Vehicle ECU intelligent management and control method and system for multi-source heterogeneous data fusion

Abstract

The application discloses a kind of vehicle ECU intelligent management and control method and system for multi-source heterogeneous data fusion, belong to vehicle ECU control technical field, pass through the analysis of the correlation between data by constructing dynamic coupling relationship matrix, identify the key data flow that exceeds preset threshold and preferentially allocate computing resources, form optimized data fusion path, judge driving intention adjustment demand by fusing environmental factors, update resource scheduling strategy to generate real-time adaptive control module management and control sequence, and use optimized training model to train the control module management and control sequence to obtain system reliability improvement index, to improve the real-time performance, adaptability and overall reliability of vehicle control system.

Description

Technical Field

[0001] This invention relates to the field of vehicle ECU control technology, and in particular discloses a vehicle ECU intelligent management and control method and system for multi-source heterogeneous data fusion. Background Technology

[0002] With the rapid development of intelligent vehicle technology, vehicles integrate a large number of electronic control units from different suppliers and with different data formats. These units continuously generate massive amounts of heterogeneous data from multiple sources, such as operating status, sensor readings, and control commands. Effectively integrating and intelligently managing this data is the core foundation for realizing advanced vehicle functions such as autonomous driving, energy efficiency optimization, and predictive maintenance. Its importance is directly related to driving safety, system reliability, and user experience.

[0003] Current mainstream vehicle data fusion and management methods often focus on the standardized processing of known data types or rely on preset fixed rules for decision-making. However, in the actual complex and ever-changing driving environment, the data interaction relationships and importance between various electronic control units are dynamically evolving. For example, in emergency braking scenarios, the data coupling between the braking system, tire sensors, and vehicle stability system increases sharply, while the data correlation of the entertainment system decreases significantly. Existing static or rule-fixed methods are unable to capture and respond to such dynamically changing coupling relationships in real time and adaptively, resulting in low data fusion efficiency or one-sided decision-making basis at critical moments.

[0004] The root of this limitation lies in two interconnected core technical challenges. The primary challenge is the dynamic nature of data coupling relationships. That is, the correlation strength and pattern between data streams generated by different electronic control units change in real time with changes in vehicle operating conditions, environment, and driving intentions. This dynamism directly leads to the second challenge: the real-time adaptability of control strategies. Because it is impossible to accurately and quickly quantify and determine which data coupling relationships are critical at the current moment, the system struggles to generate the optimal data fusion path and resource scheduling strategy that matches them. For example, when the vehicle is simultaneously performing traffic jam following and battery thermal management, the system may not be able to automatically prioritize ensuring the real-time performance and reliability of data interaction between power and thermal management related units, and instead allocate resources evenly, thus affecting overall performance.

[0005] Therefore, designing a method that can perceive the dynamic coupling relationship between multi-source heterogeneous data of a vehicle in real time and adjust the data fusion and electronic control unit control strategies autonomously and accurately accordingly, so as to ensure the efficient coordination and reliable processing of key data streams in various complex driving scenarios, has become a key issue in improving the level of intelligent vehicle control and achieving optimal system-level performance. Summary of the Invention

[0006] This invention provides a vehicle ECU intelligent management and control method and system for multi-source heterogeneous data fusion, aiming to improve the level of intelligent vehicle management and control and achieve optimal system-level performance.

[0007] One aspect of the present invention relates to a vehicle ECU intelligent control method for multi-source heterogeneous data fusion, comprising the following steps: S100. Collect multi-source heterogeneous data using a data acquisition device, process the correlation relationship between the multi-source heterogeneous data using a correlation analysis model, and obtain a dynamic coupling relationship matrix, wherein the multi-source heterogeneous data includes various vehicle operation information. S200. Based on the dynamic coupling relationship matrix, calculate the correlation strength between each control module and determine the key data flow under the current vehicle operating condition changes; S300. If the correlation strength of key data streams exceeds a preset threshold, computing resources will be allocated first to obtain an optimized data fusion path. S400: By integrating key data streams and environmental factors through an optimized data fusion path, it determines the driving intention and adjustment needs. S500 adjusts and updates resource scheduling strategies according to driving intentions to obtain a real-time adapted control module management sequence. S600: The control module's management sequence is trained using an optimized training model to obtain feedback signals under the performance optimization objective and determine the system reliability improvement index.

[0008] Further, step S100 includes: S110. The data acquisition device acquires multi-source heterogeneous data such as vehicle speed, motor speed, battery SOC, battery temperature, braking frequency, accelerator pedal opening, steering angle, ambient temperature, road conditions, passenger load, tire pressure, and battery voltage from various vehicle sensors, and uses a pre-established standardized protocol to perform unified format conversion on the multi-source heterogeneous data to obtain a standardized data set. S120. For the standardized dataset, the association rule mining algorithm is used to process the association between vehicle speed and battery SOC, the association between motor speed and accelerator pedal opening, and the association between braking frequency and road conditions, and to determine the preliminary association strength index. S130. Based on the preliminary correlation strength index, the correlation between ambient temperature and battery temperature, the correlation between steering angle and passenger load, and the correlation between battery voltage and tire pressure are incorporated to construct a dynamic update mechanism and obtain an extended correlation network. S140. If the correlation strength index in the extended correlation network exceeds the preset threshold, the dynamic coupling parameters are determined by fusing the time-varying correlation between vehicle speed and braking frequency and the time-varying correlation between battery SOC and road conditions through a time-series analysis model. S150. A matrix construction algorithm is used to process the multidimensional correlation between dynamic coupling parameters and accelerator pedal opening, ambient temperature, and battery temperature to obtain a dynamic coupling relationship matrix.

[0009] Further, step S200 includes: S210. Obtain the dynamic coupling relationship matrix and use the matrix decomposition algorithm to calculate the correlation strength between each control module; S220. Construct a data flow topology based on the correlation strength and extract working condition feature vectors; S230. If the numerical fluctuation of the working condition feature vector exceeds the preset threshold, the data flow topology is processed to obtain the flow density. S240. Identify the core transmission nodes in the topology based on the flow density to determine the key data flow under the current vehicle operating conditions.

[0010] Further, step S300 includes: S310. Obtain the correlation strength of key data streams, and use the sliding window algorithm to calculate the real-time fluctuation variance of the correlation strength to obtain the intensity fluctuation sequence. S320. Extract peak features based on the intensity fluctuation sequence. If the correlation strength corresponding to the peak feature exceeds a preset threshold, trigger the resource allocation mechanism to obtain the initial computing power allocation matrix. S330. Evaluate the load status of the current computing power node through the initial computing power allocation matrix, and determine the available link bandwidth of the computing power node; S340. Construct a priority queue based on available link bandwidth, sort the transmission tasks in the priority queue using a greedy algorithm, obtain the fusion node corresponding to the scheduling instruction set, allocate computing resources preferentially based on the distribution characteristics of the fusion node, and obtain an optimized data fusion path.

[0011] Further, step S400 includes: S410. By optimizing the data fusion path, the key data stream and the influence of environmental factors are fused, and a spatial mapping matrix is ​​constructed using the key data stream and meteorological sensing sequence. S420. Extract the road surface adhesion estimate based on the spatial mapping matrix, and generate a fused feature vector based on the road surface adhesion estimate; S430. Obtain the projection coordinates of the fused feature vector, and calculate the state transition probability based on the projection coordinates. S440. Generate the intention evolution trajectory based on the state transition probability, and determine whether the intention evolution trajectory is greater than the deviation threshold. S450. If the intention evolution trajectory is greater than the deviation threshold, it is determined that there is a need to adjust the driving intention.

[0012] Further, step S500 includes: S510. Determine the scheduling load requirement based on the driving intention adjustment needs, and obtain the upper limit of response latency based on the scheduling load requirement; S520. If the upper limit of response delay is lower than the preset delay threshold, then determine the module priority of each control module. S530. Calculate the resource occupancy quota based on the module priority, and obtain the resource adjustment increment based on the resource occupancy quota; S540: The original control logic is modified by resource adjustment increments to generate a real-time adapted control module management sequence.

[0013] Further, step S600 includes: S610. Input the control sequence from the control module into the optimization training model. The optimization training model reaches a convergent state by iteratively updating the weight matrix. S620. Based on the convergence state, obtain the feedback signal extracted by the optimized training model that is in the convergence state. S630. Compare the feedback signal with the operation log and calculate the deviation between the feedback signal and the operation log. S640. The deviation value is associated with a preset redundancy factor for quantification to determine the system reliability improvement index corresponding to the control module's control sequence. The system reliability improvement index is used to evaluate performance.

[0014] Another aspect of the present invention relates to a vehicle ECU intelligent control system for multi-source heterogeneous data fusion, used to implement the above-described vehicle ECU intelligent control method for multi-source heterogeneous data fusion, comprising: The dynamic coupling relationship matrix acquisition module is used to collect multi-source heterogeneous data using a data acquisition device, process the relationship between the multi-source heterogeneous data using an association analysis model, and obtain a dynamic coupling relationship matrix, wherein the multi-source heterogeneous data includes various vehicle operation information. The key data flow determination module is used to calculate the correlation strength between each control module based on the dynamic coupling relationship matrix, and to determine the key data flow under the current vehicle operating condition changes. The data fusion path acquisition module is used to prioritize the allocation of computing resources and obtain an optimized data fusion path if the correlation strength of key data streams exceeds a preset threshold. The driving intent adjustment demand judgment module is used to judge the driving intent adjustment demand by integrating key data streams and environmental factors through an optimized data fusion path; The control module management sequence acquisition module is used to update the resource scheduling strategy according to the driving intention and obtain the real-time adapted control module management sequence. The system reliability improvement index determination module is used to train the control module's management sequence using an optimized training model, obtain feedback signals under the performance optimization objective, and determine the system reliability improvement index.

[0015] The beneficial effects achieved by this invention are as follows: The present invention provides a vehicle ECU intelligent control method and system for multi-source heterogeneous data fusion. Addressing the business scenario where real-time fusion and resource scheduling of multi-source heterogeneous data in vehicles suffers from delayed identification of key data streams and allocation of computing resources due to dynamic changes in operating conditions, thus affecting the real-time performance and reliability of control decisions, the invention constructs a dynamic coupling relationship matrix to analyze the correlation between data, identifies key data streams exceeding a preset threshold, prioritizes the allocation of computing resources, forms an optimized data fusion path, integrates environmental factors to determine driving intentions and adjustment needs, updates resource scheduling strategies accordingly to generate a real-time adapted control module control sequence, and uses an optimized training model to train this control module control sequence to obtain system reliability improvement indicators. This achieves accurate and rapid response to key data streams and optimal dynamic allocation of computing resources under complex operating conditions, improving the real-time performance, adaptability, and overall reliability of the vehicle control system. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of the intelligent control and management system for vehicle ECUs based on multi-source heterogeneous data fusion according to the present invention. Figure 2 This is a functional block diagram of an embodiment of the vehicle ECU intelligent management and control system for multi-source heterogeneous data fusion according to the present invention.

