Electric vehicle intelligent network connection communication and edge computing optimization method

Abstract

The application discloses an electric vehicle intelligent network connection communication and edge calculation optimization method, and particularly relates to the technical field of edge calculation, and is used for solving the problem of the decline of the reliability of fusion results caused by the direct use of original confidence by edge nodes for data fusion when multi-source data streams asynchronously arrive due to the dynamic characteristics of wireless channels in the prior art; the network disturbance is quantified by constructing a dynamic topology graph, the effective survival window is determined in combination with the physical characteristics of multi-modal sensing data, the erosion degree of network disturbance on data timeliness is analyzed, then the original confidence is dynamically redistributed at the system level to generate effective confidence according to the erosion degree and the expected utility level, and finally the environment perception result is generated based on the risk minimization decision criterion and is delivered to the network-connected electric vehicle, so that the reliability of cooperative driving decision and the safety of vehicle control are improved.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and more specifically, to a method for optimizing intelligent connected communication and edge computing for electric vehicles. Background Technology

[0002] In the field of intelligent connected electric vehicles, using multimodal sensors on the vehicle to perceive the environment and uploading the data to edge computing nodes via wireless networks to achieve multi-source data fusion and collaborative decision-making is a technical approach to improve the reliability of autonomous driving systems. Existing technologies typically rely on edge nodes receiving heterogeneous perception data streams from different vehicle nodes and using data fusion algorithms to generate a global environment model to support vehicle control decisions. This technical approach involves the coordination of communication transmission and edge computing processing.

[0003] However, existing methods fail to adequately consider the implicit interference of the communication transmission process on the reliability of the data itself when dealing with the asynchronous arrival of multi-source data streams caused by the dynamic characteristics of wireless channels. Specifically, when edge nodes perform multimodal data fusion, they directly use the original confidence information attached to the data without effectively assessing and compensating for the confidence attenuation caused by transmission delays. This leads to a decrease in the reliability of the fusion results, which in turn affects the safety of subsequent vehicle decision-making and control. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides an optimization method for intelligent connected communication and edge computing of electric vehicles to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Optimization methods for intelligent connected communication and edge computing in electric vehicles include:

[0007] S1. Edge computing nodes acquire multimodal perception data and corresponding raw confidence scores uploaded by connected electric vehicles through wireless communication networks;

[0008] S2. Construct a dynamic topology graph based on the communication connection relationship between connected electric vehicles and edge computing nodes, and generate a topology entropy sequence of the dynamic topology graph;

[0009] S3. Analyze the collaborative driving task to determine the real-time environmental perception requirements, and generate the expected utility level by matching the real-time environmental perception requirements with the type of multimodal perception data and the observation target.

[0010] S4. Determine the effective survival window of each multimodal sensing data based on its physical characteristics, and analyze the degree of erosion of the effective survival window by network disturbances reflected by the topological entropy sequence.

[0011] S5. Based on the degree of erosion and the expected utility level, the original confidence scores of each multimodal sensing data are dynamically redistributed at the system level to generate effective confidence scores;

[0012] S6. Evaluate the probability of the validity of different environmental perception hypotheses and the risk of misjudgment based on the effective confidence level, generate the final environmental perception results based on the risk minimization decision criterion, and send them to the connected electric vehicles.

[0013] Furthermore, the edge computing node acquires the multimodal perception data and corresponding raw confidence scores uploaded by the connected electric vehicle through the wireless communication network, including:

[0014] The system receives data packets sent by a connected electric vehicle, wherein the data packets contain multimodal perception data collected by the on-board sensors of the connected electric vehicle and raw confidence scores generated by the on-board computing unit of the connected electric vehicle based on the quality and integrity of the perception data, wherein the multimodal perception data includes at least one of image data collected by a camera, point cloud data collected by a lidar, and radar data collected by a millimeter-wave radar.

[0015] Parse the data packets to separate the multimodal sensing data from the original confidence level, and establish the corresponding relationship between the multimodal sensing data and the original confidence level.

[0016] Furthermore, a dynamic topology graph is constructed based on the communication connection relationship between connected electric vehicles and edge computing nodes, and a topology entropy sequence of the dynamic topology graph is generated, including:

[0017] A dynamic topology graph is constructed using connected electric vehicles and edge computing nodes as vertices, and wireless communication connections between connected electric vehicles and edge computing nodes as edges.

[0018] Calculate the eigenvalues ​​of the adjacency matrix based on the adjacency matrix of the dynamic topology graph;

[0019] Calculate the topological entropy value of the dynamic topological graph based on the distribution of eigenvalues;

[0020] The steps of constructing a dynamic topology graph, calculating feature values, and calculating topology entropy values ​​are repeated at preset time intervals to generate a topology entropy sequence arranged in chronological order to quantify the unpredictability of topology changes.

[0021] Furthermore, the topological entropy value of the dynamic topological graph is calculated based on the distribution of eigenvalues, including: obtaining all eigenvalues ​​of the adjacency matrix of the dynamic topological graph; calculating the sum of squares of all eigenvalues; and using the sum of squares as the topological entropy value.

[0022] Furthermore, the collaborative driving task is analyzed to determine real-time environmental perception requirements. Based on the type of multimodal perception data and the observed targets, the expected utility level is generated by matching the real-time environmental perception requirements, including:

[0023] Analyze the cooperative driving task instructions to determine the task type and the required real-time environmental perception requirements, including the requirements for perception range, perception accuracy, and perception update frequency.

[0024] Identify the types of multimodal sensing data and the observation targets represented by the multimodal sensing data;

[0025] The types of multimodal sensing data and observation targets are matched with the real-time needs of environmental sensing, and expected utility levels representing the importance of the data are generated according to predefined matching rules.

[0026] Furthermore, identifying the type of multimodal sensing data and the observation targets represented by the multimodal sensing data includes: parsing the metadata tags in the packet header to determine the type of multimodal sensing data; and parsing the attribute information of the observation targets based on the packet payload content.

