Vehicle software cognitive decision-making method and device based on neural symbol fusion

Abstract

The invention relates to a vehicle software cognitive decision-making method and device based on neural symbol fusion. The method comprises the following steps: acquiring multi-modal data of a vehicle in real time; performing preprocessing on the multi-mode data; respectively inputting the preprocessed multi-modal data into a symbol reasoning path and a neural network path to obtain a first decision vector, a confidence coefficient of the first decision vector, a second decision vector and a confidence coefficient of the second decision vector; based on the multi-modal data, identifying a vehicle scene type through a preset random forest classifier; dynamically adjusting the weight of the confidence coefficient of the first decision vector and the weight of the confidence coefficient of the second decision vector based on the recognition result of the vehicle scene type; and carrying out weighted fusion on the first decision vector and the second decision vector, and outputting a third decision vector. Through organic combination of a symbol reasoning path, a neural network path and scene recognition, decision self-adaption, transparency and maintainability are realized.

Description

Technical Field

[0001] This invention belongs to the field of intelligent connected vehicle software technology, specifically relating to a vehicle software cognitive decision-making method and device based on neural symbol fusion. Background Technology

[0002] With the rapid development of intelligent connected vehicles, over-the-air (OTA) software download technology has become a key technology in the automotive electronics field. Currently, vehicle software rewriting decisions mainly adopt the following two technical approaches: 1. Rule-based decision-making systems: These systems make decisions through predefined rewriting rules and procedures. This system relies on a rule base built from expert experience, ensuring the interpretability of the decision-making process, but suffers from poor flexibility and insufficient adaptability. 2. Machine learning-based decision-making systems: These systems learn decision patterns from historical data through neural network models. This method has strong pattern recognition capabilities, but the decision-making process lacks interpretability and is highly dependent on the quality and quantity of data.

[0003] The existing technology has the following main technical defects: 1. Decision rigidity problem: Rule-based systems cannot adapt to complex dynamic environments. When new scenarios not covered by predefined rules arise, the system cannot make effective decisions.

[0004] 2. Black-box decision-making problem: Although neural network-based systems have strong learning capabilities, their decision-making process is not transparent, making it difficult to meet the high safety and reliability requirements of the automotive industry.

[0005] 3. Lagging knowledge updates: Both systems struggle to achieve continuous knowledge evolution and require manual intervention to update and maintain rules or models.

[0006] 4. Insufficient utilization of multi-source information: Existing systems fail to effectively integrate multimodal information such as vehicle status, software characteristics, and environmental context, resulting in limited comprehensiveness and accuracy of decision-making. Summary of the Invention

[0007] To address the problems raised in the background art, a first aspect of the present invention provides a vehicle software cognitive decision-making method based on neural symbolic fusion, comprising: real-time acquisition of multimodal data of the vehicle, the multimodal data including ECU status data, feature data of flashing software, and environmental context data; preprocessing the multimodal data; inputting the preprocessed multimodal data into a symbolic inference pathway and a neural network pathway respectively to obtain a first decision vector, a confidence level of the first decision vector, a second decision vector, and a confidence level of the second decision vector; identifying vehicle scene types based on the multimodal data using a preset random forest classifier; dynamically adjusting the weights of the confidence levels of the first and second decision vectors based on the vehicle scene type identification results; performing weighted fusion of the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence levels of the second decision vector, and outputting a third decision vector.

[0008] In some embodiments of the present invention, the step of inputting preprocessed multimodal data into a symbolic inference pathway and a neural network pathway respectively to obtain a first decision vector, a confidence level of the first decision vector, a second decision vector, and a confidence level of the second decision vector includes: inputting the preprocessed multimodal data into a symbolic inference pathway, wherein the symbolic inference pathway obtains the first decision vector and a confidence level of the first decision vector through a preset rule base, forward chain inference, and confidence level calculation; and inputting the preprocessed multimodal data into a neural network pathway, wherein the neural network pathway outputs the second decision vector and a confidence level of the second decision vector through a Transformer encoder and a temperature scaling method.

