An intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game

Abstract

This invention discloses an intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory, relating to the field of traffic control technology. The method includes: extracting structured prediction data from a demand forecast heatmap and jointly mapping it with a traffic control state set to output a traffic control constraint set; analyzing the state feasibility of the traffic control state set based on the traffic control constraint set to generate a candidate traffic control action set; evaluating the dependence of the candidate traffic control action set on vehicle multi-source perception data to generate an action dependency label set; and performing joint modeling and dynamic game theory adjudication on the candidate traffic control action set based on the credibility evolution sequence and the action dependency label set to output a cooperative scheduling scheme. This invention reduces the risk of cooperative scheduling failure and improves the overall reliability of cooperative scheduling in intelligent transportation IoT application service scenarios by using credibility evolution based on traffic control closed-loop evidence and dynamic game theory adjudication based on scheduling action dependency modeling.

Description

Technical Field

[0001] This invention relates to the field of traffic control technology, and in particular to an intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory. Background Technology

[0002] With the rapid development of autonomous driving technology and vehicle-to-everything (V2X) technology, intelligent vehicles (including autonomous taxis, logistics delivery vehicles, and shared cars) have gradually entered the commercial application stage. Especially in urban travel and last-mile delivery scenarios, the efficiency of intelligent vehicle dispatching directly affects travel experience, logistics timeliness, and the utilization rate of transportation resources. The current development trend in the field of intelligent vehicle dispatching shows the demand characteristics of "high real-time performance, high collaboration, and high intelligence," requiring intelligent transportation IoT application services to accurately match user travel or delivery needs with vehicle resources, and dynamically adapt to complex variables such as real-time road conditions, traffic control, and vehicle status.

[0003] However, in the existing urban traffic environment, intelligent vehicle dispatching faces multi-dimensional challenges, such as uneven spatial and temporal distribution of user demand (e.g., concentrated travel during morning and evening rush hours and dispersed demand in remote areas), dynamic changes in road conditions (frequent occurrences of emergencies such as congestion, accidents, and temporary traffic control), and heterogeneous vehicle resource status (differences in driving range, passenger and cargo capacity, and operational failures). Traditional dispatching solutions are no longer able to meet the needs of efficient collaboration, forming technical pain points and market gaps. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent vehicle collaborative scheduling method based on multi-source perception and dynamic game theory to solve the problems of lack of global coordination and insufficient stability of resource coordination efficiency in intelligent vehicle scheduling.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory, comprising:

[0008] Collect multi-source vehicle perception data, perform data fusion and preprocessing on the multi-source vehicle perception data, and generate a traffic operation status view;

[0009] By identifying the sources of perception in the traffic operation status view, statistically analyzing the traffic control closed-loop evidence in the traffic operation status view, forming a credible evidence sequence, calculating the credibility change trend of different sources of perception in the credible evidence sequence, and outputting the credibility evolution sequence.

[0010] Based on the credibility evolution sequence, the vehicle multi-source perception data is switched in a regular manner to generate a traffic control state set. Based on the traffic control state set, the spatiotemporal distribution of user demand is predicted using long short-term memory network and attention mechanism to generate a demand prediction heatmap.

[0011] Extract structured forecast data from the demand forecast heatmap and jointly map it with the traffic control state set to output a traffic control constraint set. Analyze the state feasibility of the traffic control state set based on the traffic control constraint set to generate a candidate traffic control action set.

[0012] The system assesses the dependence of candidate traffic control action sets on multi-source vehicle perception data, generates action dependency label sets, and performs joint modeling and dynamic game adjudication on candidate traffic control action sets based on the credibility evolution sequence and action dependency label sets, outputting a collaborative scheduling scheme.

[0013] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for generating the traffic operation status view are as follows:

[0014] Identify abnormal data in multi-source vehicle perception data and fill in missing data to generate a multi-source basic dataset;

[0015] Perform cross-source association and standardization processing on the multi-source basic dataset to generate a standardized perception dataset, and attach a unique source identifier to each vehicle's multi-source perception data;

[0016] By statistically analyzing the data update frequency and change characteristics of the standardized sensing dataset, the standardized sensing dataset is divided into different scheduling levels and structurally encapsulated to form a traffic operation status view.

[0017] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for forming a credible evidence sequence are as follows:

[0018] The traffic operation status view is classified and aggregated according to the source identifier to obtain the source-classified scheduling dataset;

[0019] Based on the source-classified scheduling dataset, the traffic operation status view is compared before and after execution, cross-source consistency is compared, and execution feedback is correlated. Multiple types of credible evidence are extracted from the traffic operation status view and organized into a credible evidence sequence in chronological order.

[0020] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for outputting the reliability evolution sequence are as follows:

[0021] The credible evidence sequence is divided into several source credible evidence subsequences according to the source identifier, and evidence compression processing is performed on the source credible evidence subsequences corresponding to each source identifier to map multiple types of credible evidence in the same scheduling period into periodic evidence scores.

[0022] Within a continuous scheduling cycle, the credibility value of the current scheduling cycle is calculated based on the periodic evidence score of the current scheduling cycle, and a credibility evolution sequence is formed according to the scheduling cycle order.

[0023] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for generating the traffic control state set are as follows:

[0024] Obtain the credibility change magnitude and stability characteristics of each source identifier in the credibility evolution sequence within the current scheduling period;

[0025] Based on the magnitude of confidence change and stability characteristics, the traffic decision weights of the multi-source perception data of vehicles from the corresponding perception sources in the traffic operation status view are quantitatively adjusted, and fusion processing is performed to form a traffic control status set.

