Vehicle motion state prediction method, device, equipment and readable storage medium

Abstract

A vehicle motion state prediction method, device, equipment and readable storage medium, the vehicle motion state prediction method includes: obtaining vehicle motion state data, using normalized vehicle motion state data to obtain first prediction data through a recurrent gate neural network; obtaining correction parameters and weight parameters of an attention selection layer through a driving style classifier; using the correction parameters and weight parameters to correct the first prediction data through the attention selection layer to obtain second prediction data; performing nonlinear correction and Kalman filter fusion correction on the second prediction data to obtain third prediction data, and using the third prediction data as the final vehicle motion state prediction result. Through this application, the vehicle motion state prediction performed by the machine learning algorithm of this application has a simple model construction and a small amount of calculation, and can be embedded in the vehicle's ECU for real-time prediction, without the need for sensors and high-precision maps, and can cover all vehicle operating conditions.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle state prediction, and in particular to a vehicle motion state prediction method, device, equipment and readable storage medium. Background Art

[0002] Currently, the vehicle dynamics model is often used to predict the operating status of a vehicle. However, most of the current vehicle dynamics models are commercial software. The vehicle kinematic model is complex to build and requires a large amount of calculation. It cannot be embedded in the vehicle's ECU (Electronic Control Unit) for real-time prediction. It also requires the cooperation of sensors and high-precision maps, and cannot cover all vehicle operating conditions. Summary of the invention

[0003] The present application provides a vehicle motion state prediction method, device, equipment and readable storage medium, aiming to solve the current technical problem that the commonly used vehicle dynamics models for predicting the vehicle's operating state are mostly commercial software, the vehicle kinematic model is complex to build, the calculation amount is large, and it cannot be embedded in the vehicle's ECU for real-time prediction. It also requires the cooperation of sensors and high-precision maps and cannot cover all vehicle operating conditions.

[0004] In a first aspect, an embodiment of the present application provides a vehicle motion state prediction method, the vehicle motion state prediction method comprising:

[0005] Acquire vehicle motion state data, and perform normalization processing to obtain normalized vehicle motion state data;

[0006] Using the normalized vehicle motion state data through a recurrent gate neural network to obtain first prediction data;

[0007] The normalized vehicle motion state data is used to obtain the correction parameters and weight parameters of the attention selection layer through the driving style classifier;

[0008] The first prediction data is corrected by using the correction parameter and the weight parameter through the attention selection layer to obtain the second prediction data;

[0009] The second prediction data is subjected to nonlinear correction and Kalman filter fusion correction to obtain third prediction data, and the third prediction data is used as the final vehicle motion state prediction result.

[0010] Optionally, the vehicle motion state data includes vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power, and the vehicle motion state data is acquired and normalized to obtain the normalized vehicle motion state data including:

[0011] Obtain vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power;

[0012] The normalized vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power are calculated by formula 1, which is:

[0013] Y = (X - Xmin) / (Xmax - Xmin);

[0014] Wherein, Y represents each normalized vehicle motion data, X represents each acquired vehicle motion state data, Xmax represents the maximum value of each vehicle motion state data, and Xmin represents the minimum value of each vehicle motion state data.

[0015] Optionally, the recurrent gate neural network includes an update gate, a reset gate and a hidden layer, and the using the normalized vehicle motion state data to obtain the first prediction data through the recurrent gate neural network includes:

[0016] According to the normalized vehicle motion state data and the output of the hidden layer at the previous moment, the output of the update gate is calculated by formula 2, which is:

[0017] Z t =σ(U z *X t +W z *H t-1 +bz);

[0018] According to the normalized vehicle motion state data and the output of the hidden layer at the previous moment, the output of the reset gate is calculated by formula three, which is:

[0019] R t =σ(U r *X t +W r *H t-1 +br);

[0020] According to the output of the update gate and the output of the reset gate, the first prediction data of the hidden layer is calculated by formula 4, and the formula 4 is:

[0021] H t =(1-Z t )*H t-1 +Z t *H' t , H' t =tanh(U h *X t +W h *R t +bh);

[0022] Among them, Z t is the update gate output, U z is the input weight parameter of the update gate, X t is the normalized vehicle motion state data, W z is the weight parameter of the hidden layer in the update gate, H t-1 is the output of the hidden layer at the previous moment at time t, bz is the correction parameter of the update gate, σ() is the activation function σ(x) = 1 / (1+e -x ), e is the natural base, R t To reset the gate output, U r is the input weight parameter of the reset gate, Wr is the weight parameter of the hidden layer in the reset gate, br is the correction parameter of the reset gate, H t is the output of the hidden layer at time t, i.e. the first predicted data, H' t is the output of the activated hidden layer at time t, tanh() is the activation function tanh(x) = (e x -e -x ) / (e x +e -x ), U h is the input weight parameter of the hidden layer, W h is to reset the weight parameter of the gate in the hidden layer, and bh is the correction parameter of the hidden layer.

