongs, and the voltage and current sampling frequency is increased to a multiple of the original frequency. Simultaneously, the real-time tracking mode of harmonic components is enabled to capture the phase change of harmonics in stages, and transient event detectors are deployed on the edge side to record voltage sag waveform segments.

[0040] The impedance disturbance test execution unit applies multi-band impedance disturbances to the adjacent node devices with the largest abnormal probability gradient changes in the fault probability distribution map.

[0041] Preferably, the multi-band impedance disturbance includes:

[0042] A sweep frequency test signal is injected through a controllable current source;

[0043] The theoretical impedance response calculation unit calculates the theoretical response impedance spectrum based on the topology model and impedance matrix;

[0044] The measured impedance response acquisition unit records the voltage of each node after the test signal is injected, and obtains the measured response impedance spectrum.

[0045] The abnormal offset determination unit calculates the distance between the theoretical response impedance spectrum and the measured response impedance spectrum, compares the distance between the measured impedance spectrum and the theoretical impedance spectrum, calculates the percentage of abnormal node offset, and marks the abnormal node as an impedance abnormality associated device when the percentage of abnormal node offset exceeds the second threshold.

[0046] Preferably, the system further includes an iterative optimization module, which specifically includes:

[0047] The data stream parsing unit parses multi-source heterogeneous sensor data carrying spatiotemporal labels in the network data stream and generates dynamic priority parameters for the constraint optimization dimension based on the spatiotemporal aggregation degree of each data packet on the transmission path.

[0048] The baseline encoding extraction unit extracts the baseline encoding of the optimized path corresponding to the cross-source query pattern in the historical record of network data streams. The baseline encoding is generated by matching the spatiotemporal coverage complementarity of different data packet transmission paths, and the baseline encoding is used as the gene expression template of the optimized path in the initial population.

[0049] The co-evolutionary unit performs co-evolution of the optimized path at the network edge nodes, driving the optimized path to iterate in the direction of minimizing conflict. Each iteration generates a path offset and updates the network adaptation parameters of the gene expression template. An adaptive evaluation function is constructed based on the correlation between the dynamic priority parameters and the data packet transmission delay.

[0050] The preloading trigger unit triggers the preloading of heterogeneous data packets according to the direction and magnitude of the path offset. The preloaded data is a set of data packets in the adjacent nodes of the network topology that match the spatiotemporal coverage area of ​​the current optimized path. The protocol compatibility of the data packets is verified by the format conversion rules in the gene expression template.

[0051] The feedback generation unit generates feedback parameters by using the conflict index output by the adaptive evaluation function and the verification results of the preloaded data, and dynamically corrects the mapping relationship between the dynamic priority parameters and the gene expression template.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] This new energy vehicle electronic control fault prediction system effectively addresses many shortcomings of existing technologies in fault prediction, location, and maintenance through the collaborative work of multiple modules. The real-time data acquisition module can acquire data from various sensors in the electronic control system in real time, comprehensively capturing system operating status information. This avoids the limitations of data from a single sensor and provides a complete and accurate data foundation for subsequent fault analysis. This allows fault prediction to be based on the overall system operation, reducing misjudgments or omissions caused by incomplete data.

[0054] The dynamic fault prediction model building module builds a model based on historical fault data, making full use of the fault patterns and characteristics contained in the historical data. Compared with the static model, the dynamic model can better adapt to the operating changes of the electronic control system under different working conditions. With the accumulation and updating of historical data, the model can continuously optimize its prediction capabilities and always maintain high prediction accuracy, ensuring the effective identification of early potential faults. This allows staff to detect abnormalities before the fault actually occurs, buying time for subsequent handling.

[0055] The fault difference calculation module inputs the real-time collected data into the dynamic fault prediction model and outputs theoretical fault indicators, providing a clear reference benchmark for subsequent anomaly analysis. This makes the subsequent comparative analysis of real-time data and theoretical indicators more targeted, enabling the rapid detection of deviations between real-time operating status and theoretical normal status, and timely detection of potential faults.

[0056] The multidimensional anomaly analysis module generates an anomaly difference matrix by comparing theoretical and measured fault indicators in multiple dimensions. This matrix can intuitively present the abnormal situation of the electronic control system from multiple dimensions. It not only covers the anomaly of a single parameter, but also reflects the correlation anomaly between various parameters. This helps staff to have a more comprehensive and in-depth understanding of the abnormal characteristics of the system, avoiding the one-sidedness of traditional single-dimensional analysis and laying the foundation for accurate fault location.

[0057] The fault probability localization module inputs the anomaly difference matrix into the spatial correlation network to generate a fault probability distribution map. This distribution map can clearly show the probability of faults occurring in each part. Based on the distribution map, staff can quickly locate high-probability fault areas and accurately pinpoint the root cause of the fault without having to conduct a comprehensive inspection of the entire electrical control system. This significantly shortens the fault localization time and improves fault handling efficiency.

