A method and system for predicting the trajectory of electric vehicles at signalized intersections.

Abstract

This invention discloses a method and system for predicting the trajectory of electric vehicles at signalized intersections, relating to the field of intelligent transportation technology. The method and system provided by this invention improve the accuracy and efficiency of electric vehicle trajectory prediction in complex intersection scenarios by utilizing Mamba to capture long-term temporal dependencies in signalized intersection videos and incorporating kinematic physical constraints using PINN. By constructing counterfactual scenarios of trajectory feature parameters for counterfactual inference prediction, the causal quantification of key factors affecting trajectory prediction results is achieved, enhancing model interpretability. Furthermore, Bayesian methods are used to perform uncertainty analysis on the Mamba-PINN fusion model, quantifying the uncertainty of trajectory prediction and improving model reliability. The optimized model exhibits strong adaptability and can provide decision-making basis and technical support for traffic signal optimization and intelligent transportation systems.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and in particular to a method, system, medium, and device for predicting the trajectory of electric vehicles at road signal-controlled intersections. Background Technology

[0002] With the popularization of electric vehicles and the rapid development of intelligent transportation systems, accurate prediction of the trajectory of electric vehicles at urban signalized intersections is a core requirement for improving traffic efficiency and ensuring driving safety. However, signalized intersections are complex scenarios, and the trajectory of electric vehicles is affected by both their own dynamic characteristics and the external environment, exhibiting strong nonlinearity, multiple constraints, and dynamic changes. Traditional trajectory prediction methods are insufficient to meet the accuracy requirements.

[0003] Existing trajectory prediction methods mainly suffer from two types of defects: First, traditional deep learning methods lack the ability to capture temporal correlation information, making them unsuitable for dynamic scenarios at signal-controlled intersections and resulting in low accuracy. Second, data-driven methods based on LSTM, Transformer, and other technologies are insufficient in capturing long sequence dependencies, have low training efficiency, and are not optimized for complex intersection scenarios and the characteristics of electric vehicles, so their accuracy still needs improvement. Furthermore, data-driven "black box" models inevitably have issues with interpretability and reliability.

[0004] Therefore, most existing technologies can only achieve basic trajectory prediction, and have problems such as low prediction accuracy, poor interpretability, insufficient reliability of prediction results, and low adaptability, making it difficult to meet the actual needs of engineering decision-making. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, medium, and device for predicting the trajectory of electric vehicles at road signal-controlled intersections to address the aforementioned technical problems.

[0006] The following technical solution is adopted in this specification: This specification provides a method for predicting the trajectory of electric vehicles at signalized intersections, including: Collect video footage of vehicle movement under different traffic conditions at road signal-controlled intersections; The target vehicle is identified from the acquired video, and the trajectory feature parameter data of the target vehicle is extracted. Based on the trajectory feature parameter data, a Mamba-PINN fusion model is trained, and a first predicted trajectory is generated through the trained Mamba-PINN fusion model. Using the trajectory feature parameter data as a baseline scenario, data-oriented intervention is performed. After obtaining the counterfactual scenario feature data, it is input into the Mamba-PINN fusion model to generate a second predicted trajectory. The importance of trajectory feature parameters is scored based on the first and second predicted trajectories, and the Mamba-PINN fusion model is optimized using the importance score results. At the same time, Bayesian methods are used to introduce uncertainty into the Mamba-PINN fusion model to optimize the model. The electric vehicle trajectory at the target road signal-controlled intersection is predicted using the optimized Mamba-PINN fusion model.

[0007] Optionally, the Mamba-PINN fusion model is trained using a combined loss function that combines data loss and physical loss.

[0008] Optionally, the trajectory feature parameters include instantaneous velocity, tangential acceleration, normal acceleration, travel angle, abscissa, and ordinate, and the steps for data orientation intervention include: Keep the remaining trajectory feature parameters unchanged; Each trajectory feature parameter data is set to a preset multiple of its original value in turn.

[0009] Optionally, the step of scoring the importance of trajectory feature parameters based on the first predicted trajectory and the second predicted trajectory specifically includes: Benchmark metrics for the baseline scenario are calculated based on the first predicted trajectory; Calculate the counterfactual index after intervention for each trajectory feature parameter based on the second predicted trajectory; The changes in the baseline index and each counterfactual index are calculated, and the changes are fused using a normalization method to obtain the importance score of each trajectory feature parameter.