[0023] Optionally, the step of correcting the first prediction data by using a correction parameter and a weight parameter through the attention selection layer to obtain the second prediction data includes:

[0024] Through the attention selection layer, the first prediction data is corrected by using the correction parameter and the weight parameter through Formula 5 to obtain the second prediction data, and the Formula 5 is:

[0025] SA=H t *Wdcc+bdcc;

[0026] Among them, SA is the second prediction data, H t is the first prediction data, Wdcc is the weight parameter, and bdcc is the correction parameter.

[0027] Optionally, performing nonlinear function correction and Kalman filter fusion correction on the second prediction data to obtain the third prediction data includes:

[0028] The second prediction data is corrected by a nonlinear function based on the feedforward neural network to obtain a correction result of the nonlinear function. The formula for the nonlinear function correction is:

[0029] F t =ReLU(SA);

[0030] According to the correction result of the nonlinear function and the actual value of the vehicle motion state data, the third prediction data is obtained by performing Kalman filter fusion correction through Formula 6, and the Formula 6 is:

[0031] P t =(1-K t )*P t - ;

[0032] Among them, F t is the correction result of the nonlinear function, ReLU() is the nonlinear function ReLU(x)=max(0,x), SA is the second prediction data, is the predicted value after Kalman filter fusion correction, that is, the third predicted data, is the predicted value before Kalman filter fusion correction, K t is the Kalman gain, Z t is the actual value of the vehicle motion state data at time t, W is the weight parameter of the linear link, B is the correction parameter of the linear link, R is the noise of the actual process, Q is the noise of the prediction process, is the prior covariance matrix, P t is the posterior covariance matrix at time t, P t-1 is the posterior covariance matrix of the previous moment before moment t.

[0033] In a second aspect, an embodiment of the present application provides a vehicle motion state prediction device, the vehicle motion state prediction device comprising:

[0034] An acquisition module is used to acquire vehicle motion state data and perform normalization processing to obtain normalized vehicle motion state data;

[0035] A first prediction module, used to obtain first prediction data by using the normalized vehicle motion state data through a recurrent gate neural network;

[0036] A classification module, used to obtain correction parameters and weight parameters of the attention selection layer through a driving style classifier using the normalized vehicle motion state data;

[0037] A second prediction module, configured to correct the first prediction data by using a correction parameter and a weight parameter through an attention selection layer to obtain second prediction data;

[0038] The correction module is used to perform nonlinear correction and Kalman filter fusion correction on the second prediction data to obtain the third prediction data, and use the third prediction data as the final vehicle motion state prediction result.

[0039] Optionally, the second prediction module is used to:

[0040] Through the attention selection layer, the first prediction data is corrected by using the correction parameter and the weight parameter through Formula 5 to obtain the second prediction data, and the Formula 5 is:

[0041] SA=H t *Wdcc+bdcc;

[0042] Among them, SA is the second prediction data, H t is the first prediction data, Wdcc is the weight parameter, and bdcc is the correction parameter.

[0043] Optionally, the correction module is used to:

[0044] The second prediction data is corrected by a nonlinear function based on a feedforward neural network to obtain a correction result of the nonlinear function. The formula for the nonlinear function correction is:

[0045] F t =ReLU(SA);

[0046] According to the correction result of the nonlinear function and the actual value of the vehicle motion state data, the third prediction data is obtained by performing Kalman filter fusion correction through Formula 6, and the Formula 6 is:

[0047] P t =(1-K t )*P t - ;

[0048] Among them, F t is the correction result of the nonlinear function, ReLU() is the nonlinear function ReLU(x)=max(0,x), SA is the second prediction data, is the predicted value after Kalman filter fusion correction, that is, the third predicted data, is the predicted value before Kalman filter fusion correction, K t is the Kalman gain, Z t is the actual value of the vehicle motion state data at time t, W is the weight parameter of the linear link, B is the correction parameter of the linear link, R is the noise of the actual process, Q is the noise of the prediction process, is the prior covariance matrix, P t is the posterior covariance matrix at time t, P t-1 is the posterior covariance matrix of the previous moment before moment t.

[0049] In the third aspect, an embodiment of the present application provides a vehicle motion state prediction device, which includes a processor, a memory, and a vehicle motion state prediction program stored in the memory and executable by the processor, wherein when the vehicle motion state prediction program is executed by the processor, the steps of the vehicle motion state prediction method described above are implemented.

