Vehicle collision risk prediction method, device and controller

Abstract

This application provides a vehicle collision risk prediction method, device, and controller, relating to the field of intelligent driving technology. The method includes: acquiring vehicle operating status data and road environment data; constructing a dynamic interactive network based on the operating status data and road environment data, wherein the interactive network uses the vehicle and other traffic participants as nodes, and the interaction relationships between nodes as edges; extracting feature representations of each node based on a graph neural network, thereby determining the influence score of each node; comparing the influence score of each node with a dynamic threshold, and generating a sparse attention weight distribution based on the comparison results; and weighted fusing the feature representations of each node based on the weight distribution to obtain the collision risk prediction result. This application, by constructing a dynamic interactive network and introducing a hierarchical attention refinement mechanism, can accurately distinguish between key risk vehicles and secondary interference targets, thereby improving the reliability of vehicle collision risk prediction.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to a method, device and controller for predicting vehicle collision risks. Background Technology

[0002] With the rapid development of vehicle intelligence and connectivity technologies, intelligent transportation systems are placing higher demands on vehicle active safety capabilities. In complex scenarios such as highway congestion and ramp merging, traffic flow density is extremely high, and dynamic interactions between vehicles are extremely frequent, leading to a significant increase in collision risk. Such accidents often have far more serious consequences than a single collision, posing a huge threat to personal safety and the socio-economic well-being. Therefore, the industry's research and development focus has gradually shifted from collision warnings for single vehicles to the accurate and dynamic prediction of collision risks in high-density, multi-vehicle interaction scenarios. This has become a core challenge and a key technological bottleneck in improving road safety.

[0003] In related technologies, deep learning models based on attention mechanisms are commonly used to predict vehicle collision risks. Specifically, onboard sensors and Vehicle-to-Everything (V2X) technology are used to collect state data and environmental information of the vehicle itself and surrounding vehicles; an attention mechanism is used to weightedly fuse information from all interacting vehicles in the scene, aiming to filter out key risk features from massive amounts of data; furthermore, the model outputs a prediction of collision risks over a future period, which provides a basis for subsequent warning or avoidance control decisions. However, the reliability of vehicle collision risk prediction results is low. Summary of the Invention

[0004] This application provides a vehicle collision risk prediction method, device, and controller to improve the problem of low reliability of vehicle collision risk prediction results in related technologies.

[0005] Firstly, this application provides a method for predicting vehicle collision risk, including:

[0006] Obtain the vehicle's operating status data and road environment data;

[0007] A dynamic interactive network is constructed based on operational status data and road environment data; the dynamic interactive network uses the vehicle itself and other traffic participants as nodes, and the interaction relationships between the nodes as edges.

[0008] Based on graph neural networks, feature representations of each node in a dynamic interactive network are extracted.

[0009] Based on the feature representation of each node, the influence score of each node is determined. The influence score is used to quantify the degree of influence of the node on the occurrence of a vehicle collision.

[0010] The influence score of each node is compared with the dynamic threshold, and a sparse attention weight distribution is generated based on the comparison results. The dynamic threshold is determined based on the running status data and road environment data.

[0011] Based on the sparse attention weight distribution, the feature representations of each node are weighted and fused to obtain the prediction results of vehicle collision risk.

[0012] In one possible implementation, feature representations of each node in a dynamic interaction network are extracted based on a graph neural network, including: iterative message passing through multiple network layers of the graph neural network; for each node, the feature representation of the node at layer L+1 is updated by aggregating the node's own feature representation at layer L and the feature representations of each of the node's neighboring nodes at layer L; performing a linear transformation on the aggregated result and processing the linear transformation result with an activation function to generate the feature representation of the node at layer L+1; where L is the layer index.

[0013] In one possible implementation, the influence score of each node is determined based on the feature representation of each node, including: multiplying the norm of the feature representation corresponding to the node with the topological centrality metric of the node in the dynamic interactive network, and using the result of the multiplication as the influence score of the node, wherein the norm of the feature representation corresponding to the node is used to quantify the urgency of the risk of the node causing a direct collision, and the topological centrality metric is used to quantify the criticality of the node in propagating collisions in the dynamic interactive network.

[0014] In one possible implementation, the operational status data includes the vehicle speed, and the road environment data includes traffic flow density. The dynamic threshold is determined as follows: the traffic flow density and the vehicle speed are input into a preset linear relationship model to obtain a preliminary threshold; based on the prediction results of vehicle collision risk over a set historical period and actual traffic condition data, a consistency score and false alarm rate are determined to assess the accuracy of the prediction results for evaluating vehicle collision risk; a reward signal is determined based on the consistency score, false alarm rate, and a preset penalty coefficient; with the goal of maximizing the reward signal, a policy gradient algorithm is used to iteratively update the model parameters of the linear relationship model to optimize the preliminary threshold online and generate a dynamic threshold.

[0015] In one possible implementation, the vehicle collision risk prediction method further includes: generating a vehicle evasion control command based on the predicted vehicle collision risk, and sending the vehicle evasion control command to the vehicle's actuator; collecting the vehicle's state feedback data in real time, and determining the execution deviation between the state feedback data and the expected state feedback data corresponding to the vehicle evasion control command; if the execution deviation exceeds a preset execution deviation threshold, triggering a replanning of the vehicle evasion control command; and simultaneously, using samples containing state feedback data and execution deviation as incremental data, employing an incremental learning algorithm to update the model parameters of the graph neural network online to optimize the accuracy of subsequent vehicle collision risk prediction.

[0016] In one possible implementation, the predicted vehicle collision risk includes a vehicle collision probability value, a predicted risk time interval, a predicted risk spatial coordinate range, and a list of risk vehicle identifiers. Based on the predicted vehicle collision risk, a vehicle evasion control command is generated, including: identifying target threat vehicles based on the risk vehicle identifier list, and determining the minimum safe distance required for the vehicle to achieve full braking within the predicted risk time interval based on the vehicle speed, the predicted risk time interval, and the vehicle's maximum deceleration, and using the minimum safe distance as the safe space boundary; constructing an optimization objective function with the safe space boundary as a hard constraint, the optimization objective function being configured to minimize the amplitude of the control command and optimize the ride comfort index within a future predicted time domain; solving the optimization objective function using a model predictive control algorithm to obtain an optimal control sequence including braking, steering, and power control quantities; and outputting the control quantity corresponding to the current moment in the optimal control sequence as the vehicle evasion control command.

[0017] In one possible implementation, in the step of constructing the dynamic interaction network, the edge weights in the dynamic interaction network are determined as follows: for any two nodes, the interaction strength is determined based on the position vectors and velocity vectors of the two nodes; the relative acceleration magnitude is determined based on the acceleration vectors of the two nodes; a scaling factor is determined based on the relative acceleration magnitude; the interaction strength is multiplied by the scaling factor, and the result of the multiplication is used as the edge weight of the two nodes; wherein, the scaling factor is positively correlated with the relative acceleration magnitude, and when the relative acceleration magnitude increases, it amplifies the interaction strength.

[0018] Secondly, this application provides a vehicle collision risk prediction device, comprising:

[0019] The acquisition module is used to acquire the vehicle's operating status data and road environment data;

[0020] The dynamic interactive network construction module is used to construct a dynamic interactive network based on operational status data and road environment data. The dynamic interactive network uses the vehicle itself and other traffic participants as nodes, and the interaction relationships between the nodes as edges.

[0021] The feature extraction module is used to extract feature representations of each node in a dynamic interactive network based on a graph neural network.

[0022] The determination module is used to determine the influence score of each node based on the feature representation of each node. The influence score is used to quantify the degree of influence of the node on the occurrence of a vehicle collision.

[0023] The comparison module is used to compare the influence score of each node with the dynamic threshold, and generate a sparse attention weight distribution based on the comparison results. The dynamic threshold is determined based on the running status data and road environment data.

[0024] The risk prediction module is used to perform weighted fusion of the feature representations of each node based on a sparse attention weight distribution to obtain the prediction result of vehicle collision risk.

[0025] Thirdly, this application provides a controller, including: a processor, and a memory communicatively connected to the processor;

[0026] Memory is used to store instructions executed by the computer;

[0027] A processor for executing computer-executable instructions stored in memory to implement any of the methods of the first aspect.

[0028] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the method of any one of the first aspects.

[0029] Fifthly, this application provides a computer program product, including a computer program that, when executed, implements the method of any one of the first aspects.

