Method and system for predicting vehicle speed in dynamic vehicle-following scenario in consideration of traffic environment

Abstract

A method and system for predicting a vehicle speed in a dynamic vehicle-following scenario in consideration of a traffic environment. The method comprises: setting in a natural driving data set a standard and a threshold of a vehicle-following process to extract vehicle-following case data; using an object detection algorithm to acquire traffic environment features related to the vehicle-following case data, extracting vehicle dynamics parameters from the vehicle-following case data, fusing the vehicle dynamics parameters with the traffic environment features, and normalizing same to obtain fused features; using a combined feature selection method to remove redundant variables from the fused features; and dividing into data sets the fused features with the redundant variables removed, training a pre-established Informer-FDR vehicle speed prediction model, and using the trained vehicle speed prediction model to acquire a vehicle speed prediction result. The present invention effectively reflects dynamic interaction features and potential traffic conflicts between drivers and traffic environments, enhances the understanding of complex driving environments by models, and improves the real-time performance and accuracy of vehicle speed prediction.

Description

Vehicle speed prediction methods and systems considering dynamic vehicle following scenarios in traffic environments

[0001] Cross-references to related applications

[0002] This application claims priority to Chinese Patent Application No. 202411552408.X, filed on November 1, 2024, entitled “Method and System for Predicting Vehicle Speed ​​Considering Dynamic Following Scenarios in Traffic Environment”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This invention relates to the field of vehicle speed prediction technology, specifically to a vehicle speed prediction method and system that considers dynamic following scenarios in traffic environments. Background Technology

[0004] Driven by the rapid development of driver assistance systems (ADAS), automotive intelligence has become a crucial solution for reducing traffic accidents and alleviating traffic congestion. Studies have shown that if ADAS can promptly warn drivers, the traffic accident rate can be reduced by 60%, highlighting the critical role of ADAS in improving road safety. Vehicle speed prediction models are particularly important in ADAS; accurately predicting vehicle speed is one of the keys to achieving efficient ADAS and intelligent driving technologies. Furthermore, rear-end collisions account for 60%-70% of traffic accidents. By accurately predicting vehicle speed and behavior, ADAS can better avoid potential collision risks, thereby reducing the incidence of rear-end collisions. Therefore, accurately predicting vehicle speed in following scenarios is of significant research and practical value for improving the effectiveness of ADAS, enhancing human-like driving capabilities, and reducing rear-end collisions.

[0005] However, achieving accurate vehicle speed prediction is a complex task. In following scenarios, drivers are not only affected by the driving behavior of the vehicle in front, but also by the complex interactions of motor vehicles, non-motor vehicles and pedestrians in the surrounding lanes, which makes the speed of the following vehicle show obvious fluctuation characteristics during the following cycle.

[0006] Existing methods rarely consider the impact of dynamic environmental factors on drivers' speed choices during following other vehicles. Furthermore, existing neural networks, in predicting vehicle speed, mostly rely on backtracking linearly significant historical information, resulting in limited receptive fields, computational redundancy, and uncertainty regarding the impact of future information on time-series predictions. Additionally, increasing network depth may slow down model training and degrade model performance. Summary of the Invention

[0007] The purpose of this invention is to address the problems in the prior art by providing a vehicle speed prediction method and system that considers dynamic vehicle following scenarios in traffic environments, thereby improving the real-time performance and accuracy of vehicle speed prediction in complex actual driving environments.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] Firstly, a method for predicting vehicle speed considering dynamic following scenarios in traffic environments is provided, including the following steps:

[0010] By setting standards and thresholds for the following process in the natural driving dataset, following case data is extracted;

[0011] The target detection algorithm is used to obtain traffic environment features related to the vehicle following case data, and the vehicle dynamic parameters in the vehicle following case data are extracted. The vehicle dynamic parameters are fused with the traffic environment features and then normalized to obtain the fused features.

[0012] The combined feature selection method is used to remove redundant variables from the fused features;

[0013] The dataset is divided into fused features after removing redundant variables. A pre-established Informer-FDR vehicle speed prediction model is trained, and the vehicle speed prediction results are obtained using the trained Informer-FDR vehicle speed prediction model.

[0014] As a preferred embodiment, in the step of extracting car-following case data by setting standards and thresholds for the car-following process in the natural driving dataset, the standards and thresholds for the car-following process are as follows, and car-following case data is selected when the following conditions are met simultaneously:

[0015] The relative speed between the two vehicles is less than 2.5 m / s;

[0016] The lateral offset between the front and rear vehicles is less than 2.5m;

[0017] Following the vehicle for more than 10 seconds;

[0018] The extracted vehicle-following case data were divided into two road types: urban roads and highways.