[0050] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a vehicle motion state prediction program is stored, wherein when the vehicle motion state prediction program is executed by a processor, the steps of the vehicle motion state prediction method as described above are implemented.

[0051] The beneficial effects brought by the technical solution provided in the embodiments of the present application include:

[0052] In the embodiment of the present application, the vehicle motion state data is obtained and normalized to obtain the normalized vehicle motion state data; the normalized vehicle motion state data is used to obtain the first prediction data through the recurrent gate neural network; the normalized vehicle motion state data is used to obtain the correction parameters and weight parameters of the attention selection layer through the driving style classifier; the first prediction data is corrected by the correction parameters and weight parameters through the attention selection layer to obtain the second prediction data; the second prediction data is corrected by nonlinear correction and Kalman filter fusion correction to obtain the third prediction data, and the third prediction data is used as the final vehicle motion state prediction result. The embodiment of the present application obtains the real-time motion state data of the vehicle, obtains the first prediction data of the vehicle motion state through the recurrent gate neural network after normalization, and corrects it through the attention selection layer, further performs nonlinear correction and Kalman filter fusion correction to obtain the final vehicle motion state prediction result. The vehicle motion state prediction performed by the machine learning algorithm of the present application has a simple model construction, a small amount of calculation, high accuracy and stability, and can be embedded in the vehicle's ECU for real-time prediction, without the cooperation of sensors and high-precision maps, and can cover all vehicle operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of a flow chart of an embodiment of a vehicle motion state prediction method of the present application;

[0054] Figure 2 For this application Figure 1 A detailed flow chart of step S10;

[0055] Figure 3 For this application Figure 1 Detailed flow chart of step S20;

[0056] Figure 4 For this application Figure 1 A detailed flow chart of step S50;

[0057] Figure 5 This is a schematic diagram of the functional modules of an embodiment of a vehicle motion state prediction device of the present application;

[0058] Figure 6 This is a schematic diagram of the hardware structure of a vehicle motion state prediction device involved in the embodiment of the present application. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0060] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0061] In a first aspect, an embodiment of the present application provides a method for predicting a vehicle motion state.

[0062] In one embodiment, referring to Figure 1 , Figure 1 This is a flow chart of an embodiment of a method for predicting vehicle motion state according to the present application. Figure 1 As shown, the vehicle motion state prediction method includes:

[0063] Step S10, acquiring vehicle motion state data, and performing normalization processing to obtain normalized vehicle motion state data.

[0064] In this embodiment, the real-time motion status data of the vehicle can be obtained through various vehicle-mounted sensors. The real-time motion status data of the vehicle may include vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power, etc. To facilitate neural network model processing, further normalization processing is performed to obtain normalized vehicle motion status data.

[0065] Step S20, using the normalized vehicle motion state data to obtain first prediction data through a recurrent gate neural network.

[0066] In this embodiment, the normalized vehicle motion state data is input into a pre-trained recurrent gate neural network model, and the first prediction data is obtained through the recurrent gate neural network.

[0067] Step S30, using the normalized vehicle motion state data to obtain correction parameters and weight parameters of the attention selection layer through a driving style classifier.

[0068] In this embodiment, different driving styles will affect the prediction of the vehicle's motion state. The driving style classifier is also a pre-trained neural network model. The driving style classifier outputs corresponding correction parameters and weight parameters of the attention selection layer according to different driving styles to correct the first prediction data.

[0069] Step S40, using the attention selection layer, the first prediction data is corrected using the correction parameter and the weight parameter to obtain the second prediction data.

[0070] In this embodiment, the attention selection layer uses the correction parameter and the weight parameter to correct the first prediction data to obtain the second prediction data, that is, the first prediction data is corrected according to different driving styles.

[0071] Step S50, performing nonlinear correction and Kalman filter fusion correction on the second prediction data to obtain third prediction data, and using the third prediction data as the final vehicle motion state prediction result.

[0072] In this embodiment, the nonlinear correction can use a linear rectifier function, which is also called the ReLU function (Rectified Linear Unit), which is an activation function commonly used in deep learning and neural networks. The ReLU function has the characteristics of simple and efficient calculation, nonlinearity, activation sparsity and convexity, and has better gradient stability, which can improve the generalization and expression ability of the model. Kalman filter fusion correction is a state estimation method. According to the difference between the measured value (i.e., the actual value) and the predicted value, the current state is corrected. The state correction equation of the Kalman filter is generally composed of a prediction step and an update step. The state correction equation of the Kalman filter is an iterative process. At each moment, prediction and update steps are performed. By continuously correcting the state estimation value, the optimal state estimation result is finally obtained. This iterative process can effectively utilize the system model and measurement information to improve the accuracy and stability of the state estimation. By performing nonlinear correction and Kalman filter fusion correction on the second predicted data, the accuracy and stability of the final vehicle motion state prediction result are improved.