[0030] The vehicle collision risk prediction method, device, and controller provided in this application acquire vehicle operating status data and road environment data; based on the operating status data and road environment data, a dynamic interactive network is constructed, wherein the dynamic interactive network uses the vehicle and other traffic participants as nodes, and the interaction relationships between the nodes as edges; feature representations of each node in the dynamic interactive network are extracted based on a graph neural network, and an influence score of each node is determined based on the feature representations of each node, which is used to quantify the degree of influence of the node on the occurrence of a vehicle collision; the influence score of each node is compared with a dynamic threshold, and a sparse attention weight distribution is generated based on the comparison results, wherein the dynamic threshold is determined based on the operating status data and road environment data; further, based on the sparse attention weight distribution, the feature representations of each node are weighted and fused to obtain the prediction result of vehicle collision risk.

[0031] In this process, by constructing a dynamic interactive network and introducing a hierarchical attention refinement mechanism, the computational overload and focus dispersion problems of traditional attention models in high-density traffic flow are effectively improved. This mechanism utilizes graph neural networks to extract deep feature representations of each node, then quantitatively evaluates the influence score of each node, and further achieves adaptive attention sparsity filtering based on dynamic thresholds and the influence scores of each node. This enables accurate differentiation between critical risk vehicles and secondary interference targets, significantly improving the real-time performance, accuracy, and robustness of risk prediction while ensuring real-time response. It not only ensures efficient processing within a limited computation cycle but also provides a highly reliable decision-making basis for vehicle collision avoidance control systems by reducing false alarm and false negative rates, thereby effectively ensuring the safety and reliability of vehicle collision avoidance systems in real-world complex traffic environments. Attached Figure Description

[0032] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0033] Figure 1 A flowchart illustrating a vehicle collision risk prediction method provided as an exemplary embodiment of this application;

[0034] Figure 2 Another flowchart illustrating the vehicle collision risk prediction method provided for an exemplary embodiment of this application;

[0035] Figure 3 A schematic diagram of a vehicle collision risk prediction device provided as an exemplary embodiment of this application;

[0036] Figure 4 A schematic diagram of the controller provided for an exemplary embodiment of this application.

[0037] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0038] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0039] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, products, or apparatus.

[0040] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0041] In related technologies, when using deep learning models based on attention mechanisms to predict vehicle collision risks, road environment data and information of the vehicle and surrounding vehicles are typically collected through onboard sensors and V2X technology. A deep neural network model based on attention mechanisms is then used for risk prediction, generating control commands to drive vehicle actuators to achieve secondary collision avoidance. During their research, the inventors discovered that when applied to ultra-high-density traffic flows, such as in congested highways or urban expressway ramp merging areas, this attention mechanism experiences "attention overload" due to the surge in the number of interacting vehicles being processed simultaneously, the highly complex and non-linear interaction relationships. Specifically, the model's attention resources are overwhelmed by a large number of concurrent interaction requests, leading to an exponential increase in its computational load. The attention weight allocation module struggles to accurately distinguish between "critical risk vehicles" posing a direct threat and "minor vehicles" merely providing background interference from dozens or even hundreds of interacting vehicles within a limited real-time computation cycle. This results in a series of serious consequences, including: 1) Directly causing response delays in the risk prediction module, preventing it from issuing warnings within the critical time window before a collision occurs, thus missing the optimal opportunity for evasive maneuvers. 2) Potentially causing distortion in prediction results, manifesting in two extremes: one is overreaction to minor vehicles, generating numerous false alarms and frequently triggering unnecessary emergency braking or steering, severely impacting driving comfort and traffic flow efficiency, and even creating new rear-end collision risks; the other is underreporting of critical risk vehicles, ignoring the truly dangerous collision threat, leading to complete failure of evasive control. This uncertainty and unreliability in prediction severely restricts the deployment and application of this solution in real-world, complex scenarios.

[0042] To address the aforementioned issues, this application provides a vehicle collision risk prediction scheme. By introducing a hierarchical attention refinement mechanism, it extracts the deep feature representation of each node through a graph neural network, thereby quantifying and evaluating the influence score of each node. An adaptive dynamic threshold is used to sparsify and filter global attention, assigning significant attention weights only to key risk vehicle nodes with influence scores higher than this dynamic threshold. This effectively solves the prediction delay or false alarm problems caused by computational resource overload and focus dispersion in traditional attention mechanisms in high-density traffic flows. This achieves accurate and real-time identification of core collision risk sources in extremely complex scenarios, significantly improving the real-time performance, accuracy, and robustness of risk prediction. It not only ensures efficient processing within a limited computation cycle but also provides a highly reliable decision-making basis for the vehicle collision avoidance control system by reducing false alarm and false negative rates, thus effectively ensuring the safety and reliability of the vehicle collision avoidance system in real-world complex traffic environments.

[0043] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0044] Figure 1 A flowchart illustrating a vehicle collision risk prediction method provided as an exemplary embodiment of this application is shown. Figure 1 As shown, the vehicle collision risk prediction method includes the following steps:

[0045] S101. Obtain the vehicle's operating status data and road environment data.

[0046] For example, the vehicle's operating status data is obtained through the vehicle's Controller Area Network (CAN) bus. At the same time, road environment data is obtained through onboard sensors such as millimeter-wave radar and lidar, as well as V2X communication modules. This road environment data includes real-time motion information of other traffic participants around the vehicle (such as vehicles, pedestrians and non-motorized vehicles), as well as static environment information. These data together constitute the raw input of the system's perception layer.

[0047] S102. Based on operational status data and road environment data, construct a dynamic interactive network; wherein, the dynamic interactive network uses the vehicle itself and other traffic participants as nodes, and the interaction relationships between the nodes as edges.

[0048] This dynamic interaction network is a mathematical model of a traffic scenario constructed in the form of a dynamic graph. It is used to abstract traffic participants and their interactions as a topological structure of nodes and edges. For example, the vehicle itself and each perceived traffic participant (such as vehicles ahead or cutting in from the side) are modeled as nodes in this dynamic interaction network. The distance and relative speed between any two nodes are calculated. If they meet preset interaction conditions (such as a distance of less than 100 meters), an edge is established between the two nodes. The weight of this edge (i.e., the strength of the interaction) can be assigned a value through an initial interaction strength function; for example, the closer the distance and the greater the speed difference, the higher the weight of the edge. This dynamic interaction network is dynamic; as the vehicle moves, the node attributes and edge relationships are dynamically updated in each calculation cycle.

[0049] S103. Extract feature representations of each node in a dynamic interactive network based on graph neural networks.

[0050] Graph neural networks (GNNs) are deep learning models specifically designed for processing graph-structured data. Feature representation refers to the high-dimensional feature vector obtained by each node after processing by the GNN, containing its own state and information about its neighboring nodes. Accordingly, the constructed dynamic interaction network is input into a pre-trained GNN, which operates through an iterative message passing mechanism. Specifically, each node (vehicle) receives feature information from its neighboring nodes (vehicles connected by edges) and aggregates this information with its own features to update its feature representation. After multiple iterations, each node ultimately obtains a deep feature vector that reflects its contextual information throughout the entire interaction network.

[0051] S104. Based on the feature representation of each node, determine the influence score of each node. The influence score is used to quantify the degree of influence of the node on the collision of vehicles.

[0052] The influence score is a scalar value used to comprehensively quantify the contribution or threat level of a vehicle node to the overall collision risk scenario. Accordingly, for each node in the dynamic interaction network, its influence score is determined based on its feature representation obtained in step S103.

[0053] In some embodiments, the influence score of each node is determined based on the feature representation of each node, including: using the norm of the feature representation corresponding to the node as the influence score of the node.

[0054] For example, the influence score of the i-th node Satisfy the following formula:

[0055]

[0056] in, This represents the feature of the node. The norm of the feature representation corresponding to the node is used to quantify the urgency of the risk of the node causing a direct collision.

[0057] In some embodiments, the influence score of each node is determined based on the feature representation of each node, including: multiplying the norm of the feature representation corresponding to the node by the topological centrality metric of the node in the dynamic interactive network, and using the result of the multiplication as the influence score of the node, wherein the topological centrality metric is used to quantify the criticality of the node in the propagation of collisions in the dynamic interactive network.

[0058] For example, the influence score of the i-th node Satisfy the following formula:

[0059]

[0060] in, This is a measure of the topological centrality of the node in a dynamic interactive network, used to quantify the criticality of the node in propagating collisions within the network.

