Vehicle driving risk prediction method based on interaction information fusion network

Abstract

The invention provides a vehicle driving risk prediction method based on an interaction information fusion network, and the method comprises the steps: obtaining a vehicle-vehicle interaction potential energy field and a vehicle-road interaction potential energy field according to the state information and road information of surrounding vehicles, the surrounding vehicles being vehicles which are located around a target vehicle and have potential influences on the target vehicle; fusing the kinematics characteristics of the vehicle-vehicle interaction potential energy field, the vehicle-road interaction potential energy field and the target vehicle by using a dynamic interaction information fusion network to obtain fusion characteristics; and predicting the fusion features by using a vehicle driving risk prediction model to obtain a prediction result. According to the method, the interaction relationship of the vehicle in the driving process can be comprehensively represented, the accuracy and interpretability of risk prediction are improved, and accurate prediction of the vehicle driving risk in a complex traffic scene is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle driving safety, and in particular to a vehicle driving risk prediction method based on an interactive information fusion network. Background Art

[0002] With the development of intelligent transportation systems, vehicle driving risk prediction has become an important technical means to ensure traffic safety and improve traffic efficiency. Traditional driving risk prediction models are mainly based on data-driven or physical model-driven approaches. Data-driven approaches predict risks by learning statistical patterns from historical data, while physical model approaches rely on mathematical modeling of vehicle motion and the traffic environment. However, these traditional methods have significant limitations when faced with complex vehicle-road collaborative environments. Data-driven approaches struggle to capture the complex interactions between vehicles, while physical model approaches struggle to adapt to dynamically changing traffic scenarios and the diversity of driving behaviors.

[0003] In a vehicle-road collaborative environment, the factors affecting vehicle driving risks become more complex and dynamic. Vehicles must not only interact with surrounding vehicles (such as following, changing lanes, overtaking, etc.), but also interact with the road environment (such as road signs, markings, obstacles, etc.). This complex interaction makes it difficult for traditional models to extract specific information features in different scenarios, thereby limiting the interpretability and adaptability of the model. For example, risk prediction models based on single sensor data often cannot fully capture the dynamic changes in vehicle-to-vehicle and vehicle-road interactions, while models based on static potential energy fields find it difficult to reflect the temporal characteristics of vehicle interactions.

[0004] Recent advances in sensor technology and data processing capabilities have made it possible to characterize the interaction between a vehicle and its surroundings by introducing potential energy field models. However, these models mostly focus on static risk assessment at a single moment in time and fail to fully consider the dynamic evolution of the interaction process. Furthermore, existing physical models cannot effectively integrate the comprehensive impact of the surrounding environment on the vehicle and therefore struggle to capture all potential risk factors. Therefore, a vehicle driving risk prediction method that can dynamically reflect the interaction between the vehicle and the traffic environment is urgently needed. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a vehicle driving risk prediction method based on an interactive information fusion network. The method can comprehensively characterize the interactive relationship between vehicles during driving, improve the accuracy and interpretability of risk prediction, and realize accurate prediction of vehicle driving risks in complex traffic scenarios.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions.

[0007] In a first aspect, the present invention provides a vehicle driving risk prediction method based on an interactive information fusion network, comprising:

[0008] Obtaining a vehicle-to-vehicle interaction potential energy field and a vehicle-to-road interaction potential energy field based on status information of surrounding vehicles and road information, wherein the surrounding vehicles are vehicles located around the target vehicle and have a potential impact on the target vehicle;

[0009] fusing the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle using a dynamic interaction information fusion network to obtain a fusion feature;

[0010] The vehicle driving risk prediction model is used to predict the fusion features to obtain a prediction result.

[0011] In some embodiments, obtaining the vehicle-to-vehicle interaction potential energy field and the vehicle-to-road interaction potential energy field based on the status information of surrounding vehicles and road information includes:

[0012] The vehicle-to-vehicle interaction potential energy field is calculated based on the actual mass, speed, acceleration, heading angle information, and position information of the target vehicle and surrounding vehicles;

[0013] The vehicle-road interaction potential energy field is obtained by calculation based on the road information.