[0009] Furthermore, the symbolic reasoning pathway obtains the first decision vector and its confidence level through a preset rule base, forward chain reasoning, and confidence level calculation, including: defining security constraints, system dependencies, and business logic for the write operation based on the rule base; generating the first decision vector based on the preprocessed multimodal data through forward chain reasoning of security constraints, system dependencies, and business logic; and calculating the confidence level of the first decision vector based on its rule matching degree, constraint satisfaction degree, and historical accuracy.

[0010] Furthermore, the neural network pathway outputs a second decision vector and its confidence level through a Transformer encoder and a temperature scaling method, including: the neural network pathway outputs a second decision vector through a Transformer encoder and forward propagation; calculates the model calibration degree through a temperature scaling method; calculates the prediction consistency based on the variance of the model prediction results; and calculates the confidence level of the second decision vector based on the model calibration degree, prediction consistency, and preset feature contribution.

[0011] In some embodiments of the present invention, the step of identifying vehicle scene types based on the multimodal data and using a preset random forest classifier includes: determining a first scene feature based on vehicle speed, voltage, and temperature in the multimodal data; determining a second scene feature based on network trends and user behavior in the multimodal data; and identifying the vehicle scene type based on the first scene feature and the second scene feature using a preset random forest classifier.

[0012] Furthermore, the dynamic adjustment of the confidence levels of the first decision vector and the second decision vector based on the vehicle scene type recognition result includes: if the vehicle scene is identified as a first scene or a second scene, then the confidence weights of the first decision vector and the second decision vector are determined according to preset values; if the vehicle scene is identified as a mixed scene, then the weights of the first decision vector and the second decision vector are determined according to the confidence ratio.

[0013] A second aspect of the present invention provides a vehicle software cognitive decision-making device based on neural symbolic fusion, comprising: an acquisition module for acquiring multimodal data of a vehicle in real time, the multimodal data including ECU status data, feature data of flashing software, and environmental context data; preprocessing the multimodal data; an input module for inputting the preprocessed multimodal data into a symbolic reasoning pathway and a neural network pathway, respectively, to obtain a first decision vector, a confidence level of the first decision vector, a second decision vector, and a confidence level of the second decision vector; a recognition module for recognizing vehicle scene types based on the multimodal data using a preset random forest classifier; an adjustment module for dynamically adjusting the weights of the confidence levels of the first and second decision vectors based on the recognition results of the vehicle scene types; and an output module for weighted fusion of the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector, and outputting a third decision vector.

[0014] A third aspect of the present invention provides an electronic device comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the vehicle software cognitive decision-making method based on neural symbol fusion provided in the first aspect of the present invention.

[0015] In a fourth aspect, the present invention provides a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the vehicle software cognitive decision-making method based on neural symbol fusion provided in the first aspect of the present invention.

[0016] The beneficial effects of this invention are: 1. Unified intelligent decision-making and interpretability: Through neural symbol fusion, it simultaneously obtains the powerful pattern recognition capabilities of deep learning and the logical rigor and interpretability of symbolic reasoning, effectively solving the black box problem of traditional AI systems.

[0017] 2. Significantly enhanced system adaptability: It can handle more than 90% of undefined scenarios, far exceeding traditional rule systems (about 60%), and can dynamically adjust decision-making strategies in real time, supporting online learning and continuous knowledge evolution.

[0018] 3. Significantly improved decision-making efficiency: The parallel processing architecture reduces the overall decision response time from 120ms using traditional methods to less than 40ms, increases system throughput by 3 times, and reduces CPU utilization by 25%. 4. Improved reliability: The decision-making accuracy has increased from 89% with traditional methods to over 96.5%, and the system's fault tolerance has been enhanced, maintaining over 85% functional integrity even under single point of failure.

[0019] 5. Enhanced practicality: Supports complex scenarios such as emergency security updates and multi-ECU collaborative flashing, is easy to deploy, is compatible with mainstream automotive protocol standards, and has 100% compatibility.