[0010] Optionally, the Mamba-PINN fusion model can be optimized using the importance score results, specifically including: The trajectory feature parameters are classified according to their importance scores; Different model weights are assigned to trajectory feature parameters at different levels; The sensitivity of the model to trajectory feature parameters at different levels and the allocation of computational resources are adjusted according to the weights.

[0011] Optionally, uncertainties can be introduced into the Mamba-PINN fusion model using a Bayesian approach to optimize the model, specifically including: A Gaussian prior distribution is assigned to each parameter of the Mamba-PINN fusion model; A Gaussian variational distribution is introduced to approximate the posterior distribution of each parameter, and the optimal variational parameters of the Mamba-PINN fusion model are obtained by optimizing the variational parameters. Optionally, it also includes: Multiple sets of data are sampled from the posterior distribution of each parameter using Monte Carlo sampling, and each set of data is input into the Mamba-PINN fusion model to obtain a set of predicted trajectories; The uncertainty of the model prediction is quantified by calculating three indicators: total average prediction variance, coverage, and uncertainty-error correlation.

[0012] This specification provides a system for predicting the trajectory of an electric vehicle at a signalized intersection, including: The acquisition module is used to acquire video footage of vehicles operating under different traffic conditions at road signal-controlled intersections; The first generation module is used to identify target vehicles from the collected video, extract trajectory feature parameter data of the target vehicles, train a Mamba-PINN fusion model based on the trajectory feature parameter data, and generate a first predicted trajectory through the trained Mamba-PINN fusion model. The second generation module is used to perform data-oriented intervention based on the trajectory feature parameter data as a benchmark scenario, obtain counterfactual scenario feature data, input it into the Mamba-PINN fusion model, and generate a second predicted trajectory. An optimization module is used to score the importance of trajectory feature parameters based on the first and second predicted trajectories, and to optimize the Mamba-PINN fusion model using the importance score results. At the same time, a Bayesian method is used to introduce uncertainty into the Mamba-PINN fusion model to optimize the model. A prediction module is used to predict the trajectory of electric vehicles at the target road signal-controlled intersection using the optimized Mamba-PINN fusion model.

[0013] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting the trajectory of electric vehicles at road signal-controlled intersections.

[0014] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for predicting the trajectory of electric vehicles at road signal-controlled intersections.

[0015] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: This specification provides a method and system for predicting the trajectory of electric vehicles at signalized intersections. By utilizing Mamba to capture long-term temporal dependencies in signalized intersection videos and incorporating kinematic physical constraints using PINN, the accuracy and efficiency of electric vehicle trajectory prediction in complex intersection scenarios are improved. Furthermore, by constructing counterfactual scenario data of trajectory feature parameters and using model counterfactual inference prediction, the causal quantification of key influencing factors in trajectory prediction is achieved, enhancing model interpretability. The Mamba-PINN fusion model is optimized using counterfactual scenarios and Bayesian methods to quantify the uncertainty of trajectory prediction, improving model reliability. The optimized model exhibits strong adaptability and can provide decision-making basis and technical support for traffic signal optimization and intelligent transportation systems. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This document provides a flowchart illustrating a method for predicting the trajectory of an electric vehicle at a road signal-controlled intersection. Figure 2 This is a schematic diagram illustrating the construction of the Mamba-PINN fusion model provided in this manual; Figure 3 This is a flowchart illustrating the steps involved in data-driven interventions provided in this manual. Figure 4 This is a flowchart illustrating the process of scoring the importance of trajectory feature parameters based on the first and second predicted trajectories, as provided in this specification. Figure 5 A flowchart illustrating the process of optimizing the Mamba-PINN fusion model based on the importance score results provided in this specification; Figure 6 This is a flowchart illustrating the process of optimizing the Mamba-PINN fusion model using a pre-defined Bayesian method, as provided in this manual. Figure 7 This is a flowchart illustrating the process of predicting uncertainty using the quantitative model provided in this manual. Figure 8 This specification provides a schematic diagram of an electric vehicle trajectory prediction system for a road signal-controlled intersection. Figure 9 This specification provides a schematic diagram of a computer device for implementing a method for predicting the trajectory of electric vehicles at a road signal-controlled intersection. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0018] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0019] This invention provides a method for predicting the trajectory of electric vehicles at signal-controlled intersections. Figure 1 The flowchart of this method is shown, and it specifically includes the following steps: S101. Collect video footage of vehicles operating under different traffic conditions at road signal-controlled intersections.