[0026] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for generating the demand prediction heatmap are as follows:

[0027] Vehicle service status, task fulfillment status, road network traffic status, and environmental event status are extracted from the traffic control status and spatiotemporally aligned to form a demand forecast dataset.

[0028] The demand forecast dataset is input into a long short-term memory network to perform temporal encoding processing, and the output is a temporal hidden state sequence.

[0029] Through the attention mechanism, attention weights are dynamically allocated to the temporal hidden state sequence under different scheduling cycles, and weighted aggregation processing is performed to generate a temporal aggregated representation vector.

[0030] Spatial mapping is performed on the time-series aggregated representation vector to generate a demand forecast heatmap.

[0031] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for outputting the traffic control constraint set are as follows:

[0032] The demand intensity, time-period demand fluctuation, and demand time sensitivity of each spatial grid are extracted from the demand forecast heatmap and calibrated according to the time dimension and spatial grid dimension to form structured forecast data.

[0033] Based on spatial grid and time dimensions, the structured prediction data and traffic control state set are spatiotemporally aligned and correlated in multiple dimensions to output a joint demand dataset;

[0034] Based on the demand-joint dataset, analyze the scheduling constraints of each spatial grid in different scheduling cycles to generate a traffic control constraint set.

[0035] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for generating the candidate traffic control action set are as follows:

[0036] Based on the traffic control constraint set, a constraint matching analysis is performed on the spatial grid corresponding to the traffic control state set and the vehicle state, road network state and task state within the scheduling cycle to form a state feasibility judgment result.

[0037] The spatial grid and scheduling cycle that satisfy the state feasibility judgment results of the traffic control state set are selected, and state mapping processing is performed on the traffic control state set within the selected spatial grid and scheduling cycle to generate a candidate traffic control action set.

[0038] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for generating the action dependency label set are as follows:

[0039] Within the scheduling cycle, the number of times each candidate traffic control action in the candidate traffic control action set references vehicle multi-source perception data from different sources, the duration of reference, and the stability of reference are statistically analyzed and quantified to generate an action perception dependency feature set.

[0040] The action-aware dependency feature set is processed by performing pattern classification to generate an action-dependent label set.

[0041] As a preferred embodiment of the intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory described in this invention, the specific steps for outputting the cooperative scheduling scheme are as follows:

[0042] Based on the action dependency tag set, identify the source identifier that each candidate traffic control action in the candidate traffic control action set depends on, and read the corresponding confidence value from the confidence evolution sequence to form a confidence mapping sequence;

[0043] Perform trustworthy consistency evaluation and constraint pruning on the trustworthy mapping sequence to form a set of trustworthy pruning scheduling actions;

[0044] The set of trusted pruning and scheduling actions is organized and summarized according to the spatial grid and scheduling cycle to generate a collaborative scheduling scheme.

[0045] The beneficial effects of this invention are as follows: By using the credibility evolution based on traffic control closed-loop evidence, the reliability of different sensing sources is dynamically characterized within a continuous scheduling cycle. This prevents sensing information from being used equivalently, and instead continuously adjusts decision weights based on historical consistency and execution feedback. This effectively suppresses the amplified interference of low-reliability information on scheduling results in scenarios of sensing anomalies, communication fluctuations, or environmental disturbances, thereby improving the stability of scheduling decisions. Simultaneously, through a dynamic game-theoretic decision-making step based on scheduling action dependency modeling, the dependency relationship of scheduling actions on multi-source sensing data is explicitly incorporated into the game and constraint analysis process. Before generating scheduling schemes, candidate actions that overly rely on low-reliability sensing sources are pruned, reducing the risk of collaborative scheduling failure and improving the overall reliability of collaborative scheduling in intelligent transportation IoT application service scenarios. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of an intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory.

[0048] Figure 2 This is a flowchart for multi-source sensing data fusion and reliability assessment.

[0049] Figure 3 A flowchart generated for dynamic game theory and collaborative scheduling.

[0050] Figure 4 A flowchart for generating intelligent demand forecasting and demand forecasting heatmaps. Detailed Implementation

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0052] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0054] Reference Figures 1-4 This is one embodiment of the present invention, which provides an intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory, including the following steps:

[0055] S1. Collect multi-source vehicle perception data, perform data fusion and preprocessing on the multi-source vehicle perception data, and generate a traffic operation status view.

[0056] S1.1 Identify abnormal data in the multi-source perception data of vehicles and fill in the missing data to generate a multi-source basic dataset.

[0057] It should be noted that in the application service scenario of intelligent transportation IoT, the multi-source vehicle perception data consists of vehicle operation status data, road network traffic status data, task scheduling service status data, and environmental and event status data. Among them, vehicle operation status data is acquired by the vehicle terminal through the on-board acquisition device, including vehicle location, driving speed, driving direction, load status, and remaining range status; road network traffic status data is collected by roadside sensing devices and traffic information release interfaces, including road segment capacity, congestion level, traffic control status, and signal timing status; task scheduling service status data is jointly collected by the task platform and the scheduling execution feedback interface, used to describe the comprehensive status of the task during the scheduling process, including task attribute information such as task release location, task service time window, and task priority, as well as scheduling execution status information such as scheduling action execution identifier, execution start status, execution interruption status, and execution completion status; environmental and event status data is collected by third-party information interfaces, including weather status, emergency event information, and information on the impact area of ​​large-scale events; timestamp identifiers and spatial location identifiers are uniformly recorded during the collection phase.