[0017] Explanation of icon numbers: 10. Dynamic coupling relationship matrix acquisition module; 20. Key data flow determination module; 30. Data fusion path acquisition module; 40. Driving intention adjustment requirement judgment module; 50. Control module management sequence acquisition module; 60. System reliability improvement index determination module. Detailed Implementation

[0018] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0019] like Figure 1 As shown, the first embodiment of this invention proposes a vehicle ECU intelligent control method for multi-source heterogeneous data fusion. The core of this method is to achieve real-time, efficient, and reliable control of the vehicle ECU system through multi-source heterogeneous data acquisition, coupling relationship modeling, key data stream identification, resource optimization allocation, driving intent adaptation, and control sequence optimization. This improves the safety, stability, and energy efficiency of vehicle operation and is applicable to ECU control scenarios for new energy vehicles and intelligent connected vehicles. The method includes the following steps: Step S100: Collect multi-source heterogeneous data using a data acquisition device, process the correlation relationship between the multi-source heterogeneous data using a correlation analysis model, and obtain a dynamic coupling relationship matrix, wherein the multi-source heterogeneous data includes various vehicle operation information.

[0020] By utilizing acquisition devices deployed in various vehicle subsystems, multi-source heterogeneous data is collected synchronously during vehicle operation. This multi-source heterogeneous data covers operational information from multiple dimensions, including vehicle power, chassis, body, and intelligent driving. A correlation analysis model is used to process the collected multi-source heterogeneous data, uncovering the inherent correlations, couplings, and constraints between different data sources. These dynamically changing correlations are quantified into a matrix form, resulting in a dynamic coupling relationship matrix. This provides core data support for subsequent calculations of correlation strength in control modules and identification of key data flows.

[0021] Data acquisition devices refer to a collection of sensors, data acquisition cards, vehicle communication modules, and edge acquisition units installed in various subsystems such as vehicle motors, chassis, body, and intelligent driving systems. They are used to continuously and multidimensionally collect vehicle operating status data, support the synchronous acquisition and transmission of high-frequency, medium-frequency, and low-frequency data, and are the core carrier for acquiring multi-source heterogeneous data.

[0022] Multi-source heterogeneous data refers to a collection of vehicle operation information that originates from different subsystems of the vehicle, has different data formats (structured data such as RPM and vehicle speed, and unstructured data such as camera images), different sampling frequencies, and different physical meanings. Specifically, it includes data related to the power system, chassis system, body system, and intelligent driving system, and is the basic data source for intelligent control of the vehicle's ECU.

[0023] Association analysis models refer to algorithmic models used to analyze the inherent correlation and coupling relationships between multi-source heterogeneous data. In this embodiment, the association analysis model based on graph attention network (GAT) is preferred, which can effectively capture the nonlinear and asymmetric coupling relationships between data and achieve accurate quantification of the correlation relationship. Conventional models in this field, such as Pearson correlation analysis model and Granger causality analysis model, can also be selected.

[0024] The dynamic coupling relationship matrix is ​​an N×N matrix (N is the total number of data sources collected) used to quantify the correlation strength and direction between data sources as they dynamically adjust with changes in vehicle operating conditions. The matrix elements take values ​​in the range of [-1, 1]. The larger the absolute value of an element, the stronger the correlation between the two corresponding data sources. A positive sign indicates a positive correlation, and a negative sign indicates a negative correlation. The matrix is ​​updated in real time with the vehicle operating conditions, accurately representing the dynamic correlation state between data.

[0025] Vehicle operation information refers to various types of data that reflect the vehicle's operating status and the working status of its subsystems. It covers multiple dimensions such as power, chassis, body, and intelligent driving, and is a core component of multi-source heterogeneous data, providing data support for ECU control and decision-making.

[0026] In this embodiment, the preset threshold used to construct the dynamic coupling relationship matrix is ​​calibrated offline using a typical operating condition database. Based on over 100 hours of real vehicle data covering operating conditions such as idling, constant speed, acceleration, deceleration, steering, emergency braking, slope, and slippery road surfaces, the 95th percentile of the data stream coupling degree is used as the baseline threshold, and the calibrated association strength threshold is 0.7. The threshold supports adaptive correction for different vehicle types; electric vehicles and passenger / commercial vehicles can automatically adjust within a range of ±0.05. The system recalculates the quantiles using newly added data every 1000km, achieving online self-updating of the threshold. The sliding window size of the time-series analysis model is fixed at 200ms, with 20 sampling points within the window and a sampling frequency of 100Hz, ensuring stable time-varying association calculations.

[0027] Thresholds and specific numerical ranges for this step: 1. Multi-source heterogeneous data acquisition parameters: Data source dimensions: 20~40 dimensions, covering powertrain system (6~10 dimensions), chassis system (5~8 dimensions), body system (4~6 dimensions), and intelligent driving system (8~15 dimensions). Power system data: Motor speed 800~6000rpm (acquisition accuracy ±10rpm), battery SOC (state of charge) 0%~100% (acquisition accuracy ±0.5%), battery SOH (state of health) 80%~100%; Chassis system data: vehicle speed 0~180km / h (collection accuracy ±0.1km / h), wheel speed 0~500rpm (collection accuracy ±1rpm), tire pressure 2.0~3.5bar (collection accuracy ±0.05bar), brake pedal travel 0~100mm, steering angle -45°~+45° (collection accuracy ±0.1°). Vehicle body system data: door status (open / closed), light status (on / off), wiper status (stop / low speed / high speed), seat sensor pressure 0~500N; Intelligent driving system data: camera image resolution 1280×720~1920×1080 pixels, lidar point cloud density 10000~100000 points / frame, millimeter-wave radar detection range 0~200m (accuracy ±0.1m), IMU attitude angle -90°~+90° (accuracy ±0.01°). Acquisition frequency: High frequency data (motor speed, wheel speed, IMU attitude) 100~1000Hz, medium frequency data (vehicle speed, brake pedal travel, steering angle) 50~100Hz, low frequency data (door status, light status, battery SOC) 10~20Hz; Data transmission latency: ≤5ms, data packet loss rate: ≤0.05%.

[0028] 2. Correlation analysis model parameters: The model is trained for 5,000 to 10,000 iterations, and the convergence threshold is a loss function value ≤ 0.001. The accuracy of association identification is ≥98%, and the false association removal rate is ≥99%.

[0029] 3. Parameters of the dynamic coupling matrix: Matrix dimension N = 20~40 (consistent with the number of data sources); The matrix elements take values ​​in the range [-1, 1], where 0~0.3 indicates a weak correlation, 0.3~0.7 indicates a moderate correlation, 0.7~1.0 indicates a strong correlation, -0.3~0 indicates a weak negative correlation, -0.7~-0.3 indicates a moderate negative correlation, and -1.0~-0.7 indicates a strong negative correlation. The matrix update cycle is 100~500ms to ensure that the data correlation changes caused by changes in vehicle operating conditions can be tracked in real time.

[0030] Step S200: Calculate the correlation strength between each control module based on the dynamic coupling relationship matrix, and determine the key data flow under the current vehicle operating condition changes.

[0031] Based on the dynamic coupling relationship matrix obtained in step S100, the correlation strength between each ECU control module of the vehicle is calculated by weighted summation algorithm to quantify the degree of mutual influence between each control module. Combining the changing characteristics of the current vehicle operating conditions (such as acceleration, deceleration, lane change, emergency braking, constant speed driving, etc.), the data stream with the greatest impact on vehicle operation safety and performance is screened from all collected data streams through operating condition matching degree analysis. The key data streams under the current operating conditions are determined, realizing the targeted focus of computing resources and improving management efficiency.

[0032] The control module refers to various control units in the vehicle ECU system, including chassis ECU, body ECU, ADAS domain controller, VCU (vehicle control unit), etc. It is responsible for receiving data and executing control commands, and is the core execution unit for intelligent management and control of vehicle ECU.

[0033] Correlation strength is a dimensionless numerical index used to quantify the degree of mutual influence and interdependence between various control modules and data streams in a vehicle. The value range is [0, 1]. The higher the value, the stronger the correlation. It is the core criterion for screening key data streams.

[0034] Vehicle operating conditions refer to the specific working states of a vehicle during operation, including acceleration, deceleration, constant speed, emergency braking, lane changing, and parking conditions. The operational needs and data correlation characteristics of a vehicle differ under different operating conditions.

[0035] Critical data streams refer to data streams that play a decisive role in vehicle operation safety, performance, and driver intent judgment under the current vehicle operating conditions, and whose correlation strength exceeds a preset threshold. They are the core objects for subsequent priority allocation of resources and data fusion.

[0036] In this embodiment, the fluctuation amplitude threshold of the operating condition feature vector is calculated using the 2-norm, with a threshold of 0.5. A value exceeding this threshold indicates a drastic change in operating conditions. The flow density threshold is 0.6; values ​​greater than or equal to this value indicate a core transmission node. The data flow topology retains only edges with an association strength ≥ 0.3. Nodes include VCU, chassis ECU, ADAS domain controller, and body ECU. The topology is reconstructed every 50ms. Matrix decomposition uses SVD decomposition, retaining 8 singular values, covering over 99% of coupling features, meeting the requirements for real-time vehicle computing.

[0037] Thresholds and specific numerical ranges for this step: 1. Parameters for calculating association strength: Calculation method: Based on the absolute values ​​of the elements of the dynamic coupling relationship matrix, a weighted sum is performed by combining the importance weights of each control module. The weight allocation is as follows: powertrain control module 0.3, chassis control module 0.25, intelligent driving control module 0.35, and vehicle comfort control module 0.1. The correlation strength ranges from [0, 1], with a calculation precision of ±0.01.

[0038] 2. Operating condition matching parameters: Operating condition matching degree calculation method: The cosine similarity algorithm is used to calculate the degree of matching between the current data stream features and typical operating condition features; The operating condition matching threshold is ≥0.8. When the matching degree between the data stream and the current operating condition exceeds this threshold, it is determined to be a data stream that is compatible with the current operating condition.

[0039] 3. Key data flow determination parameters: The correlation strength threshold is ≥0.7. When the correlation strength of a data stream exceeds this threshold and the working condition matching degree is ≥0.8, it is determined to be a critical data stream. Under a single operating condition, there are 5 to 10 critical data streams to ensure the smooth flow of core data channels while avoiding resource waste. The critical data stream filtering response time is ≤10ms, ensuring real-time adaptation to changes in operating conditions.

[0040] Step S300: If the correlation strength of the key data stream exceeds the preset threshold, computing resources are allocated first to obtain an optimized data fusion path.

[0041] If the correlation strength of the key data stream determined in step S200 exceeds a preset threshold, the system resource scheduling priority mechanism is triggered to prioritize the allocation of the vehicle ECU system's computing resources (CPU computing power, bus bandwidth, memory) to the acquisition, transmission, and fusion processing of the key data stream, ensuring the real-time performance and reliability of the key data stream. Through a global resource optimization algorithm, the current system resource load status is analyzed to avoid resource bottlenecks and to plan and obtain an optimized data fusion path that maximizes data fusion efficiency and minimizes transmission latency, providing support for the efficient fusion of multi-source data in the future.