[0020] 6. Significant economic benefits: It is estimated that after-sales service costs caused by flashing failures can be reduced by more than 60%, the manpower required for system maintenance can be reduced by 70%, and software upgrade costs can be saved by 40%. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the basic process of a vehicle software cognitive decision-making method based on neural symbol fusion in some embodiments of the present invention; Figure 2 This is a schematic diagram illustrating the specific process of a vehicle software cognitive decision-making method based on neural symbol fusion in some embodiments of the present invention. Figure 3 This is a schematic diagram of the structure of a vehicle software cognitive decision-making device based on neural symbol fusion in some embodiments of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device in some embodiments of the present invention. Detailed Implementation

[0022] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0023] Example 1

[0024] refer to Figure 1 and Figure 2In a first aspect of the present invention, a vehicle software cognitive decision-making method based on neural symbol fusion is provided, comprising: S100. acquiring multimodal data of the vehicle in real time, the multimodal data including ECU status data, feature data of flashing software, and environmental context data; preprocessing the multimodal data; S200. inputting the preprocessed multimodal data into a symbolic reasoning pathway and a neural network pathway respectively to obtain a first decision vector, a confidence level of the first decision vector, a second decision vector, and a confidence level of the second decision vector; S300. identifying vehicle scene types based on the multimodal data using a preset random forest classifier; S400. dynamically adjusting the weights of the confidence levels of the first decision vector and the second decision vector based on the vehicle scene type identification results; S500. performing weighted fusion of the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector, and outputting a third decision vector.

[0025] In step S100 of some embodiments of the present invention, multimodal data of the vehicle is acquired in real time, including ECU status data, feature data of flashing software, and environmental context data. Specifically, vehicle status data such as ECU CPU load, memory usage, power supply voltage, and network quality (packet loss rate, latency) are collected at a frequency of 100Hz via the CAN bus, and Kalman filtering is applied for smoothing and noise reduction. The binary file of the software to be flashed is parsed to statically extract the version number, digital signature, and dependency library list; its runtime API call sequence and resource usage pattern are dynamically monitored. The vehicle's geographical location and network signal strength (RSRP / RSSI) are obtained through GPS, network modules, etc., and user historical behavior data is analyzed to identify operating preferences. All the above numerical parameters are normalized to the [0,1] interval, and a 256-dimensional multimodal feature vector is constructed for use by the neural network. At the same time, key parameters (such as vehicle_speed=0) are converted into logical facts for symbolic reasoning.

[0026] In step S200 of some embodiments of the present invention, the step of inputting the preprocessed multimodal data into the symbolic inference pathway and the neural network pathway respectively to obtain the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector includes: S201. The preprocessed multimodal data is input into the symbolic reasoning pathway, which obtains the first decision vector and the confidence of the first decision vector through a preset rule base, forward chain reasoning and confidence calculation. Furthermore, the symbolic reasoning pathway obtains the first decision vector and its confidence level through a preset rule base, forward chain reasoning, and confidence level calculation, including: defining security constraints, system dependencies, and business logic for the write operation based on the rule base; generating the first decision vector based on the preprocessed multimodal data through forward chain reasoning of security constraints, system dependencies, and business logic; and calculating the confidence level of the first decision vector based on its rule matching degree, constraint satisfaction degree, and historical accuracy.

[0027] Specifically, load the first-order predicate logic rule base (including security constraints, system dependencies, and business logic rules). Input logical facts into the engine and perform forward chain reasoning. For example, verify whether the rule `safe_to_flash(Software):-vehicle_speed(0),battery_level(Level),Level>0.3` is satisfied. Output symbolic decision suggestions (such as "Allow flashing") and calculate its confidence score C_symbol (based on rule matching degree and historical accuracy).

[0028] S202. The preprocessed multimodal data is input into a neural network pathway, which outputs a second decision vector and the confidence level of the second decision vector through a Transformer encoder and a temperature scaling method.

[0029] Furthermore, the neural network pathway outputs a second decision vector and its confidence level through a Transformer encoder and a temperature scaling method, including: the neural network pathway outputs a second decision vector through a Transformer encoder and forward propagation; calculates the model calibration degree through a temperature scaling method; calculates the prediction consistency based on the variance of the model prediction results; and calculates the confidence level of the second decision vector based on the model calibration degree, prediction consistency, and preset feature contribution.

[0030] Specifically, a 256-dimensional feature vector is input into an 8-layer Transformer encoder. Feature weights are calculated using a self-attention mechanism, and the output decision probability distribution is obtained after a nonlinear transformation (e.g., [allowed: 0.75, delayed: 0.20, forbidden: 0.05]). The neural confidence C_neural (based on model calibration and prediction consistency) is calculated.