[0061] It should be noted that the above topological centrality measures It is determined based on the topological connection structure of the real-time dynamic interactive network constructed in step S102 above. It is a quantitative description of the criticality of the i-th node in the dynamic graph corresponding to the dynamic interactive network. Its specific value can be obtained by calculating the degree centrality, betweenness centrality or eigenvector centrality of the node in the dynamic graph, etc., so as to identify the key vehicle nodes that may play a hub or bridge role in the process of collision risk propagation from the network structure level.

[0062] S105. Compare the influence scores of each node with the dynamic threshold, and generate a sparse attention weight distribution based on the comparison results. The dynamic threshold is determined based on the running status data and road environment data.

[0063] Among them, the dynamic threshold is a numerical threshold that is dynamically adjusted according to changes in the real-time traffic scene; the sparse attention weight distribution refers to the selection process that assigns significant weights to only a few key nodes, while suppressing the weights of most nodes to a distribution state close to zero.

[0064] For example, the influence scores of all nodes obtained in S104 are compared with the dynamic threshold. Comparisons are made. For example, dynamic thresholds are determined in real time based on the current vehicle density on the road and the vehicle's speed. Only nodes whose scores exceed this threshold (i.e., truly high-risk critical vehicles) are assigned higher attention weights, while the weights of the remaining nodes are significantly suppressed. This process, known as "sparsening," forces the system to concentrate its limited computational resources on the few most dangerous targets.

[0065] S106. Based on the sparse attention weight distribution, the feature representations of each node are weighted and fused to obtain the prediction results of vehicle collision risk.

[0066] For example, based on a sparse attention weight distribution, the feature representations of each node are weighted and summed, and the weighted scene features are input into a prediction layer (e.g., a fully connected neural network). After passing through the Softmax function, the risk probability value is output, which satisfies, for example, the following formula:

[0067]

[0068] in, This is the output layer weight matrix. This represents the sparse attention weight for that node. This represents the probability value of a collision. Furthermore, a feature alignment mechanism is used to identify a set of key risk vehicles, thus forming a complete risk prediction result, which in turn provides a basis for subsequent adaptive avoidance control.

[0069] The vehicle collision risk prediction method provided in this application acquires the vehicle's operating status data and road environment data; based on the operating status data and road environment data, a dynamic interaction network is constructed, wherein the dynamic interaction network uses the vehicle and other traffic participants as nodes, and the interaction relationships between nodes as edges; feature representations of each node in the dynamic interaction network are extracted based on a graph neural network, and an influence score of each node is determined based on the feature representations of each node, which is used to quantify the degree of influence of the node on the occurrence of a vehicle collision; the influence score of each node is compared with a dynamic threshold, and a sparse attention weight distribution is generated based on the comparison results, wherein the dynamic threshold is determined based on the operating status data and road environment data; further, based on the sparse attention weight distribution, the feature representations of each node are weighted and fused to obtain the prediction result of the vehicle collision risk. In this process, by constructing a dynamic interaction network and introducing a hierarchical attention refinement mechanism, the computational overload and focus dispersion problems of traditional attention models in high-density traffic flow are effectively improved. This mechanism utilizes graph neural networks to extract deep feature representations of each node, thereby quantifying and evaluating the impact score of each node. Furthermore, it achieves adaptive attention sparsity filtering based on dynamic thresholds and the impact scores of each node, thus accurately distinguishing between critical risk vehicles and secondary interference targets. While ensuring real-time response, it significantly improves the real-time performance, accuracy, and robustness of risk prediction. It not only ensures efficient processing within a limited computation cycle but also provides a highly reliable decision-making basis for vehicle collision avoidance control systems by reducing false alarm and false negative rates, thereby effectively ensuring the safety and reliability of vehicle collision avoidance systems in real-world complex traffic environments.

[0070] In some embodiments, the operational status data includes the vehicle speed, and the road environment data includes traffic flow density. The dynamic threshold is determined as follows: the traffic flow density and the vehicle speed are input into a preset linear relationship model to obtain a preliminary threshold; based on the prediction results of vehicle collision risk over a set historical period and actual traffic condition data, a consistency score and false alarm rate are determined to assess the accuracy of the prediction results for evaluating vehicle collision risk; a reward signal is determined based on the consistency score, false alarm rate, and a preset penalty coefficient; with the goal of maximizing the reward signal, the model parameters of the linear relationship model are iteratively updated using a policy gradient algorithm to optimize the preliminary threshold online and generate a dynamic threshold.

[0071] For example, operational status data includes vehicle speed, acceleration, steering angle, and yaw rate; road environment data includes the position, speed, acceleration, and heading angle of surrounding vehicles, as well as road curvature, road surface adhesion coefficient, and traffic flow density.

[0072] Optionally, after acquiring the vehicle's operating status data and road environment data, the process also includes acquiring a dataset containing the operating status data and road environment data. Preprocessing operations are performed, including but not limited to data cleaning, outlier removal, timestamp alignment, and coordinate system unification, to obtain a standardized and unified dataset. .

[0073] For example, first, the data set Data cleaning and outlier removal are performed. Specifically, outliers that do not conform to physical constraints can be identified and removed through statistical detection. At the same time, sensor noise is smoothed using filtering algorithms, such as weighted moving average, which specifically satisfies the following formula:

[0074]

[0075] in, The signal is smoothed. These are the original observations. is the weighting coefficient, and m is the sliding window radius. This process ensures the continuity of the signal and reduces the interference of high-frequency noise on subsequent modeling.

[0076] Accordingly, after data cleaning, timestamp alignment is performed to synchronize all cleaned data to a unified time base. This can be achieved by resampling data streams of different frequencies using interpolation, ensuring that at time t, vehicle status, environmental perception, and V2X communication information all correspond to a consistent data snapshot. After timestamp alignment, coordinate system one is executed to unify location data from different sources to the vehicle's own coordinate system. The conversion formula is, for example:

[0077]

[0078] in, To transform the position vector into the vehicle's coordinate system, The original global coordinates, This represents the vehicle's position in global coordinates. The heading angle of this vehicle The defined rotation matrix. This transformation mechanism, through spatial coordinate transformation and heading angle compensation, ensures that the position information of all traffic participants can be uniformly and consistently expressed in the vehicle coordinate system, providing an accurate spatial reference for subsequent interaction relationship modeling.

[0079] Furthermore, data fusion is performed to integrate observations from different sensors and V2X systems. By combining weighted averaging with Kalman filtering, the processing accuracy of multi-source heterogeneous data can be effectively improved. Weighted averaging is used for initial fusion of similar redundant information, while Kalman filtering utilizes the system state transition model and observation model to achieve recursive estimation. The corresponding formulas are as follows:

[0080]

[0081] in, This is the state estimate. For prior estimation, H is the observation vector, and H is the observation matrix; The Kalman gain is determined by the ratio of the covariance of system noise to observation noise. This process ensures that the fusion results are uniformly formatted and standardized, forming a dataset that is both consistent and complete. This provides a stable and reliable foundation of input data for subsequent risk prediction.

[0082] For example, from the data set Traffic flow density extracted With the speed of this vehicle The input is a dynamic threshold. The dynamic threshold This is used to filter influence scores in hierarchical sparse attention computation. The process first performs preliminary calculations based on a linear model. Satisfy the following formula:

[0083]

[0084] in, This is a traffic density weighting parameter used to reflect the sensitivity of traffic flow density to a threshold. 'b' is the vehicle speed weighting parameter, used to reflect the adjustment effect of the vehicle speed on the threshold; 'b' is the bias term, used to provide baseline correction when the sparsity is too low or too high. It is the ratio of the number of vehicles in a unit road section to the road capacity, describing the degree of congestion in a traffic scenario, also known as traffic flow density. The longitudinal velocity of the vehicle describes its motion in the current traffic environment. This linear relationship ensures that the threshold is increased in scenarios with high traffic density or high vehicle speed, making attention calculations more sparse-oriented, thus focusing on the most critical risk vehicles and mitigating the feature redundancy problem caused by increased scene complexity.