[0042] Computing resources refer to the hardware resources in a vehicle's ECU system that can be used for data processing, transmission, and storage. These mainly include CPU computing power, vehicle bus bandwidth, and memory capacity, and serve as the basic hardware support for data fusion and control decision-making.

[0043] The resource scheduling priority mechanism refers to the mechanism in a vehicle ECU system that divides computing resources into different priorities based on the importance of data flow and control tasks, and allocates them in a differentiated manner. The core is to prioritize the resource needs of critical data flow and core control tasks.

[0044] An optimized data fusion path refers to a multi-source data fusion processing and transmission path planned for critical data flows in the vehicle's electronic and electrical architecture, which can achieve the fastest transmission speed, the lowest transmission latency, and the highest transmission reliability. Through path optimization, data congestion caused by resource contention can be avoided.

[0045] In this embodiment, resource priority allocation is triggered when the peak correlation strength of critical data flows is ≥0.7; resource priority is increased when the correlation strength fluctuation variance is ≥0.02. CPU computing power allocation: critical data flows account for 70%, non-critical data flows account for 30%; bus bandwidth: critical data flows account for 80%, non-critical data flows account for 20%. The greedy algorithm is constrained by end-to-end latency ≤20ms and packet loss rate ≤0.1%, with no more than 5 path nodes. When the computing power node load is >85%, it is marked as high load and automatically scheduled to an idle node.

[0046] Thresholds and specific numerical ranges for this step: 1. Association strength trigger threshold: consistent with the critical data flow determination threshold in step S200, i.e. ≥0.7. When the association strength of the critical data flow exceeds this threshold, the resource priority allocation mechanism is triggered. 2. Calculate resource allocation parameters: Priority division: critical data flow (priority 1), non-critical data flow (priority 2), with a resource allocation ratio of 5:1 between priority 1 and priority 2; CPU computing power allocation: Critical data flow occupies 60%~80% of CPU computing power, while non-critical data flow occupies 20%~40%; Bus bandwidth allocation: Critical data flows occupy 70%~90% of the vehicle bus bandwidth (such as CAN, Ethernet), while non-critical data flows occupy 10%~30%. Memory allocation: Critical data streams occupy 50% to 70% of memory to ensure smooth data caching and processing; Resource allocation response time is ≤10ms to ensure that critical data flows are not blocked.

[0047] 3. Optimized data fusion path parameters: Path selection algorithm: Use Dijkstra's algorithm or Q-Learning reinforcement learning algorithm to achieve optimal path search; Path latency target: End-to-end fusion latency of critical data streams ≤20ms, ensuring real-time data fusion; Path reliability targets: fused path data packet loss rate ≤ 0.1%, transmission error rate ≤ 0.05%; Path update cycle: 50~100ms, dynamically adjusted according to system resource load.

[0048] Step S400: By integrating key data streams and environmental factors through an optimized data fusion path, determine the driving intention adjustment needs.

[0049] The optimized data fusion path determined in step S300 is used to deeply integrate key data streams with the influence of external environmental factors of the vehicle. A weighted fusion algorithm is used to eliminate data redundancy and noise to obtain integrated data after fusion. Based on the integrated data after fusion, a driving intention recognition model is used to analyze the driving needs of the driver or the autonomous driving system to determine whether there is a need to adjust the driving intention, so as to provide a decision basis for subsequent resource scheduling strategy updates and control sequence adaptation.

[0050] Environmental factors refer to external environmental information that directly affects the vehicle's driving status and safety performance, in addition to the vehicle's own operating data. These mainly include distance to the vehicle in front, road slope, road curvature, weather conditions (rainy, sunny, snowy), traffic signs, and the status of surrounding vehicles. They are important auxiliary bases for judging driving intentions.

[0051] The driving intention recognition model refers to an artificial intelligence model used to analyze and fuse data and identify the driving needs of drivers or autonomous driving systems. In this embodiment, the preferred model is the LSTM (Long Short-Term Memory Network) + attention mechanism model, which can accurately capture the temporal features of driving intention. Conventional models in this field, such as RNN (Recurrent Neural Network) and CNN (Convolutional Neural Network), can also be used.

[0052] Driving intent adjustment requirements refer to the determination of the driver's or autonomous driving system's intention to change the current driving state by analyzing the fused key data stream and the influence of environmental factors. These intentions include acceleration, deceleration, left turn, right turn, lane change, emergency braking, parking, etc., and are the core basis for subsequent control sequence adjustments.

[0053] In this embodiment, a deviation of >0.2 from the intent evolution trajectory indicates a need for driving intent adjustment. Road surface adhesion estimation is graded as follows: ≥0.7 for high adhesion, 0.4~0.7 for medium adhesion, and <0.4 for low adhesion. State transition probabilities are normalized, with a sum of 1. A driving intent recognition confidence level ≥0.85 indicates a valid intent. The spatial mapping matrix is ​​a fixed 10×5 dimension, with 10 dimensions for the key data stream and 5 dimensions for environmental factors.

[0054] Thresholds and specific numerical ranges for this step: 1. Environmental factors affecting fusion parameters: Fusion metrics: Key data streams (vehicle speed, acceleration, steering angle) + environmental data (distance to the vehicle in front, road gradient, road curvature, speed limit); Environmental data collection range: distance to the vehicle in front 0~200m (accuracy ±0.1m), road slope -10%~+20% (accuracy ±0.5%), road curvature 0~0.1m. -1 Speed ​​limit: 0~120km / h; Fusion Algorithm: A weighted fusion model for multi-source heterogeneous data based on Kalman filtering, with a fusion weight of 0.7 for key data streams and 0.3 for environmental data; Fusion accuracy ≥97%, noise suppression ratio ≥40dB, fusion delay ≤15ms.

[0055] 2. Driving Intent Recognition Model Parameters: The model input sequence length is 5-10 seconds, and the input dimension is consistent with the dimension of the fused data (10-15 dimensions). The model has a recognition accuracy of ≥95% and a recognition latency of ≤20ms. Driving intention types: acceleration, deceleration, left turn, right turn, lane change, emergency braking, parking, a total of 7 categories.

[0056] 3. Threshold for determining the need to adjust driving intentions: The confidence threshold for intent recognition is ≥0.85. When the confidence threshold for recognition exceeds this threshold, it is determined to be a clear driving intent adjustment requirement. The intent is to continuously determine the duration for ≥500ms to avoid misjudgments caused by instantaneous data fluctuations.

[0057] Step S500: Adjust the resource scheduling strategy according to the driving intention to obtain the real-time adapted control module management sequence.

[0058] Based on the driving intention adjustment requirements determined in step S400, the resource scheduling strategy of the vehicle ECU system is dynamically updated, and the resource allocation ratio of each control module and each data stream is adjusted to ensure that the resource allocation is compatible with the driving intention. Combining the updated resource scheduling strategy and the current system resource load status, a real-time scheduling algorithm is used to generate a control module management sequence that can accurately adapt to the current driving intention and vehicle operating conditions, clarifying the execution order, execution time and control parameters of each control module, and realizing the coordinated management of each control module.

[0059] Resource scheduling strategy refers to a series of rules and schemes formulated by the vehicle ECU system for resource allocation, priority ranking, and task scheduling in order to efficiently utilize limited computing, storage, and communication resources. It can be dynamically updated according to driving intentions and vehicle operating conditions, and its core is to achieve optimal resource allocation.

[0060] The control module management sequence refers to the sequence of instructions arranged for each ECU control module to perform control actions in a specific order and at specific times, based on the driving intention, adjustment needs, and resource scheduling strategies. It clarifies the execution priority, execution parameters, and execution duration of each control module and is the core execution instruction for ECU control.

[0061] Real-time adaptable control module control sequence refers to the control module instruction sequence that can dynamically follow driving intentions and changes in vehicle operating conditions, and adjust the execution order and execution parameters in real time to ensure that the execution actions of the control module are accurately synchronized with driving needs and vehicle status, thereby improving the real-time performance and adaptability of control.

[0062] In this embodiment, the preset response latency threshold is 100ms; modules with latency below this value are prioritized. Module priority weights are: correlation strength 0.5, traffic density 0.3, and real-time load 0.2, with a sum of 1. Resource adjustment increments are limited to ±20% per instance to prevent system oscillations. The control module's management sequence length is fixed at 3-8 modules, covering the core control unit.

[0063] Thresholds and specific numerical ranges for this step: 1. Resource scheduling policy update parameters: Update trigger conditions: Confidence level of driving intention adjustment demand ≥ 0.8, or significant change in vehicle operating conditions (vehicle speed change > 20 km / h, steering angle change > 10°). The strategy update cycle is 50~200ms to ensure real-time adaptation of the control sequence; Resource adjustment range: The resource allocation ratio is adjusted by ≤20% for each update to avoid sudden changes in resource allocation that could lead to system instability.

[0064] 2. The control module manages sequence parameters: The sequence length consists of 3 to 8 control actions / modules, covering the core control modules required for the current driving intention; Execution timing accuracy: The execution time error of each control module is ≤5ms, ensuring the synchronization of coordinated execution; Control target: Lateral acceleration error ≤ 0.1 m / s² 2 Longitudinal acceleration error ≤ 0.1 m / s² 2 (Maintaining vehicle stability), new energy vehicles reduce electricity consumption by 5%~10% (improving energy efficiency). Sequence update period: Consistent with the resource scheduling policy update period, 50~200ms.

[0065] Step S600: Train the control module's management sequence using an optimized training model, obtain feedback signals under the performance optimization objective, and determine the system reliability improvement index.

[0066] The control module control sequence obtained in step S500 is jointly trained offline and online using an optimized training model. Through a large amount of simulation scenario data and real vehicle operation data, the execution parameters and timing logic of the control sequence are optimized so that the control sequence reaches the optimal under the preset performance optimization target. During the training process, the performance feedback signal after the system executes the control sequence is collected. Based on the feedback signal, the improvement of the system in terms of reliability, stability, energy efficiency, etc. is quantitatively calculated, the system reliability improvement index is determined, and the vehicle ECU control system is continuously optimized and evolved.

[0067] The optimized training model refers to an artificial intelligence model used to learn, optimize, and iterate the control sequence of the control module. In this embodiment, the Deep Deterministic Policy Gradient (DDPG) model or the Soft Actor Commentator (SAC) model is preferred, which is suitable for the optimized training of continuous control sequences. Alternatively, conventional models in the field such as Model Predictive Control (MPC) and reinforcement learning can also be used.

[0068] Performance optimization goals refer to the specific performance indicators that the vehicle ECU management system expects to achieve by optimizing the control sequence of the control module. These mainly include vehicle operation safety, vehicle stability, energy utilization efficiency, and system response speed, and are the core guiding principles for optimizing the control sequence.