[0019] As a preferred embodiment, the step of using an object detection algorithm to obtain traffic environment features related to vehicle-following case data includes:

[0020] The YOLOv5 object detection algorithm is used to detect other vehicles and traffic participants in adjacent lanes. Traffic density is defined based on the number of vehicles traveling in the same direction as the vehicle, as captured by the camera. Low traffic density is defined as 1-2 vehicles traveling in the same direction, medium traffic density as 3-4 vehicles traveling in the same direction, and high traffic density as more than 4 vehicles traveling in the same direction. Pedestrians and cyclists identified by the object detection algorithm are merged into a unified category, defined as vulnerable traffic participants, and used as input features for the Informer-FDR speed prediction model.

[0021] As a preferred approach, when extracting vehicle dynamics parameters from the following case data, the vehicle's own driving information is read via the CAN bus. The vehicle's own driving information includes speed, acceleration, relative distance, accelerator pedal opening, and headway.

[0022] As a preferred embodiment, the method of using combined feature selection to remove redundant variables in the fused features includes:

[0023] The correlation between candidate input features and vehicle speed is calculated using Message Integrity Check (MIC), Classification and Regression Tree (CART), Random Forest (RF), and Gradient Boosting Ensemble Learning (XGBoost) algorithms.

[0024] The average value of each candidate variable is calculated by summing the results of four different algorithms and dividing by the total number of algorithms. All variables are calculated and sorted in turn. Based on the ranking, the top few variables with the highest importance are selected as the input of the model.

[0025] As a preferred embodiment, the Informer-FDR vehicle speed prediction model is established using the following Informer deep learning model:

[0026] In the fixed-window-size scrolling prediction setting, the input for time point t is:

[0027] In the formula, L is the sample sequence length, D is the feature dimension, and in X, a single data sample is:

[0028] In the formula, This represents the input data of the i-th model at the j-th time step;

[0029] Using vehicle speed as the output indicator:

[0030] In the formula, y i This represents the output of the i-th model at each time step;

[0031] The encoder of the Informer deep learning model is composed of a multi-head sparse probabilistic self-attention mechanism and a distillation operation stacked together:

[0032] in, It is a sparse matrix containing sparsity measurements; Q and K generated during encoder training represent the query vector and key vector, respectively, V represents the value vector, and d represents the input dimension; the sparse probabilistic self-attention mechanism is defined as:

[0033] In the formula, L K 1 / L K =q(k j |q i Define the i-th query for sparsity measurement as q. i ;q i and k j They come from Q and K respectively;

[0034] The Informer deep learning model adds a convolutional layer and a max-pooling layer between each pair of self-attention layers to reduce the input length. After the previous self-attention layer, a convolutional layer with a kernel size of 3 and a stride of 1 is used to enhance the perception of contextual information features. Subsequently, a max-pooling layer with a kernel size of 3 and a stride of 2 is used to prioritize the processing of key features, thereby generating smaller but more focused feature maps for subsequent self-attention layers.

[0035] In the formula, sparse probability self-attention is included in [·]. AB In the middle; 1-D convolutional filtering is implemented through Conv(·), and the activation function is ELU(·); the encoder combines sparse probabilistic self-attention and distillation operation to generate feature maps;

[0036] The time series input to the decoder of the Informer deep learning model It is divided into two parts: the known sequence preceding the predicted point. And forecast sequences that need to mask future data

[0037] In the formula, L token L is the length of the starting marker sequence. y The length of the predicted sequence; d model For model dimensions;

[0038] The position of each input information in the sequence is encoded using the following formula:

[0039] In the formula, pos represents the position of the feature vector, and dmodel The feature vector dimension is represented by the encoding of time information in time series prediction, which is based on the importance of time information. The location and time information are integrated and input into the model.

[0040] As a preferred embodiment, the Informer-FDR vehicle speed prediction model replaces the max pooling layer of the distillation layer in the decoder of the Informer deep learning model with an average pooling layer, and replaces the convolutional layer with a dilated causal convolution. The dilated causal convolution consists of two parts: causal convolution and dilated convolution. The causal convolution is subject to time constraints, and the output at time t is convolved only with the previous time t and earlier elements to ensure causality in time-series prediction, so that the network's output information is only affected by past input information.