[0031] In step S300 of some embodiments of the present invention, identifying the vehicle scene type based on the multimodal data using a preset random forest classifier includes: S301. Based on vehicle speed, voltage, and temperature in multimodal data, determine the first scene features; based on network trends and user behavior in multimodal data, determine the second scene features; S302. Based on the first scene features and the second scene features, identify the vehicle scene type using a preset random forest classifier.

[0032] Furthermore, in step S400 of some embodiments of the present invention, the dynamic adjustment of the weights of the confidence scores of the first decision vector and the second decision vector based on the vehicle scene type recognition result includes: S401. If the vehicle scene is identified as either the first scene or the second scene, then the confidence weights of the first decision vector and the first decision vector are determined according to preset values. Specifically, for safety-critical scenarios: W_symbol=0.9, W_neural=0.1; for pattern recognition scenarios: W_symbol=0.2, W_neural=0.8; W_symbol and W_neural represent the weights of the symbolic reasoning pathway and the neural network pathway, respectively. S402. If the vehicle scene is identified as a mixed scene, the weights of the first decision vector and the second decision vector are determined according to the confidence ratio.

[0033] Specifically, in mixed scenarios, confidence ratio allocation is used: W_symbol=C_symbol / (C_symbol+C_neural), W_neural = 1 - W_symbol, Where C_symbol and C_neural represent the confidence levels of the symbolic reasoning pathway and the neural network pathway, respectively.

[0034] In step S500 of some embodiments of the present invention, the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector are weighted and fused, and a third decision vector is output.

[0035] Specifically, the output vector is standardized: Symbolic output vectorization: [Allow → 1.0, Delay → 0.0, Forbid → 0.0]; Neural output calibration: Adjust the probability distribution using temperature scaling; Ensure that the two vectors are in the same metric space.

[0036] Linear weighted fusion: Perform vector weighting: Final_Vector = W_symbol × Symbolic_Vector + W_neural × Neural_Vector; Fusion quality assessment: Calculate the entropy value of the fused vector to assess decision determinism; Final decision generation: Select the category with the highest probability as the output. For example, the output format is: {decision:"allow|delay|deny",confidence:0.xx,timestamp:"yyyy-MM-ddHH:mm:ss"} It should be noted that the cognitive decision generator formulates a detailed flashing plan (including the target ECU, resource allocation, and rollback scheme) based on the final decision. It executes the flashing operation and monitors data transmission and resource usage throughout the process, triggering a handling mechanism in case of anomalies.

[0037] To achieve continuous iteration and updates in decision-making, the system is optimized based on evaluation results: the symbolic rule base is revised or expanded; the neural network is incrementally trained using new data; and the parameters in the dynamic weight calculation model are adjusted to optimize weight allocation in similar scenarios in the future. Real-time performance: data acquisition and preprocessing ≤5ms, dual-path parallel processing ≤20ms, decision generation ≤15ms, overall system response time ≤40ms. Accuracy: decision accuracy ≥96.5%, risk identification accuracy ≥95%. Resources: memory usage ≤8GB, CPU utilization ≤70%, supports operating temperature from -40℃ to 85℃.

[0038] refer to Figure 2 In one specific embodiment of the present invention, the following steps are included: Step 10: System Initialization and Parameter Configuration. First, hardware resources are pre-allocated, including allocating 128MB of double-buffered memory for the sensing data cache to prevent data read / write conflicts, reserving 64MB of dedicated memory for policy calculation to ensure real-time computing needs, allocating 256MB of working memory for the execution process to support parallel processing tasks, reserving CPU cores 2 and 3 specifically for handling write-related computational tasks, allocating 512MB of GPU memory for neural network model inference, and setting network communication to the highest priority to ensure real-time data transmission. Then, software modules are loaded and initialized, including loading the security assessment model and completing weight initialization and inference verification tests, loading the network prediction model to establish an environmental state prediction baseline, loading the policy generation model and initializing genetic algorithm parameters, establishing a CAN bus data stream connection and configuring VCU, BMS, and gateway communication parameters, initializing the environmental sensor data stream and setting signal strength and location information acquisition frequency, and establishing a user behavior data channel and loading the historical preference database.