[0069] Feedback signals refer to various signals that the vehicle ECU system sends back after executing the control module's control sequence, regarding the execution effect, system status, and performance. These signals mainly include execution delay, control accuracy, vehicle posture error, number of fault codes, and energy consumption data, and are the core basis for model training and system reliability assessment.

[0070] The system reliability improvement index refers to a comprehensive quantitative indicator used to quantitatively evaluate the improvement in reliability, stability, and availability of the vehicle ECU intelligent control system after optimization and training compared to the original system. The value ranges from 0 to 100 points, with higher values ​​indicating a more significant improvement in system reliability.

[0071] In this embodiment, the optimized training model adopts the DDPG structure, with a 15-dimensional input layer, a 6-dimensional output layer, and two hidden layers with 64 neurons each; the learning rate is 0.001, the batch size is 32, the iterations are 10,000, and the convergence condition is that the reward change is <0.01 for 1,000 consecutive steps; the loss function is the mean squared error (MSE); the redundancy factor is fixed at 1.2; the reliability index is normalized to [0, 1] using min-max; the bias is calculated every 100ms, and the statistical period is 100 steps.

[0072] Thresholds and specific numerical ranges for this step: 1. Optimize training model parameters: Training data volume: ≥10,000 sets of driving scenario data (covering various typical working conditions), of which real vehicle data accounts for ≥40% and simulation data accounts for ≤60%; The model was trained for 10,000 to 20,000 iterations, and the convergence criterion was that the average reward value increased by less than 0.01 for 1,000 consecutive steps. Training latency is ≤500ms / time to ensure that online training does not affect the normal operation of the system.

[0073] 2. Performance optimization target parameters: Safety: Collision risk probability reduced by ≥40%, emergency braking response time ≤100ms; Stability: Vehicle body attitude error ≤ 0.5°, lateral / longitudinal acceleration error ≤ 0.1m / s² 2 ; Energy efficiency: The average power consumption of new energy vehicles is reduced by ≥7%; Response speed: The system's response delay to driving intentions is ≤30ms.

[0074] 3. Feedback signal parameters: The feedback signal acquisition frequency is 10~100Hz, and the acquisition accuracy is ±0.01. Fault code number threshold: The number of system fault codes during training is ≤1 per 10,000 steps. If this threshold is exceeded, model parameter adjustment will be triggered.

[0075] 4. System reliability improvement indicators and parameters: The indicator values ​​range from 0 to 100, with 80 to 100 being excellent, 60 to 79 being satisfactory, and <60 being unsatisfactory. Improvement target: Compared to the baseline system before optimization, the indicator score should be improved by ≥15 points; System reliability targets: Optimized system availability ≥ 99.99%, mean time between failures (MTBF) ≥ 10,000 hours; Indicator update cycle: Updated once after every 1000 executions of the same type of working condition to ensure the accuracy of the indicators.

[0076] Furthermore, the intelligent control method for vehicle ECUs oriented towards multi-source heterogeneous data fusion provided in this embodiment includes step S100 as follows: Step S110: Obtain multi-source heterogeneous data such as vehicle speed, motor speed, battery SOC, battery temperature, braking frequency, accelerator pedal opening, steering angle, ambient temperature, road conditions, passenger load, tire pressure, and battery voltage from various vehicle sensors through a data acquisition device, and perform unified format conversion on the multi-source heterogeneous data using a pre-established standardized protocol to obtain a standardized data set.

[0077] The data acquisition device obtains heterogeneous data from multiple sources from vehicle sensors, such as a speed sensor providing speed data of 80 kilometers per hour. A pre-established standardized protocol converts this data in different formats. Specifically, the standardization protocol defines that all speed units are unified to meters per second, and all percentage data are converted to floating-point numbers between 0 and 1. Through this conversion, the originally heterogeneous data is organized into a standardized data set with a consistent structure.

[0078] Step S120: For the standardized dataset, use association rule mining algorithm to process the correlation between vehicle speed and battery SOC, the correlation between motor speed and accelerator pedal opening, and the correlation between braking frequency and road conditions, and determine the preliminary correlation strength index.

[0079] The preliminary correlation strength index is determined using the following formula: (1) In formula (1), For the first Class and the The preliminary association strength index for class data, without units, is derived from the association rule mining algorithm, and its value range is [value range missing]. The larger the value, the higher the correlation between the two types of data; For data items and The frequency of occurrence is unitless and derived from statistical results of standardized datasets, with values ​​ranging from non-negative integers. For data items The frequency of occurrence of a single element is unitless and derived from statistical results of a standardized dataset, with values ​​ranging from positive integers. The data items are multi-source heterogeneous data items (such as vehicle speed, motor speed, etc.), without units, and are derived from the standardized data set in step S110. The control logic of formula (1) is designed through benchmark normalization, which eliminates the interference of the difference in the frequency of occurrence of different data items on the correlation calculation, and ensures that the correlation strength index of different data pairs is horizontally comparable; at the same time, it uses the co-occurrence frequency to characterize the correlation essence, accurately reflects the causal / correlation relationship between the two types of data, and naturally satisfies the value constraint of [0, 1], without the need for additional normalization processing. The calculation logic is self-consistent, has low complexity, can be calculated online in real time, and is adapted to the embedded operating environment of vehicle ECU.

[0080] For standardized datasets, an association rule mining algorithm is used for processing. In one implementation, this algorithm quantifies the association between variables by calculating support and confidence. For example, analysis using the association rule mining algorithm reveals that when vehicle speed consistently exceeds 100 km / h, the median rate of battery SOC decline increases significantly, thus determining the preliminary association strength between speed and fuel consumption to be 0.85.

[0081] Step S130: Based on the preliminary correlation strength index, incorporate the correlation between ambient temperature and battery temperature, the correlation between steering angle and passenger load, and the correlation between battery voltage and tire pressure to construct a dynamic update mechanism and obtain an extended correlation network.

[0082] Similarly, association rule mining algorithms analyze the co-occurrence frequency between motor speed and rapid acceleration events to obtain another set of preliminary indicators. Based on these preliminary indicators, the vehicle ECU intelligent control system incorporates more multi-dimensional relationships to build an extended association network. For example, the system analyzes whether there is a statistical correlation between an increase in ambient temperature and an increase in battery temperature readings, adding this relationship as a new node to the network. Simultaneously, the system establishes a dynamic update mechanism; for instance, it recalculates all association strength indicators every 1000 new data points collected to ensure the network reflects the vehicle's latest status.

[0083] Step S140: If the correlation strength index in the extended correlation network exceeds the preset threshold, the dynamic coupling parameters are determined by fusing the time-varying correlation between vehicle speed and braking frequency and the time-varying correlation between battery SOC and road conditions through a time-series analysis model.

[0084] The dynamic coupling parameters are determined using the following formula: (2) In formula (2), For the first Time step and The dynamic coupling parameter is unitless, derived from the time series analysis model, and its value range is [value range missing]. This characterizes the time-varying correlation strength between the two types of data; This is a static association weight, without units, configured based on experience, and its value range is [range missing]. This is used to balance the proportion of static correlations and time-varying correlations; The preliminary correlation strength index has no unit and is derived from the calculation results of formula (1); For the first Time step and The Pearson correlation coefficient, dimensionless, is derived from the time series analysis model and its value ranges from [value missing]. This characterizes the real-time linear correlation between the two types of data; The time step number is a unitless number derived from the data acquisition timestamp, and its value range is positive integer. The control logic of formula (2) is achieved through the fusion of static and dynamic dimensions. It uses the static correlation strength as a historical benchmark to avoid misjudgment of correlation caused by real-time data fluctuations, and uses the real-time correlation coefficient to capture correlation changes in dynamic operating scenarios. At the same time, it uses adjustable weights. It is adaptable to different vehicle operating conditions, has self-consistent calculation logic and low complexity, and can perform real-time online calculations, providing core quantitative support for subsequent multi-source heterogeneous data fusion and dynamic adjustment of ECU intelligent control strategies.

[0085] If a correlation strength index in the extended correlation network exceeds a preset threshold, such as the correlation strength between tire pressure and battery voltage reaching 0.9, the time-series analysis model is triggered. This time-series analysis model analyzes historical data sequences. For example, through sliding time window analysis, it discovers that within 5 minutes after a slippery road condition occurs, the average driver braking frequency exhibits a specific pattern of first rising and then falling, while the battery wear rate also fluctuates accordingly. The time-series analysis model then calculates the dynamic coupling parameters describing this time-varying correlation.

[0086] Step S150: Use a matrix construction algorithm to process the multidimensional correlation between dynamic coupling parameters and accelerator pedal opening, ambient temperature, and battery temperature to obtain a dynamic coupling relationship matrix.

[0087] The dynamic coupling matrix is ​​derived using the following formula: (3) In formula (3), For the first The dynamic coupling matrix of the time step is dimensionless, derived from a matrix construction algorithm, and has dimensions of 1. , The total number of heterogeneous data types from multiple sources, and the matrix elements are the dynamic coupling parameters of the corresponding data items; For the first Time step and The dynamic coupling parameters are unitless and are derived from the calculation results of formula (2); The total number of multi-source heterogeneous data types, without unit, is derived from the number of data acquisition items in step S110, and its value range is positive integer. The control logic of formula (3) is to realize the global and computable modeling of the multi-dimensional correlation between multi-source heterogeneous vehicle data, and update it in real time with time steps. This dynamic coupling relationship matrix integrates the originally scattered pairwise correlations into a global correlation network. The diagonal elements represent the self-coupling of data, and the off-diagonal elements represent the time-varying coupling strength between data. It not only realizes the global representation of the correlation relationship, but also accurately captures the correlation changes under the dynamic operation scenario of the vehicle. At the same time, the matrix dimension can be flexibly expanded to adapt to the multi-source data acquisition needs of different models and different scenarios, and provides core global model support for the subsequent multi-source heterogeneous data fusion and dynamic adjustment of ECU intelligent control strategy.

[0088] Finally, a matrix construction algorithm is used to process these dynamically coupled parameters. Specifically, the matrix construction algorithm treats each coupled parameter as a vector and calculates its covariance or correlation coefficient with multiple dimensions such as acceleration mode and ambient temperature, filling these coefficients into a matrix. For example, in the generated relation matrix, the first... Line number The elements of a column may represent the first... The dynamic coupling parameter and the first The correlation strength between the various rapid acceleration modes is determined, thereby obtaining a dynamic coupling relationship matrix that comprehensively describes the multi-dimensional interaction relationships within the vehicle ECU intelligent control system.

[0089] Preferably, the intelligent control method for vehicle ECUs oriented towards multi-source heterogeneous data fusion provided in this embodiment includes step S200 as follows: Step S210: Obtain the dynamic coupling relationship matrix and use the matrix decomposition algorithm to calculate the correlation strength between each control module.