[0058] Vehicle multi-source perception data is classified according to data source category and sorted by timestamp to form a vehicle perception data sequence. For the same vehicle perception data corresponding to adjacent timestamps within the same data source category in the vehicle perception data sequence, the direction of change of the vehicle perception data corresponding to adjacent timestamps and the dominant direction of change of the same vehicle perception data within the historical time window are obtained. At the same time, the magnitude of change of the vehicle perception data corresponding to adjacent timestamps is obtained, and the range of magnitude of change of the same vehicle perception data within the historical time window is calculated. When the direction of change is opposite to the dominant direction of change and the magnitude of change exceeds the range of magnitude of change, the corresponding vehicle perception data is judged as abnormal data and is removed or replaced by the interpolation result of adjacent valid data records.

[0059] After completing the anomaly data processing, the system identifies data records in the vehicle multi-source perception data that have missing fields or timestamps, and reads the valid data records corresponding to the adjacent timestamps before and after the missing position within the same data source category. For data records with missing fields, the system uses linear interpolation of adjacent valid data records to fill in the missing field values. For data records with missing timestamps, the system uses temporal interpolation of adjacent valid data records to reconstruct the data record corresponding to the missing timestamp. At the same time, the system performs spatial location continuity verification and temporal change continuity verification on the filled or reconstructed data records, and removes data records that do not meet the continuity constraints. The vehicle multi-source perception data after anomaly processing and missing data completion are then aggregated to form a multi-source basic dataset.

[0060] It should also be noted that the range of change is determined based on the statistical results of the change range of the same vehicle perception data within the historical time window. Specifically, the change range corresponding to adjacent timestamps is statistically analyzed one by one within the historical time window, and the upper and lower boundary values ​​are extracted as the range of change based on the distribution of the change range. The historical time window is updated with the scheduling cycle, so that the range of change can dynamically reflect the change characteristics of vehicle perception data under normal operating conditions.

[0061] S1.2 Perform cross-source association and standardization processing on the multi-source basic dataset to generate a standardized perception dataset, and attach a unique source identifier to each vehicle multi-source perception data.

[0062] It should be noted that the timestamp and spatial location identifiers corresponding to each vehicle perception data are read from the multi-source basic dataset. Based on the timestamp alignment and spatial location matching relationships, cross-source association processing is performed on vehicle perception data from different sources that describe the same vehicle, the same road segment, or the same task status to form a combination of associated records. The field values ​​of each vehicle perception data in the multi-source basic dataset are subjected to unified dimensional conversion and value range mapping processing to ensure that each vehicle perception data meets the unified numerical expression rules, forming vehicle perception fields. At the same time, a source identifier is attached to each vehicle perception field to represent the data collection source category. The combination of associated records, vehicle perception fields, and source identifiers are then collected to form a standardized perception dataset.

[0063] S1.3 By statistically analyzing the data update frequency and change characteristics of the standardized sensing dataset, the standardized sensing dataset is divided into different scheduling levels and structurally encapsulated to form a traffic operation status view.

[0064] It should be noted that the timestamp identifiers corresponding to each vehicle perception field are read from the standardized perception dataset, and the update interval distribution of each vehicle perception field in continuous scheduling cycles is statistically analyzed to obtain the data update frequency characteristics reflecting the speed of data updates. At the same time, the value changes of each vehicle perception field in adjacent scheduling cycles are statistically analyzed to obtain the change characteristic indicators that characterize the degree of value fluctuation and the stability of change. Vehicle perception fields with consistent trends in data update frequency characteristics and change characteristic indicators are grouped into the same update behavior category, and are uniformly organized and encapsulated according to spatial location identifiers and timestamp identifiers to form a traffic operation status view.

[0065] S2. By identifying the sources of perception in the traffic operation status view, statistically analyzing the traffic control closed-loop evidence in the traffic operation status view, forming a credible evidence sequence, calculating the credibility change trend of different sources in the credible evidence sequence, and outputting the credibility evolution sequence.

[0066] S2.1 Classify and aggregate the traffic operation status views according to the source identifier to obtain the source classification scheduling dataset.

[0067] It should be noted that the source identifiers corresponding to each vehicle perception field are read from the traffic operation status view. Vehicle operation status data, road network traffic status data, task scheduling service status data, and environmental and event status data with the same source identifier are collected and processed. During the collection and processing, the original timestamp identifiers and spatial location identifiers of each vehicle perception field are kept unchanged, and the collected vehicle perception fields are organized according to the timestamp order to form source classification scheduling datasets corresponding to different source identifiers.

[0068] S2.2 Based on the source-classified scheduling dataset, perform state comparison before and after execution, cross-source consistency comparison, and execution feedback correlation analysis on the traffic operation status view to extract multiple types of credible evidence from the traffic operation status view and organize them into a credible evidence sequence in chronological order.

[0069] It should be noted that the corresponding vehicle perception fields are read from the source classification scheduling dataset according to the source identifier, and organized into vehicle time series according to the timestamp identifier order. Under the same source identifier, for vehicle perception fields with the same spatial location identifier and the same vehicle perception field in the vehicle time series, the direction and magnitude of change between the corresponding values ​​at adjacent timestamps are statistically analyzed to form a comparison result of the before and after states to represent the continuity of time. Under the same timestamp, for vehicle perception fields with different source identifiers but the same spatial location identifier, the consistency of the direction of change and the deviation of the magnitude of change of the corresponding values ​​are compared to form a cross-source consistency comparison result to represent the consistency of multiple sources. The changes in task scheduling service status data in adjacent scheduling cycles are time-aligned and correlated with the changes in corresponding vehicle operation status data and road network traffic status data to represent the execution feedback correlation analysis results. The before and after state comparison results, cross-source consistency comparison results, and execution feedback correlation analysis results are combined according to the source identifier and timestamp identifier to generate credible evidence, and the credible evidence is arranged in chronological order to form a credible evidence sequence.