[0020] Data collection was conducted during times of clear weather and ample sunlight. For example, the intersection of Youyi Road and Wenyi Road in Xi'an was chosen as the data collection location. The collection period was during the evening rush hour on a weekday (18:30-19:00). By setting up high-position video equipment or using drone aerial photography, the driving behavior of different vehicles under different traffic conditions at the road signal control intersection was fully recorded. The collected videos were ensured to be clear and meet the requirements for subsequent processing, providing a sufficient data foundation for subsequent model training, verification, and testing.

[0021] S102. Identify the target vehicle from the collected video and extract the trajectory feature parameter data of the target vehicle. Train the Mamba-PINN fusion model based on the trajectory feature parameter data and generate the first predicted trajectory through the trained Mamba-PINN fusion model.

[0022] Before identifying vehicles, the acquired video images are first mapped to the UTM coordinate system using a four-point calibration method to complete georegistration. Then, the range of each entrance and exit lane at the intersection is defined, and the continuous position of vehicles within the range is automatically tracked and recorded. Finally, DataFromSky Viewer software is used to automatically identify and track the motion trajectory of each vehicle in the acquired video, realizing vehicle trajectory extraction. The software outputs trajectory points at 0.04-second intervals, and the information recorded for each trajectory point includes the vehicle's horizontal and vertical coordinates, speed, acceleration, and heading angle.

[0023] In other optional implementations, the acquired video is processed frame by frame using an object detection algorithm (YOLOv8n combined with SORT) to identify electric vehicles and label each vehicle with a unique Track ID. A cross-frame tracking algorithm maintains the continuity of the same vehicle ID, and trajectory data for each vehicle is extracted. The data types include Track ID, Traveled Dist, AvgSpeed, x (horizontal coordinate), y (vertical coordinate), Speed, Tan Acc (tangential acceleration), LatAcc (normal acceleration), Time (timestamp), and Angle (trajectory angle). Six features—x, y, Speed, Tan Acc, LatAcc, and Angle—are selected as trajectory feature parameters for subsequent prediction. The extracted trajectory data is then cleaned, filtering out erroneous trajectories such as those with incorrect identification or coordinate jumps. The data is grouped by Track ID and sorted by Time, ultimately generating a structured trajectory parameter data file in CSV format.

[0024] Furthermore, the structured trajectory parameter data in CSV format is read and grouped by Track ID. A sliding window approach is used to generate historical time-series sequences and future trajectory coordinate samples. The six trajectory feature parameters for prediction are normalized to eliminate the influence of dimensions. The normalization formula is: ; ; in, The characteristic mean, The characteristic standard deviation is 1e-6, which is the minimum value used to avoid a denominator of 0. This represents the normalized value of the ordinate x in the trajectory feature parameters. This represents the normalized value of the ordinate y in the trajectory feature parameters. Simultaneously, using the (x, y) coordinates of the last moment in the historical sequence as an offset, localization adjustments are made to eliminate the influence of absolute coordinates. The localization formula is: ; in, 、 The x and y coordinates are the last moment of the historical sequence, respectively; then the preprocessed data is divided into training set, validation set and test set in a ratio of 7:1:2 to construct a standardized dataset.