[0014] In some embodiments, the calculating and obtaining the vehicle-to-vehicle interaction potential energy field based on the actual mass, velocity, acceleration, heading angle information, and position information of the target vehicle and surrounding vehicles includes:

[0015] The virtual mass of the surrounding vehicles is calculated based on the actual mass and speed of the surrounding vehicles:

[0016]

[0017] Where M j For node v vj Virtual mass of node v vj is the surrounding vehicles, m j For node v vj The actual mass, v j is the speed of surrounding vehicles;

[0018] The virtual distance r between the target vehicle and the surrounding vehicles is determined based on the position information of the target vehicle, the position information and speed of the surrounding vehicles, and the size information of the target vehicle. ij :

[0019]

[0020] Where x i 、xj They are the horizontal coordinates of the target vehicle i and the surrounding vehicle j, y i 、y j are the ordinates of the target vehicle i and the surrounding vehicle j, respectively; δ1 and δ2 are the correlation coefficients of the length and width of the target vehicle, respectively; α is the speed parameter;

[0021] The vehicle-to-vehicle interaction potential energy field E is calculated based on the virtual mass, the acceleration of the surrounding vehicles, and the virtual distance. V_ij :

[0022]

[0023] Where k1, k2, and ε are all constants, N is the number of surrounding vehicles, and a j is the acceleration of the jth surrounding vehicle, θ j is the virtual distance r ij and node j The clockwise angle formed by the direction of movement;

[0024] The force F exerted by the surrounding vehicles on the target vehicle is obtained by calculating the virtual mass, the speed information of the target vehicle, and the vehicle-to-vehicle interaction potential energy field. v_ij :

[0025]

[0026] In the formula, |v i | is the target vehicle’s speed scalar, M i is the virtual mass of the target vehicle, is the velocity vector v of the target vehicle i With the field strength vector E v_ij The clockwise angle of .

[0027] In some embodiments, the road information includes road signs and markings, and calculating and obtaining the vehicle-road interaction potential energy field based on the road information includes:

[0028] Determine the road potential energy E according to the road signs and markings r :

[0029]

[0030] Where A1 is the cost constant, E r 、E ra 、E rh They are the road potential energy, the potential energy of driving forward along the center line of the lane, and the potential energy of changing lanes to the adjacent lane;

[0031] The dynamic node attraction F is obtained by calculating the direction angle between the road node connection line and the target vehicle.r_ij :

[0032]

[0033] Where, |E r | is the road potential energy, λ is the unknown coefficient, and β is the velocity vector v of the target vehicle i Connecting the road node d ij The clockwise angle of .

[0034] In some embodiments, the utilizing a dynamic interaction information fusion network to fuse the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic features of the target vehicle to obtain fusion features includes:

[0035] A dynamic potential energy field is obtained according to the kinematic characteristics of the target vehicle, the acting force and the dynamic node attraction.

[0036]

[0037] Where, is the overall characteristic state of the target vehicle at time t; is the target vehicle v vi The kinematic characteristics of is the target vehicle kinematic characteristic state, Represents the vehicle-to-vehicle interaction layer node v vi and node v vj If the vehicle-to-vehicle interaction layer node v vi and node v vj If there is interaction between is 1; if the vehicle-to-vehicle interaction layer node v vi and node v vj If there is no interaction between is 0; g(F V_ij) is the node v in the vehicle-to-vehicle interaction layer vj For node v vi The interactive influence characteristics of Represents the vehicle-road interaction layer node v ri and node v rj Connection between them, if there is a node v ri To node v rj If there is a directed edge of 1; if there is no node v ri To node v rj If there is a directed edge of =0;Ψ(F r_ij ) represents the vehicle-road interaction layer node v ri To node v rj The degree of potential field action.

[0038] In some embodiments, the vehicle travel risk prediction model includes multiple expert models, and the use of the vehicle travel risk prediction model to predict the fusion features to obtain a prediction result includes:

[0039] The routing network is used to score the multiple expert models to obtain a model score h(x):

[0040] h(x)=W r x;

[0041] Where W r is the correlation with the expert model, and x is the current input feature;

[0042] Determine the available expert model according to the model score, perform normalization on the available expert model, and determine the expert contribution value p according to the normalization result i (x):

[0043]

[0044] The prediction results of each expert model are weighted according to the expert contribution value to obtain the final prediction result y:

[0045] y=∑ i∈Β p i (x)E i (x);

[0046] Where Β is the set of expert models selected in this batch.