[0070] S2.3. Divide the credible evidence sequence into several source credible evidence subsequences according to the source identifier, and perform evidence compression processing on the source credible evidence subsequences corresponding to each source identifier to map multiple types of credible evidence in the same scheduling period into periodic evidence scores.

[0071] It should be noted that credible evidence sequences with the same source identifier and timestamps falling within the same scheduling period are grouped into a source credible evidence subsequence. Within each source credible evidence subsequence, multiple types of credible evidence extracted within the same scheduling period are sorted according to their numerical values, and the maximum and minimum values ​​in the sorting results are extracted respectively. The difference between the maximum and minimum values ​​is used as the periodic evidence score. The periodic evidence scores obtained by each source identifier within consecutive scheduling periods are sorted in order of timestamps to form a periodic evidence score sequence.

[0072] S2.4. Within a continuous scheduling cycle, based on the periodic evidence score of the current scheduling cycle, calculate the credibility value of the current scheduling cycle and form a credibility evolution sequence according to the scheduling cycle order.

[0073] It should be noted that the periodic evidence score corresponding to each source identifier is read in the order of the scheduling cycle, and the credibility value of the current scheduling cycle is calculated. The credibility values ​​of each scheduling cycle are sorted to form a credibility evolution sequence.

[0074] ;

[0075] in, The source identifier is Credible evidence during the scheduling period The credibility value; The source identifier is Credible evidence during the scheduling period Periodic evidence scoring; The source identifier is Credible evidence during the scheduling period To the scheduling cycle A set of periodic evidence scores; This represents the interval clipping operator, used to restrict the calculation result within the parentheses to a closed interval between 0 and 1; This represents a stability factor used to avoid extremely small positive values ​​where the denominator is zero. Its value ranges from 0.0000001 to 0.0001. The lower limit of 0.0000001 is used to cover scenarios where the range of the historical periodic evidence score set is close to zero, meaning that the periodic evidence score of the source identifier remains basically unchanged within a continuous scheduling period. The upper limit of 0.0001 is used to cover scenarios where the historical periodic evidence score set has small fluctuations but the overall change is still relatively low.

[0076] S3. Based on the credibility evolution sequence, perform rule-based switching on the vehicle multi-source perception data to generate a traffic control state set. Based on the traffic control state set, use long short-term memory network and attention mechanism to predict the spatiotemporal distribution of user demand and generate a demand prediction heatmap.

[0077] S3.1 Obtain the credibility change magnitude and stability characteristics of the corresponding credibility values ​​of each source identifier in the credibility evolution sequence within the current scheduling period.

[0078] It should be noted that the credibility values ​​corresponding to the current scheduling period and several consecutive scheduling periods (such as three consecutive periods) are read from the credibility evolution sequence according to the source identifier, and source credibility time segments are formed in the order of scheduling periods. Within each source credibility time segment, the change range of the credibility value is obtained by statistically analyzing the difference between the credibility value of the current scheduling period and the credibility value of the adjacent scheduling period. At the same time, the stability characteristics are obtained by statistically analyzing the fluctuation of the credibility value within the credibility time segment.

[0079] S3.2 Based on the magnitude of confidence change and stability characteristics, quantitatively adjust the traffic decision weights of the vehicle multi-source perception data from the corresponding perception sources in the traffic operation status view, and perform fusion processing to form a traffic control status set.

[0080] It should be noted that the corresponding vehicle multi-source perception data is read from the traffic operation status view according to the source identifier, and the reliability change amplitude and stability characteristics of each source identifier within the current scheduling cycle are synchronized. The reliability change amplitude and stability characteristics of each source identifier are jointly sorted to form the source reliability sorting result. Based on the source reliability sorting result, each source identifier is mapped to a traffic decision weight within a continuous value range. The traffic decision weight is then attached to the vehicle multi-source perception data of the corresponding source identifier and weighted fusion processing is performed to form a status field. The status field is then organized according to the spatial location identifier and the timestamp identifier to form a traffic control status set.

[0081] S3.3 Extract vehicle service status, task fulfillment status, road network traffic status, and environmental event status from the traffic control status set, and perform spatiotemporal alignment processing to form a demand forecast dataset.

[0082] It should be noted that, according to scheduling semantics, state fields reflecting vehicle service progress, task execution progress, road network traffic conditions, and the degree of impact of environmental events are extracted from the state fields of the traffic control state set; the state fields are aligned to the time axis of a unified scheduling cycle by timestamp identifiers, and spatial alignment is performed by spatial location identifiers, so that different state fields correspond to the same spatial location within the same scheduling cycle; the state fields after time and spatial alignment are organized in the order of scheduling cycles to form a demand forecast dataset.

[0083] It should also be noted that scheduling semantics refers to the business meaning expressed by multi-source vehicle perception data in traffic scheduling scenarios, which can be directly used to describe the current operating status and scheduling feasibility of the scheduling object.

[0084] S3.4 Input the demand forecast dataset into the Long Short-Term Memory network to perform temporal encoding processing and output the temporal hidden state sequence.