[0073] In this embodiment, the real-time motion state data of the vehicle is obtained, and after normalization, the first prediction data of the vehicle motion state is obtained through a pre-trained recurrent gate neural network. Taking into account that different driving styles will affect the prediction of the vehicle's motion state, an attention selection layer is introduced to perform corrections using corresponding correction parameters and weight parameters for different driving styles. In order to further improve the accuracy and stability of the prediction, nonlinear correction and Kalman filter fusion correction are performed to obtain the final vehicle motion state prediction result. The vehicle motion state prediction performed by the machine learning algorithm of the present application has a simple model construction, a small amount of calculation, high accuracy and stability, and can be embedded in the vehicle's ECU for real-time prediction. It does not require the cooperation of sensors and high-precision maps, and can cover all vehicle operating conditions.

[0074] Furthermore, in one embodiment, the vehicle motion state data includes vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power. Figure 2 , Figure 2 For this application Figure 1 The detailed flow chart of step S10 is as follows: Figure 2 As shown, step S10 includes:

[0075] Step S101, obtaining vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power;

[0076] Step S102, calculating the normalized vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power by formula 1, wherein formula 1 is:

[0077] Y = (X - Xmin) / (Xmax - Xmin);

[0078] Wherein, Y represents each normalized vehicle motion data, X represents each acquired vehicle motion state data, Xmax represents the maximum value of each vehicle motion state data, and Xmin represents the minimum value of each vehicle motion state data.

[0079] In this embodiment, the real-time motion state data of the vehicle can be obtained through various types of vehicle-mounted sensors. The real-time motion state data of the vehicle includes but is not limited to the vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power. In order to facilitate the processing of the neural network model, the obtained real-time motion state data of the vehicle needs to be normalized. Among them, normalization is a way to simplify calculations, that is, to transform a dimensioned expression into a dimensionless expression and become a scalar. Normalization also converts the absolute value of the physical system value into a relative value relationship. After the various vehicle motion state data obtained from the sensor are calculated and processed by formula one, the normalized vehicle motion data are obtained, which is convenient for efficient processing of the neural network model.

[0080] Further, in one embodiment, the recurrent gate neural network includes an update gate, a reset gate and a hidden layer, referring to Figure 3 , Figure 3 For this application Figure 1 The detailed flow chart of step S20 is as follows: Figure 3 As shown, step S20 includes:

[0081] Step S201, according to the normalized vehicle motion state data and the output of the hidden layer at the previous moment, the output of the update gate is calculated by formula 2, and the formula 2 is:

[0082] Z t =σ(U z *X t +W z *H t-1 +bz);

[0083] Step S202, according to the normalized vehicle motion state data and the output of the hidden layer at the previous moment, the output of the reset gate is calculated by formula 3, and the formula 3 is:

[0084] R t =σ(U r *X t +W r *H t-1 +br);

[0085] Step S203, according to the output of the update gate and the output of the reset gate, the first prediction data of the hidden layer is calculated by formula 4, and the formula 4 is:

[0086] H t =(1-Z t )*H t-1 +Z t *H' t , H' t =tanh(U h *Xt +W h *R t +bh);

[0087] Among them, Z t is the update gate output, U z is the input weight parameter of the update gate, X t is the normalized vehicle motion state data, W z is the weight parameter of the hidden layer in the update gate, H t-1 is the output of the hidden layer at the previous moment at time t, bz is the correction parameter of the update gate, σ() is the activation function σ(x) = 1 / (1+e -x ), e is the natural base, R t To reset the gate output, U r is the input weight parameter of the reset gate, Wr is the weight parameter of the hidden layer in the reset gate, br is the correction parameter of the reset gate, H t is the output of the hidden layer at time t, i.e. the first predicted data, H' t is the output of the activated hidden layer at time t, tanh() is the activation function tanh(x) = (e x -e -x ) / (e x +e -x ), U h is the input weight parameter of the hidden layer, W h is to reset the weight parameter of the gate in the hidden layer, and bh is the correction parameter of the hidden layer.