[0047] In some embodiments, the expert model is trained based on an auxiliary loss function, and the auxiliary loss function loss is:

[0048]

[0049] Where N is the total number of expert models, f i is the degree to which the expert model is selected in the current batch, P i is the weight of the expert model.

[0050] In a second aspect, the present invention further provides a vehicle driving risk prediction device based on an interactive information fusion network, comprising:

[0051] a potential energy field acquisition module, configured to acquire a vehicle-to-vehicle interaction potential energy field and a vehicle-to-road interaction potential energy field based on status information of surrounding vehicles and road information, wherein the surrounding vehicles are vehicles located around the target vehicle and have a potential impact on the target vehicle;

[0052] an information fusion module, configured to fuse the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle using a dynamic interactive information fusion network to obtain a fusion feature;

[0053] The prediction module is used to predict the fusion features using the vehicle driving risk prediction model to obtain a prediction result.

[0054] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method when executing the program.

[0055] In a fourth aspect, the present invention further provides a computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0056] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which implements the above method when executed by a processor.

[0057] Beneficial effects of the present invention: The vehicle driving risk prediction method based on the interactive information fusion network provided by the present invention obtains the vehicle-to-vehicle interaction potential energy field and the vehicle-road interaction potential energy field according to the status information of the surrounding vehicles and the road information; uses the dynamic interactive information fusion network to fuse the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field and the kinematic characteristics of the target vehicle to form a fusion feature, which comprehensively characterizes the interaction relationship between the vehicles during driving; then, the vehicle driving risk prediction model is used to predict the fusion feature, which improves the accuracy and interpretability of the risk prediction and realizes the accurate prediction of vehicle driving risks in complex traffic scenarios.

[0058] Additional aspects and advantages of the present invention will be set forth in part in the following description, will become apparent from the following description, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0060] Figure 1 One of the flow charts of a vehicle driving risk prediction method based on an interactive information fusion network provided by an embodiment of the present invention;

[0061] Figure 2A second flow chart of a vehicle driving risk prediction method based on an interactive information fusion network provided by an embodiment of the present invention;

[0062] Figure 3 A schematic diagram of a vehicle driving potential energy field provided by an embodiment of the present invention;

[0063] Figure 4 A potential energy diagram of a target vehicle in a typical lane-changing scenario provided by an embodiment of the present invention;

[0064] Figure 5 A schematic diagram of topological road network potential energy provided by an embodiment of the present invention;

[0065] Figure 6 A two-layer dynamic network diagram of a typical lane-changing process provided by an embodiment of the present invention;

[0066] Figure 7 A schematic diagram of a hybrid expert model provided by an embodiment of the present invention;

[0067] Figure 8 A schematic diagram of a switching converter architecture provided by an embodiment of the present invention;

[0068] Figure 9 A schematic diagram of performance evaluation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.

[0070] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or couplings. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0071] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which the present invention pertains. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as such herein, will not be interpreted in an idealized or overly formal sense.

[0072] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0073] Example 1

[0074] like Figure 1 and Figure 2 As shown, a vehicle driving risk prediction method based on an interactive information fusion network includes the following steps:

[0075] S1, obtaining the vehicle-to-vehicle interaction potential energy field and the vehicle-road interaction potential energy field based on the status information of surrounding vehicles and road information.

[0076] Surrounding vehicles are vehicles located around the target vehicle and potentially affecting it. In this step, key factors in the vehicle-to-vehicle interaction process include the relative position, velocity, and heading angle of the surrounding vehicles. Based on these factors, a vehicle-to-vehicle interaction layer is constructed to calculate the potential energy field of the target vehicle.

[0077] S1 includes the following sub-steps:

[0078] S11, calculating and obtaining the vehicle-to-vehicle interaction potential energy field based on the actual mass, speed, acceleration, heading angle information, and position information of the target vehicle and surrounding vehicles.