[0085] Furthermore, a reinforcement learning fine-tuner is introduced to achieve dynamic online adjustment of the threshold. The fine-tuner generates a reward signal based on the prediction accuracy over a recent period, such as the last 500 seconds. , Satisfy the following formula:

[0086]

[0087] Among them, the consistency score reflects the degree of consistency between the risk prediction results and the actual collision trend; the false alarm rate is the proportion of false high-risk alarms generated in the prediction. This is the penalty coefficient, used to balance the trade-off between accuracy and robustness in the model. The reinforcement learning fine-tuner updates parameters using the policy gradient algorithm. The adjustment range makes During operation, it can continuously adjust based on the environment and prediction performance. Among these adjustments, the policy gradient algorithm is a core method in reinforcement learning that directly optimizes the policy itself. It is defined as an algorithm that iteratively improves policy quality by calculating the gradient of the policy performance index (expected total reward) with respect to policy parameters and updating the policy parameters along the gradient direction. This mechanism, through iterative learning of the reward signal, gradually forms an adaptive threshold adjustment policy for different traffic densities and vehicle speeds, making attention allocation more efficient. It can suppress invalid interference nodes in high-density, high-speed scenarios and retain more potential risk information in low-density, low-speed scenarios, thereby improving prediction stability and reducing latency, further enhancing the real-time performance and reliability of risk prediction in complex and dynamic traffic environments.

[0088] In some embodiments, feature representations of each node in a dynamic interaction network are extracted based on a graph neural network, including: iterative message passing through multiple network layers of the graph neural network; for each node, the feature representation of the node at layer L+1 is updated in the following way: aggregating the node's own feature representation at layer L and the feature representations of each of the node's neighboring nodes at layer L; performing a linear transformation on the aggregated result and processing the linear transformation result with an activation function to generate the feature representation of the node at layer L+1; where L is the layer index.

[0089] For example, the first step is to execute the dynamic interactive network construction sub-step, specifically, at time t, the dynamic interactive network construction sub-step is executed. Represented as a dynamic image ,in, For each set of nodes, the feature vector of each node is... By speed acceleration and position vector Composition, edge set edge weight The interaction strength function is determined by the interaction strength function, which satisfies the following formula:

[0090]

[0091] in, and This represents the position vectors of node i and node j. and This represents the velocity vector between two nodes. To prevent a constant with a zero denominator, the function characterizes the interaction strength between nodes by measuring positional distance and velocity difference, making nodes that are close to each other and have large velocity differences, such as vehicles, contribute more significantly to potential risks.

[0092] Accordingly, the graph neural network feature extraction sub-step is executed to extract the dynamic graph. The input graph neural network updates node features through iterative message passing, specifically satisfying the following formula:

[0093]

[0094] in, Let i be the set of neighbors of node i. and For trainable weight matrix, For activation function, Let L be the attention weights corresponding to the Lth layer. This refers to the feature representation of a node at layer L+1. This step learns a high-dimensional representation of a node in complex interaction scenarios by aggregating neighbor features and self-features, thereby preserving the dynamic dependencies between vehicles.

[0095] In this embodiment, the multi-layer iterative message passing mechanism of a graph neural network effectively enhances the representation capability of vehicle interaction features. This structure aggregates the features of a node itself and the features of its neighboring nodes, enabling each node to integrate local topological relationship information. Through linear transformation and activation function processing, a high-dimensional feature representation containing multi-layered interaction context is gradually constructed. This hierarchical feature learning mechanism allows the model to capture the complex spatiotemporal dependencies in traffic scenarios, laying a solid feature foundation for subsequent accurate evaluation of node influence scores, thereby significantly enhancing the contextual awareness capability of risk prediction.

[0096] In some embodiments, the vehicle collision risk prediction method further includes: generating a vehicle evasion control command based on the predicted vehicle collision risk, and sending the vehicle evasion control command to the vehicle's actuator; collecting the vehicle's state feedback data in real time, and determining the execution deviation between the state feedback data and the expected state feedback data corresponding to the vehicle evasion control command; if the execution deviation exceeds a preset execution deviation threshold, triggering a replanning of the vehicle evasion control command; and simultaneously, using the samples containing state feedback data and execution deviation as incremental data, and employing an incremental learning algorithm to update the model parameters of the graph neural network online to optimize the accuracy of subsequent vehicle collision risk prediction.

[0097] For example, when the predicted collision risk includes a "high risk" prediction, a specific vehicle evasive control command is generated based on this "high risk" prediction, such as "apply 40% braking force." This command is sent in real-time via the vehicle's CAN bus to the vehicle's Electronic Stability Program (ESP) or Electric Power Steering (EPS) actuators. Simultaneously with the command issuance, onboard sensors such as wheel speed sensors or inertial measurement units (IMUs) collect real-time vehicle status feedback data, such as the vehicle's actual deceleration. Furthermore, the actual deceleration is compared with the expected deceleration (calculated from the "40% braking force" command) to calculate the execution deviation (e.g., the actual deceleration is only 80% of the expected deceleration).

[0098] Accordingly, based on a preset execution deviation threshold, such as 15% of the expected deceleration; if the execution deviation exceeds the execution deviation threshold, it indicates that the current control command has failed to achieve the expected effect (possibly due to slippery road surface, insufficient tire grip, etc.). The control system will immediately trigger replanning, for example, by using the current sensor feedback and deviation information as new inputs to quickly solve a new and more suitable vehicle evasion control command (for example, increasing the braking force to 55%) to replace the original vehicle evasion control command and ensure immediate safety.

[0099] In addition, while monitoring and controlling the execution, the status feedback data (actual deceleration) and execution deviation information are also bound to the risk prediction scenario data that led to this control (such as the dynamic graph corresponding to the dynamic interaction network at that time) to form an incremental data sample. Furthermore, an incremental learning algorithm, such as an algorithm based on online gradient descent, is used to fine-tune the model parameters of the graph neural network using this sample, so as to realize the online update of the model parameters of the graph neural network, thereby optimizing the accuracy of subsequent vehicle collision risk prediction.

[0100] This application's embodiments effectively improve the system's real-time adaptability and long-term reliability by establishing a closed-loop feedback mechanism from decision-making to execution. By comparing the expected effect of the vehicle evasion control command with the actual vehicle state's execution deviation, command replanning can be triggered immediately upon detecting a risk of control failure, ensuring the timely and effective execution of evasion actions. Simultaneously, the execution deviation data is used as incremental learning samples to optimize the graph neural network online, ensuring continuous learning and self-correction from actual interactions. This dual optimization mechanism not only guarantees the accuracy of individual control responses but also improves prediction accuracy and generalization ability in complex traffic scenarios through continuous accumulation of practical experience.

[0101] In some embodiments, the predicted vehicle collision risk includes a vehicle collision probability value, a predicted risk time interval, a predicted risk spatial coordinate range, and a list of risk vehicle identifiers. Based on the predicted vehicle collision risk, a vehicle evasion control command is generated, including: identifying target threat vehicles based on the risk vehicle identifier list, and determining the minimum safe distance required for the vehicle to achieve full braking within the predicted risk time interval based on the vehicle speed, the predicted risk time interval, and the vehicle's maximum deceleration, and using the minimum safe distance as the safe space boundary; constructing an optimization objective function with the safe space boundary as a hard constraint, the optimization objective function being configured to minimize the amplitude of the control command and optimize the ride comfort index within a future predicted time domain; solving the optimization objective function using a model predictive control algorithm to obtain an optimal control sequence including braking, steering, and power control quantities; and outputting the control quantity corresponding to the current moment in the optimal control sequence as the vehicle evasion control command.

[0102] For example, consider the prediction results of vehicle collision risk. For input, the result contains risk probability values. Risk time window (i.e., the predicted risk time interval) Potential collision zone (i.e., the predicted risk spatial coordinate range) And key risk vehicle identifiers (i.e., a list of risk vehicle identifiers). Output the optimal control command sequence Among them, real-time control quantity It was directly issued to the vehicle execution agency. Specifically, through analysis... and To identify the most threatening vehicles and their duration of operation, and to calculate the safety space boundary in conjunction with the vehicle's dynamic constraints. The safety boundary is defined as a minimum safety distance function, which specifically satisfies the following formula:

[0103]

[0104] in, For the vehicle's speed, This represents the maximum deceleration that the vehicle can achieve under the current road surface adhesion conditions. This is a risk window. This function represents the range of physical safety that a vehicle can achieve within a limited time, representing the minimum longitudinal clearance required to avoid a collision at a given speed and braking capacity.