[0041] The dilated convolution operation is as follows:

[0042] In the formula, K is the kernel size, s are the elements of X, and F i t represents the weights of the convolution kernel, t is the time step, and td·i represents the past direction.

[0043] As a preferred embodiment, the Informer-FDR vehicle speed prediction model introduces PRELU as the activation function for the distillation layer. PRELU pushes the output mean towards zero. The improved distillation operation is as follows:

[0044] In the formula, sparse probability self-attention is included in [·]. AB In the middle, the dilated causal convolution is implemented by DCConv(·), with the activation function being PReLU(·), and the final output is generated by AvgPool(·).

[0045] As a preferred embodiment, the Informer-FDR vehicle speed prediction model adds a residual connection module, which consists of three sparse probabilistic self-attention layers and two distillation layers. In order to ensure that the feature maps output by these two layers have the same shape when combined, a 1*1 convolution is used to merge the output of the first sparse probabilistic self-attention layer and the output of the third sparse probabilistic self-attention layer in the main encoder, thereby ensuring the effective transmission of feature information.

[0046] Secondly, a vehicle speed prediction system considering dynamic following scenarios in traffic environments is provided, including:

[0047] The vehicle-following case data extraction module is used to extract vehicle-following case data by setting standards and thresholds for the vehicle-following process in the natural driving dataset.

[0048] The feature fusion and normalization module is used to obtain traffic environment features related to vehicle-following case data using target detection algorithms, extract vehicle dynamic parameters from the vehicle-following case data, fuse vehicle dynamic parameters with traffic environment features, and obtain fused features after normalization.

[0049] The redundant variable removal module is used to remove redundant variables from fused features using a combined feature selection method.

[0050] The vehicle speed prediction model training and prediction output module is used to divide the dataset into fused features after removing redundant variables, train the pre-built Informer-FDR vehicle speed prediction model, and obtain the vehicle speed prediction result using the trained Informer-FDR vehicle speed prediction model.

[0051] Compared with the prior art, the present invention has at least the following beneficial effects:

[0052] By setting standards and thresholds for the following process in the natural driving dataset, following case data was effectively extracted, enabling targeted mining of following data and thus a more comprehensive understanding and analysis of following dynamic behavior in the driving environment. Object detection algorithms were used to obtain traffic environment features related to the following case data, and vehicle dynamic parameters were extracted from the data. These vehicle dynamic parameters were then fused with traffic environment features. By fusing these features, the dynamic interaction between the driver and the traffic environment and potential traffic conflicts were effectively reflected, enhancing the model's understanding of complex driving environments. The fused features, after removing redundant variables, were divided into datasets to train a pre-established Informer-FDR speed prediction model. The trained Informer-FDR speed prediction model was then used to obtain speed prediction results. Compared with other models, the Informer-FDR speed prediction model used in this invention exhibits the best prediction performance, more accurately capturing the fluctuation trend of the data sequence and showing a high degree of fit with actual data. In particular, the Informer-FDR speed prediction model has a small prediction bias at extreme points, indicating that it can more accurately capture random situations in traffic conditions.

[0053] Furthermore, this invention reduces the high computational complexity of the Informer-FDR vehicle speed prediction model by using a sparse probabilistic self-attention mechanism, which helps to solve the difficulty of applying Transformer-type deep learning models to in-vehicle platforms.

[0054] Furthermore, the Informer-FDR vehicle speed prediction model established in this invention considers the impact of dynamic traffic environment on vehicle speed fluctuations, realizes global feature extraction over long following cycles, and solves the degradation problem of deep networks.

[0055] Furthermore, this invention introduces extended causal convolution, which expands the receptive field of the Informer-FDR vehicle speed prediction model, ensuring the causality of time prediction and avoiding the loss and leakage of extreme location information.

[0056] Furthermore, the Informer-FDR vehicle speed prediction model of this invention adds a residual connection module, which accelerates the convergence of the Informer-FDR vehicle speed prediction model and enhances the stability of training. Attached Figure Description

[0057] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 is a flowchart of the vehicle speed prediction method under the dynamic following scenario of traffic environment in an embodiment of the present invention.

[0059] Figure 2 is a network structure diagram of the Informer-FDR vehicle speed prediction model established in an embodiment of the present invention;

[0060] Figure 3 is a diagram of the expanded causal convolution structure according to an embodiment of the present invention;

[0061] Figure 4 shows the test results of the vehicle speed prediction method under the dynamic following scenario in traffic environment according to the embodiment of the present invention.