[0079] S11 includes the following sub-steps:

[0080] S111: Calculate the virtual mass of the surrounding vehicles based on the actual mass and speed of the surrounding vehicles:

[0081]

[0082] Where M j For node v vj Virtual mass of node v vj is the surrounding vehicles, m j For node v vj The actual mass, v j is the speed of the surrounding vehicles, the virtual mass is used to characterize the dynamic characteristics of the vehicle, and the virtual mass M j Increases with increasing speed.

[0083] S112, determining a virtual distance r between the target vehicle and the surrounding vehicles based on the position information of the target vehicle, the position information and speed of the surrounding vehicles, and the size information of the target vehicle. ij :

[0084]

[0085] Where x i 、x j They are the horizontal coordinates of the target vehicle i and the surrounding vehicle j, y i 、y j are the ordinates of the target vehicle i and the surrounding vehicle j, δ1 and δ2 are the length and width correlation coefficients of the target vehicle, α is the speed parameter, and the virtual distance is used to characterize the relative position relationship between vehicles.

[0086] S113, calculating the vehicle-to-vehicle interaction potential energy field E according to the virtual mass, the acceleration of the surrounding vehicles, and the virtual distance V_ij :

[0087]

[0088] Where k1, k2, and ε are all constants, N is the number of surrounding vehicles, and a j is the acceleration of the jth surrounding vehicle, θ j is the virtual distance r ij and node j The clockwise angle formed by the direction of movement, vector E V_ij is the potential field of the target vehicle in the vehicle-to-vehicle interaction layer, that is, the risk intensity faced by the target vehicle in the current environment. The direction of the field strength is related to r ij Similarly, the vehicle's potential energy field is as follows Figure 3 shown.

[0089] S114, calculating the force F exerted by the surrounding vehicles on the target vehicle based on the virtual mass, the speed information of the target vehicle, and the vehicle-to-vehicle interaction potential energy field. v_ij :

[0090]

[0091] In the formula, |v i | is the target vehicle's speed scalar, is the velocity vector v of the target vehicle i With the field strength vector E v_ij The clockwise angle of the target vehicle in a typical lane-changing scenario is as follows: Figure 4 shown.

[0092] S12: Calculate and obtain the vehicle-road interaction potential energy field based on the road information.

[0093] In this step, road markings guide and restrict vehicle trajectories. Based on the high-definition lane map, a vehicle-road interaction layer is constructed to calculate the road potential field for the target vehicle.

[0094] S12 includes the following sub-steps:

[0095] S121, determining the road potential energy E according to the road signs and markings r :

[0096]

[0097] Where A1 is the cost constant, representing the safety risk that the target vehicle has to bear when driving towards the road boundary or double yellow line, E r 、E ra 、E rh They are the road potential energy, the potential energy for driving forward along the center line of the lane, and the potential energy for changing lanes to the adjacent lane. The road network potential energy is as follows: Figure 5 shown.

[0098] S122, calculate the dynamic node attraction F based on the direction angle between the road node connection line and the target vehicle r_ij :

[0099]

[0100] Where |A1| is the road potential energy, λ is the undetermined coefficient, and β is the velocity vector vi of the target vehicle and the line connecting the road node d. ij The clockwise angle of , the dynamic node attraction is used to characterize the driving constraints of the vehicle in the lane.

[0101] S2, using a dynamic interactive information fusion network to fuse the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle to obtain a fusion feature.

[0102] In this step, a two-layer dynamic network is constructed by coupling the vehicle-to-vehicle interaction layer and the vehicle-to-road interaction layer. The two-layer dynamic network of a typical lane-changing process is as follows: Figure 6 As shown in Figure 2, the dynamic potential energy field of the target vehicle is depicted. It specifically includes the following three parts:

[0103] (1) Target node v in the vehicle-to-vehicle interaction layer vi Single-point dynamic state, that is, the speed, acceleration and other characteristics of the target vehicle itself.

[0104] (2) Nodes around the vehicle-to-vehicle interaction layer v vj With the target node v viThe interaction between them, whether the driving force is received, that is, the potential energy field of the target node.

[0105] (3) Vehicle-road interaction layer node v rj With the target node v ri The connection between them, that is, the target node v in the vehicle-to-vehicle interaction layer vj Connected vehicle-road interaction layer node v ri The effect of the V2I layer on the V2I layer is reflected in how the motion and position of the vehicle-layer nodes affect the connectivity of the road-layer edges. Conversely, the V2I layer's topological structure, i.e., the reachability between nodes, constrains the motion direction and range of V2I nodes.