[0105] Accordingly, based on the above risk situation analysis, the next step is to generate a predictive control strategy. The objective function is constructed as a constraint condition, and the objective function satisfies the following formula:

[0106]

[0107] in, This represents the control input vector at the k-th step in the prediction time domain, which includes the control quantities for braking, steering, and powertrain. This item is used to limit the amplitude of control commands to prevent overly aggressive operations; Let be the comfort index function, used to describe the impact of acceleration rate of change and lateral acceleration on passenger comfort; w is the comfort weighting coefficient, used to balance safety and comfort; M is the prediction time domain length, used to determine the time range of the controller's forward-looking capability. By solving this constrained optimization objective function, the optimal control sequence is obtained. , Satisfy the following formula:

[0108]

[0109] in, This is the optimal control input for the current moment. Further, the command execution and feedback sub-steps are performed, which involves the immediate control input. The data is transmitted to the braking, steering, and powertrain systems via the vehicle's CAN bus to ensure rapid response to risks under physical constraints and safety boundaries; simultaneously, the vehicle's operating status is collected in real time and fed back to the updated data set in step S101. This creates a closed loop between vehicle collision risk prediction and vehicle avoidance control. This mechanism ensures that the system can adjust its control strategy according to the risk situation in dynamic traffic scenarios, effectively avoiding secondary collisions while maintaining an optimal balance between safety and comfort, thereby improving overall driving safety and robustness.

[0110] In some embodiments, in the step of constructing a dynamic interaction network, the weights of each edge in the dynamic interaction network are determined in the following way: for any two nodes, the interaction strength is determined based on the position vectors and velocity vectors of the two nodes; the relative acceleration magnitude is determined based on the acceleration vectors of the two nodes; a scaling factor is determined based on the relative acceleration magnitude; the interaction strength is multiplied by the scaling factor, and the result of the multiplication is used as the edge weight of the two nodes; wherein, the scaling factor is positively correlated with the relative acceleration magnitude, and when the relative acceleration magnitude increases, it amplifies the interaction strength.

[0111] For example, for any two nodes, such as node i and node j, the feature representations corresponding to node i and node j are used respectively. and The input consists of features such as velocity, acceleration, and position; the output is the enhanced interaction strength. Used to update animated graphs The edge weights. The processing is achieved by extending the original interaction strength function, resulting in enhanced interaction strength. , that is Satisfy the following formula:

[0112]

[0113] in, The original interaction strength function satisfies the following formula:

[0114]

[0115] in, and This represents the position vectors of node i and node j. and This represents the velocity vector between two nodes. To prevent a constant with a zero denominator, the function characterizes the interaction strength between nodes using positional distance and velocity difference, making nodes that are close together and have large velocity differences, such as vehicles, contribute more significantly to potential risks. The extended function then introduces an acceleration difference factor, where... and These represent the acceleration vectors of node i and node j, respectively, from the dataset. Extracted from; The modulus represents the relative acceleration, used to quantify the difference between two vehicles in acceleration or deceleration; The coefficient for adjusting the acceleration difference weight is larger, and the amplification effect of the acceleration difference on the interaction intensity is more significant.

[0116] In this embodiment, the enhanced computing mechanism considers the potential risks brought about by acceleration differences in addition to vehicle position and speed characteristics. When the acceleration changes of the two vehicles are in opposite directions or the difference is large, the interaction intensity is significantly improved, thereby highlighting the impact of emergency dynamic changes on collision risk. This enables the edge weights of the dynamic interaction network to more sensitively reflect the risk propagation characteristics of sudden events in traffic flow, allowing the subsequent graph neural network to pay more attention to high-risk interaction relationships when performing feature aggregation, further improving the response speed and accuracy of vehicle collision risk prediction in complex traffic environments.

[0117] It should be noted that the series of steps described above, from constructing the dynamic interactive network to outputting risk prediction results, can also be considered a tightly connected and inseparable organic whole, collectively constituting the complete execution flow of the vehicle collision risk prediction model. This model, through the synergistic effects of dynamic graph construction, graph neural network feature extraction, influence score calculation, dynamic threshold filtering, and sparse attention-weighted fusion, forms an end-to-end computational framework from multi-source data to risk prediction. This integrated architecture not only ensures data consistency among the various modules of the algorithm, but more importantly, it achieves accurate perception and efficient computation of high-density traffic scenarios through a hierarchical attention refinement mechanism, ultimately enabling the model to simultaneously meet the requirements of real-time performance, accuracy, and robustness.

[0118] Accordingly, the training process of this vehicle collision risk prediction model uses the input dataset. Starting with historical traffic collision scenarios and multidimensional data streams generated from virtual simulations, this dataset contains vehicle operating state vectors. Environmental perception vectors and V2X interaction features The aforementioned preprocessing operations transform the data into a standardized feature set with a unified time reference and coordinate system. This vehicle collision risk prediction model consists of an input layer, multiple hidden layers, and an output layer. The hidden layers contain graph neural network layers to capture the topological relationships between the vehicle and its environment. Nodes represent vehicle state features, and edges represent interactions between vehicles. Node representations are updated through a message passing mechanism, specifically satisfying the following formula:

[0119]

[0120] in, The feature representation of a node at layer L. That is, the feature representation of the node at the (L+1)th layer. Let i represent the set of neighboring nodes connected to node i. Let L be the weight matrix of the Lth layer. For bias terms, This is a non-linear activation function. Accordingly, after feature extraction is completed in the graph neural network, the node features are input to the attention mechanism layer. This layer calculates the attention weights between nodes using a learnable scoring function, and the attention weights satisfy the following formula:

[0121]

[0122] in, Let be the attention weights between node i and its neighbor node j, 'a' be a learnable parameter vector, and '||' denote the vector concatenation operation. The output layer maps to two types of tasks: collision probability prediction and future risk level numerical regression. The loss function is... This is a weighted combination of cross-entropy loss for classification tasks and mean squared error for regression tasks:

[0123]

[0124] in, , Indicates the true collision label of sample i. The probability predicted by the model. This represents the actual risk level. The predicted value is N, where N is the total number of training samples. and Here, represents the balancing coefficients. The optimizer uses the Adam algorithm, iteratively updating parameters based on gradient information. This update rule satisfies the following formula:

[0125]

[0126] in, Let be the model parameters for the t-th iteration. For learning rate, and These are the bias correction values ​​for the first-order and second-order moment estimates, respectively. To prevent constants with zero denominators, an early stopping strategy is introduced during training. Specifically, this involves monitoring changes in the loss on the validation set and terminating training when no improvement is achieved after several consecutive epochs. This avoids overfitting and ultimately yields a convergent weight matrix and set of bias parameters. This ensures good generalization ability in diverse traffic environments, providing reliable risk prediction input for subsequent adaptive evasion control.

[0127] Accordingly, the execution and monitoring process of vehicle evasion control commands uses the optimal control command sequence. As input, this sequence contains braking control quantities. Steering control quantity With throttle control amount The controller's internal encoding module will then... The commands are converted into low-level instructions compatible with the actuator interface and transmitted in real time to the braking system, steering actuators, and powertrain via the CAN bus, ensuring minimal command transmission latency and jitter. Furthermore, the sensor system monitors the execution process in real time, acquiring vehicle state feedback vectors. , The execution deviation is then compared with the expected command on a time-by-time basis. The execution deviation satisfies the following formula:

[0128]

[0129] in, The execution deviation at time t, The actual output control quantity of the actuator. Let be the target control variable at that moment, and the norm of the vector difference reflects the control accuracy. If Exceeding the execution deviation threshold If this happens, a dynamic adjustment mechanism is immediately triggered, which means that the vehicle evasion control command is replanned. Specifically, the current sensor feedback and deviation information are re-sent to the relevant calculation module to obtain a new vehicle evasion control command. This replaces the original vehicle evasion control commands, thereby maintaining execution stability.

[0130] Accordingly, all monitoring results provide feedback on the actual execution effectiveness. The data will be returned to the data acquisition and fusion stage as real-time closed-loop feedback, continuously correcting the accuracy of the input data and improving the reliability of subsequent prediction and control, thereby forming a closed-loop link of perception, prediction and execution, enabling the entire system to maintain high precision and high robustness when facing complex traffic environments.