[0062] Figure 5 is a comparison of the prediction performance of the Informer-FDR vehicle speed prediction model in this embodiment of the invention with other algorithm models. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, those skilled in the art can obtain other embodiments without creative effort.

[0064] As shown in Figure 1, this embodiment of the invention proposes a vehicle speed prediction method considering dynamic vehicle following scenarios in traffic environments. The method mainly includes: firstly, by setting standards and thresholds for the following process in a natural driving dataset, effective vehicle following case data is extracted; secondly, a target detection algorithm is used to obtain relevant environmental features, and vehicle dynamic parameters under the following segment are extracted from the CAN bus and fused with environmental features; then, a combined feature selection method is used to eliminate redundant variables, improve model prediction accuracy, and perform normalization processing; finally, the dataset is divided into training, testing, and validation sets, input into a pre-established Informer-FDR vehicle speed prediction model, and trained and tested to obtain the final vehicle speed prediction result. The complete vehicle speed prediction method of this embodiment of the invention mainly consists of four steps: acquiring following data, acquiring traffic environment information based on target detection algorithms, feature fusion and elimination of redundant variables, and establishing an Informer-FDR vehicle speed prediction model for vehicle speed prediction.

[0065] Step 1: Obtaining vehicle tracking data, specifically including the following steps:

[0066] Step 1.1: Install a comprehensive data acquisition system consisting of four cameras in the vehicle, including: a facial recognition camera, a pedal camera, a front-facing camera, and a wide-angle front-facing camera.

[0067] Step 1.2: Video data captured by the front-facing camera is used to identify traffic participants, while data related to the actual movement of the vehicle, such as speed, acceleration, steering wheel angle, and mileage, are extracted from the CAN-BUS system.

[0068] Step 1.3: Extract follow-up case data by setting standards and thresholds for the follow-up process. All follow-up case data must meet the specified thresholds. If any of these thresholds are not met, the corresponding event will not be included in the initial follow-up case database. The set thresholds are:

[0069] 1) The relative speed between the front and rear vehicles is less than 2.5 m / s;

[0070] 2) The lateral offset between the front and rear vehicles is less than 2.5m;

[0071] 3) The following time is greater than 10 seconds.

[0072] Furthermore, since there are significant differences in traffic conditions and driving behavior between urban roads and highways, resulting in differences in car following behavior, this embodiment of the invention divides car following cases into two road types: urban roads and highways. These serve as input features for the model, enabling the model to better understand and learn car following behavior under different road conditions.

[0073] Step 2: Obtaining traffic environment information based on object detection algorithms, specifically including the following steps:

[0074] Step 2.1: Use the YOLOv5 object detection algorithm to detect other vehicles and traffic participants in adjacent lanes. Based on the number of vehicles traveling in the same direction as the vehicle collected by the front camera, define the traffic density. Define the situation where the number of vehicles traveling in the same direction is 3 to 4 as medium traffic density, and define the situation where the number of vehicles traveling in the same direction is greater than 4 as high traffic density.

[0075] Step 2.2: In traffic environments, drivers are often influenced by individuals who may face higher risks on the road, thus affecting their driving behavior. Therefore, we merge pedestrians and cyclists identified by object detection algorithms into a unified category, named Vulnerable Traffic Participants, as input features for the model.

[0076] Step 3: Feature fusion and elimination of redundant variables, specifically including the following steps:

[0077] Step 3.1: The vehicle dynamics parameters extracted from the CAN-BUS system, including speed, acceleration, relative distance, accelerator pedal opening, and headway, are fused with the traffic environment information obtained from the target detection algorithm and normalized.

[0078] Step 3.2: Message Integrity Check (MIC), Classification and Regression Tree (CART), Random Forest (RF), and Gradient Boosting Ensemble Learning (XGBoost) algorithms are used to calculate the correlation between candidate input features and vehicle speed. The average value for each candidate variable is calculated by summing the correlation results of the four different algorithms and dividing by the total number of algorithms. All variables are calculated sequentially and ranked. Based on the ranking, the seven most important variables are selected as the model input.

[0079] Step 4: Establish an Informer-FDR vehicle speed prediction model for vehicle speed prediction. The Informer-FDR vehicle speed prediction model includes: an Informer time series prediction model, an dilated causal convolution module, and a residual connection module.

[0080] The Informer time series prediction model is an improvement on the Transformer model. It uses a multi-head probabilistic sparse self-attention mechanism to assign greater weights to important features, reducing the time complexity of the network model. At the same time, it uses distillation operations between self-attention layers to reduce the input length of the next layer model, and uses generative inference decoding to improve prediction speed, thus solving the problem of long-term dependencies in time series data.

