An automatic driving vehicle entry curve deceleration prediction method based on LSTM

Abstract

The application discloses an automatic driving vehicle entry curve deceleration prediction method based on LSTM, and aims at solving the safety and stability problems of the automatic driving vehicle in the curve driving process.The application relates to the field of automatic driving.The application comprises a data acquisition module, a data preprocessing module, a network training module and a real-time prediction module.The data acquisition module is responsible for collecting vehicle state information and road information, and outputs the information to the data preprocessing module.The data preprocessing module performs normalization processing on the vehicle state information and the road information, extracts input-output pairs by using a sliding window, and converts feature data and label data into tensor data suitable for the input of the LSTM.The network training module trains an LSTM model according to the tensor data.The real-time prediction module loads the optimal LSTM model to perform prediction, and predicts the entry curve deceleration of the automatic driving vehicle, so as to realize the safe and stable entry curve control of the automatic driving vehicle.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of automatic driving, and particularly relates to an automatic driving vehicle entry curve deceleration prediction method based on LSTM. BACKGROUND

[0002] With the rapid development of artificial intelligence, deep learning, sensor fusion and high-precision positioning technologies, automatic driving vehicle technology is gradually moving from the research and development stage to the actual road application stage. Existing automatic driving systems usually integrate laser radar, millimeter wave radar, camera, inertial navigation unit (IMU) and global positioning system (GPS) to realize the perception of the environment, the planning of the path and the motion control of the vehicle.

[0003] In actual road driving, the vehicle usually needs to implement deceleration control in advance before entering a curve to reduce the influence of centrifugal force on the stability of the vehicle, and to avoid the phenomenon of side slip and instability of the vehicle in the curve, so as to ensure the driving safety and stability during the curve passing process. During the driving process of the vehicle, the driver can make a comprehensive judgment on the curve radius, road adhesion condition and other factors based on the visual perception of the front road and driving experience, so as to make a deceleration decision in advance and realize safe curve passing. However, for the automatic driving vehicle, it is impossible to decide a reasonable entry curve deceleration based on human driving experience, which may lead to the instability of the automatic driving vehicle.

[0004] At present, acceleration vector control technology (GVC) is adopted in some mass-produced vehicle models. By monitoring the steering wheel angle and acceleration and other states, the engine output is actively adjusted to realize the transfer of load on the longitudinal axis of the vehicle, and the responsiveness and handling performance of the vehicle when entering the curve are improved. In specific implementation, the GVC system constructs an empirical formula between the lateral acceleration and the longitudinal acceleration of the vehicle based on the driving data of the driver, calculates the required entry curve deceleration a y based on the current lateral acceleration a x , and controls the engine torque output accordingly to assist the vehicle to smoothly enter the curve. However, the control strategy of GVC depends on the current state, and cannot utilize the time series information of the historical state, so it cannot predict the entry curve deceleration of the vehicle in advance, and there will be a certain response lag in vehicle control. It is difficult to realize precise control when facing complex and variable curve road conditions. Therefore, it is necessary to combine human driving data and road information to design a time series-based entry curve deceleration prediction method to accurately predict the required deceleration when entering the curve in advance, so as to provide a deceleration reference for the entry curve deceleration control of the automatic driving vehicle, and realize the safe and stable operation of the automatic driving vehicle in the curve scene. However, in the existing research, there is no relevant research on the prediction of the entry curve deceleration of the automatic driving vehicle based on the time series of the vehicle state and the road information.

[0005] Currently, some research has attempted to predict vehicle deceleration or steering adaptability in curves. For example, patent CN10946860A proposes a curve speed prediction method based on driving style. This method constructs a driving style evaluation system and combines it with fuzzy comprehensive evaluation to determine the driver type, thereby adjusting the target speed. However, this method is based on the current state of vehicle dynamics, road environment, and driving style characteristics, and does not consider the influence of temporal features on deceleration behavior. The prediction results rely more on empirical rules and are difficult to adapt to complex and dynamically changing road scenarios. Long Short-Term Memory (LSTM) networks are widely used in time-series prediction tasks in autonomous driving systems due to their advantages in temporal feature extraction and modeling. This model can mine potential temporal dependencies based on historical vehicle states, road parameters, and multi-source sensor inputs, thereby achieving dynamic prediction of vehicle deceleration upon entering a curve, providing a reference for the control system to adjust speed and path in advance. Compared with traditional control strategies, LSTM models exhibit higher generalization ability and control accuracy in dealing with complex curve changes, sudden operating conditions, and nonlinear dynamic characteristics of the system. Patent CN 109783843 A considers the time-series information of vehicle states to predict vehicle speed, thereby improving the modeling accuracy of vehicle dynamics models in simulation platforms. However, since patent CN 109783843 A does not consider road curvature information in its speed prediction, it is not suitable for cornering conditions and cannot predict deceleration upon entering a curve. Furthermore, directly differentiating the speed predicted by CN 109783843 A to obtain acceleration results in information fluctuations due to the derivative, making it difficult to guarantee the vehicle's deceleration upon entering a curve function. Therefore, patent CN 109783843 A and this invention are fundamentally different in technology. Summary of the Invention:

[0006] To address the shortcomings of existing technologies and solve the problems mentioned in the background, this invention provides an LSTM-based method for predicting deceleration upon entering a curve for autonomous vehicles. This method employs a modular design, utilizing an LSTM network to perform temporal modeling of vehicle state and road information, enabling early prediction of deceleration upon entering a curve and improving the system's responsiveness and control accuracy to dynamic changes in curves. Compared to traditional rule-based or threshold-based control methods, and strategies such as GVC that rely on engine output adjustment, this method offers significant advantages in modeling accuracy and adaptability to complex curve conditions. Through a deep learning prediction mechanism, this method exhibits stronger generalization ability and robustness, significantly improving the operational stability and control foresight of autonomous vehicles in curves.

[0007] The technical solution adopted by this invention to solve the technical problem is as follows:

[0008] The application discloses an automatic driving vehicle entry curve deceleration prediction method based on LSTM, and aims at solving the safety and stability problems of the automatic driving vehicle in the curve driving process. The application relates to the field of automatic driving. The application comprises a data collection module, a data preprocessing module, a network training module and a real-time prediction module. The data collection module is responsible for collecting vehicle state information and road information, and outputs the information to the data preprocessing module. The data preprocessing module performs normalization processing on the data collected by the data collection module, extracts input-output pairs by using a sliding window, and converts feature data and label data into tensor data suitable for the input of the LSTM. The network training module trains the LSTM model according to the tensor data, the real-time prediction module loads the optimal LSTM model to perform prediction, and the entry curve deceleration of the automatic driving vehicle is predicted.

[0009] The method comprises the following steps:

[0010] Step 1: a data collection module is designed to collect vehicle state information of an automatic driving vehicle when entering a curve, including lateral displacement y, lateral velocity v y , lateral acceleration a y , yaw rate γ, front wheel steering angle δ, and road information including heading angle ψ and road curvature κ, wherein the above information is taken as feature data of the LSTM network model and is input to the data preprocessing module, and the expected entry curve deceleration a x is taken as label data for subsequent entry curve deceleration prediction.

[0011] Step 2: a data preprocessing module is designed, and the process comprises the following substeps:

[0012] Step 2.1: the feature data and the label data are normalized by using a MinMax normalization method, and the data is scaled to the range of [0, 1] to make the model training more stable and efficient, wherein the MinMax normalization method for data preprocessing is shown in formula (1):

[0013]

[0014] Wherein, X norm is the normalized data, X is the original data collected by the data collection module, X max is the maximum value of the feature data, and X min is the minimum value of the feature data.

[0015] Step 2.2: a sliding window method is adopted, and N continuous time steps of normalized data are selected as input-output pairs of the LSTM network model, and the data is converted into tensor data as the input of the model.

[0016] Step 2.3: the tensor data required by the LSTM network for predicting the entry curve deceleration is divided into a training set, a verification set and a test set.

[0017] Step 3, design a network training module, the process of which includes the following sub-steps:

[0018] Step 3.1, define and initialize the LSTM network model, set the input dimension, output length, LSTM hidden layer unit number, LSTM network layer number, set the loss function, optimizer and learning rate scheduler.

[0019] Step 3.2, perform multiple rounds of training cycles, in each round, the LSTM network receives tensor data through forward propagation, predicts the vehicle entry speed, and calculates the prediction error between the predicted value and the true value of the vehicle entry speed. Then, the loss function is calculated according to the error between the predicted value and the true value, and the network parameters are updated; after the end of each round, the model is evaluated using the validation set to check the generalization ability and training state of the model to prevent overfitting.

[0020] Step 3.3, if the validation loss of the current round is the smallest, save the weight of the current LSTM network model as the current optimal LSTM network model.

[0021] Step 3.4, after the LSTM network training is completed, load the trained network model, use the test set to predict the entry speed and calculate the evaluation index.