[0131] For example, the system's closed-loop feedback mechanism provides feedback on the execution results. With the collected real-time data The input is the model parameter of the updated vehicle collision risk prediction model, and the output is the model parameter of the updated model. The optimized control strategy parameters, including the model predictive control (MPC) objective function weights w, are then used. This process first performs performance analysis on the feedback data, specifically by calculating the execution deviation. By analyzing deviations from vehicle status indicators, nodes exhibiting decreased control precision, increased prediction errors, or abnormal behavior are identified, thus pinpointing system performance bottlenecks. Subsequently, an incremental learning algorithm is applied to update the vehicle collision risk prediction model online, incorporating newly acquired real-time data. Feedback information Fusion, used to adjust the weight matrices of the graph neural network and the attention layer. and bias terms The updated formula is:

[0132]

[0133] in, For the current model parameter set, For learning rate, The incremental learning loss function combines cross-entropy and mean squared error to measure the deviation of new data from the model's predictions, ensuring that the model can quickly adjust after receiving new samples and adapt to the dynamic traffic environment.

[0134] Simultaneously, the control strategy parameters are optimized based on the feedback results, specifically the comfort weights in the MPC objective function J. Perform adaptive adjustments:

[0135]

[0136] Among them, by implementing deviations Passenger comfort index In combination, an incremental learning algorithm is used to solve for the new weights w', so that the control sequence is within the safety constraints. The closed-loop mechanism not only utilizes execution feedback to achieve real-time control correction, but also drives the online evolution of vehicle collision risk prediction models and control strategies, endowing the system with continuous self-learning and improvement capabilities. This ensures high performance levels in prediction accuracy, response speed, and robustness in long-term, complex, and ever-changing traffic environments.

[0137] In addition, the performance monitoring and optimization process takes system operation logs and real-time performance metrics as input, including risk prediction accuracy. With control response time The output is the optimized system configuration parameters, including dynamic thresholds. coefficient With bias The processing begins by collecting and analyzing data from the runtime logs to obtain key performance indicators. The average prediction error satisfies the following formula:

[0138]

[0139] in, Let i be the true risk probability of sample i. The model predicts the value, where N is the total number of samples, and the control response delay is also calculated. This refers to the time required for a command to be issued to the vehicle actuator and for it to reach the desired state, used to measure the system's real-time performance. These performance metrics are then input into a multi-objective optimization algorithm, and the dynamic threshold formula is adjusted accordingly. In The b parameter, along with other parameters, allows for a reduction in response latency while maintaining high prediction accuracy, thus balancing the real-time performance and precision of prediction and control.

[0140]

[0141] This optimization algorithm determines the optimal parameter combination by iteratively searching and weighing different indicators in a weighted objective function. ), and combine the optimal parameters. Update the system configuration. Furthermore, continuously monitor hardware performance metrics and software health, and automatically trigger maintenance operations (such as sensor recalibration and cache data cleanup) when performance deviations are detected. Automated tools complete closed-loop optimization from data acquisition and performance metric calculation to model parameters, control strategies, and dynamic thresholds, forming a complete self-maintenance system. This automated operation and maintenance mechanism significantly reduces the need for manual intervention while ensuring long-term system stability, fundamentally improving the overall system performance and operational reliability.

[0142] Figure 2 Another flowchart illustrating the vehicle collision risk prediction method provided for an exemplary embodiment of this application is shown. Figure 2 As shown, the vehicle collision risk prediction method includes the following steps:

[0143] S201. Obtain the vehicle's operating status data and road environment data.

[0144] For example, operational status data includes the vehicle's speed, acceleration, steering angle, and yaw rate; road environment data includes the position, speed, acceleration, and heading angle of other traffic participants in the surrounding area, as well as road curvature, road surface adhesion coefficient, and traffic flow density.

[0145] Optionally, after acquiring the vehicle's operating status data and road environment data, the process also includes acquiring a dataset containing the operating status data and road environment data. Preprocessing operations are performed, including but not limited to data cleaning, outlier removal, timestamp alignment, and coordinate system unification, to obtain a standardized and integrated dataset. .

[0146] S202. Based on operational status data and road environment data, a dynamic interactive network is constructed.

[0147] Among them, the dynamic interactive network is a mathematical model of traffic scenarios constructed in the form of a dynamic graph, with the vehicle itself and other traffic participants as nodes, and the interaction relationships between the nodes as edges.

[0148] S203. Extract feature representations of each node in a dynamic interactive network based on graph neural networks.

[0149] In some embodiments, feature representations of each node in a dynamic interaction network are extracted based on a graph neural network, including: iterative message passing through multiple network layers of the graph neural network; for each node, the feature representation of the node at layer L+1 is updated in the following way: aggregating the node's own feature representation at layer L and the feature representations of each of the node's neighboring nodes at layer L; performing a linear transformation on the aggregated result and processing the linear transformation result with an activation function to generate the feature representation of the node at layer L+1; where L is the layer index.

[0150] S204. Based on the feature representation of each node, determine the influence score of each node. The influence score is used to quantify the degree of influence of the node on the occurrence of a vehicle collision.

[0151] Specifically, the norm of the feature representation corresponding to the node is multiplied by the topological centrality metric of the node in the dynamic interactive network, and the result of the multiplication is used as the influence score of the node. The norm of the feature representation corresponding to the node is used to quantify the urgency of the risk of the node causing a direct collision, and the topological centrality metric is used to quantify the criticality of the node in the propagation of collisions in the dynamic interactive network.

[0152] S205. Compare the influence scores of each node with the dynamic threshold, and generate a sparse attention weight distribution based on the comparison results. The dynamic threshold is determined based on the running status data and road environment data.

[0153] S206. Based on the sparse attention weight distribution, the feature representations of each node are weighted and fused to obtain the prediction results of vehicle collision risk.

[0154] For example, based on a sparse attention weight distribution, the feature representations of each node are weighted and summed, and the weighted scene features are input into a prediction layer (e.g., a fully connected neural network). After passing through the Softmax function, the risk probability value is output, which satisfies, for example, the following formula:

[0155]

[0156] in, This is the output layer weight matrix. This represents the sparse attention weight for that node. This represents the probability value of a collision. Furthermore, a feature alignment mechanism is used to identify a set of key risk vehicles, thus forming a complete risk prediction result, which in turn provides a basis for subsequent adaptive avoidance control.

[0157] S207. Based on the prediction results of vehicle collision risk, generate vehicle avoidance control commands and send the vehicle avoidance control commands to the actuators of the vehicle.

[0158] Specifically, target threat vehicles are identified based on a list of risk vehicle identifiers. The minimum safe distance required for the vehicle to achieve full braking within the predicted risk time interval is determined based on the vehicle's speed, the predicted risk time interval, and the vehicle's maximum deceleration. This minimum safe distance is used as the safety space boundary. An optimization objective function is constructed using this safety space boundary as a hard constraint. This objective function is configured to minimize the amplitude of the control command and optimize ride comfort indices within a future predicted time domain. The optimization objective function is solved iteratively using a model predictive control algorithm to obtain the optimal control sequence, which includes braking, steering, and power control quantities. The control quantity corresponding to the current moment in the optimal control sequence is output as the vehicle evasion control command.

[0159] S208. Collect the vehicle's status feedback data in real time and determine the execution deviation between the status feedback data and the expected status feedback data corresponding to the vehicle avoidance control command.

[0160] S209. Determine whether the execution deviation exceeds the preset execution deviation threshold.

[0161] If so, execute S210;

[0162] If not, proceed with S201.

[0163] S210: Trigger the replanning of vehicle avoidance control commands. At the same time, use samples containing state feedback data and execution deviations as incremental data, and use an incremental learning algorithm to update the model parameters of the graph neural network online to optimize the accuracy of subsequent vehicle collision risk prediction.

[0164] In summary, this application has at least the following advantages:

[0165] I. By constructing a dynamic interactive network and introducing a hierarchical attention refinement mechanism, the computational overload and focus dispersion problems of traditional attention models in high-density traffic flow are effectively improved. This mechanism utilizes graph neural networks to extract deep feature representations of each node, then quantifies and evaluates the influence score of each node. Furthermore, based on dynamic thresholds and the influence scores of each node, adaptive attention sparsity filtering is achieved, thereby accurately distinguishing between critical risk vehicles and secondary interference targets. While ensuring real-time response, it significantly improves the real-time performance, accuracy, and robustness of risk prediction. It not only ensures efficient processing within a limited computation cycle but also provides a highly reliable decision-making basis for vehicle collision avoidance control systems by reducing false alarm and false negative rates, thus effectively ensuring the safety and reliability of vehicle collision avoidance systems in real-world complex traffic environments.