[0027] Furthermore, the effective survival window of each multimodal sensing data based on its physical characteristics is determined, and the degree of erosion of the effective survival window by network disturbances reflected by the topological entropy sequence is analyzed, including:

[0028] The physical characteristic time constraints are determined according to the type of multimodal sensing data, and the effective survival window of each multimodal sensing data is determined based on the physical characteristic time constraints.

[0029] Analyze the trend of topological entropy sequence over time to extract dynamic characteristic parameters that characterize the intensity of network disturbances;

[0030] Establish a mapping model between network disturbance intensity and the remaining value of the effective survival window;

[0031] Based on dynamic characteristic parameters and mapping relationship models, the timeliness erosion factor of network disturbances on the effective survival window is calculated;

[0032] The degree of erosion of the effective survival window by network perturbation is quantified based on the time-dependent erosion factor.

[0033] Furthermore, the original confidence levels of each multimodal sensing data are dynamically redistributed at the system level based on the degree of erosion and the expected utility level to generate effective confidence levels, including:

[0034] The confidence loss of each multimodal sensing data due to network disturbance is calculated based on the degree of erosion.

[0035] The priority weight of each multimodal sensing data in the global system is determined based on the expected utility level;

[0036] Establish a system-level confidence resource pool, the total amount of which is determined based on the sum of the confidence loss of all multimodal sensing data;

[0037] Resources in the confidence resource pool are redistributed to the corresponding multimodal sensing data according to priority weights, and the original confidence is compensated based on the amount of redistributed resources to generate effective confidence.

[0038] Furthermore, based on the effective confidence level, the probability of the validity of different environmental perception hypotheses and the risk of misjudgment are assessed. Based on the risk minimization decision criterion, the final environmental perception results are generated and distributed to connected electric vehicles, including:

[0039] Based on the effective confidence scores corresponding to the multimodal sensing data, the probability of each environmental sensing hypothesis being true is calculated.

[0040] Obtain the decision risk cost caused by misjudgment under different predefined environmental perception assumptions;

[0041] Based on the probability of each environmental perception hypothesis being true and the corresponding decision risk cost, calculate the expected risk of each environmental perception hypothesis.

[0042] The environmental perception hypothesis with the lowest expected risk is selected as the final environmental perception result, and then transmitted to the connected electric vehicle via a wireless communication network.

[0043] Furthermore, based on the probability of each environmental perception hypothesis being true and the corresponding decision risk cost, the expected risk of each environmental perception hypothesis is calculated, including multiplying the probability of each environmental perception hypothesis being true by its decision risk cost, and the product is the expected risk of the environmental perception hypothesis.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. By constructing a dynamic topology graph and generating a topology entropy sequence, the perturbation of wireless communication networks is accurately quantified, which can effectively assess the impact of network dynamic characteristics on data transmission quality. By establishing an effective survival window model for multimodal sensing data and combining it with network perturbation analysis, the problem of data credibility decay caused by transmission delay can be accurately identified, providing a scientific basis for subsequent confidence redistribution. Based on the dual mechanism of network state perception and data timeliness assessment, the accuracy of edge computing nodes in judging the reliability of environmental sensing data is significantly improved.

[0046] 2. A system-level dynamic confidence redistribution method is adopted, which adaptively adjusts the confidence weight of each data based on data utility and network erosion degree. This ensures the dominant role of high-value perception data in the decision-making process. By combining confidence assessment with the cost of misjudgment through the decision criterion of minimizing risk, the optimal environmental perception result is generated. This not only improves the reliability of cooperative driving decisions but also ensures the safety of vehicle control. It effectively solves the problem of fusion of multi-source heterogeneous data in dynamic network environments and provides more reliable environmental perception capabilities for intelligent connected electric vehicles. Attached Figure Description

[0047] Figure 1 This is a flowchart of the electric vehicle intelligent connected communication and edge computing optimization method of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example: Figure 1 The present invention provides an optimization method for intelligent connected communication and edge computing in electric vehicles, comprising:

[0050] S1. Edge computing nodes acquire multimodal perception data and corresponding raw confidence scores uploaded by connected electric vehicles through wireless communication networks;

[0051] S2. Construct a dynamic topology graph based on the communication connection relationship between connected electric vehicles and edge computing nodes, and generate a topology entropy sequence of the dynamic topology graph;

[0052] S3. Analyze the collaborative driving task to determine the real-time environmental perception requirements, and generate the expected utility level by matching the real-time environmental perception requirements with the type of multimodal perception data and the observation target.

[0053] S4. Determine the effective survival window of each multimodal sensing data based on its physical characteristics, and analyze the degree of erosion of the effective survival window by network disturbances reflected by the topological entropy sequence.

[0054] S5. Based on the degree of erosion and the expected utility level, the original confidence scores of each multimodal sensing data are dynamically redistributed at the system level to generate effective confidence scores;

[0055] S6. Evaluate the probability of the validity of different environmental perception hypotheses and the risk of misjudgment based on the effective confidence level, generate the final environmental perception results based on the risk minimization decision criterion, and send them to the connected electric vehicles.

[0056] S1. Edge computing nodes acquire multimodal sensing data and corresponding raw confidence scores uploaded by connected electric vehicles via wireless communication networks. The specific implementation is as follows:

[0057] Edge computing nodes receive data packets sent by connected electric vehicles via a wireless communication network. These data packets are encapsulated using a standard protocol format conforming to vehicle communication standards and consist of a protocol header and payload data. The protocol header includes metadata fields such as a data source identifier, timestamp, and data type identifier. The payload data includes multimodal perception data collected in real-time by the connected electric vehicle's onboard sensors and raw confidence scores generated by the onboard computing unit. The multimodal perception data includes, but is not limited to, at least one of the following: image data collected by cameras, point cloud data collected by LiDAR, and radar data collected by millimeter-wave radar. This data is transmitted to the onboard computing unit via the vehicle network for preprocessing. The initial confidence level is generated by the onboard computing unit based on the quality and integrity of the perceived data. The quality assessment includes quantitative analysis of parameters such as signal-to-noise ratio, resolution, and occlusion degree of sensor data, while the integrity assessment includes data frame integrity verification and judgment of the effectiveness of sensor coverage. The onboard computing unit performs weighted calculations on the quality and integrity indicators through preset scoring rules. For example, the signal-to-noise ratio indicator is quantified into a score of 0-100, the resolution is graded and scored according to pixel density or point cloud density, and the occlusion degree is scored inversely according to the proportion of occlusion area. Finally, these scores are weighted and summed according to preset weights (e.g., quality indicator weight 0.6, integrity indicator weight 0.4), and the calculation result is normalized to a value between 0 and 1 as the initial confidence level, where 0 represents completely unreliable and 1 represents completely reliable.