[0025] The pre-built Mamba-PINN fusion model was then trained and validated using training and validation sets. The Mamba-PINN fusion model, constructed from Mamba encoders and PINN multi-step predictors, is a neural network component based on a state-space model, primarily used for efficiently processing long sequence data. It replaces the self-attention mechanism in traditional Transformers by introducing a Selective State-Space Model (SSM), thereby significantly reducing computational complexity while maintaining the ability to model long-distance dependencies. PINN (Physics-Informed Neural Networks) is a deep learning framework that embeds physical laws into the neural network training process. It commonly uses fully connected neural networks, with spatiotemporal coordinates as input and physical fields (such as temperature and velocity) as output. By introducing governing equations (such as partial differential equations, PDEs) as regularization terms into the loss function, the model adheres to known physical laws while fitting data, thus improving generalization ability and reducing dependence on large amounts of labeled data. The specific process of using the Mamba-PINN fusion model in this invention is as follows: (1) Mamba encoder: The preprocessed 6-dimensional temporal trajectory feature parameter data is mapped to a high-dimensional latent space through linear projection. The projection formula is as follows: ,in For projection weights, As a bias term, the 6-dimensional features are mapped to a 256-dimensional latent space; temporal dependencies are captured through a 3-layer Mamba network, with the forward propagation formula for each Mamba layer being... Finally, after layer normalization, a regular feature representation is output. ; (2) Multi-step predictor: The last step feature output by the Mamba encoder and the coordinates of the last moment of the historical sequence are used as inputs. The input features for each step prediction are obtained through a concatenation operation. ,in The coordinates are the coordinates at the current moment; the future trajectory coordinates are predicted step by step through a three-layer fully connected network. The forward propagation formula of the fully connected network is: ; ; ; Construction of the Mamba-PINN fusion model Figure 2 As shown, a complete future trajectory sequence is generated through 30 iterative predictions. Then, a combined loss function, combining data loss and physical loss, is designed to train the constructed Mamba-PINN fusion model, balancing prediction accuracy and physical plausibility. (1) Data loss: The mean squared error (MSE) is used to measure the deviation between the predicted trajectory and the actual trajectory. The formula is as follows: ,in To predict trajectory coordinates for the model, These are the coordinates of the actual trajectory.

[0026] (2) Physical losses: The speed and acceleration of the electric vehicle are constrained to not exceed reasonable upper limits. The formulas for calculating speed and acceleration are as follows: ; in, The time step is given; the physical loss formula is... , where 20 is the upper limit of velocity (m / s) and 4 is the upper limit of acceleration (m / s²). The ReLU function is used to penalize predicted values ​​that exceed the constraints.

[0027] (3) Total loss: ; in The weights for the physical loss are used to balance the data loss and the physical loss. The AdamW optimizer and training set are used to train the fusion model, with a learning rate of 1e-4 and a weight decay of 1e-4. Gradient clipping is used to prevent gradient explosion, and an early stopping mechanism is introduced to avoid overfitting. When the validation set loss does not decrease for 20 consecutive rounds, training is terminated and the optimal model is saved.

[0028] The test set data is then input into the optimal model to generate the first predicted trajectory. Based on the first predicted trajectory and the corresponding real trajectory, core evaluation metrics are calculated, including average displacement error (ADE), final displacement error (FDE), and inefficiency (MR). The calculation formulas for these metrics are as follows: ; ; ; in, For the sample size, To predict the step size, Let the coordinates of the true trajectory of the i-th sample at step t in the test set be . The indicator function is used. Testing showed that the model's ADE = 0.2412m, FDE = 0.4886m, and MR = 0.0000%, validating the high accuracy and effectiveness of the constructed and trained Mamba-PINN fusion model. This invention significantly improves the accuracy and efficiency of electric vehicle trajectory prediction at signalized intersections by fusing efficient Mamba sequence modeling with PINN physical constraints.

[0029] S103. Using trajectory feature parameter data as the baseline scenario, perform data-oriented intervention, obtain counterfactual scenario feature data, and input it into the Mamba-PINN fusion model to generate a second predicted trajectory.

[0030] Specifically, such as Figure 3 As shown, the trajectory feature parameters include instantaneous velocity, tangential acceleration, normal acceleration, travel angle, abscissa, and ordinate. The steps for data orientation intervention include: S1021. Keep the remaining trajectory feature parameter data unchanged; S1022. Sequentially set each trajectory feature parameter data to a preset multiple of its original value.