[0088] In this embodiment, the recurrent gate neural network uses GRU (Gate Recurrent Unit), which is a special RNN (Recurrent Neural Network). GRU combines the forget gate and the input gate into a single update gate. Since there is one less gate, the parameters of the model are reduced, and the cell state and the hidden state are also merged, making the model simpler, so that the vehicle state prediction method of the present application can be embedded in the vehicle ECU for real-time prediction. The historical vehicle motion state data can be normalized and time-series constructed in advance to obtain a training data set, and the training data set is used to train the recurrent gate neural network. When the recurrent gate neural network converges, a trained recurrent gate neural network model is obtained, and at the same time, the weight parameters and correction parameters of the update gate, reset gate and hidden layer are obtained, including the input weight parameter U of the update gate. z , the weight parameter W of the hidden layer in the update gate z , update the gate's correction parameter bz, reset the gate's input weight parameter U r, the weight parameter Wr of the hidden layer in the reset gate, the correction parameter br of the reset gate, the input weight parameter U of the hidden layer h , reset the weight parameter W of the gate in the hidden layer h , and the correction parameter bh of the hidden layer. When training the recurrent gate neural network, the prediction duration can be set, that is, the recurrent gate neural network is used to predict the vehicle motion state after how long.

[0089] Furthermore, in one embodiment, step S40 includes:

[0090] Through the attention selection layer, the first prediction data is corrected by using the correction parameter and the weight parameter through Formula 5 to obtain the second prediction data, and the Formula 5 is:

[0091] SA=H t *Wdcc+bdcc;

[0092] Among them, SA is the second prediction data, H t is the first prediction data, Wdcc is the weight parameter, and bdcc is the correction parameter.

[0093] In this embodiment, the correction parameters and weight parameters are obtained by using the normalized vehicle motion state data through the driving style classifier. The first prediction data is corrected by using the correction parameters and weight parameters through the attention selection layer. That is, the first prediction data output by the recurrent gate neural network is corrected according to different driving styles, so that the obtained second prediction data is more accurate.

[0094] Further, in one embodiment, referring to Figure 4 , Figure 4 For this application Figure 1 The detailed flow chart of step S50 is as follows: Figure 4 As shown, step S50 includes:

[0095] Step S501, performing nonlinear function correction on the second prediction data based on a feedforward neural network to obtain a correction result of the nonlinear function, wherein the formula for the nonlinear function correction is:

[0096] F t =ReLU(SA);

[0097] Step S502, according to the correction result of the nonlinear function and the actual value of the vehicle motion state data, the third prediction data is obtained by performing Kalman filter fusion correction through Formula 6, and the Formula 6 is:

[0098] P t =(1-K t )*P t- ;

[0099] Among them, F t is the correction result of the nonlinear function, ReLU() is the nonlinear function ReLU(x)=max(0,x), SA is the second prediction data, is the predicted value after Kalman filter fusion correction, that is, the third predicted data, is the predicted value before Kalman filter fusion correction, K t is the Kalman gain, Z t is the actual value of the vehicle motion state data at time t, W is the weight parameter of the linear link, B is the correction parameter of the linear link, R is the noise of the actual process, Q is the noise of the prediction process, is the prior covariance matrix, P t is the posterior covariance matrix at time t, P t-1 is the posterior covariance matrix of the previous moment before moment t.

[0100] In this embodiment, the linear rectification function is also called the ReLU function (Rectified Linear Unit), which is an activation function commonly used in deep learning and neural networks. If the input is greater than 0, the value provided as the input is directly returned. If the input is 0 or less, 0 is returned. The nonlinear function correction can make the prediction model have better gradient stability, and can improve the generalization and expression ability of the prediction model. Kalman filter fusion correction is a state estimation method. According to the difference between the measured value (ie, the actual value) and the predicted value, the current state is corrected. The state correction equation of the Kalman filter generally consists of a prediction step and an update step. The state correction equation of the Kalman filter is an iterative process. At each moment, prediction and update steps are performed. By continuously correcting the state estimation value, the optimal state estimation result is finally obtained. This iterative process can effectively utilize the system model and measurement information to improve the accuracy and stability of the state estimation. By performing nonlinear correction and Kalman filter fusion correction on the second prediction data, the accuracy and stability of the final vehicle motion state prediction result are improved. Among them, the actual value Z of the vehicle motion state data at time t t It can be obtained through vehicle-mounted sensors. The weight parameter W of the linear link and the correction parameter B of the linear link are obtained after training the neural network model. The system disturbance and error at each moment, that is, noise, is divided into the noise of the actual process and the noise of the prediction process. The covariance matrix represents the error covariance between the system state estimate and the actual state, which is divided into a priori covariance matrix and a posteriori covariance matrix.

[0101] In a second aspect, an embodiment of the present application also provides a vehicle motion state prediction device.