[0058] The adaptive maintenance strategy module configures maintenance parameters based on the fault probability distribution map, enabling it to develop personalized maintenance plans for different fault probability areas and fault types. This eliminates the drawbacks of traditional fixed-cycle maintenance and experience-based maintenance. For areas with high fault probability, targeted maintenance parameters can be configured first to strengthen maintenance efforts; for areas with low fault probability, the maintenance frequency and intensity can be adjusted appropriately to avoid over-maintenance. This ensures maintenance effectiveness while reducing maintenance costs, extending the lifespan of the electronic control system, and improving the overall operational reliability and user experience of new energy vehicles. Attached Figure Description

[0059] Figure 1 This is a timing diagram of the new energy vehicle electronic control fault prediction system described in this invention;

[0060] Figure 2 Flowchart illustrating the working principle of the module for building dynamic fault prediction models;

[0061] Figure 3 This is a flowchart illustrating the working principle of the multidimensional anomaly analysis module.

[0062] Figure 4 A flowchart illustrating the working principle of the fault probability localization module. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1 This invention provides a new energy vehicle electronic control fault prediction system, the system comprising:

[0065] The dynamic fault prediction model construction module trains and generates a prediction model based on historical fault data, which outputs theoretical fault indicators. The fault difference calculation module inputs real-time sensor data into the model and outputs theoretical fault indicator values. The multi-dimensional anomaly analysis module compares the theoretical fault indicators with the measured fault indicators in multiple dimensions, generating an anomaly difference matrix that includes time-domain, frequency-domain, and sequence features. The fault probability localization module inputs this matrix into a spatial correlation network and, combined with the electronic control topology, simulates the anomaly propagation path to generate a fault probability distribution map with labeled probability values. The adaptive maintenance strategy module adjusts maintenance parameters based on the probability distribution map, including increasing the sampling frequency, harmonic tracking, and impedance disturbance testing.

[0066] Example 1: See Figure 2The system details the operational procedures of the dynamic fault prediction model construction module and the theoretical index calculation unit. First, it initiates a historical data feature mining unit to perform multi-dimensional decomposition processing on historical fault data stored in the database. This historical data includes parameters such as voltage, current, and temperature recorded by various sensors under different operating conditions, along with their corresponding fault event records. During processing, signal decomposition technology is used to separate the data sequence of each sensor into steady-state and transient components, calculating the energy ratio of each. For example, for the motor controller current signal, its steady-state component reflects the current characteristics under normal load, while the transient component contains current fluctuation information during start-up, shutdown, or load abrupt changes. These energy ratios form a mapping relationship with specific fault types, constructing a correlation matrix between sensor data and fault modes. Due to the diversity of vehicle operating conditions, this unit also employs a mode alignment algorithm to align fault sequences at different vehicle speeds and ambient temperatures along the time axis, eliminating the impact of operating condition differences on feature extraction.

[0067] After historical data feature mining is completed, the dynamic fault prediction model generation unit begins operation. This unit adopts a hybrid prediction network architecture, with its core comprising three collaborative sub-units. The time-series prediction unit based on equipment aging curves first analyzes the patterns of equipment performance degradation in historical data, establishing an aging trend model with operating time as the independent variable. This model outputs basic fault prediction values ​​for key components of the electronic control system (such as power devices and capacitors), with the predicted values ​​exhibiting non-linear changes over time. A feature correction network embedding a frequency domain attention mechanism then calibrates the basic prediction values. This network focuses on addressing signal distortion issues caused by sensor loosening, electromagnetic interference, etc., identifying abnormal frequency components through frequency domain analysis and calculating their impact on the prediction results. The feature compensator, as a dynamic adjustment component, continuously receives real-time sensor data streams and dynamically adjusts the prediction weights of each sensor based on data quality assessment results. For example, when abnormal fluctuations are detected in the data from a temperature sensor, the weight of that sensor in the prediction model is automatically reduced, and compensation prediction is performed using data from a more correlated current sensor.

[0068] The theoretical index calculation unit, serving as the application output of the model, receives sensor data from the electronic control system transmitted by the real-time data acquisition module. This unit incorporates an adaptive filter processor, employing different filtering parameters for different vehicle operating phases (e.g., stationary, acceleration, constant speed, deceleration). During vehicle acceleration, the filter primarily suppresses high-frequency noise caused by powertrain vibrations; during constant speed, it mainly eliminates measurement drift caused by ambient temperature fluctuations. The filtered data is then fed into a spatiotemporal feature fusion algorithm, which analyzes not only the data characteristics of individual sensors but also the correlations between multiple sensors. For example, it synchronously analyzes the balance of the motor's three-phase current, the phase relationship between the DC bus voltage and the inverter output current, and the conduction delay between the radiator temperature and the junction temperature of power devices. By establishing a multi-parameter correlation model, the algorithm outputs theoretical fault indices encompassing the fluctuation range of normal operating conditions. These indices exist in the form of dynamic thresholds, with upper and lower limits automatically adjusted over time. For example, for capacitor value decay faults, the upper threshold of the theoretical fault index gradually decreases over operating time, reflecting performance changes after equipment aging.

[0069] The entire implementation process employs a distributed computing architecture. Historical data feature mining is completed on cloud servers, dynamic fault prediction model generation is deployed on edge computing nodes, and theoretical indicator calculations are performed in real time on the in-vehicle terminal. A model update mechanism periodically uploads new fault data collected by the in-vehicle terminal to the cloud to refresh the historical database and optimize prediction model parameters. This architecture ensures both sufficient model training and meets the low-latency requirements of real-time prediction. Encryption protocols are used to protect fault data security during data transmission, and operation logs are recorded for all processing steps for auditing and traceability.