[0022] Step 4, design a real-time prediction module, load the optimal LSTM network model obtained from the network training module, use the vehicle state information and road information collected by the data acquisition module, and generate tensor data after processing by the data preprocessing module, to predict the vehicle entry speed in real time, for the safe and stable entry control of the autonomous vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a method flowchart of the present application.

[0024] Figure 2 is a data acquisition module flowchart of the present application. DETAILED DESCRIPTION

[0025] The present application will be described in detail below with reference to the accompanying drawings.

[0026] The application provides an automatic driving vehicle entry curve deceleration prediction method based on an LSTM, which comprises a data acquisition module, a data preprocessing module, a network training module and a real-time prediction module; the data acquisition module is responsible for acquiring vehicle state information and road information and outputting the same to the data preprocessing module; the data preprocessing module performs normalization processing on the data acquired by the data acquisition module, extracts input-output pairs using a sliding window, and converts feature data and label data into tensor data suitable for LSTM input; the network training module trains an optimal LSTM model using the processed data, the real-time prediction module loads the optimal network model and data for prediction, and the entry curve deceleration of the automatic driving vehicle is predicted. Figure 1 The application specifically comprises the following steps with reference to the schematic diagram of

[0027] Step 1, designing a data acquisition module

[0028] The data acquisition module refers to the schematic diagram of Figure 2 The process comprises the following sub-steps:

[0029] Step 1.1, constructing a sine working condition simulation scene through a Carsim platform, which is used for simulating typical road curvature disturbance and vehicle dynamic response behavior, and outputting an ideal path trajectory containing lateral displacement y ref and heading angle Meanwhile, a model predictive controller is used to control the CarSim vehicle model to track the ideal trajectory, and vehicle state information when the automatic driving vehicle enters a curve is collected.

[0030] Step 1.2, constructing a model predictive controller, specifically as follows:

[0031] A two-wheel two-degree-of-freedom vehicle model, a linear tire model and a tracking error model are used to design a model predictive control (MPC) based active front wheel steering controller system model, wherein the two-wheel two-degree-of-freedom vehicle model of the MPC is formula (2), which is converted into a two-degree-of-freedom motion differential equation form (3) of an automobile:

[0032]

[0033] Wherein, k1 is the lateral stiffness of the front wheel tire, indicating the size of the lateral force generated per unit side slip angle; k2 is the lateral stiffness of the rear wheel tire; β is the vehicle mass center side slip angle, indicating the included angle between the vehicle body movement direction and the vehicle head direction; u is the longitudinal speed of the vehicle; ω r is the yaw rate of the vehicle, i.e., the rotation rate of the vehicle around the vertical axis; δ is the front wheel steering angle, which is the control input; m is the total mass of the vehicle; a and b are the distances from the vehicle mass center to the front axle and the rear axle, respectively; is the lateral acceleration of the vehicle mass center; I z is the moment of inertia of the vehicle around the mass center; denotes the time derivative of yaw rate, i.e., yaw angular acceleration;

[0034]

[0035] The tracking error model extracts the key error information for controller design by comparing the current motion state of the vehicle with the reference path state. Let the current lateral position of the vehicle be y, the heading angle be The ideal lateral position on the corresponding reference path is y ref , and the ideal heading angle is The lateral deviation e y and the heading deviation are defined as:

[0036]

[0037] To accurately describe the error trend and improve the prediction performance of the controller, the error derivative term, i.e., the error rate of change, is further introduced:

[0038]

[0039] where denotes the rate of change of the heading angle deviation; ω r is the yaw rate of the vehicle; u is the longitudinal speed of the vehicle; κ is the curvature of the reference path; denotes the rate of change of the lateral deviation; is the heading angle deviation; v is the disturbance term, which is usually used to represent modeling errors, external disturbances, or dynamic factors not considered by the controller;

[0040] Combining equation (3) and equation (5), the state-space form of the MPC controller prediction model is obtained as follows:

[0041]

[0042] where is the state variable, the control input u represents the front wheel steering angle δ, d = κ, and the prediction output A, B u , B d , C are the system state matrix, control input matrix, disturbance matrix, and output matrix, respectively, and are as follows:

[0043]

[0044]

[0045] In the MPC prediction model, all are constant matrices within the prediction horizon;

[0046] In model predictive control, the controller performs rolling optimization and control input update with a fixed sampling period, so the system model must be expressed in discrete time form; in order to apply the continuous-time state-space model to the digital controller, it needs to be discretized in time; the zero-order hold method is used to discretize equation (3) to obtain the discrete-time model:

[0047]

[0048] where, is the state variable increment, the control input increment is Δδ, and the predicted output is

[0049] In the design of the MPC controller, in order to obtain the optimal control input sequence in the future period of time, a quadratic programming (QP) problem needs to be solved online. The QP optimizer takes the discretized system state-space model as the basis, inputs the current system state as the initial state, combines the desired path and control objectives, and calculates an optimal control input sequence that satisfies the constraints by minimizing the cost function in a limited prediction horizon:

[0050]

[0051] where, represents the optimal front wheel steering angle control input at the kth step of the prediction horizon, and N is the prediction horizon length;

[0052] Although the QP solver outputs a complete control input sequence U * , only the first control input in the sequence is used as the execution instruction at the current time under the MPC framework and is input to the Carsim platform to control the actual motion of the vehicle.

[0053] Finally, the model predictive controller takes the desired path as the reference trajectory, combines the current state information of the vehicle, and constructs a QP optimization problem based on the discrete state-space model to dynamically solve the optimal control sequence and extract the first control input value which is used as the control signal to the Carsim simulation platform for vehicle dynamics model execution, thereby achieving high-precision control of the front wheel steering angle and ensuring the stability of the curve.

[0054] Step 1.3, based on the Carsim platform, construct a vehicle dynamics model, input the front wheel steering angle δ output by the model predictive controller as the input signal, and input the current vehicle longitudinal speed u and path curvature κ; after the model runs, output the dynamic response data of the vehicle, including the lateral displacement y, lateral speed v y , and lateral acceleration a ylateral acceleration a, yaw rate γ, front wheel steering angle δ, road curvature κ and heading angle ψ, for representing the vehicle trajectory response performance;

[0055] Step 1.4, setting up an acceleration conversion module, calculating the expected value of the entry curve deceleration a y according to the lateral acceleration a x and the driving state, for reflecting the required speed correction amplitude of the vehicle under path deviation, to support the longitudinal motion control modeling;

[0056] Using the lateral acceleration change rate as the adjustment basis, through first-order filter smoothing processing, dynamically adjusting the longitudinal braking acceleration input, realizing the active adjustment of the vehicle posture, and the specific control law expression is as follows:

[0057]

[0058] Where: G xc is the longitudinal braking acceleration; G y is the lateral acceleration; is the lateral acceleration change rate; C xy is the control gain; T s is the first-order filter time constant.

[0059] Step 1.5, collecting vehicle running state information and road information, the lateral displacement y, lateral velocity v y , lateral acceleration a y , yaw rate γ, front wheel steering angle δ, road curvature κ and heading angle ψ output by the Carsim vehicle model, and the expected entry curve deceleration a x together as feedback variables, input to the data preprocessing module, for building time series samples and supporting subsequent training and prediction of the LSTM network model.

[0060] Step 2, designing a data preprocessing module, the process includes the following sub-steps:

[0061] Step 2.1, normalizing the feature data and label data by the MinMax normalization method, scaling the data to the range of [0, 1], so that the model training is more stable and efficient, and the MinMax normalization method for data preprocessing is realized by formula (12),

[0062]

[0063] Where, X norm is the normalized data, X is the original data collected by the data acquisition module, X max is the maximum value of the feature data, and X min is the minimum value of the feature data;

[0064] Step 2.2, slice the time series data to generate the input-output pairs required by the model using the fixed-length sliding window algorithm, each sliding window contains four time steps of data continuously t = 1, 2, 3, 4, and they are used to predict the entry curve deceleration of the subsequent time step, the sliding window can help the model capture the dynamic characteristics in the time series, especially the time sequence dependence of the vehicle state.

[0065] Step 2.3, convert the normalized data into tensor data and adjust the dimensions according to the requirements of the LSTM network structure, the input tensor data is a three-dimensional tensor, which is used as the input of the model;

[0066] Step 2.4, divide the data set into training set, validation set and test set according to the set proportion (80% training set, 10% validation set, 10% test set), divide the data set according to the set proportion (80% training set, 10% validation set, 10% test set), prepare for subsequent training and evaluation.

[0067] Step 3, design the network training module, the process includes the following sub-steps;

[0068] Step 3.1, define and initialize the LSTM network model, the input is three-dimensional tensor data, the dimension is (N, C, T), where N represents the batch size, C = 8 represents the feature dimension (corresponding to lateral displacement y, lateral velocity v y , lateral acceleration a y , yaw rate γ, front wheel steering angle δ, road curvature κ, heading angle ψ and entry curve deceleration a x ), T = 4 represents the input sequence length (i.e. the model predicts the deceleration at the next time based on the data of the last 4 historical time).