[0106] Specifically, the dynamic potential energy field is obtained according to the kinematic characteristics of the target vehicle, the action force and the dynamic node attraction.

[0107]

[0108] Where, is the overall characteristic state of the target vehicle at time t, is the target vehicle v vi The kinematic characteristics of is the target vehicle v vi The kinematic characteristics of is the target vehicle kinematic characteristic state, Represents the vehicle-to-vehicle interaction layer node v vi and node v vj If the vehicle-to-vehicle interaction layer node v vi and node v vj If there is interaction between is 1; if the vehicle-to-vehicle interaction layer node v vi and node v vj If there is no interaction between is 0; g(F v_ij ) is the node v in the vehicle-to-vehicle interaction layer vj For node v vi The interactive influence characteristics of Represents the vehicle-road interaction layer node v ri and node v rj Connection between them, if there is a node v ri To node v rj If there is a directed edge of 1; if there is no node v ri To node v rj If there is a directed edge of =0;Ψ(F r_ij ) represents the vehicle-road interaction layer node vri To node v rj The degree of potential field action.

[0109] S3, using a vehicle driving risk prediction model to predict the fusion features to obtain a prediction result.

[0110] The vehicle travel risk prediction model includes multiple expert models, and S3 includes the following sub-steps:

[0111] S31, scoring the multiple expert models using a routing network to obtain a model score h(x):

[0112] h(x)=W r x;

[0113] Where W r is the correlation with the expert model, and x is the input feature of the current input.

[0114] In this step, a hybrid expert model is introduced. The hybrid expert model is as follows: Figure 7 As shown in Figure 1, by dynamically selecting the most appropriate expert model, precise learning and optimization are performed for different interaction scenarios. To dynamically select the expert model that best suits the current input, a score must first be calculated for each expert model. The scoring mechanism is implemented using a routing network, whose output represents each expert model's adaptability to the current input.

[0115] S32, determining an available expert model based on the model score, normalizing the available expert model, and determining the expert contribution value p based on the normalization result. i (x):

[0116]

[0117] Specifically, the softmax activation function is used to normalize the available N experts, from the set The top k most suitable experts are selected to process the input. The contribution of each expert is determined by its gating value p i (x) determines that the larger the gate value, the stronger the influence of the expert.

[0118] S33, weighting the prediction results of each expert model according to the expert contribution value to obtain the final prediction result y:

[0119] y=∑ i∈B p i (x)E i (x);

[0120] Where B is the set of expert models selected in this batch.

[0121] In addition to multiple expert models, the vehicle driving risk prediction model also includes an attention layer, a residual connection, and a normalization layer.

[0122] To prevent some expert models from being over-enhanced during training while other expert models are under-trained, this paper introduces an auxiliary loss function. The auxiliary loss function achieves balanced training by limiting the activation frequency of each expert model. The specific formula is as follows:

[0123]

[0124] An auxiliary loss term, loss, is introduced into the total loss of the model. This auxiliary loss is calculated by calculating the relationship between the activation frequency of each expert and its importance, that is, it is measured by the weighted dot product of the expert importance vector f and P. The vector f reflects the degree to which each expert is selected in the current batch, while the vector P represents the weight of the expert.

[0125] Specifically, the auxiliary loss function loss is:

[0126]

[0127] Where N is the total number of expert models, f i is the degree to which the expert model is selected in the current batch, P i is the weight of the expert model, and B is the set of all expert models selected in this batch.

[0128] like Figure 8 As shown, an embodiment of the present application also provides a switching converter, which includes an encoder and a decoder, the encoder including a multi-head attention module, a residual module, a normalization layer, a hybrid expert module, etc., and the decoder including a multi-head attention module, a residual module, a normalization layer, a cross attention module, a hybrid expert module, a linear layer, etc.