[0058] After receiving a data packet, the edge computing node first parses the protocol header according to the communication protocol specification, extracting the data source identifier and timestamp for data packet management and timing ordering. The data source identifier uses a unique encoding format, such as a combination of vehicle identification code and sensor number. The timestamp is accurate to the millisecond level to ensure the accuracy of data timing. Subsequently, the multimodal sensing data and the original confidence score in the payload data are separated according to the data type identifier field. The data type identifier is parsed using a predefined encoding mapping table; for example, encoding value 1 represents image data, encoding value 2 represents point cloud data, and encoding value 3 represents radar data. During the parsing process, a cyclic redundancy check (CRC) mechanism is used to verify the integrity of the data packet transmission. The check polynomial uses the standard CRC-32 algorithm. When the check fails, the data packet is discarded, and a feedback mechanism is used to request the sender to retransmit, ensuring data reliability.

[0059] After data packet parsing, the edge computing nodes establish a correspondence between multimodal sensing data and the original confidence scores. This correspondence is achieved by creating a data management table. Each record in the table contains five fields: data source identifier, timestamp, multimodal sensing data storage address pointer, original confidence score value, and data type identifier. The data source identifier and timestamp together form a composite primary key, ensuring the uniqueness of the correspondence between each multimodal sensing data record and the original confidence score. The storage address pointer points to the actual storage location of the multimodal sensing data in memory, managed using a linked list structure for dynamic expansion. The edge computing nodes can quickly query this table using a hash index mechanism, retrieving the original confidence score corresponding to a specific multimodal sensing data record in real time, supporting subsequent data processing. All parsed multimodal sensing data and original confidence scores are stored with timestamps in the distributed cache of the edge computing nodes, and are managed by partitioning according to the data source identifier. Each partition has an independent memory space and read / write locks, facilitating on-demand access and processing in subsequent steps.

[0060] The entire data acquisition process employs a multi-stage pipeline architecture, with data packet reception, parsing, and storage operations executed in parallel pipelines to improve processing efficiency. Edge computing nodes monitor the cache status in real time, and when the data volume exceeds a preset threshold (e.g., cache utilization reaches 80%), a data eviction mechanism is activated, prioritizing the eviction of the earliest accessed data using a Least Recently Used (LRU) algorithm to ensure system real-time performance. Simultaneously, the system incorporates an anomaly handling mechanism; when erroneous data packets are received consecutively more than a certain number (e.g., 5 times), a connection diagnostic process is automatically triggered to check the wireless communication link status. Through these methods, edge computing nodes achieve efficient and reliable acquisition and management of multimodal perception data and raw confidence scores uploaded by connected electric vehicles, providing a complete and reliable data foundation for subsequent processing steps. The time overhead of all processing steps is recorded and monitored, ensuring that single data packet processing latency is controlled within milliseconds, meeting real-time requirements.

[0061] S2. Construct a dynamic topology graph based on the communication connection relationship between connected electric vehicles and edge computing nodes, and generate a topology entropy sequence for the dynamic topology graph. The specific implementation is as follows:

[0062] Edge computing nodes construct a dynamic topology graph based on the communication connections between connected electric vehicles (EVs) and edge computing nodes. In practice, each EV and each edge computing node is treated as a vertex in the topology graph, and the vertex set contains all entities involved in communication. Vertices are labeled with unique identifiers: the identifier for an EV is generated from its vehicle identification number (VIN), and the identifier for an edge computing node is generated from its network address. Edges between vertices represent wireless communication connections; an edge is established between corresponding vertices only if an active wireless communication link exists between the EV and the edge computing node. The edge weights are quantified using wireless signal quality metrics, such as a value mapped from 0 to 1 based on the received signal strength index (RSI), where 0 represents the worst signal quality and 1 represents the best signal quality. The dynamic topology graph is stored using an adjacency list data structure, updating the state of vertices and edges in real time. The adjacency list is updated immediately when a new communication connection is established or an existing connection is broken.

[0063] The eigenvalues ​​of the adjacency matrix are calculated based on the adjacency matrix of the dynamic topology graph. First, the adjacency list is converted into an adjacency matrix representation, which is an n×n square matrix, where n is the number of vertices. Matrix element Aij represents the connection between vertex i and vertex j; Aij takes the weight of the edge when an edge exists, and a value of 0 when no edge exists. The diagonal element Aii is uniformly set to 0, indicating that vertices are not self-connected. Numerical methods are used to calculate the eigenvalues ​​of the adjacency matrix, such as iteratively solving for the roots of the characteristic polynomial using the QR algorithm. A convergence threshold is set during the iteration of the QR algorithm; the calculation terminates when the change in eigenvalues ​​between two consecutive iterations is less than this threshold. The calculated eigenvalues ​​include both real and complex eigenvalues. After taking the modulus of all eigenvalues, subsequent processing is performed to ensure that all eigenvalues ​​are non-negative real numbers.

[0064] The topological entropy of a dynamic topological graph is calculated based on the distribution of eigenvalues. After obtaining all eigenvalues ​​of the adjacency matrix, the sum of the squares of all eigenvalues ​​is calculated as the topological entropy. Specifically, for each eigenvalue, its modulus is first determined, then the square of that modulus is calculated, and finally, all squares are summed to obtain the topological entropy. This calculation process ensures that the topological entropy is a non-negative real number, reflecting the complexity of the network topology; a larger value indicates a more complex network connection. After the topological entropy is calculated, it is normalized by dividing by the square of the number of vertices to make the result comparable. For example, when the number of vertices is 10, dividing by 100 ensures the normalized entropy is between 0 and 1.