[0090] The correlation strength between each control module is calculated using the following formula: (4) In formula (4), For the first The and the first The correlation strength between control modules, dimensionless, derived from the SVD matrix factorization algorithm, has a value range of [value missing]. A higher value indicates a higher degree of correlation between modules; For the first , The left singular vector corresponding to the control module is dimensionless, derived from a matrix factorization algorithm, and has a dimension of . ; The dynamic coupling relationship matrix is ​​unitless and is derived from the result of formula (3). The F-norm operator is a unitless operator used to calculate the F-norm of a matrix, quantifying the overall magnitude of matrix elements. For the first The transpose of the left singular vector of the control module is unitless and comes from the matrix transpose operation. It is used for dimension adaptation in matrix multiplication to realize the inner product operation of eigenvectors. The control logic of formula (4) realizes the accurate mapping from the coupling of multi-source data layer to the association of ECU module layer. It not only extracts the global coupling features of the module through matrix decomposition, but also realizes the intuitive quantification of the association strength through F norm. The calculation logic is rigorous and has low complexity. It can be calculated online in real time, providing core quantitative support for the collaborative management, fault diagnosis and strategy optimization of ECU multi-module.

[0091] The vehicle ECU intelligent control system first acquires a dynamic coupling relationship matrix, which originates from previous correlation processing of multi-source vehicle data. Specifically, a matrix factorization algorithm is used to calculate the correlation strength between each control module. Matrix factorization is a technique that decomposes a high-dimensional matrix into low-dimensional factors, such as non-negative matrix factorization, representing the dynamic coupling relationship matrix as the product of two low-rank matrices. Through this decomposition, the algorithm extracts latent features, for example, decomposing the dynamic coupling relationship matrix into U and V, where U represents the implicit representation of the control module and V represents the representation of the correlation dimension. During the calculation, the vehicle ECU intelligent control system iteratively optimizes and minimizes the reconstruction error, for example, by updating the values ​​of U and V through gradient descent, thereby quantifying the strength between modules. For example, the correlation strength between the motor control module and the braking module is calculated to be 0.75, indicating the closeness of their data interaction. This method allows for the separation of independent components from complex matrices, facilitating subsequent analysis.

[0092] Step S220: Construct the data flow topology based on the correlation strength and extract the working condition feature vector.

[0093] The operating condition feature vector is extracted using the following formula: (5) In formula (5), For the first The time-step condition feature vector, dimensionless, originates from data flow topology analysis, and has a dimension of [missing information]. , To control the total number of modules, the element represents the strength of the association between the corresponding modules; For the first Time step , The correlation strength between control modules is dimensionless and is derived from the calculation results of formula (4); The total number of control modules, without units, derived from the vehicle ECU control architecture, and its value range is a positive integer; The vector transpose operator converts row vectors into column vectors to adapt to the dimensional requirements of subsequent matrix operations and model inputs. The control logic of formula (5) aggregates the scattered inter-module relationships into globally unified operating condition features, fully covering the coupling state between all control modules, and realizing the global and quantitative representation of vehicle operating conditions; at the same time, the transpose operator adapts to the dimensional requirements of subsequent matrix operations and model inputs, with low computational complexity and real-time online generation, providing core standardized feature inputs for ECU intelligent control, operating condition identification, and strategy optimization.

[0094] A data flow topology is constructed based on the correlation strength, and operating condition feature vectors are extracted. The data flow topology is a directed graph structure, where nodes represent control modules and edge weights represent correlation strength. For example, during construction, the vehicle ECU intelligent management system uses threshold filtering to filter weak correlations, generating a connected graph. Next, operating condition feature vectors are extracted from the data flow topology. For example, a graph embedding algorithm is used to convert the topology into a vector representation, where the vector dimensions correspond to operating conditions such as speed and battery SOC, and the numerical values ​​reflect the current state.

[0095] Step S230: If the numerical fluctuation of the working condition feature vector exceeds the preset threshold, then process the data flow topology to obtain the flow density.

[0096] The data flow density is obtained from the topology using the following formula: (6) In formula (6), For the first The data flow density at each time step is a unitless density derived from a topology analysis algorithm, with a value range of [value missing]. A larger value indicates more intensive data transmission. This is a set of edges that flow from the data flow to the topology. It is unitless and originates from the topology construction result in step S220. The elements of the set are the associated edges between control modules. The total number of topological edges, unitless, derived from topological structure statistics, and its value ranges to positive integers; For the first Time step , The correlation strength between control modules is dimensionless and is derived from the calculation results of formula (4); For the summation operator, the set of opposite edges The correlation strengths of all modules in the topology are summed. The control logic of formula (6) is based on the set of edges from the data flow to the topology. It sums the correlation strengths between modules corresponding to all topological edges using a summation operator, and then normalizes the result by dividing by the total number of topological edges to obtain the first result. The flow density at each time step. This formula transforms the global relational state of the topology into a single scalar, quantifying the density of data transmission and the level of system coupling. It triggers calculations for scenarios where the fluctuation of the operating condition feature vector exceeds a threshold, accurately capturing the topological flow mutations when the vehicle's operating conditions change drastically. It has low computational complexity and can be calculated online in real time, providing core quantitative support for the dynamic adjustment of ECU control strategies, load assessment, and fault early warning.

[0097] If the fluctuation range of the operating condition feature vector exceeds a preset threshold, the data flow topology is processed to obtain the flow density. The fluctuation range is evaluated by calculating the Euclidean distance of the vectors. If it exceeds 0.5, the flow of edges in the topology is analyzed, for example, the amount of data packets transmitted per unit time is counted to obtain a density value such as 2.3 MB per second.

[0098] Step S240: Identify the core transmission nodes in the topology based on the flow density to determine the key data flow under the current vehicle operating conditions.

[0099] By identifying data flow density to the core transmission nodes in the topology, the critical data flow under changing vehicle operating conditions is determined. Core transmission nodes are selected through density sorting; for example, the node with the highest density is the central processing unit. This identifies the critical flow from sensors to the controller, ensuring the vehicle responds effectively to changes in operating conditions. Through this process, precise management of vehicle data is achieved.

[0100] Furthermore, the intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion provided in this embodiment includes step S300: Step S310: Obtain the correlation strength of key data streams, and use the sliding window algorithm to calculate the real-time fluctuation variance of the correlation strength to obtain the intensity fluctuation sequence.

[0101] The real-time variance of the correlation strength is derived using the following formula: (7) In formula (7), For the first The real-time variance of the association strength at each time step, dimensionless, derived from the sliding window algorithm, with a value range of [value missing]. A larger value indicates a more drastic fluctuation in the correlation strength. This is the size of the sliding window, without units, based on empirical configuration, and its value range is positive integers (e.g., ...). ); For the first The key data stream correlation strength at each time step is dimensionless and is derived from the calculation results of formula (4). This represents the average association strength within the sliding window, is dimensionless, and is derived from sliding window statistics. The calculation formula is as follows: The control logic of formula (7) is based on the correlation strength of key data streams within the sliding window. It first calculates the average correlation strength within the window as a benchmark, and then normalizes it using the sum of squared deviations and the window size to obtain the result. Real-time fluctuation variance of the time step. This formula transforms the dynamic change of correlation strength into a single scalar, quantifying the severity of fluctuations. It balances the real-time performance and noise resistance of fluctuation detection through an adjustable sliding window, with low computational complexity and real-time online calculation capability. It provides core quantitative support for ECU operating condition identification, dynamic adjustment of control strategies, and fault early warning.

[0102] The vehicle ECU intelligent control system first acquires the correlation strength of key vehicle data streams, such as the strength value between the motor control module and the sensor module, calculated by real-time monitoring of data interaction frequency. Specifically, a sliding window algorithm is used to calculate the real-time fluctuation variance of the correlation strength, resulting in an intensity fluctuation sequence. The sliding window algorithm is a time series processing technique that defines a fixed-size window that slides across the data stream, for example, a window size of 10 seconds. Within each window, the algorithm calculates the mean and variance of the correlation strength. The process involves collecting intensity samples within the window, such as five consecutive intensity values: 0.6, 0.7, 0.65, 0.72, and 0.68. Then, the average of these intensity samples (0.67) is calculated, and the variance is obtained to be approximately 0.0025, thus generating an intensity fluctuation sequence that reflects the trend of intensity changes. This method facilitates the capture of real-time dynamics, ensuring that the intensity fluctuation sequence accurately represents the data fluctuations under vehicle operating conditions.

[0103] Step S320: Extract peak features based on the intensity fluctuation sequence. If the correlation strength corresponding to the peak feature exceeds a preset threshold, trigger the resource allocation mechanism to obtain the initial computing power allocation matrix.

[0104] The initial computing power allocation matrix is ​​obtained using the following formula: (8) In formula (8), This is the initial computing power allocation matrix, which is dimensionless and derived from the resource allocation mechanism. Its dimensions are... , To control the number of modules, This refers to the number of computing nodes. For the first The control module is assigned to the first The initial computing power of a computing node, measured in TOPS (trillion operations per second), is derived from the resource allocation mechanism and its value ranges from [value missing]. ; This is the basic computing power quota, measured in TOPS, derived from the computing power node configuration, and its value range is a positive real number. For the first The correlation strength of the key data streams corresponding to the control module is unitless and comes from the calculation result of formula (4); A preset threshold for association strength is provided; it has no unit, is based on empirical configuration, and its value range is [value range missing]. ; The maximum computing power of the computing power node is expressed in TOPS, derived from hardware configuration, and its value range is positive real numbers. The control logic of formula (8) is based on the correlation strength of key data streams. When the correlation strength of the control module exceeds the preset threshold, a basic computing power quota is allocated to the module; otherwise, 0 computing power is allocated, and the computing power allocation result is filled into the correspondence between the module and the computing power node. The initial computing power allocation matrix is ​​defined. This formula enables precise on-demand allocation of computing power resources, allocating computing power only to modules with high load and high coordination requirements, thus avoiding resource waste. At the same time, matrix modeling achieves a global representation of the computing power allocation status, with low computational complexity and real-time online allocation, providing a core foundational model for dynamic scheduling and load balancing of ECU computing power.

[0105] Peak features are extracted from the intensity fluctuation sequence. If the correlation strength corresponding to the peak feature exceeds a preset threshold, a resource allocation mechanism is triggered to obtain an initial computing power allocation matrix. Specifically, peak feature extraction is achieved through a local maximum detection algorithm. For example, a peak point, such as the position with an intensity of 0.85, is identified in the intensity fluctuation sequence. If it exceeds the threshold of 0.8, the resource allocation mechanism is activated. This initial allocation mechanism involves constructing an initial computing power allocation matrix to represent computing power resources. For example, the rows of the initial computing power allocation matrix correspond to nodes such as processor 1 and processor 2, and the columns correspond to task types such as data fusion and transmission. The initial values ​​are filled based on historical load, thus providing a foundation for subsequent optimization.

[0106] Step S330: Evaluate the load status of the current computing power node through the initial computing power allocation matrix and determine the available link bandwidth of the computing power node.

[0107] The load status of the current computing nodes is assessed by using an initial computing power allocation matrix to determine the available link bandwidth of the computing nodes. Specifically, the assessment process includes summing the matrix elements to calculate the total load of each node. For example, the total load of processor 1 is 15 units. If the threshold is 20, then the available bandwidth is 5 units. This determination helps to identify bottleneck nodes.