[0069] The LSTM network contains 2 layers of LSTM structure, each layer has 128 hidden units, the loss function is mean square error (MSE), and the optimizer is AdamW (the initial learning rate is set to 0.01, and the weight decay coefficient is 0.001);

[0070] Step 3.2, execute the model training process. Set the total training period to 150 rounds;

[0071] In each round, the tensor data will be input into the LSTM model, the model will predict the entry curve deceleration of the autonomous vehicle through forward propagation, and calculate the error between the predicted value and the true value of the vehicle entry curve deceleration. Then, the loss function is calculated according to the error between the predicted value and the true value, and the network parameters are updated.

[0072] In the LSTM model, the input at time step t includes the hidden state h t of the previous time step and the input x tThese inputs are processed by different gating mechanisms. Specifically, the key gating mechanisms of LSTM are as follows:

[0073] forget gate f t : controls the proportion of historical information to be retained, the formula is as follows,

[0074] f t = σ(W f · [h t-1 , x t ]+b f ) (13)

[0075] where W f is the weight matrix of the forget gate, b f is the bias term, σ is the sigmoid activation function, h t-1 is the hidden state at the previous time, x t is the input at the current time, and the output value is between 0 and 1, which determines how much information to forget;

[0076] input gate i t : controls how the current input affects the memory cell, the formula is as follows,

[0077] i t = σ(W i · [h t-1 , x t ]+b i ) (14)

[0078] where W i is the weight matrix of the input gate, b i is the bias term, which determines the effectiveness of the current input;

[0079] candidate memory cell generates candidate new memory content, the formula is as follows,

[0080]

[0081] where W c is the weight matrix of the candidate memory cell, b c is the bias term, and tanh is the hyperbolic tangent activation function;

[0082] memory cell update C t : integrates the outputs of the forget gate and the input gate to update the memory cell, the formula is as follows,

[0083]

[0084] where f t controls the retained historical information, and i t controls the addition of new information;

[0085] Output gate o t : Control the output of hidden state, the formula is as follows,

[0086] o t = σ(W o · [h t-1 , x t ] + b o ) (17)

[0087] Wherein, W0is the weight matrix of the output gate, b o is the bias term, and the output value determines the contribution degree of the current hidden state;

[0088] Final hidden state calculation h t : The current hidden state of the memory cell after being processed by the output gate, the formula is as follows,

[0089] h t = o t ⊙ tanh(C t ) (18)

[0090] Wherein, o t is the output of the output gate, and C t is the current memory cell;

[0091] Final output calculation: At each time step, LSTM generates the output of the current time, which is the combination of current time step information and historical information. Through the fusion of long-term historical state and instantaneous feature, the model can dynamically capture the required deceleration trend before turning, and realize high-precision prediction of the dynamic characteristics of turning.

[0092] After each round, the model will enter the verification mode, evaluate the model performance using the validation set, and record the current validation loss value. At this time, the model will calculate the gradient through back propagation, and update the model parameters using the optimization algorithm. In the training process, we use gradient clipping to prevent gradient explosion, and the clipping threshold is set to 0.15. In addition, after each round of training, we also update the learning rate scheduler according to the set strategy to ensure the stability of the training process;

[0093] Step 3.3, if the validation loss of the current round is lower than the historical optimal value, update the optimal network model record, and save the model parameters obtained in this round of training;

[0094] Step 3.4, after training, load the optimal network model, use the test set to make prediction and calculate various evaluation indexes, mean squared error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and determination coefficient (R 2), which can help us quantify the predictive ability of the model and ensure that the model can accurately predict the entry speed and make reliable control decisions in practical applications;

[0095] Step 4, design a real-time prediction module, load the optimal LSTM network model obtained by the network training module, use the vehicle state information and road information collected by the data collection module, and generate tensor data after processing by the data preprocessing module, to predict the vehicle entry speed in real time.

[0096] Step 5, design a PID controller module, which includes the following sub-steps:

[0097] Step 5.1, the vehicle entry speed predicted by the optimal LSTM network model in step 4 as the target control quantity at the current time. This value represents the desired entry speed required by the vehicle before entering the curve area, which is used to ensure that the vehicle passes through the curve at a safe speed and improves stability and handling.