[0129] The characteristic states of the target vehicle and surrounding vehicles are acquired, and their virtual masses are calculated. The virtual distances between the target vehicle and all surrounding vehicles and obstacles at each moment are calculated and substituted into the potential energy field formula to calculate the target vehicle's potential field. The target vehicle's potential energy field is used to calculate the magnitude of the interactive fusion forces acting on the target vehicle, thereby quantifying the risk faced by the target vehicle. These interactive fusion forces and the target vehicle's own dynamics data are used as inputs to a switching converter. Processed by the converter's encoder and decoder, the converter learns different interactive behaviors, thereby predicting the target vehicle's future driving risks.

[0130] like Figure 9As shown in the figure, the driving risk of the target vehicle changes in the lane changing scenario. The driving risk of the target vehicle shows a trend of first increasing and then decreasing, which is consistent with the changing trend of the actual vehicle driving risk.

[0131] The method provided by the present invention exhibits a clear trend of increasing and decreasing risk. In a real-world driving scenario, a stationary obstacle vehicle 2 appears in front of the target vehicle. The target vehicle avoids the obstacle by first decelerating, then slowly accelerating to change lanes, ultimately successfully completing the lane change and resuming acceleration. During this process, the collision risk rapidly increases when the target vehicle approaches the obstacle vehicle while not decelerating. As the target vehicle begins to decelerate, the risk continues to rise, but the rate of increase slows. After entering the slowly accelerating lane change phase, the risk increase increases again. After the lane change is completed, the risk gradually decreases and ultimately disappears after the lane change is successful. This effect is generally consistent with the trend shown by the curve of the method provided by the present invention. The ITTC model reflects similar formal risk changes, but exhibits smaller overall results compared to the method provided by the present invention. During the lane change process, the normalized risk value of the ITTC model is, on average, 0.1783 lower than that of the method provided by the present invention. At the point where the risk difference is the largest, the method provided by the present invention is 0.2887 higher than the ITTC indicator, demonstrating the accuracy of the method provided by the present invention. In addition to the improvement in accuracy, the method provided by the present invention also shows higher sensitivity. At the 6th frame, the vehicle begins to decelerate. At this time, the method provided by the present invention reaches a larger risk value, while ITTC is still at a lower risk.

[0132] In addition, as shown in Table 1, the method provided by the present invention has the lowest MAE and RMSE compared with other methods (such as multiple linear regression, ridge regression, LSTM, BiLSTM, TCN, and Transformer), and R 2 The highest, thus it can be seen that the method provided by the present invention has the best performance.

[0133] Table 1 Performance comparison table

[0134]

[0135] The vehicle driving risk prediction method based on an interactive information fusion network, provided by embodiments of the present invention, utilizes on-board sensors to acquire driving status information (including speed, acceleration, heading angle, etc.) of surrounding vehicles and road sign and marking information to construct a potential energy field for the vehicle itself. This vehicle-to-vehicle interaction potential energy field is fused with the vehicle-to-road interaction potential energy field via a generalized force model to form a dynamic interactive information network. Subsequently, a routing mechanism and a hybrid expert model are introduced, combined with a Transformer architecture to design a switching transformer to model and predict vehicle driving risk. The hybrid expert model dynamically selects the most appropriate expert network for precise learning and optimization in different interaction scenarios. Furthermore, an auxiliary loss function is used to constrain the expert network, further enhancing the model's robustness and adaptability. The model can switch to different expert modes based on specific traffic scenarios, enabling accurate prediction of complex interactive behaviors. This dual optimization architecture of a dynamic interactive information fusion network and a hybrid expert module, combined with a routing mechanism and auxiliary loss function technology, effectively improves the accuracy and interpretability of vehicle driving risk prediction, demonstrating significant advantages in complex traffic scenarios.

[0136] Example 2

[0137] Based on Example 1, this Example 2 provides a vehicle driving risk prediction device based on an interactive information fusion network. The vehicle driving risk prediction device based on an interactive information fusion network corresponds to the above-mentioned vehicle driving risk prediction method based on an interactive information fusion network, and specifically includes:

[0138] a potential energy field acquisition module, configured to acquire a vehicle-to-vehicle interaction potential energy field and a vehicle-to-road interaction potential energy field based on status information of surrounding vehicles and road information, wherein the surrounding vehicles are vehicles located around the target vehicle and have a potential impact on the target vehicle;

[0139] an information fusion module, configured to fuse the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle using a dynamic interactive information fusion network to obtain a fusion feature;

[0140] The prediction module is used to predict the fusion features using the vehicle driving risk prediction model to obtain a prediction result.