[0070] The model's interpretability is achieved through feature importance assessment. The system records the contribution of each sensor feature in the prediction process and displays the main influencing factors in a visual format. When the prediction results are output, a brief explanation of the judgment criteria is provided simultaneously, such as "the current fault prediction is mainly based on the abnormal temperature rise rate and changes in current harmonic content." This design facilitates technicians' understanding of the model's decision-making logic and provides a reference for subsequent maintenance decisions. The fault prediction model is validated using a time-series cross-validation method, dividing historical data into training and test sets in chronological order to ensure that the validation process conforms to the temporal characteristics of actual data generation. Validation metrics include prediction accuracy, false alarm rate, and early warning time. All metric calculations take into account the natural performance degradation caused by equipment aging. Model parameter optimization employs a heuristic search algorithm to control computational complexity while ensuring prediction accuracy, meeting the computing power constraints of the vehicle terminal.

[0071] Example 2: See Figure 3The system involves the complete workflow of a multi-dimensional anomaly analysis module. This module is responsible for comparing and analyzing theoretical and measured fault indicators at multiple levels and from multiple perspectives, ultimately generating a difference matrix that comprehensively reflects the abnormal state of the system. Upon system startup, the time-domain cumulative deviation calculation unit begins operation. This unit receives the theoretical fault indicator sequence from the upstream module and the real-time acquired measured fault indicator sequence. Due to potential asynchronous issues in data acquisition, the unit first uses a time series alignment algorithm to perform time-series matching between the two sets of sequences. The algorithm aligns the two sequences on the time axis by finding the optimal path, eliminating comparison errors caused by inconsistent sampling time points. After alignment, the unit sets a configurable sliding time window, which moves gradually along the time axis. Within each window, the cumulative deviation between the theoretical and measured values ​​is calculated. This calculation not only considers simple differences but also introduces a weighted accumulation mechanism, assigning higher weights to persistent deviations and appropriately reducing the weights to occasional fluctuations. After each window's calculation, a deviation value is generated, and the calculation results of all windows are arranged in chronological order to form a time-domain deviation vector. This vector reflects the accumulation and changing trend of system anomalies over time.

[0072] The frequency domain energy shift detection unit performs analysis simultaneously. This unit performs frequency domain transformation on both theoretical and measured fault index data, extracting the fault harmonic components. Multi-scale frequency band analysis technology is employed to decompose the harmonic signal into different frequency sub-bands. The system's preset sensitive frequency bands cover the typical operating frequency range of key equipment in the electronic control system, including motor drive frequencies, switching power supply operating frequencies, and their harmonic regions. Within each sensitive frequency band, the unit calculates the energy density of both theoretical and measured indices. Energy density quantization uses an integral method to calculate the total energy of the signal within a specific frequency band. Subsequently, the unit calculates the relative shift between the two energy densities, obtained through ratio calculation, reflecting the energy change of the measured signal relative to the theoretical expectation. Each order of harmonics generates an independent shift index, and all indices are arranged from low to high frequency, forming a frequency domain shift vector. This vector reveals the abnormal distribution characteristics of the system in the frequency dimension.

[0073] The sequence similarity assessment unit analyzes from a structural feature perspective. This unit employs a structural matching algorithm to compare the overall morphological similarity between theoretical and measured sequences. The algorithm first identifies key feature points in both sequences, including extreme points, inflection points, and abrupt change points. Phase synchronization errors are calculated for these feature points to analyze the degree of matching at corresponding temporal positions. Simultaneously, the algorithm calculates the divergence metric of the sequence interval distribution to assess the differences in amplitude distribution between the two sequences. To comprehensively assess similarity, the unit also introduces a dynamic time warping algorithm to overcome the scaling and deformation of sequences along the time axis. All similarity assessment results are quantified into a multi-dimensional sequence similarity vector, which reflects the abnormal state of the system from a structural feature perspective.

[0074] The anomaly difference matrix generation unit is responsible for integrating the results of the first three types of analysis. This unit receives the time-domain deviation vector, frequency-domain offset vector, and sequence similarity vector, and performs tensor concatenation. The concatenated data contains three dimensions: device number, timestamp, and difference type. Due to the differences in the units and numerical ranges of different vectors, the unit performs feature importance-weighted normalization. During the processing, different weight coefficients are assigned to different types of difference features based on historical data analysis results. Features with higher importance (such as persistent time-domain deviations or energy shifts in specific frequency bands) retain more original information during normalization. The final generated anomaly difference matrix is ​​a third-order tensor, whose dimensional structure is device number multiplied by timestamp multiplied by difference type.

[0075] The entire implementation process adopts a pipelined architecture, with the four units connected by data buffers to ensure the continuity and stability of the processing. A quality control mechanism is implemented to verify the integrity of the output data from each unit, automatically triggering recalculation upon detecting data anomalies. Timestamps for all processing stages are strictly synchronized to ensure that analytical results from different dimensions accurately correspond to the same point in time. During matrix generation, the system records the parameter settings and intermediate results for each processing step, forming a complete processing log for easy subsequent traceability and analysis.