[0039] Step 20: Multimodal Data Synchronous Acquisition and Preprocessing. This involves precise acquisition of vehicle status data. CPU load rate (accuracy ±1%), memory usage (sampling interval 50ms), power supply voltage and fluctuation range (accuracy ±0.1V), and network communication quality (packet loss rate accuracy 0.1%, transmission delay accuracy 1ms, available bandwidth) of each ECU are synchronously acquired via the CAN bus at a frequency of 100Hz. A Kalman filter algorithm (process noise covariance Q=0.01, observation noise covariance R=1.0, 5 iterations of filtering) is applied to smooth the raw data. Deep software feature extraction includes static feature extraction (parsing the ELF header structure of the software binary file to extract the version number and build timestamp, verifying the digital signature using the RSA-2048 algorithm, analyzing the dependency library list to construct a version compatibility matrix, and extracting security level identifiers and permission requirement lists) and dynamic behavior monitoring (monitoring API call sequences to record call frequency and time distribution, analyzing memory usage patterns to detect memory leak risks, recording exception handling logs to classify and statistically analyze error code frequency, and monitoring CPU usage fluctuations to identify abnormal peak patterns). The comprehensive environmental context perception includes external environment perception (GPS positioning data update frequency of 1Hz, accuracy of ±3 meters, 4G / 5G signal strength RSRP sampling frequency of 10Hz, threshold range of -110dBm to -60dBm, WiFi signal strength RSSI sampling frequency of 5Hz, threshold range of -90dBm to -30dBm, weather condition identification such as temperature, humidity and rainfall, road type identification such as highway, urban road and rural road) and user behavior analysis (analyzing driving data of the past 30 days to identify rapid acceleration and emergency braking behavior patterns, statistically analyzing the usage frequency of each functional module to build a user preference profile, and analyzing historical swipe operation records to identify user acceptance thresholds). Finally, data standardization and feature engineering are performed, including numerical parameter normalization (using the min-max method to map parameters to the [0,1] interval, and using the 3σ principle to remove and interpolate outliers), constructing a 256-dimensional multimodal feature vector (32-dimensional vehicle state features, 128-dimensional software features, and 96-dimensional environmental context features), and generating logical fact assertions (discretizing continuous variables into logical assertions and establishing a mapping table between fact assertions and the original data).

[0040] Step 30: Parallel processing of neural symbols through dual pathways includes deep execution of symbolic rule reasoning. First, a first-order predicate logic rule library (containing security constraint rules, system dependency rules, and business logic rules) is loaded from the SQLite database and its syntactic correctness and logical consistency are verified. Then, forward chain reasoning is executed through the Drools rule engine (logical fact assertions are input, pattern matching is performed, and rule execution is triggered to generate new reasoning facts, iterating until no new facts are generated). Finally, the confidence score is calculated (C_symbol is obtained by weighting the rule matching degree, constraint satisfaction degree, and historical accuracy at a ratio of 0.4:0.4:0.2). The neural network inference process first performs layer normalization and position encoding on the 256-dimensional feature vector. Then, it performs forward propagation through an 8-layer Transformer encoder (each layer contains an 8-head self-attention mechanism, a feedforward neural network, and residual connections, using the GELU activation function). The output layer generates a decision probability distribution through global average pooling and fully connected layers. Finally, the confidence score is evaluated (C_neural is obtained by weighting the model calibration, prediction consistency, and feature importance in a ratio of 0.5:0.3:0.2).

[0041] Step 40: Attention Fusion and Intelligent Decision-Making. First, dynamic weight adaptive calculation is performed. Scene features (such as safety-critical features and pattern recognition features) are analyzed in real time, and a random forest classifier is used for scene classification. Weights are then dynamically allocated (W_symbol=0.9, W_neural=0.1 for safety-critical scenes, W_symbol=0.2, W_neural=0.8 for pattern recognition scenes, and weights allocated according to confidence level for mixed scenes). Then, weighted fusion is performed. After vectorizing the symbolic output and calibrating the neural output, linear weighted fusion is performed. (Final_Vector=W_symbol×Symbolic_Vector+W_neural×Neural_Vector) and evaluate the fusion quality, finally generating a decision (outputting the result of allowing, delaying or prohibiting, along with confidence level and timestamp).