[0141] For specific details, please refer to the description of the vehicle driving risk prediction method based on the interactive information fusion network, which will not be repeated here.

[0142] Example 3

[0143] Embodiment 3 of the present invention provides an electronic device including a memory and a processor. The processor and the memory communicate with each other. The memory stores program instructions executable by the processor. The processor calls the program instructions to execute a vehicle driving risk prediction method based on an interactive information fusion network. The method includes the following process steps:

[0144] Obtaining a vehicle-to-vehicle interaction potential energy field and a vehicle-to-road interaction potential energy field based on status information of surrounding vehicles and road information, wherein the surrounding vehicles are vehicles located around the target vehicle and have a potential impact on the target vehicle;

[0145] fusing the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle using a dynamic interaction information fusion network to obtain a fusion feature;

[0146] The vehicle driving risk prediction model is used to predict the fusion features to obtain a prediction result.

[0147] Example 4

[0148] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, a vehicle driving risk prediction method based on an interactive information fusion network is implemented. The method includes the following steps:

[0149] Obtaining a vehicle-to-vehicle interaction potential energy field and a vehicle-to-road interaction potential energy field based on status information of surrounding vehicles and road information, wherein the surrounding vehicles are vehicles located around the target vehicle and have a potential impact on the target vehicle;

[0150] fusing the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle using a dynamic interaction information fusion network to obtain a fusion feature;

[0151] The vehicle driving risk prediction model is used to predict the fusion features to obtain a prediction result.

[0152] Example 5

[0153] Embodiment 5 of the present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor, a vehicle driving risk prediction method based on an interactive information fusion network is implemented. The method includes the following process steps:

[0154] Obtaining a vehicle-to-vehicle interaction potential energy field and a vehicle-to-road interaction potential energy field based on status information of surrounding vehicles and road information, wherein the surrounding vehicles are vehicles located around the target vehicle and have a potential impact on the target vehicle;

[0155] fusing the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle using a dynamic interaction information fusion network to obtain a fusion feature;

[0156] The vehicle driving risk prediction model is used to predict the fusion features to obtain a prediction result.

[0157] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0158] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for method or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The method and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Ordinary technicians in this field can understand and implement it without expending creative work.

[0159] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A vehicle driving risk prediction method based on an interactive information fusion network, characterized in that: include: Obtaining a vehicle-to-vehicle interaction potential energy field and a vehicle-to-road interaction potential energy field based on status information of surrounding vehicles and road information, wherein the surrounding vehicles are vehicles located around the target vehicle and have a potential impact on the target vehicle; fusing the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle using a dynamic interaction information fusion network to obtain a fusion feature; The vehicle driving risk prediction model is used to predict the fusion features to obtain a prediction result.

2. The method according to claim 1, characterized in that The obtaining of the vehicle-to-vehicle interaction potential energy field and the vehicle-to-road interaction potential energy field based on the status information of surrounding vehicles and the road information includes: The vehicle-to-vehicle interaction potential energy field is calculated based on the actual mass, speed, acceleration, heading angle information, and position information of the target vehicle and surrounding vehicles; The vehicle-road interaction potential energy field is obtained by calculation based on the road information.

3. The method according to claim 2, characterized in that The calculating and obtaining the vehicle-to-vehicle interaction potential energy field based on the actual mass, speed, acceleration, heading angle information, and position information of the target vehicle and surrounding vehicles includes: The virtual mass of the surrounding vehicles is calculated based on the actual mass and speed of the surrounding vehicles: Where M j For node v vj Virtual mass of node v vj is the surrounding vehicles, m j For node v vj The actual mass, v j is the speed of surrounding vehicles; The virtual distance r between the target vehicle and the surrounding vehicles is determined based on the position information of the target vehicle, the position information and speed of the surrounding vehicles, and the size information of the target vehicle. ij : Where x i 、x j They are the horizontal coordinates of the target vehicle i and the surrounding vehicle j, y i 、y j are the ordinates of the target vehicle i and the surrounding vehicle j, respectively; δ1 and δ2 are the length and width correlation coefficients of the target vehicle, respectively; α is the speed parameter; The vehicle-to-vehicle interaction potential energy field E is calculated based on the virtual mass, the acceleration of the surrounding vehicles, and the virtual distance. V_ij : Where k1, k2, and ε are all constants, N is the number of surrounding vehicles, and a j is the acceleration of the jth surrounding vehicle, θ j is the virtual distance r ij and node j The clockwise angle formed by the direction of movement; The force F exerted by the surrounding vehicles on the target vehicle is obtained by calculating the virtual mass, the speed information of the target vehicle, and the vehicle-to-vehicle interaction potential energy field. v_ij : In the formula, |v i | is the target vehicle’s speed scalar, M i is the target vehicle virtual mass, is the velocity vector v of the target vehicle i With the field strength vector E v_ij The clockwise angle of .