[0166] Second, the multi-layer iterative message passing mechanism of graph neural networks effectively enhances the representation capability of vehicle interaction features. This structure aggregates the features of a node itself and the features of its neighboring nodes, enabling each node to integrate local topological relationship information. Then, through linear transformation and activation function processing, a high-dimensional feature representation containing multi-layered interaction context is gradually constructed. This hierarchical feature learning mechanism allows the model to capture the complex spatiotemporal dependencies in traffic scenarios, laying a solid feature foundation for subsequent accurate assessment of node influence scores, thereby significantly enhancing the context-awareness capability of risk prediction.

[0167] Third, by establishing a closed-loop feedback mechanism from decision-making to execution, the system's real-time adaptability and long-term reliability are effectively improved. By comparing the expected effect of vehicle evasion control commands with the actual vehicle state's execution deviation, command replanning can be triggered immediately upon detecting control failure risk, ensuring timely and effective execution of evasion actions. Simultaneously, execution deviation data is used as incremental learning samples to optimize the graph neural network online, ensuring continuous learning and self-correction from actual interactions. This dual optimization mechanism not only guarantees the accuracy of individual control responses but also improves prediction accuracy and generalization ability in complex traffic scenarios through continuous accumulation of practical experience.

[0168] Fourth, by enhancing the computational mechanism, the potential risks brought about by acceleration differences are considered in addition to vehicle position and speed characteristics. When the acceleration changes of the two vehicles are in opposite directions or the difference is large, the interaction intensity is significantly enhanced, thereby highlighting the impact of emergency dynamic changes on collision risk. This enables the edge weights of the dynamic interaction network to more sensitively reflect the risk propagation characteristics of sudden events in traffic flow, allowing the subsequent graph neural network to pay more attention to high-risk interaction relationships when performing feature aggregation, further improving the response speed and accuracy of vehicle collision risk prediction in complex traffic environments.

[0169] Fifth, compared with existing technologies, this technology effectively solves the problem of prediction delay or false alarm caused by the overload of computing resources and the dispersion of focus in traditional attention mechanisms in high-density traffic flow. It achieves the technical effect of accurately and in real time identifying the core collision risk source in extremely complex scenarios, enabling the system to dynamically balance the breadth and depth of attention according to specific scenarios, and realize the adaptive improvement of the robustness and sensitivity of the vehicle collision risk prediction model under different traffic loads. It effectively breaks through the limitations of traditional open-loop or fixed-parameter systems, enabling the entire system to continuously learn and evolve from actual interactions, thereby achieving stable optimization of long-term performance and a high degree of adaptability to changes in vehicle status and environment. In addition, it systematically improves the technical challenges of the entire chain from perception and cognition to decision execution, and ultimately achieves the core technical effect of significantly reducing the incidence of collisions and secondary accidents in highly dynamic and uncertain real traffic environments.

[0170] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0171] Figure 3 A schematic diagram of a vehicle collision risk prediction device provided as an exemplary embodiment of this application. Figure 3 As shown, the vehicle collision risk prediction device 30 includes an acquisition module 31, a dynamic interactive network construction module 32, a feature extraction module 33, a determination module 34, a comparison module 35, and a risk prediction module 36, wherein:

[0172] The acquisition module 31 is used to acquire the vehicle's operating status data and road environment data;

[0173] The dynamic interactive network construction module 32 is used to construct a dynamic interactive network based on the operating status data and road environment data; wherein, the dynamic interactive network uses the vehicle and other traffic participants as nodes, and the interaction relationship between each node as edges;

[0174] Feature extraction module 33 is used to extract feature representations of each node in a dynamic interactive network based on graph neural networks;

[0175] The determination module 34 is used to determine the influence score of each node based on the feature representation of each node. The influence score is used to quantify the degree of influence of the node on the vehicle collision.

[0176] The comparison module 35 is used to compare the influence score of each node with the dynamic threshold, and generate a sparse attention weight distribution based on the comparison results. The dynamic threshold is determined based on the running status data and road environment data.

[0177] The risk prediction module 36 is used to perform weighted fusion of the feature representations of each node based on the sparse attention weight distribution to obtain the prediction result of vehicle collision risk.

[0178] In one possible implementation, the feature extraction module 33 can be specifically used to: perform iterative message passing on the dynamic interaction network through multiple network layers of the graph neural network; for each node, update the feature representation of the node at layer L+1 in the following way: aggregate the node's own feature representation at layer L and the feature representations of each of the node's neighboring nodes at layer L; perform a linear transformation on the aggregated result, and process the linear transformation result with an activation function to generate the feature representation of the node at layer L+1; where L is the layer index.

[0179] In one possible implementation, the determining module 34 may be specifically used to: multiply the norm of the feature representation corresponding to the node by the topological centrality metric of the node in the dynamic interactive network, and use the result of the multiplication as the influence score of the node, wherein the norm of the feature representation corresponding to the node is used to quantify the urgency of the risk of the node causing a direct collision, and the topological centrality metric is used to quantify the criticality of the node in propagating collisions in the dynamic interactive network.

[0180] In one possible implementation, the operational status data includes the vehicle speed, and the road environment data includes traffic flow density. The dynamic threshold is determined as follows: the traffic flow density and the vehicle speed are input into a preset linear relationship model to obtain a preliminary threshold; based on the prediction results of vehicle collision risk over a set historical period and actual traffic condition data, a consistency score and false alarm rate are determined to assess the accuracy of the prediction results for evaluating vehicle collision risk; a reward signal is determined based on the consistency score, false alarm rate, and a preset penalty coefficient; with the goal of maximizing the reward signal, a policy gradient algorithm is used to iteratively update the model parameters of the linear relationship model to optimize the preliminary threshold online and generate a dynamic threshold.

[0181] In one possible implementation, the risk prediction module 36 can be specifically used to: generate vehicle evasion control commands based on the prediction results of vehicle collision risks, and send the vehicle evasion control commands to the actuators of the vehicle; collect the status feedback data of the vehicle in real time, and determine the execution deviation between the status feedback data and the expected status feedback data corresponding to the vehicle evasion control commands; if the execution deviation exceeds a preset execution deviation threshold, trigger the replanning of the vehicle evasion control commands; at the same time, use the samples containing status feedback data and execution deviations as incremental data, and use an incremental learning algorithm to update the model parameters of the graph neural network online to optimize the accuracy of subsequent vehicle collision risk prediction.

[0182] In one possible implementation, the predicted vehicle collision risk includes a vehicle collision probability value, a predicted risk time interval, a predicted risk spatial coordinate range, and a list of risk vehicle identifiers. The risk prediction module 36 can also be used to: determine target threat vehicles based on the risk vehicle identifier list, and determine the minimum safe distance required for the vehicle to achieve full braking within the predicted risk time interval based on the vehicle speed, the predicted risk time interval, and the vehicle's maximum deceleration, and use the minimum safe distance as the safe space boundary; construct an optimization objective function with the safe space boundary as a hard constraint, the optimization objective function being configured to minimize the amplitude of the control command and optimize the ride comfort index within a future predicted time domain; solve the optimization objective function in a rolling manner through a model predictive control algorithm to obtain the optimal control sequence including braking, steering, and power control quantities; and output the control quantity corresponding to the current moment in the optimal control sequence as the vehicle avoidance control command.

[0183] In one possible implementation, in the step of constructing the dynamic interaction network, the weights of each edge in the dynamic interaction network are determined as follows: for any two nodes, the interaction strength is determined based on the position vectors and velocity vectors of the two nodes; the relative acceleration magnitude is determined based on the acceleration vectors of the two nodes; a scaling factor is determined based on the relative acceleration magnitude; the interaction strength is multiplied by the scaling factor, and the result of the multiplication is used as the edge weight of the two nodes; wherein, the scaling factor is positively correlated with the relative acceleration magnitude, and when the relative acceleration magnitude increases, it amplifies the interaction strength.

[0184] The vehicle collision risk prediction device provided in this application embodiment can execute the technical solution shown in the above vehicle collision risk prediction method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0185] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0186] It should be further noted that although the steps in the flowchart are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0187] It should be noted that the above-described device embodiments are merely illustrative, and the device of this application can also be implemented in other ways; and it should be understood that the division of the various modules of the above device is only a logical functional division, and in actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can all be implemented in software through processing element calls; they can all be implemented in hardware; or some modules can be implemented by processing element calls to software, and some modules can be implemented in hardware. For example, the risk prediction module can be a separately established processing element, or it can be integrated into a chip of the above device. Alternatively, it can be stored as program code in the memory of the above device, and the risk prediction module's function can be called and executed by a processing element of the above device. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through hardware integrated logic circuits in the processor element or software instructions.