[0065] The process of constructing a dynamic topology graph, calculating feature values, and calculating topology entropy values ​​is repeated at preset time intervals. The time interval is set according to application requirements, for example, 100 milliseconds, and a high-precision timer triggers each topology entropy calculation. Each calculation generates a topology entropy value, which is stored chronologically to form a topology entropy sequence. The sequence is stored in a circular buffer, the size of which is set according to the time window requirements, for example, storing the topology entropy values ​​of the most recent 1000 time points, corresponding to a 100-second time window. The topology entropy sequence is used to quantify the unpredictability of topology changes. By analyzing the statistical characteristics of the sequence, such as calculating the variance of entropy values ​​between adjacent time points, a larger variance indicates a more drastic topology change. The entire process is performed in real time, with edge computing nodes continuously monitoring changes in network topology status, providing a data foundation for subsequent analysis of network disturbances. The time complexity of all calculation processes has been optimized to ensure completion within the set time intervals, meeting real-time requirements. When the number of vertices is large, a distributed computing method is used to distribute the feature value calculation task to multiple computing units for parallel execution, ensuring computational efficiency. At the same time, an exception handling mechanism is set up so that when the eigenvalue calculation fails to converge, a backup algorithm such as the power iteration method is used to perform the calculation to ensure the reliability of the system.

[0066] S3. Analyze the cooperative driving task to determine the real-time environmental perception requirements. Based on the type of multimodal perception data and the observation target, match the real-time environmental perception requirements to generate the expected utility level. The specific implementation is as follows:

[0067] Edge computing nodes parse collaborative driving task instructions to determine the task type and required real-time environmental perception needs. These instructions are received via the vehicle-to-cloud communication interface and encapsulated in a structured data format, containing three main fields: task identifier, task parameters, and performance requirements. The task type is determined by parsing the task identifier field. The task identifier uses a predefined encoding scheme; for example, code 001 represents a lane-changing assistance task, code 002 represents an intersection passage task, and code 003 represents a platooning task. Real-time environmental perception requirements are extracted from the performance requirements field, including three dimensions: perception range requirements, perception accuracy requirements, and perception update frequency requirements. Perception range requirements are expressed in meters, e.g., 200 meters, indicating the environmental information to be monitored within this distance range. Perception accuracy requirements are expressed in meters, representing the error tolerance, e.g., 0.5 meters, indicating the maximum permissible deviation between the perception result and the actual location. Perception update frequency requirements are expressed in Hertz, e.g., 10 Hertz, indicating the number of times the perception result needs to be updated per second. These requirements together constitute a complete description of the real-time environmental perception needs, providing a quantitative benchmark for subsequent data utility evaluation.

[0068] The system identifies the types of multimodal sensing data and the observed targets represented by this data. The type of multimodal sensing data is determined by parsing the metadata tags in the data packet header. These metadata tags use a fixed format and include sensor type encoding and data format description fields. Sensor type encoding uses enumerated values; for example, encoding 1 represents camera data, encoding 2 represents LiDAR data, and encoding 3 represents millimeter-wave radar data. Data format descriptions include parameters such as resolution, frame rate, and data volume. The attribute information of the observed targets is obtained by parsing the data packet payload content, and feature extraction algorithms are used to analyze the sensing data content. For image data, convolutional neural network target detection algorithms are used to identify observed targets such as vehicles, pedestrians, and traffic signs, along with their bounding box positions. For point cloud data, Euclidean clustering is used for point cloud segmentation to identify obstacle outlines and motion states. For radar data, fast Fourier transform is used to process the echo signals to obtain target distance and velocity information. All identification results are accompanied by a confidence score, indicating the reliability of the identification results. The confidence score is calculated using the softmax function and ranges from 0 to 1.

[0069] The system matches the types and observation targets of multimodal sensing data with real-time environmental sensing requirements, generating expected utility levels representing the importance of the data based on predefined matching rules. The matching rules are stored in a multidimensional decision table, where row dimensions correspond to different observation target types and column dimensions correspond to different real-time environmental sensing requirement dimensions. Each cell stores a matching score, preset based on expert experience and determined through multiple rounds of expert review using the Delphi method. The score ranges from 0 to 1, representing the importance of that type of data to that requirement dimension. For example, for lane-changing assistance tasks, the score for observing vehicles approaching from behind is set to 0.9 in the perception range dimension, 0.8 in the perception accuracy dimension, and 0.7 in the update frequency dimension. The matching process calculates a weighted total score, with the weights of each requirement dimension dynamically adjusted according to the task type. The weight allocation scheme is determined using the analytic hierarchy process (AHP). For example, safety-critical tasks are assigned a higher weight to perception accuracy (0.5), while dynamic scenario tasks are assigned a higher weight to update frequency (0.4). The weighted total score is calculated using the linear weighted summation formula: Total score = w1×s1 + w2×s2 + w3×s3, where w1, w2, and w3 are weight coefficients that satisfy w1 + w2 + w3 = 1, and s1, s2, and s3 are the matching scores for each dimension.

[0070] Based on the weighted total score range mapped to the expected utility level, two thresholds are set to divide the total score interval into three levels: 0 to 0.3 is mapped to the low utility level, 0.3 to 0.7 to the medium utility level, and 0.7 to 1.0 to the high utility level. The threshold values ​​are determined through historical data analysis, and the optimal split point is determined after clustering historical matching results using the k-means clustering algorithm. After the expected utility level is generated, it is associated and stored with the corresponding multimodal perception data, using a key-value pair structure. The key is a unique identifier for the data packet, and the value is the utility level value. The system maintains a matching rule base and supports online updates to the matching rules. When a new driving scenario appears, the decision table content is updated using an incremental learning algorithm. An anomaly handling mechanism is also set up. When encountering an undefined observation target type, the k-nearest neighbor algorithm is used to select the most similar existing rule for processing. The similarity calculation uses cosine similarity measurement, and the new target type is recorded for subsequent rule base expansion. All matching decision-making processes are logged, including input parameters, matching rule versions, calculation results, and timestamps. The logs are stored in a circular buffer, retaining the most recent 1000 records for auditing purposes. Through this method, the system achieves accurate evaluation of the utility of multimodal perception data, ensuring that limited communication resources are prioritized for the most critical environmental perception data for cooperative driving tasks, providing a reliable input basis for subsequent confidence level reallocation.