[0042] Step 50: Decision Execution and Full-Process Monitoring includes intelligent generation of flashing plans, determination of the target ECU list and flashing order based on dependency graphs, optimization of resource allocation (CPU, memory, and network bandwidth), and formulation of rollback plans (target recovery time < 5 seconds). Fine-grained monitoring of the execution process includes real-time status tracking (data transmission progress, system resource usage, and key parameter monitoring) and anomaly detection and handling (threshold alarms, pattern recognition, and automatic recovery).

[0043] Step 60: Feedback learning and continuous optimization involve multi-dimensional effect evaluation, including technical effects (decision accuracy, system performance impact, and resource utilization efficiency) and business effects (user satisfaction, service quality, and economic benefits). Intelligent knowledge base updates include symbol rule base optimization (correcting erroneous rules, expanding the rule base, and adjusting confidence levels), neural network model updates (incremental learning, online learning, and version management), and fusion parameter tuning (adjusting weight calculation parameters and confidence level methods). Long-term system maintenance includes establishing a performance monitoring system (key indicator dashboards, alarm thresholds, and regular reports) and improving adaptive capabilities (new scenario learning, anomaly handling optimization, and user experience improvement).

[0044] Step 70: Technical parameters ensure real-time performance (end-to-end response time ≤ 40ms), accuracy indicators (decision accuracy ≥ 96.5%, risk identification accuracy ≥ 95%, system availability ≥ 99.9%) and resource constraints (memory usage ≤ 8GB, CPU utilization ≤ 70%, storage space ≤ 256GB, power consumption ≤ 25W).

[0045] Example 2

[0046] refer to Figure 3 In a second aspect, the present invention provides a vehicle software cognitive decision-making device 1 based on neural symbolic fusion, comprising: an acquisition module 11, configured to acquire multimodal data of a vehicle in real time, the multimodal data including ECU status data, feature data of flashing software, and environmental context data; preprocessing the multimodal data; an input module 12, configured to input the preprocessed multimodal data into a symbolic reasoning pathway and a neural network pathway respectively to obtain a first decision vector, a confidence level of the first decision vector, a second decision vector, and a confidence level of the second decision vector; an identification module 13, configured to identify vehicle scene types based on the multimodal data using a preset random forest classifier; an adjustment module 14, configured to dynamically adjust the weights of the confidence levels of the first decision vector and the second decision vector based on the identification results of the vehicle scene types; and an output module 15, configured to perform weighted fusion of the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector, and output a third decision vector.

[0047] The input module 12 includes: a first input unit for inputting preprocessed multimodal data into a symbolic inference pathway, wherein the symbolic inference pathway obtains a first decision vector and a confidence level of the first decision vector through a preset rule base, forward chain inference, and confidence level calculation; and a second input unit for inputting preprocessed multimodal data into a neural network pathway, wherein the neural network pathway outputs a second decision vector and a confidence level of the second decision vector through a Transformer encoder and a temperature scaling method.

[0048] Example 3

[0049] refer to Figure 4 A third aspect of the present invention provides an electronic device comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the vehicle software cognitive decision-making method based on neural symbol fusion of the first aspect of the present invention.

[0050] Electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 502 or a program loaded from storage device 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. An input / output (I / O) interface 505 is also connected to bus 504.

[0051] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, hard disks; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 4 Each box shown can represent a device or multiple devices as needed.

[0052] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by a processing device 501, it performs the functions defined in the methods of embodiments of this disclosure. It should be noted that the computer-readable medium described in embodiments of this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0053] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more computer programs, which, when executed by the electronic device, cause the electronic device to: Computer program code for performing the operations of embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, C++, and Python—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0054] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A vehicle software cognitive decision-making method based on neural symbol fusion, characterized in that, include: Real-time acquisition of multimodal data of the vehicle, including ECU status data, feature data of flashing software, and environmental context data; preprocessing of the multimodal data; The preprocessed multimodal data is input into the symbolic reasoning pathway and the neural network pathway respectively to obtain the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector. Based on the multimodal data, vehicle scene types are identified using a preset random forest classifier; Based on the vehicle scene type recognition results, the weights of the confidence scores of the first decision vector and the second decision vector are dynamically adjusted. The first decision vector, its confidence level, the second decision vector, and its confidence level are weighted and fused to output the third decision vector.