[0085] It should be noted that the Long Short-Term Memory (LSTM) network is trained based on the demand prediction dataset formed within historical scheduling cycles. During training, the demand prediction dataset corresponding to consecutive scheduling cycles is read from the historical demand prediction dataset according to spatial location identifiers as the trainable input sequence, and the demand prediction dataset corresponding to the input sequence in subsequent scheduling cycles is used as the supervised target sequence. The prediction error between the trainable input sequence and the supervised target sequence is obtained, and the gradient of the network parameters is obtained by backpropagation of the prediction error. The network parameters are iteratively updated using gradient descent update rules. Training is stopped when the prediction error no longer decreases, and the network parameters corresponding to the minimum prediction error are saved for use in the trained LSM network.

[0086] The state fields of corresponding spatial locations within consecutive scheduling cycles are read from the demand forecast dataset according to their spatial location identifiers, and then arranged into a time-series input sequence according to their timestamps. This time-series input sequence is then fed into a Long Short-Term Memory (LSTM) network cycle by cycle. Within each scheduling cycle, the LSM network performs linear transformations and nonlinear mappings on the state fields of the current scheduling cycle and the historical memory vectors stored within the LSM network, respectively, to obtain the gating coefficients for the forget gate, input gate, and output gate. The historical memory vectors are weighted and retained using the forget gate gating coefficients, and the time-series input sequence is weighted and written using the input gate gating coefficients. The weighted retention result and the weighted writing result are then combined to form the updated memory vector corresponding to the current scheduling cycle. The updated memory vector is then gated and output based on the output gate gating coefficients to generate the hidden vector corresponding to the current scheduling cycle. The hidden vectors obtained for the same spatial location identifier within consecutive scheduling cycles are organized according to their timestamps to form a hidden vector subsequence. Finally, the hidden vector subsequences corresponding to each spatial location identifier are aggregated to form a time-series hidden state sequence.

[0087] It should also be noted that the historical memory vector is the historical information of the previous scheduling cycle that is retained within the Long Short-Term Memory network.

[0088] S3.5. Through the attention mechanism, attention weights are dynamically allocated to the temporal hidden state sequences under different scheduling cycles, and weighted aggregation processing is performed to generate temporal aggregated representation vectors.

[0089] It should be noted that, from the temporal hidden state sequence, the hidden vector subsequences obtained for the corresponding spatial positions within the continuous scheduling period are read according to the spatial position identifier. The correlation score between each hidden vector in the hidden vector subsequence and the other hidden vectors in the same hidden vector subsequence is obtained. The correlation scores within the same hidden vector subsequence are normalized to obtain the attention weight of each hidden vector in the corresponding scheduling period. Based on the attention weight, the hidden vectors in the hidden vector subsequence are weighted and summed to obtain the temporal aggregated representation vector.

[0090] The expression for calculating the relevance score is:

[0091] ;

[0092] in, Indicates the first The correlation score between the hidden vector of each scheduling cycle and the other hidden vectors in the same hidden vector subsequence. Indicates the first The hidden vectors corresponding to each scheduling period are transposed. As a result, Indicates the first The hidden vector corresponding to each scheduling cycle Indicates the number of scheduling cycles. The L2 norm of a vector is used to uniformly measure the overall strength of hidden vectors, ensuring that correlation calculations only reflect the structural similarity between hidden vectors.

[0093] S3.6 Perform spatial mapping processing on the time-series aggregated representation vector to generate a demand forecast heatmap.

[0094] It should be noted that the spatial location identifier of the temporal aggregation representation vector is mapped to a spatial grid with a fixed grid side length. For each spatial grid, the corresponding temporal aggregation representation vector is input into the spatial mapping function within each scheduling cycle. The spatial mapping function performs dimensional projection and nonlinear constraint processing on the temporal aggregation representation vector and outputs a spatial grid-level demand prediction field. The spatial grid-level demand prediction field includes the demand quantity prediction value (from the vector component in the temporal aggregation representation vector used to represent the trend of demand scale change) and the demand time sensitivity level prediction value (from the vector component in the temporal aggregation representation vector used to represent the strength of demand constraint on the time window). The spatial grid-level demand prediction fields output by each spatial grid within the same scheduling cycle are filled into a two-dimensional matrix according to the spatial grid index and a timestamp identifier corresponding to the scheduling cycle is added to form a demand prediction heatmap.

[0095] S4. Extract the structured forecast data from the demand forecast heatmap and perform joint mapping with the traffic control state set to output the traffic control constraint set. Analyze the state feasibility of the traffic control state set based on the traffic control constraint set and generate a set of candidate traffic control actions.

[0096] S4.1 Extract the demand intensity, time-period demand fluctuation, and demand time sensitivity of each spatial grid from the demand forecast heatmap, and calibrate them according to the time dimension and spatial grid dimension to form structured forecast data.

[0097] It should be noted that the predicted demand quantity and the predicted demand time sensitivity level for each spatial grid in each scheduling cycle are read from the demand forecasting heatmap, and the predicted demand quantity is taken as the demand intensity. Within the same spatial grid, the difference in the predicted demand quantity for adjacent scheduling cycles is obtained in the order of scheduling cycles and aggregated into a fluctuation sequence. The statistical value of the fluctuation sequence is taken as the demand fluctuation of the time period. The predicted demand time sensitivity level is taken as the demand time sensitivity. Spatial grid index, scheduling cycle index and timestamp identifier are added to the demand intensity, the demand fluctuation of the time period and the demand time sensitivity respectively to form structured forecast data.