[0031] Typical electric vehicle trajectories from the test set were selected as benchmark scenarios, and a benchmark scenario dataset was constructed using the data of the six trajectory feature parameters mentioned above. For the benchmark scenario, a univariate control method was used to keep other trajectory feature parameters constant, and targeted interventions were applied to the six prediction trajectory feature parameters in sequence to construct counterfactual scenarios and obtain six sets of counterfactual scenario feature data accordingly. In specific implementation, interventions can be applied only to the dynamic features Speed, Tan Acc, Lat Acc, and Angle, while the x and y coordinates are not affected. The intervention rules are as follows: the value of Speed ​​after intervention is 1.2 times the original value, the value of Tan Acc after intervention is 1.2 times the original value, the value of Lat Acc after intervention is 1.2 times the original value, and the value of Angle after intervention is increased by 0.1 from the original value.

[0032] Then, the acquired multiple sets of counterfactual scenario feature data are input into the trained Mamba-PINN fusion model to generate multiple corresponding second prediction trajectories under the counterfactual scenario.

[0033] S104. The importance of the trajectory feature parameters is scored based on the first and second predicted trajectories, and the Mamba-PINN fusion model is optimized using the importance score results and the preset Bayesian method.

[0034] like Figure 4 As shown, in step S104, the importance scoring of trajectory feature parameters based on the first and second predicted trajectories specifically includes: S1031. Calculate the benchmark indicators of the benchmark scenario based on the first predicted trajectory; S1032. Calculate the counterfactual index after intervention for each trajectory feature parameter based on the second predicted trajectory; S1033. Calculate the changes in the baseline index and each counterfactual index, and use a normalization method to fuse the changes to obtain the importance score of each trajectory feature parameter.

[0035] First, the aforementioned benchmark scene dataset is input into the trained Mamba-PINN fusion model, and the corresponding first predicted trajectory is output as the benchmark predicted trajectory. Based on the benchmark predicted trajectory and the corresponding real trajectory, the benchmark scene is calculated. and As a benchmark indicator; Then, the trajectory feature parameter data of multiple sets of counterfactual scenarios obtained from targeted intervention are input into the optimal Mamba-PINN model to obtain the predicted trajectories of multiple counterfactual scenarios. Similarly, the ADE and FDE of each counterfactual scenario are calculated. , As a counterfactual indicator; calculate the change between the baseline indicator and each counterfactual indicator. 、 Finally, a normalization method is used to fuse the changes, and the importance score of each trajectory feature parameter is obtained. The calculation formula is as follows: ; in, , , respectively, represent the maximum changes in ADE and FDE after intervention of all trajectory feature parameters, and 0.6 and 0.4 are the weights of the changes in ADE and FDE, respectively.

[0036] Furthermore, such as Figure 5 As shown, in step S104, the Mamba-PINN fusion model is optimized using the importance score results, specifically including: S1041. Classify the trajectory feature parameters according to their importance scores; S1042. Different model weights are assigned to trajectory feature parameters corresponding to different levels. S1043. Adjust the sensitivity of the model to trajectory feature parameters at different levels and allocate computational resources according to the weights.

[0037] First, each trajectory feature parameter is classified according to its importance score: a score ≥ 0.7 is a key influencing feature, 0.3 ≤ score < 0.7 is a medium influencing feature, and a score < 0.3 is a weak influencing feature. The results are shown in Table 1. Then, a feature influence quantification report is output to clarify the degree of influence of each feature on the trajectory prediction results and to identify key influencing factors, providing a basis for model optimization.

[0038] Table 1

[0039] Then, by strengthening key influencing features and simplifying weak influencing features, specifically by assigning different model weights to trajectory feature parameters of different levels, the Mamba-PINN model structure and training process are further optimized. During model training, the optimizer adjusts the magnitude of parameter updates based on the weights of trajectory features, giving higher attention and weight to key influencing features, thereby improving prediction accuracy. Simultaneously, based on the results of counterfactual inference, the model adjusts its sensitivity and allocation of computational resources on feature parameters of different levels by strengthening high-weight key influencing features and weakening the expressive power of low-weight weak influencing features, ensuring efficient capture of the most critical dynamic features. This step of optimizing the Mamba-PINN model by scoring the importance of trajectory feature parameters enables the Mamba-PINN model to perform more accurate trajectory predictions in various dynamic and complex traffic environments, achieving optimal performance improvement at the feature level. Furthermore, the causal quantification of feature parameters through counterfactual inference enhances the interpretability of model predictions.