[0108] Step S340: Construct a priority queue based on available link bandwidth, sort the transmission tasks in the priority queue using a greedy algorithm, obtain the fusion node corresponding to the scheduling instruction set, allocate computing resources preferentially based on the distribution characteristics of the fusion node, and obtain an optimized data fusion path.

[0109] The optimized data fusion path is derived using the following formula: (9) In formula (9), The optimized data fusion path is unitless, derived from a greedy algorithm, and is an ordered sequence of nodes, including the priority and transmission order of the fusion nodes. It is the set of all feasible fusion paths, unitless, and derived from the data flow topology; For nodes and The correlation strength between them is dimensionless and comes from the calculation results of formula (4); For nodes and The available link bandwidth between them, in Mbps, is derived from the bandwidth assessment result in step S330, and the value range is a positive real number. For nodes and The transmission delay between links, measured in milliseconds, originates from link status detection and takes the value of a positive real number. This is the maximum value operator, which has no unit and is used to select the path with the maximum objective function. The control logic of formula (9) uses the inter-node correlation strength, available link bandwidth, and transmission delay as core factors to construct the single-link objective function. The algorithm quantifies the overall performance of candidate paths using a summation operator, and then selects the optimal path from all feasible paths using a maximum value operator. This formula integrates multiple dimensions of indicators, including business priority, transmission capacity, and real-time performance. It achieves rapid path optimization through a greedy algorithm, exhibiting low computational complexity and real-time online execution. This provides core quantitative support for the dynamic scheduling of data fusion transmission and the priority allocation of computing resources, ensuring the transmission performance of high-priority fusion tasks and adapting to the real-time data fusion needs of dynamic vehicle operation scenarios.

[0110] A priority queue is constructed based on available link bandwidth. A greedy algorithm is used to sort the transmission tasks in the priority queue, obtain the fusion nodes corresponding to the scheduling instruction set, and allocate computing resources preferentially based on the distribution characteristics of the fusion nodes to obtain an optimized data fusion path. Specifically, the priority queue arranges tasks in descending order of bandwidth. For example, task A, which requires 3 units of bandwidth, is placed at the top. The greedy algorithm allocates resources one by one to ensure maximum efficiency. Fusion nodes, such as the central gateway, are given priority in resource allocation based on a distribution such as a star topology, thereby optimizing the transmission delay from the sensor to the controller. Through the above process, efficient scheduling of vehicle data is achieved.

[0111] Preferably, the intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion provided in this embodiment includes step S400 as follows: Step S410: By integrating key data streams and environmental factors through an optimized data fusion path, a spatial mapping matrix is ​​constructed using key data streams and meteorological sensing sequences.

[0112] The space mapping matrix is ​​constructed using the following formula: (10) In formula (10), This is a spatial mapping matrix, unitless, derived from the data fusion path, with dimensions of [missing information]. , As a key data flow dimension, From the perspective of environmental factors; This is a key data flow matrix, unitless, derived from the key data flow in step S410, with dimensions of [missing information]. , This represents the number of sampling points; This is an environmental factor (meteorological sensing sequence) matrix, without units, derived from the environmental data in step S410, with dimensions of [missing information]. ; As a transpose operator, the environmental factor matrix is... Transpose the matrix to match the dimensional alignment requirements of matrix multiplication and achieve deep fusion. The control logic of formula (10) is based on the key data flow matrix and the environmental factor matrix. First, the environmental factor matrix is ​​transposed using the transpose operator to achieve temporal dimensional alignment of the two types of data. Then, the key data flow and environmental factors are deeply fused through matrix multiplication to construct... The 3D spatial mapping matrix maps scattered vehicle operation data and environmental data into unified global spatial features, enabling deep fusion modeling of multi-source heterogeneous data.

[0113] The vehicle ECU intelligent control system first integrates key vehicle data streams, such as speed and steering angle, with environmental factors, such as the effects of rain and snow, through an optimized data fusion path. Specifically, this fusion process involves collecting meteorological sensing sequences, such as temperature and humidity data sequences obtained from onboard sensors, and then combining them with the key data streams to construct a spatial mapping matrix. In this matrix, rows represent spatial location points, such as road coordinates, and columns represent fused feature values, such as the slip coefficient, thus forming a comprehensive representation of the vehicle's surrounding environment.

[0114] Step S420: Extract the road surface adhesion estimate based on the spatial mapping matrix, and generate a fused feature vector through the road surface adhesion estimate.

[0115] The road surface adhesion estimate is obtained using the following formula: (11) In formula (11), For the first The estimated road surface adhesion at the time step is unitless, derived from the spatial mapping matrix, and its value ranges from [value range missing]. The higher the value, the stronger the road grip. For the first The vehicle speed at the time step, in km / h, is derived from the speed sensor data in step S110. For the first The vehicle acceleration at the time step, in m / s², is derived from velocity data differential calculation. For the first The ambient temperature at the time step, in °C, is derived from the environmental sensor data in step S110. For the first The road condition level at the time step has no unit and is derived from the road sensor data in step S110. Its value range is... (0 represents slippery, 1 represents dry); The regression coefficients are unitless, derived from model training, and their values ​​are real numbers. They are used to fit the influence of each parameter on the adhesion force. The control logic of formula (11) uses vehicle speed, acceleration, ambient temperature, and road condition level as core input parameters. A linear regression model is constructed through preset regression coefficients to calculate the road adhesion force estimate at time step t. The dispersed vehicle state, environmental state, and road condition parameters are integrated into a single scalar to quantify the road grip level and provide core working condition basis for ECU chassis control and braking strategy optimization.

[0116] The process of extracting road surface adhesion estimates from a spatial mapping matrix is ​​achieved by analyzing the distribution of elements in the matrix. Specifically, road surface adhesion estimate refers to the estimated value of the friction force between the road surface and the tire. It is affected by environmental factors such as water accumulation or ice. During extraction, the system scans the matrix to find low-value areas. For example, if the coefficient of friction of a road segment corresponding to a certain row in the spatial mapping matrix is ​​less than 0.5, the estimate is defined as low adhesion. This estimate generates a fusion feature vector, which integrates the estimated data with other perceived information such as wind speed to form a multi-dimensional vector for subsequent analysis, thereby providing a more accurate road surface condition reference in vehicle control.

[0117] Step S430: Obtain the projection coordinates of the fused feature vector, and calculate the state transition probability based on the projection coordinates.

[0118] The state transition probability is obtained by the following formula: (12) In formula (12), From state Leap to The probability, dimensionless, is derived from the projection calculation of the fused feature vector and its value ranges from [value missing]. ,and ; For the first , The vehicle status at each time step (such as acceleration, braking, steering, etc.) has no unit and is derived from the vehicle operating status classification. This is a set of vehicle states, which is unitless and originates from the state classification results. This is a weight vector, without units, derived from model training, and its dimension is the feature dimension. This is a state-action feature vector, without units, derived from the fused feature vector, and its dimension is the feature dimension; For the first The driving action of a time step has no unit and originates from vehicle control commands. This is the transpose of the weight vector, used for dimension adaptation in matrix multiplication, to achieve weighted projection of the feature vectors. It is an exponential function operator used to convert weighted projection values ​​into non-negative probability weights, thus avoiding the problem of negative probability; For the summation operator, for all possible next states The weights are summed to normalize the probability. The control logic of formula (12) is based on the state-action feature vector, which is weighted and projected through the weight vector. Then, the projected value is converted into a non-negative probability weight through the exponential function. Finally, the transition probability from the current state to the next state is obtained by summing and normalizing. The multi-source fusion features are mapped into a probability model of state transition, providing the core probability basis for ECU state prediction and strategy optimization.

[0119] After obtaining the projected coordinates of the fused feature vector, the state transition probability is calculated. Specifically, the projected coordinates are the coordinate points that map the vector to a low-dimensional space. For example, the vector is projected onto a two-dimensional plane through principal component analysis to obtain coordinates such as (0.3, 0.7). Then, the transition probability is calculated based on these coordinates, which is the probability of the vehicle transitioning from the current state, such as normal driving, to a potential slip state. The probability value is evaluated using a Markov chain model, such as 0.4, thereby capturing dynamic changes in driving.

[0120] Step S440: Generate the intention evolution trajectory based on the state transition probability, and determine whether the intention evolution trajectory is greater than the deviation threshold.

[0121] The deviation from the intended evolutionary trajectory is calculated using the following formula: (13) In formula (13), For the first The deviation of the intended evolution trajectory at each time step, dimensionless, derived from trajectory comparison calculations, with a value range of [value missing]. The larger the value, the farther the deviation from the reference trajectory; For the first The actual intention of the time step is to evolve the trajectory vector, which is dimensionless and comes from the state transition probability sequence; For the first The reference intention evolution trajectory vector for each time step is dimensionless and derived from the standard driving condition database. The 2-norm (Euclidean distance) operator is a unitless operator used to calculate the spatial distance between two vectors. The control logic of formula (13) is based on the actual intention evolution trajectory vector and the reference intention evolution trajectory vector, and calculates the spatial distance between the two vectors using the 2-norm (Euclidean distance) operator to obtain the first... Trajectory deviation at time steps. This formula maps the dispersed sequence of state transition probabilities to a single scalar, quantifying the degree of deviation between actual driving intention and standard operating conditions. It accurately captures significant trajectory deviations through Euclidean distance, with low computational complexity and real-time online calculation capabilities. It provides core quantitative support for ECU abnormal driving intention detection, safety warnings, and active control, effectively improving vehicle driving safety and intelligence.

[0122] The intention evolution trajectory is generated based on the state transition probability, and it is determined whether it exceeds a deviation threshold. Specifically, the intention evolution trajectory is a path curve simulated based on a probability sequence; for example, the intention evolution trajectory shows the evolution from a straight intention to a sharp turn.

[0123] Step S450: If the intention evolution trajectory is greater than the deviation threshold, it is determined that there is a need to adjust the driving intention.

[0124] If the curve deviates from the standard path by more than a threshold, such as 0.2, it is determined that there is a need to adjust driving intention, thereby triggering the intervention of the assistance system to improve safety. Through the above process, real-time monitoring of driving intention is achieved.

[0125] Furthermore, the intelligent control method for vehicle ECUs oriented towards multi-source heterogeneous data fusion provided in this embodiment includes step S500 as follows: Step S510: Determine the scheduling load requirement based on the driving intention adjustment needs, and obtain the upper limit of response delay based on the scheduling load requirement.

[0126] Dispatch load demand refers to the total amount of computational and control tasks that the vehicle control system must perform to meet adjustments in driving intent. Specifically, when the vehicle's ECU intelligent management system determines that a change from a straight-ahead intention to an emergency obstacle avoidance intention is needed, this adjustment demand is translated into a series of specific control command generation and execution tasks, such as path replanning, steering torque calculation, and brake pressure modulation. These tasks collectively constitute the dispatch load demand. The response latency limit refers to the maximum time allowed to complete this series of tasks, and it directly relates to the timeliness of control intervention.