[0098] S4.2. Based on the spatial grid and time dimensions, the structured prediction data and traffic control state set are spatiotemporally aligned and correlated in multiple dimensions to output a joint demand dataset.

[0099] It should be noted that the spatial location identifiers in the traffic control state set are mapped to the corresponding spatial grid indexes in the structured prediction data, and the traffic control state set is re-indexed according to the spatial grid index and the scheduling cycle index; spatiotemporal alignment processing is performed using the spatial grid index and the scheduling cycle index as alignment keys to obtain aligned record pairs with consistent spatial grid index and scheduling cycle index; within each aligned record pair, the state field of the traffic control state set is retained according to the source identifier, and the demand intensity, time period demand fluctuation, and demand time sensitivity are concatenated and associated with the state field at the field level to form a joint demand dataset containing the spatial grid index, scheduling cycle index, timestamp identifier, demand intensity, time period demand fluctuation, demand time sensitivity, state field, and source identifier.

[0100] S4.3. Based on the demand-joint dataset, analyze the scheduling constraints of each spatial grid in different scheduling cycles to generate a traffic control constraint set.

[0101] It should be noted that the following data is retrieved from the joint demand dataset: demand intensity, time-period demand fluctuation, and demand time sensitivity, along with the vehicle location, speed, load status, remaining range status, road segment capacity, congestion level, traffic control status, signal timing status, task service time window, task priority, scheduling action execution identifier, execution start status, execution interruption status, and execution completion status. Constraint quantification processing is performed on the extracted data, including mapping demand time sensitivity to service timeliness level, jointly mapping congestion level and traffic control status to traffic restriction level, jointly mapping remaining range status and load status to task capacity restriction level, jointly mapping task service time window and scheduling cycle index to time window satisfaction level, and jointly mapping scheduling action execution identifier and execution completion status to fulfillment reliability level. Consistency summarization is performed on the traffic restriction level, task capacity restriction level, time window satisfaction level, fulfillment reliability level, and service timeliness level obtained for each spatial grid identifier under each scheduling cycle index, outputting a traffic control constraint set containing reachability constraints, time window constraints, capacity constraints, fulfillment constraints, and service timeliness constraints.

[0102] S4.4. Based on the traffic control constraint set, perform a constraint matching analysis on the spatial grid corresponding to the traffic control state set and the vehicle state, road network state and task state within the scheduling cycle to form a state feasibility judgment result.

[0103] It should be noted that, according to the spatial grid identifier and scheduling cycle index, the corresponding accessibility constraints, time window constraints, carrying capacity constraints, performance constraints, and service timeliness constraints are read from the traffic control constraint set; simultaneously, vehicle operation status data, road network traffic status data, and task scheduling service status data under the same spatial grid identifier and scheduling cycle index are read from the traffic control status set; the vehicle operation status data is matched item by item with the accessibility constraints and carrying capacity constraints, the road network traffic status data is matched item by item with the traffic restriction constraints, and the task scheduling service status data is matched item by item with the time window constraints, performance constraints, and service timeliness constraints; in each constraint matching process, the judgment result of whether the matching is satisfied or not is recorded, and the matching results of all constraints within the same spatial grid and scheduling cycle are summarized. When all constraints are satisfied, a feasible status identifier is output; when any constraint is not satisfied, an infeasible status identifier is output, forming a status feasibility judgment result.

[0104] S4.5. Select the spatial grid and scheduling cycle that meet the feasibility judgment results of the traffic control state set, and perform state mapping processing on the traffic control state set within the selected spatial grid and scheduling cycle to generate a candidate traffic control action set.

[0105] It should be noted that, according to the spatial grid identifier and scheduling cycle index, combinations of spatial grids and scheduling cycles that meet the feasibility constraints are selected from the state feasibility judgment results, and the corresponding vehicle operation status data, road network traffic status data, and task scheduling service status data are read from the traffic control status set. Based on the status fields of location, driving direction, and service occupancy status reflected in the vehicle operation status data, corresponding vehicle operation adjustment action descriptions are constructed. Based on the status fields of traffic capacity, congestion level, and control status reflected in the road network traffic status data, corresponding path and traffic adjustment action descriptions are constructed. Based on the status fields of task priority, execution progress, and execution status reflected in the task scheduling service status data, corresponding task scheduling adjustment action descriptions are constructed. The action description fields such as vehicle operation adjustment action descriptions, path and traffic adjustment action descriptions, and task scheduling adjustment action descriptions are combined and mapped to form a candidate traffic control action set.

[0106] S5. Evaluate the dependence of the candidate traffic control action set on vehicle multi-source perception data, generate an action dependency label set, and perform joint modeling and dynamic game adjudication on the candidate traffic control action set based on the credibility evolution sequence and the action dependency label set, and output a collaborative scheduling scheme.

[0107] S5.1 During the scheduling cycle, statistically analyze and quantify the number of times each candidate traffic control action in the candidate traffic control action set references vehicle multi-source perception data from different sources, the duration of reference, and the stability of reference, and generate an action perception dependency feature set.