2. The vehicle software cognitive decision-making method based on neural symbol fusion according to claim 1, characterized in that, The step of inputting the preprocessed multimodal data into the symbolic inference pathway and the neural network pathway respectively to obtain the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector includes: The preprocessed multimodal data is input into the symbolic reasoning pathway, which obtains the first decision vector and the confidence level of the first decision vector through a preset rule base, forward chain reasoning and confidence calculation. The preprocessed multimodal data is input into a neural network pathway, which outputs a second decision vector and the confidence level of the second decision vector through a Transformer encoder and a temperature scaling method.

3. The vehicle software cognitive decision-making method based on neural symbol fusion according to claim 2, characterized in that, The symbolic reasoning pathway obtains the first decision vector and its confidence level through a preset rule base, forward chain reasoning, and confidence calculation. Based on the rule base, define the security constraints, system dependencies, and business logic for the write operation; Based on the preprocessed multimodal data, a first decision vector is generated through forward chain reasoning based on security constraints, system dependencies, and business logic. The confidence level of the first decision vector is calculated based on the rule matching degree, constraint satisfaction degree, and historical accuracy of the first decision vector.

4. The vehicle software cognitive decision-making method based on neural symbol fusion according to claim 2, characterized in that, The neural network pathway, through a Transformer encoder and temperature scaling method, outputs a second decision vector and the confidence level of the second decision vector, including: The neural network pathway outputs a second decision vector through a Transformer encoder and forward propagation. The model calibration degree is calculated using the temperature scaling method; Calculate prediction consistency based on the variance of the model prediction results; The confidence level of the second decision vector is calculated based on model calibration degree, prediction consistency and preset feature contribution degree.

5. The vehicle software cognitive decision-making method based on neural symbol fusion according to claim 1, characterized in that, The process of identifying vehicle scene types based on the multimodal data using a preset random forest classifier includes: Based on vehicle speed, voltage, and temperature from multimodal data, the features of the first scene are determined; Based on network trends and user behavior in multimodal data, the characteristics of the second scenario are determined; Based on the first scene features and the second scene features, the vehicle scene type is identified by using a preset random forest classifier.

6. The vehicle software cognitive decision-making method based on neural symbol fusion according to claim 5, characterized in that, The dynamic adjustment of the confidence scores of the first and second decision vectors based on the vehicle scene type recognition results includes: If the vehicle scene is identified as either the first scene or the second scene, the confidence weights of the first decision vector and the first decision vector are determined according to preset values. If the vehicle scene is identified as a mixed scene, the weights of the first decision vector and the second decision vector are determined based on the confidence ratio.

7. A vehicle software cognitive decision-making device based on neural symbol fusion, characterized in that, include: The acquisition module is used to acquire multimodal data of the vehicle in real time, including ECU status data, feature data of the flashing software, and environmental context data. The multimodal data is preprocessed; The input module is used to input the preprocessed multimodal data into the symbolic reasoning pathway and the neural network pathway respectively, to obtain the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector. The identification module is used to identify vehicle scene types based on the multimodal data using a preset random forest classifier; The adjustment module is used to dynamically adjust the confidence scores of the first decision vector and the weights of the confidence scores of the second decision vector based on the recognition results of the vehicle scene type. The output module is used to perform weighted fusion of the first decision vector, the confidence level of the first decision vector, the second decision vector, and the confidence level of the second decision vector, and output the third decision vector.

8. The vehicle software cognitive decision-making device based on neural symbol fusion according to claim 7, characterized in that, The input module includes: The first input unit is used to input the preprocessed multimodal data into the symbolic reasoning path. The symbolic reasoning path obtains the first decision vector and the confidence of the first decision vector through a preset rule base, forward chain reasoning and confidence calculation. The second input unit is used to input the preprocessed multimodal data into the neural network path, which outputs the second decision vector and the confidence level of the second decision vector through the Transformer encoder and temperature scaling method.

9. An electronic device, comprising: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the vehicle software cognitive decision-making method based on neural symbol fusion as described in any one of claims 1 to 6.

10. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by the processor, it implements the vehicle software cognitive decision-making method based on neural symbol fusion as described in any one of claims 1 to 6.

Images