[0071] S4. Determine the effective survival window of each multimodal sensing data based on its physical characteristics, and analyze the degree of erosion of the effective survival window by network disturbances reflected by the topological entropy sequence. Specifically, this is implemented as follows:

[0072] Edge computing nodes determine the corresponding physical characteristic time constraints based on the type of multimodal sensing data, and then determine the effective survival window for each type of multimodal sensing data based on these constraints. The type of multimodal sensing data is determined by parsing the metadata tag in the data packet header. This metadata tag includes a sensor type code and a data feature description field. The sensor type code uses predefined enumerated values, such as 1 for camera, 2 for LiDAR, and 3 for millimeter-wave radar. Physical characteristic time constraints are predefined based on sensor characteristics and application scenario requirements. A mapping table between data types and time constraint values ​​is established. This mapping table is obtained through statistical analysis of a large amount of experimental data, comprehensively considering sensor sampling periods, data processing latency, and the real-time requirements of the application scenario. For example, the time constraint value for image data acquired by a camera is 100 milliseconds, for point cloud data acquired by LiDAR it is 50 milliseconds, and for radar data acquired by millimeter-wave radar it is 80 milliseconds. These time constraint values ​​are stored in a configuration file and support dynamic updates. The effective survival window is calculated based on a timestamp. The start time is the data acquisition timestamp, and the end time is the data acquisition timestamp plus the corresponding time constraint value. The timestamp accuracy reaches the millisecond level, using an international standard time format. The effective survival window is represented in the form of a time interval, such as [start_time, end_time], where start_time is the data acquisition time and end_time is start_time plus the time constraint value. All time calculations are stored using 64-bit integers to ensure calculation accuracy.

[0073] The trend of topological entropy sequence over time is analyzed to extract dynamic characteristic parameters characterizing the intensity of network disturbances. The topological entropy sequence is derived from the time-ordered sequence of topological entropy values ​​generated in step S2, stored in a circular buffer with a capacity of 1000 data points, using a first-in, first-out (FIFO) management strategy. The analysis process employs a sliding window mechanism, with the window size set to 20 consecutive topological entropy values ​​based on application requirements, and the window sliding step size being one topological entropy value. The dynamic characteristic parameters include three indices: the mean, variance, and rate of change of the topological entropy values. The mean reflects the average intensity of network disturbances and is obtained by calculating the arithmetic mean of all topological entropy values ​​within the window; the variance reflects the degree of fluctuation in network disturbances and is obtained by calculating the sum of squared deviations of each topological entropy value from the mean and then dividing by the window size; the rate of change reflects the speed of change of network disturbances and is obtained by calculating the average of the first-order differences of adjacent topological entropy values ​​within the window. All characteristic parameters are normalized to a value between 0 and 1 using the min-max scaling method, and the normalized parameters are obtained based on historical data statistics.

[0074] A mapping model is established between network disturbance intensity and the remaining value of the effective survival window. Network disturbance intensity is quantified using dynamic characteristic parameters. A weighted combination approach is used to integrate the mean, variance, and rate of change into a single disturbance intensity index. Weights are allocated according to the importance of the characteristic parameters, and the relative importance weights of each parameter are determined using the analytic hierarchy process (AHP). For example, the mean has a weight of 0.5, variance 0.3, and rate of change 0.2, with a sum of 1. The remaining value of the effective survival window is defined as the ratio of the time interval from the current time point to the end time of the effective survival window to the total time constraint value, representing the degree of data validity remaining. The calculation formula is (end_time - current_time) / (end_time - start_time). The mapping model uses a predefined linear decay function y = kx + b, where x is the network disturbance intensity, y is the time-related erosion factor, k is the slope parameter, and b is the intercept parameter. The parameters k and b were determined by fitting experimental data. The optimal parameter values ​​were obtained by linear regression analysis of historical data using the least squares method. The value of k represents the rate at which network disturbances erode data validity and is negative. The value of b represents the initial data validity guarantee value under the ideal condition of zero network disturbances and is positive. For example, the regression analysis yielded k = -0.8 and b = 0.95.

[0075] Based on dynamic feature parameters and a mapping relationship model, the time-dependent erosion factor of network disturbances on the effective survival window is calculated. The dynamic feature parameters are input into the mapping relationship model. First, the network disturbance intensity *x* is calculated using the formula *x = u1 × mean + u2 × variance + u3 × change_rate*, where *u1*, *u2*, and *u3* are weight coefficients, and *mean*, *variance*, and *change_rate* are normalized feature parameter values. Then, the network disturbance intensity *x* is substituted into the linear decay function *y = kx + b* to calculate the time-dependent erosion factor *y*. Boundary conditions are set during the calculation process: when the calculated *y* value is less than 0, it is forcibly set to 0; when it is greater than 1, it is forcibly set to 1, ensuring that the time-dependent erosion factor always remains between 0 and 1. The time-dependent erosion factor represents the degree of data validity decay caused by network disturbances; the smaller the value, the more severe the erosion. A value of 1 indicates no erosion, and a value of 0 indicates complete erosion. The calculation process is executed in real time, updating the calculation results immediately whenever a new topological entropy value is generated, with the calculation frequency consistent with the topological entropy sequence update frequency.