[0081] The Informer-FDR model in this embodiment of the invention is an improvement on the Informer model. It replaces the conventional convolutional layers in the distillation layers between attention blocks in the original encoder with dilated causal convolutions, and introduces average pooling layers to improve model performance. The activation function of the distillation layers is set to the parameterized modified linear unit PRELU to accelerate model convergence. Furthermore, residual connection modules are added to allow information to be transferred across layer depths. This design not only improves training stability but also mitigates the degradation problem of deep networks. The overall network structure is shown in Figure 2.

[0082] Step 4.1: Build the Informer deep learning model.

[0083] The Informer time series prediction model uses a multi-head probabilistic sparse self-attention mechanism to assign greater weights to important features, reducing the time complexity of the network model. At the same time, it uses distillation operations between self-attention layers to reduce the input length of the next layer model, and uses generative inference decoding to improve prediction speed, thus solving the problem of long-term dependencies in time series data.

[0084] In the fixed-window-size scrolling prediction setting, the input for time point t is:

[0085] In the formula, L is the sample sequence length, D is the feature dimension, and in X, a single data sample is:

[0086] In the formula, This represents the input data of the i-th model at the j-th time step. Generally, vehicle speed is used as the output metric, and the label Y is denoted as:

[0087] In the formula, y i This represents the output of the i-th model at each time step.

[0088] In Informer, the encoder is composed of a multi-head sparse probabilistic self-attention mechanism and a distillation operation. Traditional self-attention mechanisms require double dot product calculations, which is a major factor contributing to their slow prediction training speed. The sparse probabilistic self-attention mechanism, on the other hand, reduces time complexity and memory usage by selecting a few high-scoring dot product pairs and ignoring the others with lower scores. The specific operation is as follows:

[0089] in, It is a sparse matrix containing sparsity measurements; Q and K, generated during encoder training, represent the query vector and key vector, respectively. Furthermore, V represents the value vector, and d represents the input dimension.

[0090] The sparse probabilistic self-attention mechanism is defined as follows:

[0091] In the formula, L K 1 / L K =q(k j |q i Define the i-th query for sparsity measurement as q. i ;q i and k j The input lengths of the query vector and the key vector are respectively derived from Q and K. In practice, the input lengths of the query vector and the key vector are usually equal in the self-attention computation, making the total time and space complexity of the sparse probability self-attention O(LlnL). The sparse probability self-attention selects the results with higher scores in the sparsity metric and computes the corresponding scores.

[0092] Informer adds a convolutional layer and a max-pooling layer between each pair of self-attention layers to reduce the input length. After the previous self-attention layer, a convolutional layer with a kernel size of 3 and a stride of 1 is used to enhance the perception of contextual features. Subsequently, a max-pooling layer with a kernel size of 3 and a stride of 2 is used to prioritize the processing of key features, thereby generating smaller but more focused feature maps for subsequent self-attention layers.

[0093] In the formula, sparse probability self-attention is included in [·]. AB In the 1-D model, convolutional filtering is implemented using Conv(·), with the activation function being ELU(·). The encoder combines sparse probabilistic self-attention and distillation operations to generate feature maps.

[0094] The time series input to the decoder It is divided into two parts: the known sequence preceding the predicted point. And forecast sequences that need to mask future data The formula is as follows:

[0095] In the formula, L token L is the length of the starting marker sequence. y The length of the predicted sequence; d model For model dimensions.

[0096] In the vehicle speed prediction problem, time series information represents the vehicle's operating state at each moment, and the order of the data is crucial. Since the Informer model does not have iterative operations like the RNN model, it is necessary to encode the position of each input information in the sequence. The specific process is shown in the formula:

[0097] In the formula: pos represents the position of the feature vector, d model The feature vector dimension is represented by the encoding of time information in time series prediction, which is based on the importance of time information. The location and time information are integrated and input into the model.

[0098] To address several key drawbacks of the Informer model when applied to vehicle speed prediction in a following scenario: traditional convolutional layers can only recall historical information of linear size; stacking self-attention blocks with standard convolutional layers may lead to redundant and meaningless computations and result in the leakage of future information, this invention solves these problems by introducing dilated causal convolution.

[0099] Step 4.2: Replace the max pooling layer in the distillation layer of the Informer model's decoder with an average pooling layer, and replace the regular convolutional layer with a dilated causal convolution. The structure of the dilated causal convolution is shown in Figure 3.