[0127] Step S520: If the upper limit of response delay is lower than the preset delay threshold, then determine the module priority of each control module.

[0128] The module priority of each control module is determined by the following formula: (14) In formula (14), For the first The module priority of the control module is unitless, derived from multi-dimensional weighted calculation, and its value range is [value range missing]. A larger value indicates a higher priority. For the first The normalized value of the correlation strength of the control module is dimensionless and comes from the normalized calculation result of formula (4); For the first The normalized value of the data flow density of the control module is unitless and comes from the normalized calculation result of formula (6); For the first The real-time load normalization value of the control module is dimensionless and comes from the load detection of the computing nodes. These are weighting coefficients, unitless, derived from empirical configuration, and satisfy... The range of values ​​is The control logic of formula (14) takes the correlation strength of the control module, data flow density, and real-time load as three core indicators as inputs. First, it normalizes each indicator to eliminate the influence of dimensions, and then constructs a linear weighted model through preset weight coefficients to calculate the module priority. This formula integrates multi-dimensional indicators of business needs, transmission needs, and computing power needs. It achieves scene adaptive scheduling through adjustable weights, has low computational complexity, and can be calculated online in real time. It provides core priority basis for ECU computing power allocation, data scheduling, and control command sorting, effectively improving the scheduling efficiency and real-time performance of the system.

[0129] For stability control measures implemented on slippery surfaces to prevent skidding, the upper limit of the response delay is set at 50 milliseconds. If the preset general delay threshold is 100 milliseconds, it indicates that the current operating conditions place higher demands on the real-time performance of the control system. Determining module priorities relies on analyzing the criticality and dependencies of the control tasks. In one embodiment, the vehicle ECU intelligent management system divides control modules into critical path modules and auxiliary modules. For example, the electronic stability program module, directly responsible for vehicle yaw stability control, is assigned the highest priority, while the module responsible for updating the infotainment system interface is assigned the lowest priority. This division ensures that critical safety control functions receive priority resources when computing resources are scarce. Calculating resource occupancy quotas based on module priorities involves allocating a defined proportion of computing unit time slices or memory bandwidth to each module. Specifically, the vehicle ECU intelligent management system can employ a weighted allocation algorithm. For example, if the priority weight of the electronic stability program module is 0.6, the electric power steering module is 0.3, and other modules share 0.1, then the resource occupancy quotas will be allocated according to this weight ratio.

[0130] Step S530: Calculate the resource occupancy quota based on the module priority, and obtain the resource adjustment increment based on the resource occupancy quota.

[0131] Resource occupancy quotas are derived using the following formula: (15) In formula (15), For the first The resource consumption quota of the control module, without units (computing power / bandwidth ratio), is derived from priority normalization calculation, and its value range is [value missing]. ; For the first The module priority of the control module has no unit and is derived from the calculation result of formula (14); This represents the total system resource quota, without units, derived from system resource allocation, and its value range is [range missing]. (Total percentage is 1); The total number of control modules is unitless and derived from the vehicle ECU control architecture. The control logic of formula (15) is based on the priority of each control module. First, the proportion of the module's priority to the total priority is calculated, then multiplied by the total system resource limit to obtain the result. Resource occupancy quotas for control modules. This formula maps dispersed module priorities to explicit resource allocation ratios, realizing the scheduling logic of prioritizing resource allocation for high-priority modules. By summing and normalizing, it strictly satisfies the allocation ratio constraints, has low computational complexity, and can be calculated online in real time. It provides a core quota basis for ECU resource scheduling, load balancing, and computing power / bandwidth allocation, effectively improving the system's resource utilization and scheduling efficiency.

[0132] The incremental resource adjustment is derived using the following formula: (16) In formula (16), For the first The resource adjustment increment of the control module is unitless, derived from resource quota comparison calculation, and its value range is [missing value]. ; This is an adjustment coefficient, dimensionless, derived from empirical configuration, and its value range is [range missing]. , used to control the adjustment range; For the first The target resource occupancy quota of the control module has no unit and is derived from the calculation result of formula (15); For the first The original resource occupancy quota of the control module is unitless and comes from the original control logic configuration. The control logic of formula (16) is based on the target resource quota and the original resource quota of the control module. First, the difference between the two (target - original) is calculated to quantify the resource demand gap / redundancy. Then, the difference is scaled by the adjustment coefficient to obtain the first... The incremental resource adjustment of the control module maps the dispersed quota differences into a clear resource adjustment range, realizing dynamic closed-loop adjustment of resources and providing a core adjustment basis for ECU resource scheduling and load balancing.

[0133] Resource adjustment increments are calculated based on the difference between the current system load and the resource occupancy quota, and are used to dynamically adjust the resource quotas of each module. Adjusting the original control logic through resource adjustment increments means that the system will dynamically adjust its task scheduling strategy.

[0134] Step S540: Modify the original control logic by adjusting resources incrementally to generate a real-time adapted control module management sequence.

[0135] The control module's management sequence is generated using the following formula: (17) In formula (17), This is a control module management sequence, without units, derived from the adjustment incremental correction results, and is an ordered sequence containing the identifier of each control module and the updated resource usage quota; For the first The identifier for each control module has no unit and is derived from the control module number. For the first The target resource occupancy quota of the control module has no unit and is derived from the calculation result of formula (15); For the first The resource adjustment increment of the control module is unitless and comes from the calculation result of formula (16). The control logic of formula (17) is based on the target resource quota of each control module. The original quota is corrected by the resource adjustment increment, the updated resource occupancy quota is calculated, and then an ordered control sequence is generated according to the module priority. This formula maps the dispersed module priority, resource quota, and adjustment increment into a unified scheduling sequence. The control logic is optimized in a closed loop through incremental correction. The priority order ensures the scheduling priority of high priority modules. The calculation complexity is low and it can be executed online in real time. It provides the core scheduling basis for ECU resource scheduling, control command execution, and system adaptive control, and effectively improves the system's operating efficiency and real-time performance.

[0136] The original control logic dictated that all modules would run at a fixed cycle. The revised logic, however, allows high-priority modules to run preemptively with shorter cycles and more frequent preemptive actions, while temporarily suspending or extending the running cycles of low-priority modules. The resulting control sequence is a detailed task execution schedule that specifies when each control module will start, for how long it will run, and when it will be suspended in the next control cycle, thus ensuring that control intentions with high real-time requirements are executed accurately and promptly.

[0137] Preferably, the intelligent control method for vehicle ECUs oriented towards multi-source heterogeneous data fusion provided in this embodiment includes step S600 as follows: Step S610: Input the control sequence of the control module into the optimization training model. The optimization training model reaches a convergence state by iteratively updating the weight matrix.

[0138] The iteratively updated weight matrix is ​​obtained using the following formula: (18) In formula (18), For the first , The weight matrix of the next iteration is unitless and comes from the optimized training model. Its dimension is the feature dimension × the output dimension. The learning rate is unitless and derived from the model's hyperparameter configuration; its value range is [range missing]. ; This is the gradient of the loss function with respect to the weights, which is dimensionless and comes from the backpropagation algorithm; it is used to update the weights. is the loss function, which is unitless and derived from the model definition (such as mean squared error loss), used to measure the model prediction error. The control logic of formula (18) is based on the gradient descent algorithm. Taking the current weight matrix as the benchmark, the weights are iteratively updated along the opposite direction of the gradient of the loss function with the learning rate as the step size, gradually reducing the model prediction error until the weight matrix reaches the convergence state, thereby realizing the adaptive fitting of the optimized training model to the control module's control sequence.

[0139] The vehicle ECU intelligent management system first imports the generated control module control sequence as input data into an optimization training model. This optimization training model is based on a neural network architecture and aims to optimize the response efficiency of the vehicle control system through a continuous iterative process. For example, the optimization training model uses gradient descent to iteratively update the weight matrix. Specifically, after the control module control sequence is input, the optimization training model initializes the weight matrix with random values, then calculates the loss function in each iteration cycle, and adjusts the weight values ​​through backpropagation until the loss function value is lower than a preset threshold, thus achieving convergence. This iterative update ensures that the model can gradually adapt to the control requirements of different driving scenarios. For example, in congested urban areas, the optimization training model learns through multiple iterations how to prioritize resource allocation to the braking module to reduce response latency.

[0140] Step S620: Based on the convergence state, obtain the feedback signal extracted by the optimized training model that is in the convergence state.

[0141] The feedback signal is the output vector extracted from the converged optimized training model, which represents the predictive performance evaluation of the control module's control sequence.

[0142] Step S630: Compare the feedback signal with the operation log and calculate the deviation between the feedback signal and the operation log.

[0143] The deviation between the feedback signal and the operation log is calculated using the following formula: (19) In formula (19), For the first The average deviation between the time step feedback signal and the operation log is a unitless value derived from comparative calculation, and its range is [value missing]. A larger value indicates a more significant deviation. For the first The first time step feedback signal 3D features, unitless, derived from step S620 optimization training model extraction; For the first The first step of the time step run log Dimensional feature, without unit, derived from vehicle ECU operation log; This represents the total number of feature dimensions, which is unitless and originates from the feature extraction configuration. Its value range is positive integers. For the summation operator, the summation operation is performed by traversing all feature dimensions and accumulating the sum. The control logic of formula (19) is based on the feedback signal features and the operation log features. First, the absolute difference of each dimension feature is calculated to eliminate the cancellation of positive and negative deviations. Then, the differences of all dimensions are summed and averaged to obtain the first... The average deviation value at each time step. This formula maps dispersed multi-dimensional feature differences to a single scalar, ensures the unbiasedness of deviation quantification through absolute value operations, and adapts to different feature dimensions through average normalization. It has low computational complexity, can be calculated online in real time, and provides a core evaluation basis for model effectiveness verification and iterative optimization of control strategies, effectively improving the accuracy of model training and the stability of system operation.

[0144] The vehicle ECU intelligent control system compares the feedback signal with the vehicle operation log, which records the timestamp and module status during actual execution. The Euclidean distance is calculated as the deviation value. For example, if the log shows that the braking response is 80 milliseconds while the feedback signal predicts 70 milliseconds, the deviation is 10 milliseconds. This allows for the quantification and optimization of the difference between the training model's prediction and the actual response.

[0145] Step S640: Quantify the deviation value by associating it with a preset redundancy factor to determine the system reliability improvement index corresponding to the control module's control sequence. The system reliability improvement index is used to evaluate performance.