[0040] Furthermore, such as Figure 6 As shown, in step S104, a Bayesian method is used to introduce uncertainty into the Mamba-PINN fusion model to optimize the model, specifically including: S1051. Assign Gaussian prior distributions to each parameter of the Mamba-PINN fusion model; S1052. Introduce a Gaussian variational distribution to approximate the posterior distribution of each parameter, and obtain the optimal variational parameters of the Mamba-PINN fusion model by optimizing the Gaussian variational parameters.

[0041] Specifically, this refers to the parameter set consisting of the projection weights of the Mamba encoder, the state transition matrix, and the weights and bias terms of the fully connected layers of the multi-step predictor. All are assigned a Gaussian prior distribution get , Let be the identity matrix, and 0.01 be the variance of the standard deviation; then, a Gaussian variational distribution is introduced to approximate the parameters. posterior distribution , and For variational parameters, construct the objective function. To optimize variational parameters To obtain the optimal variational parameters of the model In the formula ] indicates based on variational distribution The log-likelihood expectation, where These are the parameters of the model. It is the input data. It is observation data; The variational distribution represents the distribution of model parameters. An assumed probability distribution; Kullback-Leibler divergence measures the variational distribution. With prior distribution The differences between them.

[0042] In the Bayesian Mamba-PINN model, the objective function is... By optimizing variational parameters To approximate the posterior distribution of the parameters. First, the variational distribution. It is introduced with an approximate true posterior distribution, while the prior distribution is... It is set to a Gaussian distribution. The objective is to maximize the variational lower bound, i.e., by optimizing... and This allows the variational distribution to fit the log-likelihood of the data to the greatest extent possible while maintaining consistency with the prior distribution. The optimization process includes maximizing the expected log-likelihood. and minimizing the KL divergence between the variational distribution and the prior distribution This allows the variational distribution to better approximate the posterior distribution. By introducing the uncertainties of the Bayesian prior distribution and the approximate posterior distribution into the parameters of the Mamba-PINN model, and through subsequent calculation and analysis of the model uncertainty index, the stability and reliability of the model are assessed.

[0043] like Figure 7 As shown, after step S104, the electric vehicle trajectory prediction method for road signal-controlled intersections of the present invention further includes: S1061. Multiple sets of data are sampled from the posterior distribution of each parameter through Monte Carlo sampling. Each set of data is input into the Mamba-PINN fusion model to obtain a set of predicted trajectories. S1062. The uncertainty of the model prediction is quantified by calculating three indicators: total average prediction variance, coverage, and uncertainty-error correlation.

[0044] Multiple sets of trajectory feature parameter data are sampled from the posterior distribution of the parameters using Monte Carlo (MC) sampling. Each set of parameters is input into the Mamba-PINN model to obtain a set of predicted trajectories. Then, the overall average prediction variance and coverage uncertainty-error correlation are calculated based on the predicted trajectories. (1) Mean prediction variance: ; It is used to measure the overall uncertainty of the prediction results; the smaller the variance, the more stable the prediction. (2) Coverage: ; in The mean of the MC sampled predicted trajectory is 1.96, which is the Z score of the 95% confidence interval. A CR close to 95% indicates that the confidence interval setting is effective. (3) Uncertainty-error correlation: ; in, The square of the prediction error, This is the variance and error sequence after expansion. The closer the value is to 1, the better the uncertainty reflects the prediction error and the better the prediction reliability.

[0045] The Mamba-PINN model is optimized using Bayesian methods. A Bayesian Mamba-PINN model is then constructed based on the model. The uncertainty of the Bayesian Mamba-PINN model prediction is quantified using three indicators: average prediction variance, coverage, and uncertainty-error correlation, thereby ensuring the reliability of the prediction results.

[0046] S105. Predict the electric vehicle trajectory at the target road signalized intersection using the optimized Mamba-PINN fusion model.

[0047] To address the characteristics of electric vehicles and the complex scenarios at signalized intersections, the Mamba-PINN fusion model is optimized using the aforementioned counterfactual scenario and Bayesian methods. The optimized Mamba-PINN fusion model is then used to predict the trajectories of electric vehicles at the target road signalized intersection, effectively improving the accuracy and efficiency of electric vehicle trajectory prediction in complex scenarios. The optimized model exhibits strong adaptability and is easily implemented in engineering, providing decision-making support for engineering applications. Furthermore, it offers core technical support for intelligent traffic signal optimization, ensuring the safety of electric vehicle traffic, and constructing intelligent transportation systems.