[0100] Dilated causal convolution consists of two parts: causal convolution and dilated convolution. Unlike traditional convolution, causal convolution is subject to time constraints. It convolves the output at time t only with elements from the previous layer at time t and earlier elements to ensure causality in time-series prediction, ensuring that the network's output is only influenced by past input information. The dilated convolution operation is shown below:

[0101] In the formula, K is the kernel size, s are the elements of X, and F i t represents the weights of the convolution kernel, t is the time step, and td·i represents the past direction.

[0102] Building upon this, a parameterized modified linear unit (PRELU) is introduced as the activation function for the distillation layer. Compared to the exponential linear unit (ELU), the parameterized PRELU pushes the output mean closer to zero without significantly increasing computational complexity or overfitting risk, thus promoting faster model convergence. The improved distillation operation is shown in the equation:

[0103] In the formula, the sparse probability ProbSparse self-attention is included in [·]. AB In the middle, the dilated causal convolution is implemented by DCConv(·), with the activation function being PReLU(·), and the final output is generated by AvgPool(·).

[0104] Step 4.3: Add a residual connection module, which consists of three sparse probabilistic self-attention layers and two distillation layers. To ensure that the feature maps output by these two layers have the same shape when combined, this embodiment of the invention uses a 1*1 convolution to merge the outputs of the first and third sparse probabilistic self-attention layers in the main encoder, ensuring effective transmission of feature information.

[0105] To verify the vehicle speed prediction performance of the method described in this invention in a following scenario, the method was trained using the following dataset constructed in this embodiment to predict vehicle speed after 2 seconds. The absolute error between the actual and predicted values ​​for each frame was calculated, and the test results and errors are shown in Figure 4. As can be seen from the figure, the prediction error of the method in this embodiment remains consistently low, with only a few significant errors occurring, indicating a high degree of fit with the actual vehicle speed.

[0106] Furthermore, the superiority of this invention is verified by comparing it with other mainstream time-series prediction methods. Figure 5 shows a comparison of the performance of the method described in this invention with other vehicle speed prediction methods in a specific vehicle-following segment. As can be seen from the figure, GRU's prediction performance is poor, while Autoformer, Transformer, and LSTM can capture the trend of data fluctuations and some extreme values, but they exhibit a certain degree of bias in prediction. Compared with other models, the Informer-FDR model has the best prediction performance, more accurately capturing the fluctuation trend of the data sequence, and has a higher degree of fit with actual data. In particular, the Informer-FDR model has a smaller prediction bias at extreme points, indicating that it can more accurately capture random situations in traffic conditions.

[0107] Another embodiment of the present invention proposes a vehicle speed prediction system considering dynamic following scenarios in traffic environments, comprising:

[0108] The vehicle-following case data extraction module is used to extract vehicle-following case data by setting standards and thresholds for the vehicle-following process in the natural driving dataset.

[0109] The feature fusion and normalization module is used to obtain traffic environment features related to vehicle-following case data using target detection algorithms, extract vehicle dynamic parameters from the vehicle-following case data, fuse vehicle dynamic parameters with traffic environment features, and obtain fused features after normalization.

[0110] The redundant variable removal module is used to remove redundant variables from fused features using a combined feature selection method.

[0111] The vehicle speed prediction model training and prediction output module is used to divide the dataset into fused features after removing redundant variables, train the pre-built Informer-FDR vehicle speed prediction model, and obtain the vehicle speed prediction result using the trained Informer-FDR vehicle speed prediction model.

[0112] Another embodiment of the present invention provides an electronic device, comprising: a memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the vehicle speed prediction method considering a dynamic following scenario in a traffic environment.

[0113] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the vehicle speed prediction method considering dynamic following scenarios in traffic environments.

[0114] For example, the instructions stored in the memory can be divided into one or more modules / units. These modules / units are stored in a computer-readable storage medium and executed by the processor to complete the vehicle speed prediction method for dynamic following scenarios considering traffic conditions as described in this invention. The one or more modules / units can be a series of computer-readable instruction segments capable of performing specific functions, which describe the execution process of the computer program on the server.

[0115] The electronic device may be a smartphone, laptop, PDA, or cloud server, among other computing devices. It may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the electronic device may also include more or fewer components, or combinations of certain components, or different components; for example, it may also include input / output devices, network access devices, buses, etc.

[0116] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0117] The memory can be an internal storage unit of the server, such as a hard drive or RAM. Alternatively, it can be an external storage device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory can include both internal and external storage units. The memory is used to store computer-readable instructions and other programs and data required by the server. It can also be used to temporarily store data that has been output or will be output.