[0108] It should be noted that within each scheduling cycle, the action description fields corresponding to each candidate traffic control action are read one by one from the candidate traffic control action set. Based on the vehicle operation status data, road network traffic status data, and task scheduling service status data explicitly associated in the action description fields, the set of vehicle multi-source perception data on which the current candidate traffic control action depends is determined. According to the source identifiers attached to the vehicle multi-source perception data, the number of times each candidate traffic control action references vehicle multi-source perception data corresponding to different source identifiers within the current scheduling cycle is counted. At the same time, within the continuous scheduling cycle, the number of scheduling cycles in which the same candidate traffic control action continuously references vehicle multi-source perception data of the same source identifier is counted to obtain the reference duration. Based on whether the reference of the current candidate traffic control action to vehicle multi-source perception data of the same source identifier is interrupted or frequently switched within the continuous scheduling cycle, the reference stability is quantified. The reference count, reference duration, and reference stability are combined according to the candidate traffic control action and the source identifier to form an action perception dependency feature set.

[0109] S5.2 Execute pattern classification processing on the action perception dependent feature set to generate action dependent label set.

[0110] It should be noted that, from the action perception dependency feature set, the corresponding three types of action features—citation count, citation duration, and citation stability—are read one by one according to the candidate traffic control actions. The action features are then combined according to the source identifier dimension to form an action dependency feature vector. Action dependency feature vectors that show consistency in the distribution structure of citation count, the distribution structure of citation duration, and the change pattern of citation stability are merged into the same dependency pattern category. A unique action dependency label is assigned to each dependency pattern category. The candidate traffic control actions are associated with and aggregated with their corresponding action dependency labels to form an action dependency label set.

[0111] S5.3. Based on the action dependency tag set, identify the source identifier that each candidate traffic control action in the candidate traffic control action set depends on, and read the corresponding confidence value from the confidence evolution sequence to form a confidence mapping sequence.

[0112] It should be noted that candidate traffic control actions and their corresponding action dependency tags are read one by one from the action dependency tag set, and one or more source identifiers that each candidate traffic control action depends on in the current scheduling cycle are determined based on the action dependency tags; the credibility values ​​of the source identifiers in the corresponding scheduling cycle are read from the credibility evolution sequence, and the candidate traffic control actions, source identifiers and scheduling cycle identifiers are combined and mapped, and then aggregated according to the order of action identifiers to obtain a credibility mapping sequence.

[0113] S5.4 Perform trusted consistency assessment and constraint pruning on the trusted mapping sequence to form a trusted pruning scheduling action set.

[0114] It should be noted that the source identifiers corresponding to the candidate traffic control actions are read one by one from the trusted mapping sequence, and the trustworthiness values ​​of each source identifier in the current scheduling cycle are also read. Among the multiple source identifiers corresponding to the same candidate traffic control action, the trustworthiness values ​​are sorted in order of magnitude, and the difference between the maximum and minimum trustworthiness values, as well as the degree of concentration of the trustworthiness values ​​in the sorted sequence, are calculated to characterize the trustworthiness consistency of the multi-source perception information on which the current candidate traffic control action depends.

[0115] When the reliability values ​​in the ranking results show a concentrated distribution and the difference is within a stable range (not exceeding the historical statistical fluctuation range), it is determined that the current candidate traffic control action meets the reliability consistency requirement in the current scheduling cycle; when the reliability values ​​in the ranking results show a significant discrete distribution or the difference exceeds the stable range, it is determined that the current candidate traffic control action does not meet the reliability consistency requirement in the current scheduling cycle; based on the consistency assessment results, candidate traffic control actions that do not meet the reliability consistency requirement are removed from the candidate traffic control action set to form a reliable pruning scheduling action set.

[0116] S5.5 Organize and summarize the set of trusted pruning and scheduling actions according to the spatial grid and scheduling cycle to generate a collaborative scheduling scheme.

[0117] It should be noted that the spatial grid identifier and scheduling cycle identifier corresponding to each trusted trimmed scheduling action are read one by one from the trusted trimmed scheduling action set. The trusted trimmed scheduling actions are then grouped according to the spatial grid identifier. Within each spatial grid group, the trusted trimmed scheduling actions are further organized in chronological order according to the scheduling cycle identifier. Within the same spatial grid and the same scheduling cycle, multiple trusted trimmed scheduling actions are deduplicated and conflict-resolved, retaining the scheduling action combinations that satisfy traffic control constraints and do not conflict with each other. The scheduling action combinations formed by each spatial grid in each scheduling cycle are summarized according to the spatial grid identifier and the scheduling cycle identifier to form a collaborative scheduling scheme.

[0118] In summary, this invention achieves the following: Firstly, by dynamically characterizing the reliability of different sensing sources within a continuous scheduling cycle based on the credibility evolution of traffic control closed-loop evidence, sensing information is no longer used equivalently but its decision weights are continuously adjusted based on historical consistency and execution feedback. This effectively suppresses the amplified interference of low-reliability information on scheduling results and improves the stability of scheduling decisions in scenarios involving sensing anomalies, communication fluctuations, or environmental disturbances. Secondly, through a dynamic game-theoretic decision-making step based on scheduling action dependency modeling, the dependency relationship of scheduling actions on multi-source sensing data is explicitly incorporated into the game and constraint analysis process. Before generating scheduling schemes, candidate actions that overly rely on low-reliability sensing sources are pruned, reducing the risk of collaborative scheduling failure and improving the overall reliability of collaborative scheduling in intelligent transportation IoT application service scenarios.