[0076] The erosion degree of network disturbances on the effective survival window is quantified based on a time-sensitive erosion factor. The erosion degree is calculated by the difference between the time-sensitive erosion factor and the initial validity, where the initial validity is 1, and the erosion degree is defined as 1-y. The erosion degree value ranges from 0 to 1, with a larger value indicating more severe erosion; a value of 0 indicates no erosion, and a value of 1 indicates complete erosion. The erosion degree calculation results are associated and stored with the corresponding multimodal sensing data using a key-value pair structure. The key is a unique identifier for the data packet, and the value is the erosion degree value, stored in a newly defined erosion degree field. The system periodically updates the mapping relationship model parameters, refitting the k and b values ​​using a sliding window least squares method. The sliding window contains the most recent 100 sets of valid data to adapt to changes in the network environment. An anomaly detection mechanism is also implemented: when 10 consecutive time-sensitive erosion factors are less than 0.3, a network status alarm is triggered, indicating a potential serious network problem requiring manual intervention. All computation processes are logged in detail, including input parameters, intermediate results, and final output. The logs are stored in a structured format, containing fields such as timestamp, computation node identifier, data packet identifier, feature parameter values, network disturbance intensity, timeliness erosion factor, and erosion degree. The most recent 1000 records are retained for auditing purposes. Through this method, the system achieves accurate quantification of the impact of network disturbances, providing a reliable basis for evaluating the effectiveness of multimodal sensing data and accurate input parameters for subsequent confidence reassignment.

[0077] S5. Based on the degree of erosion and the expected utility level, the original confidence scores of each multimodal sensing data are dynamically redistributed at the system level to generate effective confidence scores. The specific implementation is as follows:

[0078] Edge computing nodes calculate the confidence loss of each multimodal sensing data due to network disturbances based on the degree of erosion. The degree of erosion, derived from the quantification result obtained in step S4, represents the impact of network disturbances on data validity, ranging from 0 to 1, where 0 represents no erosion and 1 represents complete erosion. The confidence loss is calculated by multiplying the degree of erosion by the original confidence value, which is the confidence value carried in the data packets uploaded by the connected electric vehicle in step S1. This value is generated by the onboard computing unit based on the quality and integrity of the sensing data. The specific calculation formula is: Confidence Loss = Erosion Degree × Original Confidence. This calculation ensures that the confidence loss is proportional to the impact of network disturbances and does not exceed the original confidence value. Boundary conditions are set during the calculation process: when the calculation result is less than 0, it is taken as 0; when it is greater than the original confidence value, the original confidence value is taken, ensuring that the confidence loss always falls between 0 and the original confidence value. All calculations use 32-bit floating-point precision to ensure accuracy. For example, if the original confidence level of a certain multimodal sensing data is 0.8 and the erosion level is 0.3, then the confidence loss is 0.24. The calculation result is stored in a newly added confidence loss field, which establishes a relationship with the corresponding multimodal sensing data. It is stored in a key-value pair structure, where the key is a unique identifier for the data packet and the value is the confidence loss value.

[0079] The priority weights of each multimodal perception data point within the system are determined based on its expected utility level. The expected utility level, derived from the data importance assessment results generated in step S3, is categorized into three levels: low utility, medium utility, and high utility. Priority weights are determined using a predefined mapping table. This table is set according to the importance and urgency of the cooperative driving task, and the relative importance of each utility level is determined through expert evaluation and the analytic hierarchy process (AHP). The mapping table is defined using an enumeration method; for example, a high utility level maps to a weight value of 0.6, a medium utility level to a weight value of 0.3, and a low utility level to a weight value of 0.1. Weight values ​​are normalized to ensure that the sum of the priority weights of all multimodal perception data is 1. The weight allocation scheme considers the impact of different utility levels of data on driving safety, assigning higher weights to safety-critical data. Priority weight calculation is performed in real-time; when the expected utility level changes, the weight values ​​are updated immediately, with atomic operations used to ensure data consistency. Weight values ​​are stored in a priority weight field, establishing a correspondence with the multimodal perception data.

[0080] A system-level confidence resource pool is established, with its total size determined by the sum of the confidence losses of all multimodal sensing data. The system-level confidence resource pool is a virtual set of confidence resources, its total size obtained by accumulating the confidence losses of all current multimodal sensing data. The total resource pool size is the sum of the confidence losses of each data point. The resource pool employs a dynamic management mechanism, updating its total size in real time whenever new multimodal sensing data arrives or existing data exceeds the effective survival window. An upper limit threshold is set for the total resource pool size, for example, not exceeding 50% of the sum of all original confidence scores. When this threshold is exceeded, a resource reclamation mechanism is initiated, prioritizing the reclamation of confidence resources from low-utility data. The resource pool status is monitored in real time, recording historical operation logs of resource allocation and reclamation. The logs include fields such as timestamp, operation type, and resource change amount. The resource pool is initialized to a total size of 0, dynamically changing as data arrives and expires.

[0081] Resources in the confidence resource pool are redistributed to the corresponding multimodal sensing data according to priority weights. The original confidence score is compensated based on the redistributed resources to generate an effective confidence score. The redistribution process follows a weighted proportional allocation principle. The compensation amount for each multimodal sensing data point is calculated as follows: Compensation Amount = Priority Weight × Total Resource Pool Amount. The effective confidence score is calculated by adding the compensation amount to the original confidence score, using the formula: Effective Confidence Score = Original Confidence Score + Compensation Amount. A validity check is implemented during the calculation process to ensure that the effective confidence score does not exceed 1. A score greater than 1 is used, and a score less than 0 is used. The generated effective confidence score replaces the original confidence score and serves as the basis for subsequent environmental sensing decisions. The entire redistribution process is executed periodically, with the execution cycle set according to application requirements, for example, every 100 milliseconds. The redistribution algorithm employs a fairness guarantee mechanism to ensure that high-priority data receives more resources while low-priority data also receives basic protection. All reallocation operations are logged in detail, including timestamps, data identifiers, original confidence levels, compensation amounts, and effective confidence levels, for subsequent performance analysis and optimization. This system-level dynamic reallocation mechanism achieves optimal allocation of limited confidence resources, improving the reliability of multimodal sensing data under network disturbances. The system periodically evaluates the reallocation effect, comparing changes in confidence distribution before and after reallocation to optimize weight allocation strategies and resource pool management parameters.