[0098] Step 5.2, collect the actual longitudinal acceleration of the vehicle at the current time on the Carsim simulation platform and the predicted value to get the longitudinal error signal e(t):

[0099]

[0100] Where, positive error means that the current braking is insufficient and needs to increase the brake pedal input; negative error means that the current deceleration is too large and needs to reduce the braking instruction.

[0101] Step 5.3, based on the classic PID controller structure, real-time calculation of brake pedal opening control quantity u(t), the control law is as follows:

[0102]

[0103] Where, u(t) is the current brake pedal opening control instruction; K p ,K i ,K d are the proportional, integral and differential control gains respectively; e(t) is the real-time longitudinal error value; the controller parameters are obtained by debugging the vehicle dynamics model to ensure response speed and stability.

[0104] Step 5.4, send the calculated braking instruction u(t) to the vehicle dynamics model of the Carsim simulation platform as the current braking control input (brake pedal opening), and the new entry speed of the vehicle in the simulation environment after responding to the control instruction The feedback is sent to the control module as the input of the next control cycle to realize the closed-loop control of longitudinal entry deceleration.

Claims

1. An LSTM-based automatic driving vehicle entry curve deceleration prediction method, characterized in that: The method comprises a data acquisition module, a data preprocessing module, a network training module and a real-time prediction module; the data acquisition module is responsible for collecting vehicle state information and road information and outputting them to the data preprocessing module; The data preprocessing module normalizes the data collected by the data acquisition module, uses a sliding window to extract input-output pairs, and converts feature data and label data into tensor data suitable for LSTM input; The network training module trains the LSTM model according to the tensor data, and the real-time prediction module loads the optimal network model for prediction, and the prediction output automatically predicts the entry speed of the autonomous vehicle; The data acquisition module is used for collecting vehicle state information of the automatic driving vehicle when entering a curve, including lateral displacement y, lateral speed v y , lateral acceleration a y , yaw rate γ, front wheel steering angle δ, and road information including heading angle ψ and road curvature κ, wherein the above information is input to the data preprocessing module as feature data of the LSTM network model. An acceleration conversion module is provided to calculate an entry curve deceleration desired value a y based on the lateral acceleration a x and the running state; the desired entry curve deceleration a x is then used as label data for subsequent entry curve deceleration prediction. The real-time prediction of the vehicle entry curve deceleration The target control quantity of the PID controller at the current time; collecting a vehicle actual longitudinal acceleration at a current time and differentiating the predicted value to obtain a longitudinal error signal e(t); Wherein, the positive error indicates that the current braking is insufficient, and the brake pedal input needs to be increased; the negative error indicates that the current deceleration is too large, and the braking instruction needs to be reduced; Based on the structure of the classic PID controller, the brake pedal opening control quantity is calculated in real time; Wherein, u(t) is the current brake pedal opening control instruction; K p , i , d P, I and D control gains, respectively; e(t) is the real-time longitudinal error value; The calculated braking instruction u(t) is sent to the vehicle dynamics model as the braking control input at the current time, and the new entry speed of the vehicle in the simulation environment after responding to the control instruction is fed back to the control module as the input of the next control period, realizing the closed-loop control of the longitudinal entry speed.

2. The LSTM-based entry speed prediction method for an autonomous vehicle according to claim 1, characterized in that: The data preprocessing module normalizes the feature data and label data by the MinMax normalization method, scales the data to the range [0, 1], and preprocesses the data by the MinMax normalization method as shown in the formula: wherein X norm is normalized data, X is raw data collected by the data collection module, X max is the maximum value of the feature data, X min is the minimum value of the feature data; Then, the sliding window method is used to select N continuous time steps of normalized data as the input-output pairs of the LSTM network model, and convert them into tensor data as the input of the model; finally, the tensor data is divided into training set, validation set and test set; The network training module comprises a forward propagation and a backward propagation; during forward propagation, the LSTM network model is defined and initialized, and in each round, the LSTM network receives the tensor data and predicts the entry speed of the autonomous vehicle through forward propagation, while calculating the error between the predicted value and the true value of the vehicle entry speed; during backward propagation, the loss function is calculated according to the error between the predicted value and the true value, and the network parameters are updated; after the LSTM network training is completed, the trained network model is loaded, the test set is used to predict the entry speed and calculate the evaluation index; The real-time prediction module loads the optimal LSTM network model obtained by the network training module, uses the vehicle state information and road information collected by the data acquisition module, and generates tensor data after processing by the data preprocessing module, to predict the entry speed in real time, which is used for safe and stable entry control of the autonomous vehicle.

Images