[0076] The matrix data is stored in a compressed format, minimizing storage space while retaining key information. The data access interface supports fast querying and retrieval by device, time range, or difference type. The system also provides visualization tools to display the 3D matrix data in various view formats, including time series plots, spectrum plots, and 3D scatter plots, assisting technicians in intuitively understanding system anomaly characteristics. During processing, the system monitors computing resource usage in real time and dynamically adjusts the analysis granularity. When the system load is high, the time window step size is appropriately increased or the number of frequency bands analyzed is reduced; when the load is low, the analysis accuracy and granularity are improved. This adaptive mechanism ensures stable processing performance under different operating conditions. All algorithm parameters can be adjusted through configuration files, enabling the system to adapt to the characteristics of different vehicle models and configurations of electronic control systems. Parameter adjustment records are included in the version management system to ensure traceability of each change. The final generated anomaly difference matrix not only contains numerical difference data but also includes metadata information, including data source, processing time, and confidence index. This metadata provides necessary background information and usage guidance for subsequent modules using the matrix data. The matrix data output adopts a standardized format to ensure seamless integration with the downstream fault probability location module.

[0077] Example 3: See Figure 4 This document describes the complete workflow of the fault probability localization module. Based on the anomaly difference matrix output by the multi-dimensional anomaly analysis module and combined with the topology of the electronic control system, this module generates a fault probability distribution map by simulating anomaly propagation paths, achieving precise location of faulty equipment. The electronic control topology modeling unit is initiated first. This unit reads the vehicle's electronic control system configuration information, including the type, location, and connection relationships of all electronic control nodes. Based on this information, a directed graph structure is constructed, where nodes represent electronic control devices (such as motor controllers, battery management systems, and on-board chargers), and edges represent electrical connection lines. Each edge is labeled with actual line impedance parameters, including resistance, inductance, and capacitance values, all derived from the equipment's technical specifications. The unit also introduces reverse current constraints at power access points, considering the actual flow direction of current in the power grid. Based on this data, the unit generates two key matrices: an impedance matrix Z describing the impedance characteristics between nodes, and a node admittance matrix Y describing the current-voltage relationship between nodes. These two matrices comprehensively characterize the electrical characteristics of the electronic control system.

[0078] The anomaly propagation simulation unit receives the anomaly difference matrix from upstream and maps it to the corresponding nodes in the topology graph. The mapping process is based on the correspondence between device numbers, assigning the anomaly difference value of each device to the corresponding node in the topology graph. Then, an anomaly propagation inference algorithm based on a graph neural network is initiated. The algorithm first calculates the attenuation factor of the anomaly current based on the node impedance parameters; this factor characterizes the degree of attenuation of the anomaly signal during transmission.

[0079]

[0080] Where: α ij Z represents the abnormal decay factor from node i to node j. ij Ri represents the path impedance magnitude from node i to node j, and R0 is the system characteristic impedance reference value. The algorithm employs a multi-head attention mechanism to capture cross-regional anomaly correlation characteristics, with each attention head focusing on a specific type of anomaly propagation pattern. The Monte Carlo method is used to simulate the random diffusion path of anomalous currents in the topological network. Each simulation randomly selects the propagation direction based on the anomaly value and connection relationship of the current node, recording the propagation path and the nodes traversed.

[0081] The probability distribution generation unit statistically analyzes all Monte Carlo simulation results and calculates the frequency of abnormal currents occurring on each line during the simulation. Combining this with line impedance parameters, it calculates the residence probability of abnormal currents on the lines. Impedance matching is considered in the calculation; lines with lower impedance are more prone to accumulating abnormal currents. A fault probability distribution map covering the entire network is generated. This map is based on the topology, with each node and edge labeled with its calculated probability value. The system sets a preset abnormal residence probability threshold and automatically identifies a set of suspicious lines whose probability values ​​exceed the threshold. These suspicious lines are sorted from highest to lowest probability value, forming a priority list.

[0082] The physical area location unit performs spatial clustering analysis on the fault probability distribution map, using a density clustering algorithm to identify anomalous probability clusters. The algorithm clusters nodes based on the electrical distance and the similarity of their probability values; nodes with closer electrical distances and more similar probability values ​​are grouped into the same cluster. The physical boundary of each cluster is delineated based on the vehicle's actual location information and the electronic control topology connections. Boundary delineation considers the physical installation location and electrical connections of the equipment to ensure physical spatial continuity within each cluster.

[0083] The output of this module contains two structured datasets: a set of suspicious device identifiers and a sequence of main propagation paths for anomalies. Suspicious device identifiers are generated by binding and matching the electronically controlled nodes connected to the suspicious path set with the actual equipment on the vehicle. Each identifier includes the device number, device type, and probability value. The main propagation path for anomalies is generated by recording the node paths and their order during each round of propagation in the Monte Carlo simulation, counting the frequency of each path across all simulated paths, and selecting the path sequence with the highest cumulative frequency as the main path. The output path sequence is a structured, ordered list of nodes, containing the node number, type, and sequential position within the path.