[0082] S6. Assess the probability of success and risk of misjudgment of different environmental perception hypotheses based on the effective confidence level, generate the final environmental perception results based on the risk minimization decision criterion, and distribute them to connected electric vehicles. The specific implementation is as follows:

[0083] Edge computing nodes calculate the probability of each environmental perception hypothesis's validity based on the effective confidence scores corresponding to multimodal perception data. Environmental perception hypotheses refer to descriptions of possible states of the current driving environment, including but not limited to specific scenarios such as the type of obstacle ahead, distance range, and motion state. The probability calculation employs a weighted fusion method, with the following specific process: First, for each environmental perception hypothesis, the effective confidence scores of all multimodal perception data supporting the hypothesis are multiplied by their corresponding weight coefficients and summed to obtain the original probability value of the hypothesis; then, the original probability values ​​of all environmental perception hypotheses are summed to obtain a total; finally, the original probability value of each hypothesis is divided by this total to obtain the normalized probability of validity. The effective confidence score comes from the confidence score value generated in step S5 after system-level dynamic reallocation, ranging from 0 to 1, representing the reliability assessment result of each multimodal perception data point in the current network environment. Weighting coefficients are pre-set based on the sensor type and observation angle of the data source. The relative importance weights of each data source are determined through the analytic hierarchy process (AHP) and expert evaluation. For example, visual data has a weight of 0.4, radar data has a weight of 0.3, and lidar data has a weight of 0.3, with the sum of all weighting coefficients being 1. The above normalization process ensures that the sum of the probabilities of all environmental perception hypotheses is 1. The probability ranges from 0 to 1, with a higher value indicating a higher likelihood of the hypothesis being true. The calculation results are stored in the environmental perception hypothesis probability field, establishing a correlation with the corresponding hypothesis description information, and a hash table structure is used for fast querying and updating.

[0084] This system acquires the decision-making risk costs resulting from misjudgments under different predefined environmental perception assumptions. These costs are predefined using a cost matrix, where rows represent the actual environmental state, columns represent the perceived decision outcome, and element values ​​represent the cost of the corresponding misjudgment. Cost values ​​are set based on the severity of the potential consequences of misjudgments, determined through expert evaluation and historical accident data analysis, and undergo multiple rounds of expert review using the Delphi method to reach consensus. For example, the cost of misjudging an obstacle as a pedestrian is set to 100, the cost of misjudging it as a vehicle is set to 50, and the cost of a correct judgment is set to 0. These cost values ​​are expressed as dimensionless relative values, with higher values ​​indicating more severe risks. The cost matrix is ​​dynamically adjusted according to different driving scenarios; for example, different cost matrices are used for highway and urban road scenarios. Highway scenarios focus more on long-distance obstacle recognition, while urban road scenarios focus more on pedestrian and non-motorized vehicle recognition. The cost matrix is ​​stored in a configuration file, supporting online updates and version management. Each update records a change log including the modification time, content, and reason.

[0085] The expected risk of each environmental perception hypothesis is calculated based on its probability of validity and the corresponding decision risk cost. The expected risk is obtained by multiplying the probability of validity by the decision risk cost, as shown in the formula: Expected risk equals probability of validity multiplied by decision risk cost, where the decision risk cost is obtained from the cost matrix and corresponds to the potential risk cost of each hypothesis as a decision outcome. This calculation is based on risk decision theory, combining probabilistic and cost-based assessments to provide a comprehensive risk assessment index for each hypothesis. The calculation process iterates through all environmental perception hypotheses, calculating the expected risk value for each hypothesis separately. 32-bit floating-point precision is used to ensure accuracy. The expected risk value represents the average risk loss that may result from choosing that hypothesis as a decision outcome; a higher value indicates higher risk, and a lower value indicates lower risk. The calculation results are stored in the expected risk field, establishing a correspondence with the environmental perception hypotheses, and stored in an array structure for subsequent comparison and selection. An exception handling mechanism is implemented during the calculation process; when a value overflows or an illegal value occurs, a default risk value is used instead, and an exception log is recorded.

[0086] The environmental perception hypothesis with the lowest expected risk is selected as the final environmental perception result. The selection process employs a linear search algorithm, traversing the expected risk values ​​of all environmental perception hypotheses and identifying the hypothesis corresponding to the lowest expected risk value. The algorithm's time complexity is O(N), where N is the number of hypotheses, ensuring selection is completed within real-time requirements. When multiple hypotheses have the same minimum expected risk value, a two-tier selection strategy is used, prioritizing the hypothesis with the higher probability of success; if the probabilities are also the same, the hypothesis with the most recent timestamp is selected. The selection result serves as the final environmental perception decision, including a complete description of the environmental state and a confidence assessment. The decision result is encapsulated in a structured data format, including fields such as timestamp, scenario identifier, hypothesis description, risk value, and probability of success. A selection log is recorded during the decision-making process, including the risk values ​​of all candidate hypotheses and the reasons for the final selection, for subsequent auditing and analysis.

[0087] The final environmental perception results are transmitted to the connected electric vehicle via a wireless communication network. The transmitted data uses a compressed encoding format to reduce the amount of data transmitted, and the compression algorithm employs lossless compression to ensure data integrity. The data transmission protocol uses a reliable transmission mechanism, including data verification and retransmission mechanisms, to ensure that data packets arrive intact. After receiving the data, the connected electric vehicle makes corresponding driving decisions based on the perception results, such as deceleration, lane changing, or emergency braking. These adjustments are implemented through the vehicle control system. The entire processing is executed in real time, with the total latency from receiving multimodal perception data to issuing the final decision controlled within 100 milliseconds, meeting the real-time requirements of cooperative driving. The system logs all decision-making processes, including input data, intermediate results, and final decisions. The logs are stored in a circular buffer, retaining the most recent 1000 decision records for subsequent performance evaluation and algorithm optimization. Through the above-mentioned risk-minimizing decision-making mechanism, the accuracy and reliability of environmental perception are improved while ensuring driving safety, providing high-quality environmental perception services for connected electric vehicles.