[0188] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a System-On-a-Chip (SOC).

[0189] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is 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, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. 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 that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Video Discs, DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0190] Figure 4 A schematic diagram of the controller provided for an exemplary embodiment of this application. (See attached diagram.) Figure 4 As shown, the controller 40 in this embodiment includes:

[0191] At least one processor 41; and a memory 42 communicatively connected to the at least one processor;

[0192] The memory 42 stores instructions that can be executed by at least one processor 41 to cause the controller to perform the method as described in any of the above embodiments.

[0193] Alternatively, the memory 42 can be either standalone or integrated with the processor 41.

[0194] The memory 42 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0195] The processor 41 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. Specifically, when implementing the vehicle collision risk prediction method described in the foregoing method embodiments, the controller may be, for example, an electronic device with processing capabilities such as a server.

[0196] Optionally, the controller may also include a communication interface 43. In a specific implementation, if the communication interface 43, memory 42, and processor 41 are implemented independently, they can be interconnected via a bus to communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.

[0197] Optionally, in a specific implementation, if the communication interface 43, memory 42 and processor 41 are integrated on a single chip, then the communication interface 43, memory 42 and processor 41 can communicate through an internal interface.

[0198] The implementation principle and technical effects of the controller provided in this embodiment can be found in the foregoing embodiments, and will not be repeated here.

[0199] This application also provides a computer program product, including a computer program, which, when executed, implements the method steps as described in the above method embodiments. The specific implementation and technical effects are similar and will not be repeated here.

[0200] This application also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed, they are used to implement the method steps as described in the above method embodiments. The specific implementation methods and technical effects are similar and will not be repeated here.

[0201] The aforementioned computer-readable storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0202] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in a vehicle collision risk prediction device.

[0203] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units 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 indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0204] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0205] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0206] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part 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 of the various embodiments of this invention. 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.

[0207] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0208] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for predicting vehicle collision risk, characterized in that, include: Obtain the vehicle's operating status data and road environment data; The operating status data includes the vehicle speed, and the road environment data includes traffic flow density; Based on the operational status data and the road environment data, a dynamic interaction network is constructed; wherein, the dynamic interaction network uses the vehicle itself and other traffic participants as nodes, and the interaction relationships between the nodes as edges; Feature representations of each node in the dynamic interaction network are extracted based on graph neural networks; Based on the feature representation of each node, an influence score for each node is determined, and the influence score is used to quantify the degree of influence of the node on the occurrence of a vehicle collision. The influence score of each node is compared with a dynamic threshold, and a sparse attention weight distribution is generated based on the comparison results. Based on the sparse attention weight distribution, the feature representations of each node are weighted and fused to obtain the prediction result of vehicle collision risk. The dynamic threshold is determined in the following way: The traffic flow density and the vehicle speed are input into a preset linear relationship model to obtain a preliminary threshold. Based on the predicted results of vehicle collision risk over a historical time period and actual traffic data, a consistency score and false alarm rate are determined for the accuracy of the predicted results used to assess vehicle collision risk. A reward signal is determined based on the consistency score, the false alarm rate, and the preset penalty coefficient. With the goal of maximizing the reward signal, the model parameters of the linear relationship model are iteratively updated using a policy gradient algorithm to optimize the initial threshold online and generate the dynamic threshold.

2. The vehicle collision risk prediction method according to claim 1, characterized in that, The step of extracting feature representations of each node in the dynamic interaction network based on a graph neural network includes: The dynamic interaction network is iteratively message-passed through multiple network layers of the graph neural network. For each of the nodes, the feature representation of the node at layer L+1 is updated in the following way: Aggregate the node's own feature representation at layer L with the feature representations of each of the node's neighboring nodes at layer L; The aggregated result is subjected to a linear transformation, and the result is processed by an activation function to generate the feature representation of the node in the (L+1)th layer; where L is the layer index.

3. The vehicle collision risk prediction method according to claim 2, characterized in that, The determination of the influence score of each node based on the feature representation of each node includes: The norm of the feature representation corresponding to the node is multiplied by the topological centrality metric of the node in the dynamic interactive network, and the result of the multiplication is used as the influence score of the node. The norm of the feature representation corresponding to the node is used to quantify the urgency of the risk of the node causing a direct collision, and the topological centrality metric is used to quantify the criticality of the node in the dynamic interactive network in the propagation of collisions.

4. The vehicle collision risk prediction method according to any one of claims 1 to 3, characterized in that, Also includes: Based on the predicted vehicle collision risk, a vehicle avoidance control command is generated and sent to the actuator of the vehicle. The status feedback data of the vehicle is collected in real time, and the execution deviation between the status feedback data and the expected status feedback data corresponding to the vehicle evasion control command is determined. If the execution deviation exceeds a preset execution deviation threshold, the vehicle evasion control command will be re-planned. Meanwhile, the samples containing the state feedback data and the execution deviation are used as incremental data, and the model parameters of the graph neural network are updated online using an incremental learning algorithm to optimize the accuracy of subsequent vehicle collision risk prediction.

5. The vehicle collision risk prediction method according to claim 4, characterized in that, The predicted vehicle collision risk includes a vehicle collision probability value, a predicted risk time interval, a predicted risk spatial coordinate range, and a list of risk vehicle identifiers. Based on the predicted vehicle collision risk, a vehicle avoidance control command is generated, including: Based on the risk vehicle identification list, target threat vehicles are identified, and based on the vehicle speed, the predicted risk time interval, and the vehicle's maximum deceleration, the minimum safe distance required for the vehicle to achieve full braking within the predicted risk time interval is determined, and the minimum safe distance is used as the safe space boundary. Using the safety space boundary as a hard constraint, an optimization objective function is constructed. The optimization objective function is configured to minimize the amplitude of the control command and optimize the ride comfort index within a future prediction time domain. By using a model predictive control algorithm, the optimization objective function is solved in a rolling manner to obtain the optimal control sequence that includes braking, steering and power control quantities; The control quantity corresponding to the current moment in the optimal control sequence is output as the vehicle evasion control command.

6. The vehicle collision risk prediction method according to any one of claims 1 to 3, characterized in that, In the step of constructing the dynamic interactive network, the weights of each edge in the dynamic interactive network are determined in the following way: For any two nodes among the nodes, the interaction strength is determined based on the position vector and velocity vector of the two nodes; Based on the acceleration vectors of the two nodes, the magnitude of the relative acceleration is determined; The scaling factor is determined based on the magnitude of the relative acceleration; Multiply the interaction strength by the scaling factor, and use the result of the multiplication as the edge weight of the two nodes; The scaling factor is positively correlated with the magnitude of the relative acceleration. When the magnitude of the relative acceleration increases, it amplifies the interaction intensity.

7. A vehicle collision risk prediction device, characterized in that, include: The acquisition module is used to acquire the vehicle's operating status data and road environment data; The operating status data includes the vehicle speed, and the road environment data includes traffic flow density; A dynamic interactive network construction module is used to construct a dynamic interactive network based on the operating status data and the road environment data; wherein, the dynamic interactive network uses the vehicle and other traffic participants as nodes, and the interaction relationships between the nodes as edges; The feature extraction module is used to extract feature representations of each node in the dynamic interaction network based on a graph neural network; The determination module is used to determine the influence score of each node based on the feature representation of each node, wherein the influence score is used to quantify the degree of influence of the node on the occurrence of a vehicle collision; The comparison module is used to compare the influence score of each node with a dynamic threshold, and generate a sparse attention weight distribution based on the comparison results. The risk prediction module is used to perform weighted fusion of the feature representations of each node based on the sparse attention weight distribution to obtain the prediction result of vehicle collision risk. The dynamic threshold is determined in the following way: The traffic flow density and the vehicle speed are input into a preset linear relationship model to obtain a preliminary threshold. Based on the predicted results of vehicle collision risk over a historical time period and actual traffic data, a consistency score and false alarm rate are determined for the accuracy of the predicted results used to assess vehicle collision risk. A reward signal is determined based on the consistency score, the false alarm rate, and the preset penalty coefficient. With the goal of maximizing the reward signal, the model parameters of the linear relationship model are iteratively updated using a policy gradient algorithm to optimize the initial threshold online and generate the dynamic threshold.

8. A controller, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory is used to store computer-executed instructions; The processor is configured to execute the computer execution instructions to implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed, is used to implement the method as described in any one of claims 1 to 6.

Images