[0102] In one embodiment, referring to Figure 5 , Figure 5 This is a functional module diagram of an embodiment of a vehicle motion state prediction device of the present application, such as Figure 5 As shown, the vehicle motion state prediction device includes:

[0103] The acquisition module 10 is used to acquire the vehicle motion state data and perform normalization processing to obtain the normalized vehicle motion state data;

[0104] A first prediction module 20, configured to obtain first prediction data using the normalized vehicle motion state data through a recurrent gate neural network;

[0105] A classification module 30, for obtaining correction parameters and weight parameters of the attention selection layer through a driving style classifier using the normalized vehicle motion state data;

[0106] A second prediction module 40, configured to modify the first prediction data using a correction parameter and a weight parameter through an attention selection layer to obtain second prediction data;

[0107] The correction module 50 is used to perform nonlinear correction and Kalman filter fusion correction on the second prediction data to obtain third prediction data, and use the third prediction data as the final vehicle motion state prediction result.

[0108] Furthermore, in one embodiment, the vehicle motion state data includes vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power, and the acquisition module 10 is used to:

[0109] Obtain vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power;

[0110] The normalized vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power are calculated by formula 1, which is:

[0111] Y = (X - Xmin) / (Xmax - Xmin);

[0112] Wherein, Y represents each normalized vehicle motion data, X represents each acquired vehicle motion state data, Xmax represents the maximum value of each vehicle motion state data, and Xmin represents the minimum value of each vehicle motion state data.

[0113] Further, in one embodiment, the recurrent gate neural network includes an update gate, a reset gate and a hidden layer, and the first prediction module 20 is used to:

[0114] According to the normalized vehicle motion state data and the output of the hidden layer at the previous moment, the output of the update gate is calculated by formula 2, which is:

[0115] Z t =σ(U z *X t +W z *H t-1 +bz);

[0116] According to the normalized vehicle motion state data and the output of the hidden layer at the previous moment, the output of the reset gate is calculated by formula three, which is:

[0117] R t =σ(U r *X t +W r *H t-1 +br);

[0118] According to the output of the update gate and the output of the reset gate, the first prediction data of the hidden layer is calculated by formula 4, and the formula 4 is:

[0119] H t =(1-Z t )*H t-1 +Z t *H' t , H' t =tanh(U h *X t +W h *R t +bh);

[0120] Among them, Z t is the update gate output, U z is the input weight parameter of the update gate, X t is the normalized vehicle motion state data, W z is the weight parameter of the hidden layer in the update gate, H t-1 is the output of the hidden layer at the previous moment at time t, bz is the correction parameter of the update gate, σ() is the activation function σ(x) = 1 / (1+e -x ), e is the natural base, R t To reset the gate output, U r is the input weight parameter of the reset gate, Wr is the weight parameter of the hidden layer in the reset gate, br is the correction parameter of the reset gate, H t is the output of the hidden layer at time t, i.e. the first predicted data, H' t is the output of the activated hidden layer at time t, tanh() is the activation function tanh(x) = (e x -e-x ) / (e x +e -x ), U h is the input weight parameter of the hidden layer, W h is to reset the weight parameter of the gate in the hidden layer, and bh is the correction parameter of the hidden layer.

[0121] Furthermore, in one embodiment, the second prediction module 40 is used to:

[0122] Through the attention selection layer, the first prediction data is corrected by using the correction parameter and the weight parameter through Formula 5 to obtain the second prediction data, and the Formula 5 is:

[0123] SA=H t *Wdcc+bdcc;

[0124] Among them, SA is the second prediction data, H t is the first prediction data, Wdcc is the weight parameter, and bdcc is the correction parameter.

[0125] Furthermore, in one embodiment, the correction module 50 is used to:

[0126] The second prediction data is corrected by a nonlinear function based on the feedforward neural network to obtain a correction result of the nonlinear function. The formula for the nonlinear function correction is:

[0127] F t =ReLU(SA);

[0128] According to the correction result of the nonlinear function and the actual value of the vehicle motion state data, the third prediction data is obtained by performing Kalman filter fusion correction through Formula 6, and the Formula 6 is:

[0129] P t =(1-K t )*P t - ;

[0130] Among them, F t is the correction result of the nonlinear function, ReLU() is the nonlinear function ReLU(x)=max(0,x), SA is the second prediction data, is the predicted value after Kalman filter fusion correction, that is, the third predicted data, is the predicted value before Kalman filter fusion correction, K t is the Kalman gain, Z t is the actual value of the vehicle motion state data at time t, W is the weight parameter of the linear link, B is the correction parameter of the linear link, R is the noise of the actual process, Q is the noise of the prediction process, is the prior covariance matrix, P t is the posterior covariance matrix at time t, P t-1 is the posterior covariance matrix of the previous moment before moment t.