[0048] based on Figure 1 The proposed method for predicting electric vehicle (EV) trajectories at signalized intersections utilizes Mamba to capture long-term temporal dependencies in intersection videos and incorporates kinematic physical constraints using PINN, thereby improving the accuracy and efficiency of EV trajectory prediction in complex intersection scenarios. By constructing counterfactual scenario data of trajectory feature parameters and model counterfactual inference, the method quantifies the causal impact of key trajectory prediction factors, enhancing model interpretability. Furthermore, the method optimizes the Mamba-PINN fusion model using counterfactual scenarios and Bayesian methods, achieving interpretability and uncertainty quantification in trajectory prediction, thus improving model reliability. The optimized model exhibits strong adaptability, addresses existing technical pain points, and provides technical support for traffic signal optimization, demonstrating significant engineering application value and theoretical innovation.

[0049] When applying the electric vehicle trajectory prediction method for road signal-controlled intersections provided in this manual, it is not necessary to rely on... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this manual does not impose any restrictions on it.

[0050] The above describes one or more embodiments of a method for predicting the trajectory of an electric vehicle at a road signal-controlled intersection. Based on the same idea, this specification also provides a corresponding system for predicting the trajectory of an electric vehicle at a road signal-controlled intersection, such as... Figure 7 As shown.

[0051] Figure 8 This specification provides a schematic block diagram of an electric vehicle trajectory prediction system for a road signal-controlled intersection. The system includes: The acquisition module 701 is used to acquire video of vehicle operation under different traffic conditions at road signal-controlled intersections; The first generation module 702 is used to identify the target vehicle from the acquired video, extract the trajectory feature parameter data of the target vehicle, train the Mamba-PINN fusion model based on the trajectory feature parameter data, and generate the first predicted trajectory through the trained Mamba-PINN fusion model. The second generation module 703 is used to perform data-oriented intervention based on the trajectory feature parameter data as a benchmark scenario, and after obtaining the counterfactual scenario feature data, input it into the Mamba-PINN fusion model to generate a second predicted trajectory. The optimization module 704 is used to score the importance of trajectory feature parameters based on the first predicted trajectory and the second predicted trajectory, and at the same time, it uses a Bayesian method to introduce uncertainty into the Mamba-PINN fusion model to optimize the model. Prediction module 705 is used to predict the trajectory of electric vehicles at a target road signal-controlled intersection using an optimized Mamba-PINN fusion model.

[0052] Specific limitations regarding the electric vehicle trajectory prediction system for signalized intersections can be found in the limitations of the electric vehicle trajectory prediction method for signalized intersections described above, and will not be repeated here. Each module in the aforementioned electric vehicle trajectory prediction system for signalized intersections can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0053] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1A method for predicting the trajectory of electric vehicles at road signal-controlled intersections is provided.

[0054] This instruction manual also provides Figure 8 The schematic diagram of the computer device shown is as follows: Figure 8 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 A method for predicting the trajectory of electric vehicles at road signal-controlled intersections is provided.

[0055] This specification provides a method and system for predicting the trajectory of electric vehicles at signalized intersections. By utilizing Mamba to capture long-term temporal dependencies in signalized intersection videos and incorporating kinematic physical constraints using PINN, the accuracy and efficiency of electric vehicle trajectory prediction in complex intersection scenarios are improved. Furthermore, by constructing counterfactual scenario data of trajectory feature parameters and using model counterfactual inference prediction, the causal quantification of key influencing factors in trajectory prediction is achieved, enhancing model interpretability. The Mamba-PINN fusion model is optimized using counterfactual scenarios and Bayesian methods to quantify the uncertainty of trajectory prediction, improving model reliability. The optimized model exhibits strong adaptability and can provide decision-making basis and technical support for traffic signal optimization and intelligent transportation systems.