[0084] The entire implementation process adopts a layered processing architecture. Electrical control topology modeling and anomaly propagation simulation are performed on edge computing nodes, while probability distribution generation and physical region localization are completed on cloud servers. Data transmission uses a compressed encoding format; topology information uses a graph-based encoding algorithm, and probability data uses a floating-point compressed format. A version management mechanism is established, recording any changes to the topology model in a change log to ensure consistency between the model and the actual situation. Multiple verification mechanisms are introduced during processing. After topology modeling is completed, the correctness of the impedance and admittance matrices is verified through circuit simulation. During anomaly propagation simulation, reasonableness check rules are set to prevent propagation paths that violate circuit laws. In the probability calculation stage, cross-validation methods are used to ensure the reliability of statistical results. All verification results are recorded in the audit log.

[0085] The visualization component converts the fault probability distribution map into a 3D visualization model, where node size represents the probability value, edge thickness represents the abnormal current intensity, and color intensity represents the difference in anomaly type. Technicians can view detailed probability data for any node and the associated anomaly propagation path through the interactive interface. The system supports historical data playback, allowing for the reproduction of the dynamic process of anomaly propagation.

[0086] The output data uses a standardized JSON format and includes complete metadata information. Each device record in the suspicious device identifier set includes the device ID, device name, subsystem to which it belongs, installation location, probability value, and timestamp. The anomaly propagation main path sequence record includes the path ID, path score, node list, and the anomaly contribution of each node. This data is pushed to the downstream adaptive maintenance strategy module via a message queue service and is also persistently stored in a time-series database for subsequent analysis.

[0087] The module continuously monitors changes in the electrical control system's status during operation. When a topology change is detected (such as equipment replacement or line modification), a topology model update process is automatically triggered. The update process includes model verification, data migration, and historical data re-labeling to ensure the accuracy and continuity of fault probability calculations. The system also establishes a feedback mechanism to compare actual maintenance results with predicted results to optimize the probability calculation model parameters.

[0088] Example 4: This describes the specific operation flow of the adaptive maintenance strategy module. Based on the analysis results of the fault probability distribution map, this module dynamically adjusts system monitoring parameters and performs proactive diagnostic tests. When the abnormal probability value of a certain area in the fault probability distribution map exceeds a preset abnormal dwell probability threshold (e.g., 0.85), the system automatically triggers a mode switching command. This command is sent to the monitoring terminal belonging to the target area, executing three key operations: increasing the voltage and current sampling frequency from the fundamental frequency of 1kHz to 4kHz; enabling the real-time harmonic component tracking mode, performing phase change detection for harmonics below the 13th order; and deploying a transient event detector on the edge computing node, configured to capture voltage sag events lasting longer than 50μs, and recording waveform segments for 10 cycles before and after the event.

[0089] The impedance disturbance test execution unit applies multi-band impedance disturbances to adjacent nodes with the largest changes in abnormal probability gradient in the fault probability distribution diagram. The test process adopts a four-stage protocol: In the first stage, a frequency sweep test signal is injected through a controllable current source, covering a frequency range of 100Hz to 10kHz, selecting 128 test frequency points at logarithmic intervals, with a constant amplitude current signal injected for 200ms at each frequency point. In the second stage, the theoretical response impedance spectrum is calculated based on the impedance matrix in the electrification topology model, taking into account line distribution parameters and the equivalent circuit model of the equipment. In the third stage, the voltage response of each node is synchronously acquired during the test signal injection, and the impedance amplitude and phase of the corresponding frequency points are extracted by Fourier transform to form the measured response impedance spectrum. In the fourth stage, abnormal offset is determined by calculating the dynamic time warping distance between the theoretical spectrum and the measured spectrum. This distance metric reflects the overall shape difference between the two impedance curves. The distance metric between the impedance spectrum of the equipment under test and the theoretical reference spectrum is compared, and the percentage of abnormal node offset is calculated according to the following formula (see Table 1).

[0090] Table 1: Impedance Spectrum Anomaly Shift Judgment Table.

[0091] Equipment Number Test frequency band (Hz) Theoretical impedance magnitude (Ω) Measured impedance modulus (Ω) Phase offset (degrees) Offset percentage (%) status flags EPC-07 1250-1300 0.85 1.12 +15.6 31.8 abnormal EPC-07 3250-3300 1.20 1.18 -2.3 1.7 normal EPC-12 7800-7850 2.05 3.11 +28.9 51.7 abnormal EPC-15 9500-9550 0.98 0.95 -1.2 3.1 normal

[0092] When the percentage offset of a specific frequency band exceeds a second threshold (e.g., 30%), the system automatically marks the device as an impedance anomaly associated device. The marking results are spatially correlated with the fault probability distribution map. If the same device is marked as abnormal in both diagnostic methods, its maintenance priority is raised to the highest level.

[0093] The data stream generated during mode switching employs a hierarchical storage strategy: routine monitoring data is retained for 72 hours, high-frequency sampling data for 24 hours, and transient event waveform fragments are permanently stored in a dedicated database. Data indexing establishes multi-dimensional associations, supporting queries by time range, device number, and anomaly type. Harmonic phase abrupt change records contain the phase angle change trajectory of the fifth harmonic component, recording a set of phase angle data every 10ms. An event marker is generated when a phase jump exceeding 15 degrees between adjacent sampling points is detected.