[0131] Among them, the functional implementation of each module in the above-mentioned vehicle motion state prediction device corresponds to the various steps in the above-mentioned vehicle motion state prediction method embodiment, and its functions and implementation processes are no longer repeated here one by one.

[0132] In a third aspect, an embodiment of the present application provides a vehicle motion state prediction device.

[0133] Reference Figure 6 , Figure 6 Schematic diagram of the hardware structure of the vehicle motion state prediction device involved in the embodiment of the present application. In the embodiment of the present application, the vehicle motion state prediction device may include a processor, a memory, a communication interface and a communication bus.

[0134] The communication bus may be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0135] The communication interface includes an input / output (I / O) interface, a physical interface, and a logical interface, etc., which are used to interconnect the devices inside the vehicle motion state prediction device, and an interface used to interconnect the vehicle motion state prediction device with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, an optical fiber interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.

[0136] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0137] The processor may be a general-purpose processor, and the general-purpose processor may call the vehicle motion state prediction program stored in the memory and execute the vehicle motion state prediction method provided by the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the vehicle motion state prediction program is called may refer to the various embodiments of the vehicle motion state prediction method of the present application, and will not be repeated here.

[0138] Those skilled in the art will understand that Figure 6 The hardware structure shown in the figure does not constitute a limitation on the present application, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0139] In a fourth aspect, an embodiment of the present application also provides a readable storage medium.

[0140] A vehicle motion state prediction program is stored on a readable storage medium of the present application, wherein when the vehicle motion state prediction program is executed by a processor, the steps of the vehicle motion state prediction method as described above are implemented.

[0141] Among them, the method implemented when the vehicle motion state prediction program is executed can refer to the various embodiments of the vehicle motion state prediction method of the present application, and will not be repeated here.

[0142] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0143] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0144] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0145] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0146] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0147] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0148] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A vehicle motion state prediction method, It is characterized in that The vehicle motion state prediction method comprises: Acquire vehicle motion state data, and perform normalization processing to obtain normalized vehicle motion state data; Using the normalized vehicle motion state data through a recurrent gate neural network to obtain first prediction data; The normalized vehicle motion state data is used to obtain the correction parameters and weight parameters of the attention selection layer through the driving style classifier; The first prediction data is corrected by using the correction parameter and the weight parameter through the attention selection layer to obtain the second prediction data; Performing nonlinear correction and Kalman filter fusion correction on the second prediction data to obtain third prediction data, and using the third prediction data as the final vehicle motion state prediction result; The performing of nonlinear function correction and Kalman filter fusion correction on the second prediction data to obtain the third prediction data comprises: The second prediction data is corrected by a nonlinear function based on a feedforward neural network to obtain a correction result of the nonlinear function. The formula for the nonlinear function correction is: ； According to the correction result of the nonlinear function and the actual value of the vehicle motion state data, the third prediction data is obtained by performing Kalman filter fusion correction through Formula 6, and the Formula 6 is: ， ， ， ， ； in, is the correction result of the nonlinear function, ReLU() is the nonlinear function ReLU(x)=max(0,x), SA is the second prediction data, is the predicted value after Kalman filter fusion correction, that is, the third predicted data, is the predicted value before Kalman filter fusion correction, K t is the Kalman gain, Z t is the actual value of the vehicle motion state data at time t, W is the weight parameter of the linear link, B is the correction parameter of the linear link, R is the noise of the actual process, Q is the noise of the prediction process, is the prior covariance matrix, is the posterior covariance matrix at time t, is the posterior covariance matrix of the previous moment before moment t.

2. The vehicle motion state prediction method according to claim 1, It is characterized in that The vehicle motion state data includes vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power. The vehicle motion state data is obtained and normalized to obtain the normalized vehicle motion state data including: Obtain vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power; The normalized vehicle speed, brake pedal travel, accelerator pedal travel, engine speed and engine power are calculated by formula 1, which is: Y = (X-Xmin) / (Xmax-Xmin); Wherein, Y represents each normalized vehicle motion data, X represents each acquired vehicle motion state data, Xmax represents the maximum value of each vehicle motion state data, and Xmin represents the minimum value of each vehicle motion state data.