[0056] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0057] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for predicting the trajectory of an electric vehicle at a signalized intersection, characterized in that, include: Collect video footage of vehicle movement under different traffic conditions at road signal-controlled intersections; The target vehicle is identified from the acquired video, and the trajectory feature parameter data of the target vehicle is extracted. Based on the trajectory feature parameter data, a Mamba-PINN fusion model is trained, and a first predicted trajectory is generated through the trained Mamba-PINN fusion model. Using the trajectory feature parameter data as a baseline scenario, data-oriented intervention is performed. After obtaining counterfactual scenario data, it is input into the Mamba-PINN fusion model to generate a second predicted trajectory. The importance of trajectory feature parameters is scored based on the first and second predicted trajectories, and the Mamba-PINN fusion model is optimized using the importance score results. At the same time, Bayesian methods are used to introduce uncertainty into the Mamba-PINN fusion model to optimize the model. Predict the electric vehicle trajectory at the target road signalized intersection using the optimized Mamba-PINN fusion model.

2. The road signal-controlled intersection electric vehicle trajectory prediction method of claim 1, wherein, The Mamba-PINN fusion model is trained using a combined loss function that combines data loss and physical loss.

3. The road signal-controlled intersection electric vehicle trajectory prediction method of claim 1, wherein, The trajectory feature parameters include instantaneous velocity, tangential acceleration, normal acceleration, travel angle, abscissa, and ordinate. The steps for data orientation intervention include: Keep the remaining trajectory feature parameters unchanged; Each trajectory feature parameter data is set to a preset multiple of its original value in turn.

4. The road signal-controlled intersection electric vehicle trajectory prediction method of claim 3, wherein, The step of scoring the importance of trajectory feature parameters based on the first predicted trajectory and the second predicted trajectory specifically includes: Benchmark metrics for the baseline scenario are calculated based on the first predicted trajectory; Calculate the counterfactual index after intervention for each trajectory feature parameter based on the second predicted trajectory; The changes in the baseline index and each counterfactual index are calculated, and the changes are fused using a normalization method to obtain the importance score of each trajectory feature parameter.

5. The road signal-controlled intersection electric vehicle trajectory prediction method of claim 4, wherein, The Mamba-PINN fusion model is optimized using the importance score results, specifically including: The trajectory feature parameters are classified according to their importance scores; Different model weights are assigned to trajectory feature parameters at different levels; The sensitivity of the model to trajectory feature parameters at different levels and the allocation of computational resources are adjusted according to the weights.

6. The road signal-controlled intersection electric vehicle trajectory prediction method of claim 4, wherein, The method of using Bayesian methods to introduce uncertainties into the Mamba-PINN fusion model to optimize the model specifically includes: A Gaussian prior distribution is assigned to each parameter of the Mamba-PINN fusion model; By introducing a Gaussian variational distribution to approximate the posterior distribution of each parameter, the optimal variational parameters of the Mamba-PINN fusion model are obtained by optimizing the Gaussian variational parameters.

7. The road signal-controlled intersection electric vehicle trajectory prediction method of claim 6, wherein, Also includes: Multiple sets of data are sampled from the posterior distribution of each parameter using Monte Carlo sampling, and each set of data is input into the Mamba-PINN fusion model to obtain a set of predicted trajectories; The uncertainty of the model prediction is quantified by calculating three indicators: total average prediction variance, coverage, and uncertainty-error correlation.

8. An electric vehicle trajectory prediction system for a signalized road intersection, characterized by, include: The acquisition module is used to acquire video footage of vehicles operating under different traffic conditions at road signal-controlled intersections; The first generation module is used to identify target vehicles from the collected video, extract trajectory feature parameter data, train a Mamba-PINN fusion model based on the trajectory feature parameter data, and generate a first predicted trajectory through the trained Mamba-PINN fusion model. The second generation module is used to perform data-oriented intervention based on the trajectory feature parameter data as a benchmark scenario, obtain counterfactual scenario feature data, input it into the Mamba-PINN fusion model, and generate a second predicted trajectory. An optimization module is used to score the importance of trajectory feature parameters based on the first predicted trajectory and the second predicted trajectory, and to optimize the Mamba-PINN fusion model using the importance score results. At the same time, a Bayesian method is used to introduce uncertainty into the Mamba-PINN fusion model to optimize the model. The prediction module is used to predict the trajectory of electric vehicles at the target road signal-controlled intersection using an optimized Mamba-PINN fusion model.

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

10. A computer device, comprising: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 7.

Images