[0146] The system reliability improvement index is derived from the following formula: (20) In formula (20), This is a system reliability improvement indicator, without units, derived from quantification, and its value range is [value range missing]. The closer the value is to 1, the more significant the improvement in system reliability. For the first The average deviation value of the time step is unitless and comes from the calculation result of formula (19); This represents the total number of time steps, with no unit, derived from the statistical period configuration, and its value range is positive integers. This is a redundancy factor, dimensionless, derived from empirical configuration, and its value range is [range missing]. This is used to quantify the effect of redundant design on suppressing deviations; It is a normalization operator with no unit, used to convert deviations into reliability indicators. The weighted average deviation term is calculated by weighting the deviations of all time steps within the statistical period, with redundancy factors as weights, to quantify the global average deviation level within the period. The control logic of formula (20) is based on the average deviation of each time step within the statistical period. The deviation is weighted and suppressed by redundancy factors to calculate the global average weighted deviation within the period. Then, the deviation level is converted into a system reliability improvement index by subtracting normalization from 1, realizing the quantitative mapping from multi-time-step deviations to a single reliability index, and providing a core quantitative basis for the performance evaluation of the control method.

[0147] The deviation value is quantified by correlating it with a preset redundancy factor. This redundancy factor, a safety margin coefficient (e.g., 1.2) based on historical data, amplifies the deviation to assess potential risks. Finally, a percentage value is determined as the reliability improvement index for the vehicle ECU intelligent control system, used for overall performance evaluation. Through this process, the vehicle ECU intelligent control system achieves dynamic optimization of the control module, improving vehicle safety and adaptability.

[0148] Please see Figure 2 This invention provides a vehicle ECU intelligent control system for multi-source heterogeneous data fusion, used to implement the aforementioned vehicle ECU intelligent control method for multi-source heterogeneous data fusion. It includes a dynamic coupling relationship matrix acquisition module 10, a key data flow determination module 20, a data fusion path acquisition module 30, a driving intention adjustment requirement judgment module 40, a control module control sequence acquisition module 50, and a system reliability improvement index determination module 60. The dynamic coupling relationship matrix acquisition module 10 is used to collect multi-source heterogeneous data using a data acquisition device, process the correlation relationships between the multi-source heterogeneous data using a correlation analysis model, and obtain a dynamic coupling relationship matrix. The multi-source heterogeneous data includes various vehicle operation information. The key data flow determination module 20 is used to determine the dynamic coupling relationship matrix based on the data fusion path. The system employs a matrix to calculate the correlation strength between control modules and determine key data flows under changing vehicle operating conditions. A data fusion path acquisition module 30 prioritizes allocating computing resources to obtain an optimized data fusion path if the correlation strength of key data flows exceeds a preset threshold. A driving intent adjustment requirement judgment module 40 integrates key data flows with environmental factors through the optimized data fusion path to determine driving intent adjustment requirements. A control module management sequence acquisition module 50 updates resource scheduling strategies based on driving intent adjustment requirements to obtain a real-time adapted control module management sequence. A system reliability improvement index determination module 60 trains the control module management sequence using an optimized training model, obtains feedback signals under performance optimization objectives, and determines the system reliability improvement index.

[0149] The vehicle ECU intelligent control method and system for multi-source heterogeneous data fusion provided in this embodiment, compared with the prior art, analyzes the correlation between data by constructing a dynamic coupling relationship matrix, identifies key data streams exceeding a preset threshold and prioritizes the allocation of computing resources, forms an optimized data fusion path, integrates environmental factors to determine driving intention adjustment needs, updates resource scheduling strategies accordingly to generate a real-time adapted control module control sequence, and uses an optimized training model to train the control module control sequence to obtain system reliability improvement indicators. Thus, it achieves accurate and rapid response to key data streams and optimal dynamic allocation of computing resources under complex working conditions, improving the real-time performance, adaptability and overall reliability of the vehicle control system.

[0150] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for intelligent control of vehicle ECUs oriented towards multi-source heterogeneous data fusion, characterized in that, Includes the following steps: S100. Collect multi-source heterogeneous data using a data acquisition device, process the correlation relationship between the multi-source heterogeneous data using a correlation analysis model, and obtain a dynamic coupling relationship matrix, wherein the multi-source heterogeneous data includes various vehicle operation information. S200. Based on the dynamic coupling relationship matrix, calculate the correlation strength between each control module and determine the key data flow under the current vehicle operating condition changes; S300. If the correlation strength of the key data stream exceeds a preset threshold, computing resources are allocated first to obtain an optimized data fusion path. S400: By integrating the key data streams with the influence of environmental factors through an optimized data fusion path, the driving intention adjustment needs are determined; S500: Adjust the resource scheduling strategy according to the driving intention to obtain a real-time adapted control module management sequence; S600. The control module's management sequence is trained using an optimized training model to obtain feedback signals under the performance optimization objective and to determine the system reliability improvement index.

2. The intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion according to claim 1, characterized in that, Step S100 includes: S110. The data acquisition device acquires multi-source heterogeneous data such as vehicle speed, motor speed, battery SOC, battery temperature, braking frequency, accelerator pedal opening, steering angle, ambient temperature, road conditions, passenger load, tire pressure, and battery voltage from various vehicle sensors, and performs unified format conversion on the multi-source heterogeneous data using a pre-established standardized protocol to obtain a standardized data set. S120. For the standardized data set, an association rule mining algorithm is used to process the association between vehicle speed and battery SOC, the association between motor speed and accelerator pedal opening, and the association between braking frequency and road conditions to determine preliminary association strength indicators. S130. Based on the preliminary correlation strength index, the correlation between ambient temperature and battery temperature, the correlation between steering angle and passenger load, and the correlation between battery voltage and tire pressure are incorporated to construct a dynamic update mechanism and obtain an extended correlation network. S140. If the correlation strength index in the extended correlation network exceeds a preset threshold, the dynamic coupling parameters are determined by fusing the time-varying correlation between vehicle speed and braking frequency and the time-varying correlation between battery SOC and road conditions through a time-series analysis model. S150. A matrix construction algorithm is used to process the multidimensional correlation between the dynamic coupling parameters and the accelerator pedal opening, the ambient temperature, and the battery temperature to obtain a dynamic coupling relationship matrix.

3. The intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion according to claim 2, characterized in that, Step S200 includes: S210. Obtain the dynamic coupling relationship matrix and use the matrix decomposition algorithm to calculate the correlation strength between each control module; S220. Construct a data flow topology based on the correlation strength and extract working condition feature vectors; S230. If the numerical fluctuation of the working condition feature vector exceeds a preset threshold, the data flow topology is processed to obtain the flow density. S240. Identify the core transmission node in the data flow topology based on the flow density, and determine the key data flow under the current vehicle operating condition changes.

4. The intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion according to claim 1, characterized in that, Step S300 includes: S310. Obtain the correlation strength of the key data streams, and use the sliding window algorithm to calculate the real-time fluctuation variance of the correlation strength to obtain the intensity fluctuation sequence. S320. Extract peak features based on the intensity fluctuation sequence. If the correlation strength corresponding to the peak features exceeds a preset threshold, trigger the resource allocation mechanism to obtain the initial computing power allocation matrix. S330. Evaluate the load status of the current computing power node through the initial computing power allocation matrix, and determine the available link bandwidth of the computing power node; S340. Construct a priority queue based on the available link bandwidth, sort the transmission tasks in the priority queue using a greedy algorithm, obtain the fusion node corresponding to the scheduling instruction set, allocate computing resources preferentially based on the distribution characteristics of the fusion node, and obtain an optimized data fusion path.

5. The intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion according to claim 4, characterized in that, Step S400 includes: S410. By optimizing the data fusion path, the key data stream and the influence of environmental factors are fused, and a spatial mapping matrix is ​​constructed using the key data stream and meteorological sensing sequence. S420. Extract the road surface adhesion estimate based on the spatial mapping matrix, and generate a fused feature vector using the road surface adhesion estimate; S430. Obtain the projection coordinates of the fused feature vector, and calculate the state transition probability based on the projection coordinates; S440. Generate an intention evolution trajectory based on the state transition probability, and determine whether the intention evolution trajectory is greater than the deviation threshold. S450. If the intention evolution trajectory is greater than the deviation threshold, it is determined that there is a need to adjust the driving intention.

6. The intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion according to claim 1, characterized in that, Step S500 includes: S510. Determine the scheduling load requirement based on the driving intention adjustment needs, and obtain the upper limit of response latency based on the scheduling load requirement; S520. If the upper limit of the response delay is lower than the preset delay threshold, then determine the module priority of each control module. S530. Calculate the resource occupancy quota based on the module priority, and obtain the resource adjustment increment based on the resource occupancy quota; S540. The original control logic is corrected by adjusting the resource increment to generate a real-time adapted control module management sequence.

7. The intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion according to claim 6, characterized in that, Step S600 includes: S610. The control module control sequence is input into the optimization training model, and the optimization training model reaches a convergent state by iteratively updating the weight matrix. The iteratively updated weight matrix is ​​obtained using the following formula: ； in, For the first , The weight matrix of the next iteration. For learning rate, The gradient of the loss function with respect to the weights. The loss function; S620. Based on the convergence state, obtain the feedback signal extracted by the optimized training model in the convergence state; S630. Compare the feedback signal with the operation log, and calculate the deviation value between the feedback signal and the operation log; S640. The deviation value is associated with a preset redundancy factor for quantification processing to determine the system reliability improvement index corresponding to the control sequence of the control module, wherein the system reliability improvement index is used to evaluate performance.

8. The intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion according to claim 7, characterized in that, In step S630, the deviation between the feedback signal and the operation log is obtained using the following formula: ； in, For the first The average deviation between the time step feedback signal and the operation log. For the first The first time step feedback signal dimensional features, For the first The first step of the time step run log dimensional features, This represents the total number of feature dimensions.

9. The intelligent control method for vehicle ECUs based on multi-source heterogeneous data fusion according to claim 8, characterized in that, In step S640, the system reliability improvement index is obtained using the following formula: ； in, As an indicator for improving system reliability, For the first The average deviation of the time step. To count the total number of time steps, As a redundancy factor, This is the normalization operator.

10. A vehicle ECU intelligent control system for multi-source heterogeneous data fusion, used to implement the vehicle ECU intelligent control method for multi-source heterogeneous data fusion as described in any one of claims 1 to 9, characterized in that, include: The dynamic coupling relationship matrix acquisition module is used to collect multi-source heterogeneous data using a data acquisition device, process the relationship between the multi-source heterogeneous data using an association analysis model, and obtain a dynamic coupling relationship matrix, wherein the multi-source heterogeneous data includes various vehicle operation information. The key data flow determination module is used to calculate the correlation strength between each control module based on the dynamic coupling relationship matrix, and to determine the key data flow under the current vehicle operating condition changes. The data fusion path acquisition module is used to prioritize the allocation of computing resources and obtain an optimized data fusion path if the correlation strength of the key data stream exceeds a preset threshold. The driving intent adjustment demand judgment module is used to determine the driving intent adjustment demand by fusing the key data streams and the influence of environmental factors through an optimized data fusion path. The control module management sequence acquisition module is used to update the resource scheduling strategy according to the driving intention adjustment requirements to obtain a real-time adapted control module management sequence. The system reliability improvement index determination module is used to train the control sequence of the control module using an optimized training model, obtain feedback signals under the performance optimization target, and determine the system reliability improvement index.

Images