[0118] It should be noted that the information interaction and execution process between the above-mentioned module units are based on the same concept as the method embodiment. For details on their specific functions and technical effects, please refer to the method embodiment section. They will not be repeated here.

[0119] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, 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 computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0121] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0122] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for predicting vehicle speed considering dynamic following scenarios in traffic environments, characterized in that, Includes the following steps: By setting standards and thresholds for the following process in the natural driving dataset, following case data is extracted; The target detection algorithm is used to obtain traffic environment features related to the vehicle following case data, and the vehicle dynamic parameters in the vehicle following case data are extracted. The vehicle dynamic parameters are fused with the traffic environment features and then normalized to obtain the fused features. The combined feature selection method is used to remove redundant variables from the fused features; The dataset is divided into fused features after removing redundant variables. A pre-established Informer-FDR vehicle speed prediction model is trained, and the vehicle speed prediction results are obtained using the trained Informer-FDR vehicle speed prediction model. The Informer-FDR vehicle speed prediction model replaces the max pooling layer of the distillation layer in the decoder of the Informer deep learning model with an average pooling layer, and changes the convolutional layer to a dilated causal convolution. The dilated causal convolution consists of two parts: causal convolution and dilated convolution. The causal convolution is subject to time constraints, and the output at time t is convolved only with the previous time t and earlier elements to ensure causality in time-series prediction, so that the network's output information is only affected by past input information. The dilated convolution operation is as follows: In the formula, K is the kernel size, s are the elements of X, and F i t represents the weights of the convolution kernel, t is the time step, and td·i represents the past direction. The Informer-FDR vehicle speed prediction model introduces PRELU as the activation function for the distillation layer. PRELU pushes the output mean towards zero. The improved distillation operation is as follows: In the formula, sparse probability self-attention is included in [·]. AB In the middle, the dilated causal convolution is implemented by DCConv(·), with the activation function being PReLU(·), and the final output is generated by AvgPool(·). The Informer-FDR vehicle speed prediction model adds a residual connection module, which consists of three sparse probabilistic self-attention layers and two distillation layers. In order to ensure that the feature maps output by these two layers have the same shape when they are added together, a 1*1 convolution is used to merge the output of the first sparse probabilistic self-attention layer and the output of the third sparse probabilistic self-attention layer in the main encoder to ensure the effective transmission of feature information.

2. The vehicle speed prediction method considering dynamic following scenarios in traffic environment as described in claim 1, characterized in that, In the step of extracting car-following case data by setting standards and thresholds for the car-following process in the natural driving dataset, the standards and thresholds for the car-following process are as follows, and car-following case data is selected when the following conditions are met: The relative speed between the two vehicles is less than 2.5 m / s; The lateral offset between the front and rear vehicles is less than 2.5m; Following the vehicle for more than 10 seconds; The extracted vehicle-following case data were divided into two road types: urban roads and highways.

3. The vehicle speed prediction method considering dynamic following scenarios in traffic environment as described in claim 1, characterized in that, The steps for obtaining traffic environment features related to vehicle-following case data using object detection algorithms include: The YOLOv5 object detection algorithm is used to detect other vehicles and traffic participants in adjacent lanes. Traffic density is defined based on the number of vehicles traveling in the same direction as the vehicle, as captured by the camera. Low traffic density is defined as 1-2 vehicles traveling in the same direction, medium traffic density as 3-4 vehicles traveling in the same direction, and high traffic density as more than 4 vehicles traveling in the same direction. Pedestrians and cyclists identified by the object detection algorithm are merged into a unified category, defined as vulnerable traffic participants, and used as input features for the Informer-FDR speed prediction model.

4. The vehicle speed prediction method considering dynamic following scenarios in traffic environment as described in claim 1, characterized in that, When extracting vehicle dynamics parameters from the following case data, the vehicle's own driving information is read through the CAN bus. The vehicle's own driving information includes speed, acceleration, relative distance, accelerator pedal opening, and headway.

5. The vehicle speed prediction method considering dynamic vehicle following scenarios in traffic environment according to claim 1, characterized in that, The method of removing redundant variables from fused features using combined feature selection includes: The correlation between candidate input features and vehicle speed is calculated using Message Integrity Check (MIC), Classification and Regression Tree (CART), Random Forest (RF), and Gradient Boosting Ensemble Learning (XGBoost) algorithms. The average value of each candidate variable is calculated by summing the results of four different algorithms and dividing by the total number of algorithms. All variables are calculated and sorted in turn. Based on the ranking, the top few variables with the highest importance are selected as the input of the model.