3. The vehicle motion state prediction method according to claim 1, It is characterized in that The recurrent gate neural network includes an update gate, a reset gate and a hidden layer, and the first prediction data is obtained by using the normalized vehicle motion state data through the recurrent gate neural network, including: According to the normalized vehicle motion state data and the output of the hidden layer at the previous moment, the output of the update gate is calculated by formula 2, which is: Z t =σ(U z *X t +W z *H t-1 +bz); According to the normalized vehicle motion state data and the output of the hidden layer at the previous moment, the output of the reset gate is calculated by formula three, which is: R t =σ(U r *X t +W r *H t-1 +br); According to the output of the update gate and the output of the reset gate, the first prediction data of the hidden layer is calculated by formula 4, and the formula 4 is: H t =(1-Z t )*H t-1 +Z t *H' t ,H' t =tanh(U h *X t +W h *R t +bh); Among them, Z t is the update gate output, U z is the input weight parameter of the update gate, X t is the normalized vehicle motion state data, W z is the weight parameter of the hidden layer in the update gate, H t-1 is the output of the hidden layer at the previous moment at time t, bz is the correction parameter of the update gate, σ() is the activation function σ(x)=1 / (1+e -x ), e is the natural base, R t To reset the gate output, U r is the input weight parameter of the reset gate, Wr is the weight parameter of the hidden layer in the reset gate, br is the correction parameter of the reset gate, H t is the output of the hidden layer at time t, i.e. the first predicted data, H' t is the output of the activated hidden layer at time t, tanh() is the activation function tanh(x)=(e x -e -x ) / ( e x + e -x ), U h is the input weight parameter of the hidden layer, W h is to reset the weight parameter of the gate in the hidden layer, and bh is the correction parameter of the hidden layer.

4. The vehicle motion state prediction method according to claim 1, It is characterized in that The method of correcting the first prediction data by using the correction parameter and the weight parameter through the attention selection layer to obtain the second prediction data includes: Through the attention selection layer, the first prediction data is corrected by using the correction parameter and the weight parameter through Formula 5 to obtain the second prediction data, and the Formula 5 is: SA=H t *Wdcc+bdcc; Among them, SA is the second prediction data, H t is the first prediction data, Wdcc is the weight parameter, and bdcc is the correction parameter.

5. A vehicle motion state prediction device, It is characterized in that The vehicle motion state prediction device comprises: An acquisition module is used to acquire vehicle motion state data and perform normalization processing to obtain normalized vehicle motion state data; A first prediction module, used to obtain first prediction data by using the normalized vehicle motion state data through a recurrent gate neural network; A classification module, used to obtain correction parameters and weight parameters of the attention selection layer through a driving style classifier using the normalized vehicle motion state data; A second prediction module, configured to correct the first prediction data by using a correction parameter and a weight parameter through an attention selection layer to obtain second prediction data; A correction module, used for performing nonlinear correction and Kalman filter fusion correction on the second prediction data to obtain third prediction data, and using the third prediction data as the final vehicle motion state prediction result; The correction module is used to: The second prediction data is corrected by a nonlinear function based on a feedforward neural network to obtain a correction result of the nonlinear function. The formula for the nonlinear function correction is: ； According to the correction result of the nonlinear function and the actual value of the vehicle motion state data, the third prediction data is obtained by performing Kalman filter fusion correction through Formula 6, and the Formula 6 is: ， ， ， ， ； in, is the correction result of the nonlinear function, ReLU() is the nonlinear function ReLU(x)=max(0,x), SA is the second prediction data, is the predicted value after Kalman filter fusion correction, that is, the third predicted data, is the predicted value before Kalman filter fusion correction, K t is the Kalman gain, Z t is the actual value of the vehicle motion state data at time t, W is the weight parameter of the linear link, B is the correction parameter of the linear link, R is the noise of the actual process, Q is the noise of the prediction process, is the prior covariance matrix, is the posterior covariance matrix at time t, is the posterior covariance matrix of the previous moment before moment t.

6. The vehicle motion state prediction device according to claim 5, It is characterized in that The second prediction module is used to: Through the attention selection layer, the first prediction data is corrected by using the correction parameter and the weight parameter through Formula 5 to obtain the second prediction data, and the Formula 5 is: SA=H t *Wdcc+bdcc; Among them, SA is the second prediction data, H t is the first prediction data, Wdcc is the weight parameter, and bdcc is the correction parameter.

7. A vehicle motion state prediction device, It is characterized in that The vehicle motion state prediction device includes a processor, a memory, and a vehicle motion state prediction program stored in the memory and executable by the processor, wherein when the vehicle motion state prediction program is executed by the processor, the steps of the vehicle motion state prediction method as described in any one of claims 1 to 4 are implemented.

8. A readable storage medium, It is characterized in that The readable storage medium stores a vehicle motion state prediction program, wherein when the vehicle motion state prediction program is executed by a processor, the steps of the vehicle motion state prediction method according to any one of claims 1 to 4 are implemented.

Images