[0094] Impedance spectrum data generated from impedance disturbance testing is stored as structured records, each containing complete impedance characteristics at 128 frequency points. The system establishes a historical baseline database of impedance spectra, comparing each test result longitudinally with the impedance spectrum under normal operating conditions to calculate the offset trend in characteristic frequency bands. If the same device shows an offset exceeding 20% ​​in the same frequency band during three consecutive tests, a preventative maintenance alert is triggered. Safety protection mechanisms are implemented during the testing process. The current injection amplitude is dynamically limited based on the device's rated current, not exceeding 20% ​​of the rated value. The injected signal is synchronized with the device's operating status in real time, avoiding switching instants of power devices. Equipment temperature changes are monitored during testing; if the temperature rise rate exceeds 5℃ / min, the test is immediately terminated. All operation commands are transmitted via encrypted channels, and execution results are returned to the central system to generate audit logs.

[0095] The maintenance strategy decision engine integrates mode switching monitoring results and impedance test data to generate four types of maintenance instructions: For equipment with detected harmonic phase abrupt changes, it is recommended to check the gate drive circuit of the power device; for equipment with impedance anomalies, it is recommended to check the connector contact status and capacitor aging; for equipment with both transient events and impedance shifts, it is recommended to replace the entire power module; for equipment without detected anomalies, the next inspection interval is extended. The instruction set is distributed to the maintenance terminal via the vehicle-to-everything (V2X) network, simultaneously updating the equipment's predictive model parameters. After performing maintenance operations, maintenance personnel must enter the actual faulty component and handling measures into the system. This data is used to verify the accuracy of predictions and optimize the settings of the abnormal dwell probability threshold and impedance shift threshold. After each maintenance task is completed, the system automatically re-evaluates the equipment's failure probability curve for the next three operating cycles and dynamically adjusts the trigger frequency of the monitoring strategy.

[0096] Example 5: The complete workflow of the iterative optimization module is described. This module is responsible for continuously optimizing the multi-source heterogeneous sensor data streams transmitted in the system, improving data transmission efficiency and processing performance through dynamic path adjustment and data preloading mechanisms. The data stream parsing unit first performs real-time parsing of the network data streams. These data streams carry spatiotemporal tags and include data from various heterogeneous data sources such as voltage sensors, current sensors, and temperature sensors. During parsing, the unit analyzes the spatiotemporal clustering characteristics of each data packet along the transmission path, i.e., the temporal arrival distribution and spatial source distribution characteristics of the data packets. Based on these characteristics, a dynamic priority parameter is generated. The calculation of this parameter considers the urgency of the data packets, the data volume, and the load status of the destination node. High-priority data packets typically have tighter timing constraints and higher real-time requirements; for example, fault warning data is assigned the highest priority.

[0097] The baseline encoding extraction unit extracts the baseline encodings of optimized paths corresponding to cross-source query patterns from the historical records of network data streams. These encodings are generated by analyzing the spatiotemporal coverage complementarity of different data packet transmission paths, i.e., finding combinations of data packets that can form complementary advantages in transmission time and spatial paths. The extracted baseline encodings serve as gene expression templates for optimized paths in the initial population, with each template representing a validated effective transmission path pattern. These templates contain information such as path node sequences, transmission timing arrangements, and protocol conversion rules.

[0098] The co-evolutionary unit performs a co-evolutionary process to optimize paths at network edge nodes. This unit drives the optimization path iteratively towards the direction that minimizes conflict, generating a path offset and updating the network adaptation parameters of the gene expression template in each iteration. The fitness evaluation function is constructed based on the correlation between dynamic priority parameters and packet transmission delay, and its expression is:

[0099]

[0100] in: This represents the output value of the adaptive evaluation function. This represents the dynamic priority weight of the i-th data packet. This represents the theoretical transmission delay of the i-th data packet. This represents the actual additional latency of the i-th data packet. A higher function value indicates better path optimization. During each iteration, the system adjusts the direction and magnitude of the path offset based on the adaptive evaluation results.

[0101] The preloading trigger unit initiates a heterogeneous data packet preloading operation based on the direction and magnitude of the path offset. The preloaded data is selected from a set of data packets in adjacent nodes of the network topology that match the spatiotemporal coverage area of ​​the current optimized path. These data packets undergo protocol compatibility verification using format conversion rules in the gene expression template, ensuring that data packets from different protocols can be transmitted collaboratively. The preloading mechanism employs a predictive loading strategy, predicting potentially needed subsequent data packets based on historical data transmission patterns and caching them in advance at edge nodes.

[0102] The feedback generation unit generates feedback parameters based on the conflict index output by the adaptive evaluation function and the validation results of preloaded data. The conflict index reflects the degree of resource contention on the data transmission path, while the preload validation results indicate the success rate of packet protocol conversion. These feedback parameters are used to dynamically correct the mapping relationship between dynamic priority parameters and gene expression templates, forming a closed-loop optimization mechanism. The feedback adjustment process employs a gradual adjustment strategy, fine-tuning the mapping parameters each time based on the latest transmission results.

[0103] The entire implementation process adopts a distributed architecture. Data stream parsing and baseline encoding extraction are completed at the central node, while co-evolution and preloading triggering are executed at the edge nodes. The system establishes a unified time synchronization mechanism, with all nodes using Coordinated Universal Time (UTC) timestamps to ensure consistency of spatiotemporal labels. Data transmission employs compression encoding technology to reduce network bandwidth consumption.