[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent vehicle cooperative scheduling based on multi-source perception and dynamic game, characterized in that: include, The process involves collecting multi-source vehicle sensing data, fusing and preprocessing this data, and generating a traffic operation status view. The specific steps are as follows: Identify abnormal data in multi-source vehicle perception data and fill in missing data to generate a multi-source basic dataset; Perform cross-source association and standardization processing on the multi-source basic dataset to generate a standardized perception dataset, and attach a unique source identifier to each vehicle's multi-source perception data; By statistically analyzing the data update frequency and change characteristics of the standardized sensing dataset, the standardized sensing dataset is divided into different scheduling levels and encapsulated in a structured manner to form a traffic operation status view. By identifying the sources of perception in the traffic operation status view, statistically analyzing the traffic control closed-loop evidence in the traffic operation status view, forming a credible evidence sequence, calculating the credibility change trend of different sources in the credible evidence sequence, and outputting the credibility evolution sequence, the specific steps are as follows. The credible evidence sequence is divided into several source credible evidence subsequences according to the source identifier, and evidence compression processing is performed on the source credible evidence subsequences corresponding to each source identifier to map multiple types of credible evidence in the same scheduling period into periodic evidence scores. Within a continuous scheduling cycle, based on the periodic evidence score of the current scheduling cycle, the credibility value of the current scheduling cycle is calculated, and a credibility evolution sequence is formed according to the scheduling cycle order. Based on the reliability evolution sequence, rule-based switching is performed on the multi-source perception data of vehicles to generate a traffic control state set. The specific steps are as follows. Obtain the credibility change magnitude and stability characteristics of each source identifier in the credibility evolution sequence within the current scheduling period; Based on the magnitude and stability characteristics of credibility changes, the traffic decision weights of the multi-source perception data of vehicles from the corresponding perception sources in the traffic operation status view are quantitatively adjusted, and fusion processing is performed to form a traffic control status set. Based on the traffic control state set, the spatiotemporal distribution of user demand is predicted using long short-term memory networks and attention mechanisms, generating a demand prediction heatmap. The structured forecast data from the demand forecast heatmap is extracted and jointly mapped with the traffic control state set to output the traffic control constraint set. The specific steps are as follows. The demand intensity, time-period demand fluctuation, and demand time sensitivity of each spatial grid are extracted from the demand forecast heatmap and calibrated according to the time dimension and spatial grid dimension to form structured forecast data. Based on spatial grid and time dimensions, the structured prediction data and traffic control state set are spatiotemporally aligned and correlated in multiple dimensions to output a joint demand dataset; Based on the demand joint dataset, analyze the scheduling constraints of each spatial grid in different scheduling cycles to generate a traffic control constraint set; Based on the analysis of the traffic control constraint set, the feasibility of the traffic control state set is analyzed, and a set of candidate traffic control actions is generated. The system assesses the dependence of candidate traffic control action sets on multi-source vehicle perception data, generates action dependency label sets, and performs joint modeling and dynamic game adjudication on candidate traffic control action sets based on the credibility evolution sequence and action dependency label sets, outputting a collaborative scheduling scheme.

2. The intelligent vehicle collaborative scheduling method based on multi-source perception and dynamic game according to claim 1, characterized in that: The specific steps for forming a credible evidence sequence are as follows: The traffic operation status view is classified and aggregated according to the source identifier to obtain the source-classified scheduling dataset; Based on the source-classified scheduling dataset, the traffic operation status view is compared before and after execution, cross-source consistency is compared, and execution feedback is correlated. Multiple types of credible evidence are extracted from the traffic operation status view and organized into a credible evidence sequence in chronological order. 3.The intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game according to claim 1, wherein: The specific steps for generating the demand forecast heatmap are as follows: Vehicle service status, task fulfillment status, road network traffic status, and environmental event status are extracted from the traffic control status and spatiotemporally aligned to form a demand forecast dataset. The demand forecast dataset is input into a long short-term memory network to perform temporal encoding processing, and the output is a temporal hidden state sequence. Through the attention mechanism, attention weights are dynamically allocated to the temporal hidden state sequence under different scheduling cycles, and weighted aggregation processing is performed to generate a temporal aggregated representation vector. Spatial mapping is performed on the time-series aggregated representation vector to generate a demand forecast heatmap.

4. The intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory as described in claim 3, characterized in that: The specific steps for generating the candidate traffic control action set are as follows: Based on the traffic control constraint set, a constraint matching analysis is performed on the spatial grid corresponding to the traffic control state set and the vehicle state, road network state and task state within the scheduling cycle to form a state feasibility judgment result. The spatial grid and scheduling cycle that satisfy the state feasibility judgment results of the traffic control state set are selected, and state mapping processing is performed on the traffic control state set within the selected spatial grid and scheduling cycle to generate a candidate traffic control action set.

5. The intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory as described in claim 4, characterized in that: The specific steps for generating the action-dependent tag set are as follows. Within the scheduling cycle, the number of times each candidate traffic control action in the candidate traffic control action set references vehicle multi-source perception data from different sources, the duration of reference, and the stability of reference are statistically analyzed and quantified to generate an action perception dependency feature set. The action-aware dependency feature set is processed by performing pattern classification to generate an action-dependent label set.

6. The intelligent vehicle cooperative scheduling method based on multi-source perception and dynamic game theory as described in claim 5, characterized in that: The specific steps of the output collaborative scheduling scheme are as follows: Based on the action dependency tag set, identify the source identifier that each candidate traffic control action in the candidate traffic control action set depends on, and read the corresponding confidence value from the confidence evolution sequence to form a confidence mapping sequence; Perform trustworthy consistency evaluation and constraint pruning on the trustworthy mapping sequence to form a set of trustworthy pruning scheduling actions; The set of trusted pruning and scheduling actions is organized and summarized according to the spatial grid and scheduling cycle to generate a collaborative scheduling scheme.

Images