4. The method according to claim 3, characterized in that The road information includes road signs and markings, and the calculating and obtaining the vehicle-road interaction potential energy field according to the road information includes: Determine the road potential energy E according to the road signs and markings r : Where A1 is the cost constant, E r 、E ra 、E rh They are the road potential energy, the potential energy of driving forward along the center line of the lane, and the potential energy of changing lanes to the adjacent lane; The dynamic node attraction F is obtained by calculating the direction angle between the road node connection line and the target vehicle. r_ij : Where, |E r | is the road potential energy, λ is the unknown coefficient, and β is the velocity vector v of the target vehicle i Connecting the road node d ij The clockwise angle of .

5. The method according to claim 4, characterized in that The method of using a dynamic interactive information fusion network to fuse the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle to obtain a fusion feature includes: A dynamic potential energy field is obtained according to the kinematic characteristics of the target vehicle, the acting force and the dynamic node attraction. Where, is the overall characteristic state of the target vehicle at time t; is the target vehicle v vi The kinematic characteristics of is the target vehicle kinematic characteristic state, Represents the vehicle-to-vehicle interaction layer node v vi and node v vj If the vehicle-to-vehicle interaction layer node v vi and node v vj If there is interaction between is 1; if the vehicle-to-vehicle interaction layer node v vi and node v vj If there is no interaction between is 0; g(F v_ij ) is the node v in the vehicle-to-vehicle interaction layer vj For node v vi The interactive influence characteristics of Represents the vehicle-road interaction layer node v ri and node v rj Connection between them, if there is a node v ri To node v rj If there is a directed edge of 1; if there is no node v ri To node v rj If there is a directed edge of =0;Ψ(F r_ij ) represents the vehicle-road interaction layer node v ri To node v rj The degree of potential field action.

6. The method according to claim 1, characterized in that The vehicle travel risk prediction model includes multiple expert models, and the vehicle travel risk prediction model is used to predict the fusion features to obtain a prediction result, including: The routing network is used to score the multiple expert models to obtain a model score h(x): h(x)=W r ·x; Where W r is the correlation with the expert model, and x is the current input feature; Determine the available expert model according to the model score, perform normalization on the available expert model, and determine the expert contribution value p according to the normalization result i (x): The prediction results of each expert model are weighted according to the expert contribution value to obtain the final prediction result y: y=∑ i∈Β p i (x)E i (x); Where Β is the set of expert models selected in this batch.

7. The method according to claim 6, characterized in that The expert model is trained based on an auxiliary loss function, and the auxiliary loss function loss is: Where N is the total number of expert models, f i is the degree to which the expert model is selected in the current batch, P i is the weight of the expert model.

8. A vehicle driving risk prediction device based on an interactive information fusion network, characterized in that: include: a potential energy field acquisition module, configured to acquire a vehicle-to-vehicle interaction potential energy field and a vehicle-to-road interaction potential energy field based on status information of surrounding vehicles and road information, wherein the surrounding vehicles are vehicles located around the target vehicle and have a potential impact on the target vehicle; an information fusion module, configured to fuse the vehicle-to-vehicle interaction potential energy field, the vehicle-road interaction potential energy field, and the kinematic characteristics of the target vehicle using a dynamic interactive information fusion network to obtain a fusion feature; The prediction module is used to predict the fusion features using the vehicle driving risk prediction model to obtain a prediction result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The device stores a computer program, which implements the method according to any one of claims 1 to 7 when executed by a processor.

Images