[0088] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0089] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0090] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0091] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0092] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0093] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0095] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0097] In conclusion, 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. An optimization method for intelligent connected communication and edge computing in electric vehicles, characterized in that, include: S1. Edge computing nodes acquire multimodal perception data and corresponding raw confidence scores uploaded by connected electric vehicles through wireless communication networks; S2. Construct a dynamic topology graph based on the communication connection relationship between connected electric vehicles and edge computing nodes, and generate a topology entropy sequence of the dynamic topology graph; S3. Analyze the collaborative driving task to determine the real-time environmental perception requirements, and generate the expected utility level by matching the real-time environmental perception requirements with the type of multimodal perception data and the observation target. S4. Determine the effective survival window for each multimodal sensing data based on its physical characteristics, and analyze the degree of erosion of the effective survival window by network disturbances reflected by the topological entropy sequence, including: The physical characteristic time constraints are determined according to the type of multimodal sensing data, and the effective survival window of each multimodal sensing data is determined based on the physical characteristic time constraints. Analyze the trend of topological entropy sequence over time to extract dynamic characteristic parameters that characterize the intensity of network disturbances; Establish a mapping model between network disturbance intensity and the remaining value of the effective survival window; Based on dynamic characteristic parameters and mapping relationship models, the timeliness erosion factor of network disturbances on the effective survival window is calculated; The degree of erosion of the effective survival window by network perturbation is quantified based on the time-dependent erosion factor. S5. Based on the degree of erosion and the expected utility level, the original confidence scores of each multimodal sensing data are dynamically redistributed at the system level to generate effective confidence scores; S6. Evaluate the probability of the validity of different environmental perception hypotheses and the risk of misjudgment based on the effective confidence level, generate the final environmental perception results based on the risk minimization decision criterion, and send them to the connected electric vehicles.

2. The electric vehicle intelligent connected communication and edge computing optimization method according to claim 1, characterized in that, Edge computing nodes acquire multimodal sensing data and corresponding raw confidence scores uploaded by connected electric vehicles via wireless communication networks, including: The system receives data packets sent by a connected electric vehicle, wherein the data packets contain multimodal perception data collected by the on-board sensors of the connected electric vehicle and raw confidence scores generated by the on-board computing unit of the connected electric vehicle based on the quality and integrity of the perception data, wherein the multimodal perception data includes at least one of image data collected by a camera, point cloud data collected by a lidar, and radar data collected by a millimeter-wave radar. Parse the data packets to separate the multimodal sensing data from the original confidence level, and establish the corresponding relationship between the multimodal sensing data and the original confidence level.

3. The electric vehicle intelligent connected communication and edge computing optimization method according to claim 1, characterized in that, A dynamic topology graph is constructed based on the communication connections between connected electric vehicles and edge computing nodes, and a sequence of topological entropy of the dynamic topology graph is generated, including: A dynamic topology graph is constructed using connected electric vehicles and edge computing nodes as vertices, and wireless communication connections between connected electric vehicles and edge computing nodes as edges. Calculate the eigenvalues ​​of the adjacency matrix based on the adjacency matrix of the dynamic topology graph; Calculate the topological entropy value of the dynamic topological graph based on the distribution of eigenvalues; The steps of constructing a dynamic topology graph, calculating feature values, and calculating topology entropy values ​​are repeated at preset time intervals to generate a topology entropy sequence arranged in chronological order to quantify the unpredictability of topology changes.

4. The electric vehicle intelligent connected communication and edge computing optimization method according to claim 3, characterized in that, The calculation of the topological entropy value of a dynamic topological graph based on the distribution of eigenvalues ​​includes: obtaining all eigenvalues ​​of the adjacency matrix of the dynamic topological graph; calculating the sum of squares of all eigenvalues; and using the sum of squares as the topological entropy value.

5. The electric vehicle intelligent connected communication and edge computing optimization method according to claim 1, characterized in that, The collaborative driving task is analyzed to determine real-time environmental perception requirements. Based on the type of multimodal perception data and the observed targets, the expected utility level is generated by matching the real-time environmental perception requirements, including: Analyze the cooperative driving task instructions to determine the task type and the required real-time environmental perception requirements, including the requirements for perception range, perception accuracy, and perception update frequency. Identify the types of multimodal sensing data and the observation targets represented by the multimodal sensing data; The types of multimodal sensing data and observation targets are matched with the real-time needs of environmental sensing, and expected utility levels representing the importance of the data are generated according to predefined matching rules.

6. The electric vehicle intelligent connected communication and edge computing optimization method according to claim 5, characterized in that, Identifying the type of multimodal sensing data and the observation targets represented by the multimodal sensing data includes: parsing the metadata tags in the packet header to determine the type of multimodal sensing data; and parsing the attribute information of the observation targets based on the packet payload content.

7. The electric vehicle intelligent connected communication and edge computing optimization method according to claim 1, characterized in that, Based on the degree of erosion and the expected utility level, the original confidence scores of each multimodal sensing data are dynamically redistributed at the system level to generate effective confidence scores, including: The confidence loss of each multimodal sensing data due to network disturbance is calculated based on the degree of erosion. The priority weight of each multimodal sensing data in the global system is determined based on the expected utility level; Establish a system-level confidence resource pool, the total amount of which is determined based on the sum of the confidence loss of all multimodal sensing data; Resources in the confidence resource pool are redistributed to the corresponding multimodal sensing data according to priority weights, and the original confidence is compensated based on the amount of redistributed resources to generate effective confidence.

8. The electric vehicle intelligent connected communication and edge computing optimization method according to claim 1, characterized in that, Based on the effective confidence level, the probability of the validity of different environmental perception hypotheses and the risk of misjudgment are assessed. The final environmental perception results are generated based on the risk minimization decision criterion and distributed to connected electric vehicles, including: Based on the effective confidence scores corresponding to the multimodal sensing data, the probability of each environmental sensing hypothesis being true is calculated. Obtain the decision risk cost caused by misjudgment under different predefined environmental perception assumptions; Based on the probability of each environmental perception hypothesis being true and the corresponding decision risk cost, calculate the expected risk of each environmental perception hypothesis. The environmental perception hypothesis with the lowest expected risk is selected as the final environmental perception result, and then transmitted to the connected electric vehicle via a wireless communication network.

9. The electric vehicle intelligent connected communication and edge computing optimization method according to claim 8, characterized in that, Based on the probability of each environmental perception hypothesis being true and the corresponding decision risk cost, the expected risk of each environmental perception hypothesis is calculated, including multiplying the probability of each environmental perception hypothesis being true by its decision risk cost, and the product is the expected risk of the environmental perception hypothesis.

Images