6. The vehicle speed prediction method considering dynamic following scenarios in traffic environment as described in claim 1, characterized in that, The Informer-FDR vehicle speed prediction model is built as an Informer deep learning model in the following manner: In the fixed-window-size scrolling prediction setting, the input for time point t is: In the formula, L is the sample sequence length, D is the feature dimension, and in X, a single data sample is: In the formula, This represents the input data of the i-th model at the j-th time step; Using vehicle speed as the output indicator: In the formula, y i This represents the output of the i-th model at each time step; The encoder of the Informer deep learning model is composed of a multi-head sparse probabilistic self-attention mechanism and a distillation operation stacked together: in, It is a sparse matrix containing sparsity measurements; Q and K generated during encoder training represent the query vector and key vector, respectively, V represents the value vector, and d represents the input dimension; the sparse probabilistic self-attention mechanism is defined as: In the formula, L K 1 / L K =q(k j |q i Define the i-th query for sparsity measurement as q. i ;q i and k j They come from Q and K respectively; The Informer deep learning model adds a convolutional layer and a max-pooling layer between each pair of self-attention layers to reduce the input length. After the previous self-attention layer, a convolutional layer with a kernel size of 3 and a stride of 1 is used to enhance the perception of contextual information features. Subsequently, a max-pooling layer with a kernel size of 3 and a stride of 2 is used to prioritize the processing of key features, thereby generating smaller but more focused feature maps for subsequent self-attention layers. In the formula, sparse probability self-attention is included in [·]. AB In the middle; 1-D convolutional filtering is implemented through Conv(·), and the activation function is ELU(·); the encoder combines sparse probabilistic self-attention and distillation operation to generate feature maps; The time series input to the decoder of the Informer deep learning model It is divided into two parts: the known sequence preceding the predicted point. And forecast sequences that need to mask future data In the formula, L token L is the length of the starting marker sequence. y The length of the predicted sequence; d model For model dimensions; The position of each input information in the sequence is encoded using the following formula: In the formula, pos represents the position of the feature vector, and d model The feature vector dimension is represented by the encoding of time information in time series prediction, which is based on the importance of time information. The location and time information are integrated and input into the model.

7. A vehicle speed prediction system considering dynamic vehicle following scenarios in traffic environments, characterized in that, include: The vehicle-following case data extraction module is used to extract vehicle-following case data by setting standards and thresholds for the vehicle-following process in the natural driving dataset. The feature fusion and normalization module is used to obtain traffic environment features related to vehicle-following case data using target detection algorithms, extract vehicle dynamic parameters from the vehicle-following case data, fuse vehicle dynamic parameters with traffic environment features, and obtain fused features after normalization. The redundant variable removal module is used to remove redundant variables from fused features using a combined feature selection method. The vehicle speed prediction model training and prediction output module is used to divide the dataset into fused features after removing redundant variables, train the pre-built Informer-FDR vehicle speed prediction model, and obtain the vehicle speed prediction result using the trained Informer-FDR vehicle speed prediction model. The Informer-FDR vehicle speed prediction model replaces the max pooling layer of the distillation layer in the decoder of the Informer deep learning model with an average pooling layer, and changes the convolutional layer to a dilated causal convolution. The dilated causal convolution consists of two parts: causal convolution and dilated convolution. The causal convolution is subject to time constraints, and the output at time t is convolved only with the previous time t and earlier elements to ensure causality in time-series prediction, so that the network's output information is only affected by past input information. The dilated convolution operation is as follows: In the formula, K is the kernel size, s are the elements of X, and F i t represents the weights of the convolution kernel, t is the time step, and td·i represents the past direction. The Informer-FDR vehicle speed prediction model introduces PRELU as the activation function for the distillation layer. PRELU pushes the output mean towards zero. The improved distillation operation is as follows: In the formula, sparse probability self-attention is included in [·]. AB In the middle, the dilated causal convolution is implemented by DCConv(·), with the activation function being PReLU(·), and the final output is generated by AvgPool(·). The Informer-FDR vehicle speed prediction model adds a residual connection module, which consists of three sparse probabilistic self-attention layers and two distillation layers. In order to ensure that the feature maps output by these two layers have the same shape when they are added together, a 1*1 convolution is used to merge the output of the first sparse probabilistic self-attention layer and the output of the third sparse probabilistic self-attention layer in the main encoder to ensure the effective transmission of feature information.

Images