[0104] During optimization, the system maintains a path optimization knowledge base, storing historically successful optimization path cases and corresponding performance metrics. Each time a path optimization is performed, the system first searches the knowledge base for optimization solutions for similar scenarios as initial solutions, accelerating the optimization convergence process. The knowledge base is periodically cleaned and optimized, retaining only the most representative optimization cases. All data packet transmissions are encrypted, and gene expression template transmissions are verified using digital signatures. Preloaded data has an expiration period; unused preloaded data that expires is automatically deleted. A transmission quality monitoring system is established to monitor metrics such as data packet loss rate, transmission latency, and bandwidth utilization in real time.

[0105] The visual monitoring interface displays the real-time status of the optimization process, including the current transmission path topology, packet traffic distribution, and optimization progress indicators. Technical personnel can view historical optimization records, analyze optimization trend, and manually adjust optimization parameter weights. The iterative optimization module establishes data interfaces with other system modules. It receives data priority information from the fault prediction module and outputs the optimized data transmission plan to the maintenance strategy module. All interface data uses a standardized format to ensure inter-module collaboration. When a network topology change or node failure is detected, the optimization process is automatically reinitialized. In case of data transmission anomalies, a backup transmission path is activated to ensure reliable transmission of critical data. All abnormal events are logged in detail for subsequent analysis and system improvement.

[0106] The optimization effect is evaluated using a multi-indicator comprehensive evaluation method, including indicators such as data transmission timeliness, bandwidth utilization, and node load balancing. The evaluation results are fed back into the optimization algorithm to adjust the parameter weights of the adaptive evaluation function. The system periodically generates optimization performance reports, showing the optimization effect and improvement suggestions for each time period.

[0107] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 process, method, article, or apparatus.

[0108] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fault prediction system for electronic control systems of new energy vehicles, comprising: The real-time data acquisition module is used to acquire sensor data from the electronic control system of new energy vehicles in real time. The dynamic fault prediction model building module constructs a dynamic fault prediction model based on historical fault data. The fault difference calculation module inputs the real-time collected sensor data of the electronic control system into the dynamic fault prediction model and outputs theoretical fault indicators. The multidimensional anomaly analysis module compares the theoretical fault indicators with the measured fault indicators in multiple dimensions to generate an anomaly difference matrix, specifically including: The time-domain cumulative deviation calculation unit performs a sliding comparison between theoretical and measured fault indicators within a preset time window, uses a time series alignment algorithm to align asynchronously sampled fault sequences, calculates the cumulative deviation within each window, and generates a time-domain deviation vector. The frequency domain energy shift detection unit decomposes the fault harmonic components of theoretical and measured indicators using a frequency domain decomposition algorithm, calculates the energy spectral density ratio, extracts the energy shift index of each harmonic, and constructs a frequency domain shift vector. The sequence similarity evaluation unit matches the distance distribution between theoretical indices and measured sequences based on the structure matching algorithm, calculates the phase synchronization error at amplitude abrupt change points, quantifies the divergence measure of the interval distribution, and generates a sequence similarity vector. The anomaly difference matrix generation unit concatenates the time-domain deviation vector, frequency-domain offset vector, and sequence similarity vector into tensors, and eliminates dimensional differences through normalization processing weighted by feature importance. The output is a third-order anomaly difference matrix with dimensions of device number multiplied by timestamp multiplied by difference type. The frequency domain energy shift detection unit specifically includes: In the frequency domain analysis stage, the fault harmonic components of theoretical and measured indicators are first extracted. Multi-scale frequency band analysis is used to decompose each group of harmonic signals and extract the energy distribution characteristics within the preset sensitive frequency band interval. The sensitive frequency band interval covers the operating frequency range of typical electrical control equipment. The energy density quantization unit quantifies the energy density of theoretical and measured indicators within the sensitive frequency band, and extracts the offset indicators of each order harmonic based on the relative offset between the two. The frequency domain offset vector construction unit summarizes the energy offset results of all orders and constructs the frequency domain offset vector. The fault probability localization module inputs the anomaly difference matrix into a spatial correlation network to generate a fault probability distribution map. The adaptive maintenance strategy module configures maintenance parameters according to the fault probability distribution map. Specifically, it includes: when the abnormal probability value of a certain area in the fault probability distribution map exceeds the preset abnormal residence probability threshold, it sends a mode switching command to the monitoring terminal to which the area belongs, executes to increase the voltage and current sampling frequency to a multiple of the original frequency, simultaneously enables the real-time tracking mode of harmonic components, captures the phase change of harmonics, deploys transient event detectors on the edge side, and records voltage sag waveform segments. The impedance disturbance test execution unit applies multi-band impedance disturbances to the adjacent node devices with the largest changes in abnormal probability gradient in the fault probability distribution map. The multi-band impedance disturbance includes: A sweep frequency test signal is injected through a controllable current source; The theoretical impedance response calculation unit calculates the theoretical response impedance spectrum based on the topology model and impedance matrix; The measured impedance response acquisition unit records the voltage of each node after the test signal is injected, and obtains the measured response impedance spectrum. The abnormal offset determination unit calculates the distance between the theoretical response impedance spectrum and the measured response impedance spectrum, compares the distance between the measured impedance spectrum and the theoretical impedance spectrum, calculates the percentage of abnormal node offset, and marks the abnormal node as an impedance abnormality associated device when the percentage of abnormal node offset exceeds the second threshold.

2. The new energy vehicle electronic control fault prediction system according to claim 1, characterized in that, The dynamic fault prediction model construction module specifically includes: The historical data feature mining unit performs multi-dimensional decomposition processing on historical fault data, extracts the energy ratio of steady-state and transient components of fault features, establishes a correlation matrix between sensor data and fault modes, and aligns fault sequence patterns under different operating conditions. The dynamic fault prediction model generation unit inputs the processed historical fault data into the hybrid prediction network. The hybrid prediction network includes a time series prediction unit based on the equipment aging curve to generate basic fault prediction values, a feature correction network with embedded frequency domain attention mechanism to correct prediction deviations caused by sensor distortion, and a feature compensator to dynamically adjust prediction weights based on real-time sensor data. The theoretical index calculation unit inputs the real-time collected sensor data of the electronic control system into the dynamic fault prediction model to obtain theoretical fault indices.

3. The new energy vehicle electronic control fault prediction system according to claim 2, characterized in that, The theoretical index calculation unit includes the following steps: Based on adaptive filtering during the operation of the electronic control system, measurement noise caused by environmental interference is eliminated; The correlation features of voltage, current, and temperature are integrated through a spatiotemporal feature fusion algorithm; The output includes theoretical fault indicators within the normal operating condition fluctuation range, which are dynamically updated as the equipment ages.

4. The new energy vehicle electronic control fault prediction system according to claim 1, characterized in that, The fault probability location module specifically includes: The electronic control topology modeling unit constructs a topology diagram of the connection relationship between electronic control nodes based on vehicle location information, marks the line impedance parameters between each node, and superimposes the reverse current constraint condition of the power access point on the topology diagram to generate an electrified topology model including the impedance matrix and the node admittance matrix. The anomaly propagation simulation unit maps the anomaly difference matrix to the corresponding node of the electronic control topology model, performs anomaly propagation inference based on graph neural network, and the calculation of the anomaly propagation inference includes calculating the attenuation factor of the abnormal current based on the node impedance parameters, capturing cross-regional anomaly correlation features through multi-head attention mechanism, and simulating the diffusion path of the abnormal current in the topology network using Monte Carlo method. The probability distribution generation unit counts the frequency of abnormal current occurrences in simulated propagation for each line, calculates the abnormal current residence probability value in combination with the line impedance parameters, generates a fault probability distribution map covering the entire network, and marks the set of suspicious lines whose probability values ​​exceed the preset abnormal residence probability threshold. The physical region positioning unit performs spatial clustering analysis on the fault probability distribution map, identifies abnormal probability clusters, and delineates the physical boundaries of the fault based on the vehicle location and the connection relationship between the electronic control topology.

5. The new energy vehicle electronic control fault prediction system according to claim 4, characterized in that, The fault probability localization module also includes outputs including suspicious device identifiers and abnormal propagation main paths; The suspicious device identifier is based on the electronic control node connected to the suspicious line set, which binds the electronic control node to the actual equipment of the vehicle to form a suspicious device identifier set; The main path of anomaly propagation is obtained by recording the node paths and their order in each round of propagation during the anomaly diffusion process in Monte Carlo simulation. Among all simulated paths, the frequency of occurrence of each path is counted, and the path sequence with the highest cumulative frequency is selected as the main path of anomaly propagation. The output sequence of the main path of anomaly propagation is a structured and ordered list of nodes.

6. The new energy vehicle electronic control fault prediction system according to claim 1, characterized in that, It also includes an iterative optimization module, which specifically includes: The data stream parsing unit parses multi-source heterogeneous sensor data carrying spatiotemporal labels in the network data stream and generates dynamic priority parameters for the constraint optimization dimension based on the spatiotemporal aggregation degree of each data packet on the transmission path. The baseline encoding extraction unit extracts the baseline encoding of the optimized path corresponding to the cross-source query pattern in the historical record of network data streams. The baseline encoding is generated by matching the spatiotemporal coverage complementarity of different data packet transmission paths, and the baseline encoding is used as the gene expression template of the optimized path in the initial population. The co-evolutionary unit performs co-evolution of the optimized path at the network edge nodes, driving the optimized path to iterate in the direction of minimizing conflict. Each iteration generates a path offset and updates the network adaptation parameters of the gene expression template. An adaptive evaluation function is constructed based on the correlation between the dynamic priority parameters and the data packet transmission delay. The preloading trigger unit triggers the preloading of heterogeneous data packets according to the direction and magnitude of the path offset. The preloaded data is a set of data packets in the adjacent nodes of the network topology that match the spatiotemporal coverage area of ​​the current optimized path. The protocol compatibility of the data packets is verified by the format conversion rules in the gene expression template. The feedback generation unit generates feedback parameters by using the conflict index output by the adaptive evaluation function and the verification results of the preloaded data, and dynamically corrects the mapping relationship between the dynamic priority parameters and the gene expression template.

Images