Tracked vehicle driving state prediction model training method, prediction method and product

Abstract

The invention discloses a tracked vehicle driving state prediction model training method, a tracked vehicle driving state prediction method and a tracked vehicle driving state prediction product, and relates to the technical field of tracked vehicle intelligent driving. Inputting the measurable control vector data into the trained transmission-track model to obtain actual control vector data; acquiring state vector data of the tracked vehicle at the previous moment, and sequentially calculating prior estimations of the state vector data and the measurement vector data to obtain prior state estimation and prior measurement estimation; converting the priori state estimation, the priori measurement estimation and the actual control vector data into a differential sequence; and training the recurrent neural network by using the differential sequence to obtain a tracked vehicle driving state prediction model. According to the method, the neural network with a good nonlinear fitting characteristic is combined into the Kalman filtering algorithm, effective state estimation of a strong nonlinear system is realized, a Kalman filtering state estimation principle is integrated, and to-be-learned parameters of the neural network are reduced.

Description

Technical Field

[0001] The present application relates to the field of intelligent driving technology for tracked vehicles, and in particular to a training method, a prediction method and a product for predicting a driving state of a tracked vehicle. Background Art

[0002] With the development of sensing and control technologies, tracked vehicles, as an important special vehicle, are developing towards intelligence. As they travel on complex and changeable ground, the driving state of tracked vehicles changes rapidly, and dangerous conditions such as steering, side slipping, and even rollover may occur. Therefore, in order to achieve high-precision and safe driving control, it is necessary to obtain the driving state of tracked vehicles with high credibility. Although sensors such as GNSS and IMU can directly measure state quantities such as position, yaw angular velocity, and acceleration, since tracked vehicles mostly travel in harsh off-road environments with gravel, rain, and bumps, the sensor noise cannot be ignored, so it is necessary to use state estimation methods to obtain accurate and reliable driving states.

[0003] At present, state estimation methods based on mechanism models, such as Kalman filtering and its variants, require the establishment of a kinematic or dynamic model of the tracked vehicle before combining sensor information for state estimation. They are suitable for linear or weakly nonlinear systems that can be modeled. However, when a tracked vehicle is traveling, there is dynamic coupling between the directions, the coupling of state variables in each direction is obvious, and the motion nonlinearity is strong. The track force model is highly nonlinear because it usually uses the integration of time and position based on the modeling of ground shear displacement. At the same time, the drastic changes in ground conditions and random disturbances caused by tracked vehicles will lead to mismatch between the vehicle nominal model and the actual controlled object. Strong nonlinearity and model mismatch lead to poor results of state estimation methods based on mechanism models. Data-driven state estimation methods do not need to establish complex high-precision models, but use data to learn the relationship between system input and output. They have good applicability, but lack interpretability. At the same time, the black box characteristics will lead to many parameters to be learned and long training time. Summary of the invention

[0004] The purpose of this application is to provide a tracked vehicle driving state prediction model training method, prediction method and product, which can realize effective state estimation of strongly nonlinear systems, and the neural network has fewer parameters to be learned, thereby improving training efficiency.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for training a tracked vehicle driving state prediction model, the method comprising:

[0007] Acquire measurement vector data and measurable control vector data of the tracked vehicle; the measurement vector data includes: longitudinal acceleration, lateral acceleration and yaw rate of the tracked vehicle; the measurable control vector data includes: left and right drive motor torques, left and right driving wheel speeds and transmission gear position;

[0008] Input the measurable control vector data into the trained transmission-track model to obtain actual control vector data; the actual control vector data includes: the lateral and longitudinal forces exerted on the track by the ground, the steering drive torque and the steering resistance torque;

[0009] Acquire the state vector data of the tracked vehicle at the last moment; the state vector data includes: the predicted values ​​of the longitudinal velocity, lateral velocity and yaw angular velocity of the tracked vehicle;

[0010] The Kalman recursive framework is used to calculate the prior estimate of the state vector data at the current moment to obtain the prior state estimate;

[0011] Calculating a priori estimates of the measurement vector data using a Kalman recursive framework to obtain a priori measurement estimate;

[0012] Converting the a priori state estimate, the a priori measurement estimate and the actual control vector data into a differential sequence;

[0013] The differential sequence is used to train a recurrent neural network to obtain a tracked vehicle driving state prediction model; the output of the tracked vehicle driving state estimator is state vector prediction data; the state vector prediction data is the predicted real-time longitudinal speed, lateral speed and yaw angular velocity of the tracked vehicle.

[0014] Optionally, the training method of the transmission-track model specifically includes:

[0015] Acquire driving data of the tracked vehicle; the driving data includes: left and right driving motor torque training data, left and right driving wheel speed training data and transmission gear position training data;

[0016] Performing force analysis on the tracked vehicle to obtain vehicle force data; the vehicle force data includes: training data of lateral and longitudinal forces of the track of the tracked vehicle, training data of steering drive torque and training data of steering resistance torque;

[0017] The driving data is used as input and the vehicle force data is used as output to train a neural network and obtain a transmission-track model.

[0018] Optionally, the neural network is a long short-term memory neural network.

[0019] Optionally, the loss function of the recurrent neural network is:

[0020] in, represents the loss function, Δy k Indicates the measurement update difference.

[0021] Optionally, the recurrent neural network includes: three independent LSTM units and several FC layers;

[0022] Three independent LSTM units are used to learn the process noise covariance matrix, the state covariance matrix, and the measurement covariance matrix, respectively.

[0023] Optionally, the Kalman recursive framework is a Kalman recursive framework based on a mechanism model;

[0024] The processing of the Kalman recursive framework includes:

[0025] Build a tracked vehicle dynamics model:

[0026]

[0027] Where m is the vehicle mass, I z is the turning inertia of the vehicle, B is the track center distance, v x is the longitudinal velocity, v y is the lateral speed, ω is the steering angular velocity, i=1,2 represents the left and right tracks respectively, F xi is the longitudinal driving force of the ground acting on the tracks on both sides, F yi is the lateral force of the ground acting on the tracks on both sides, T di is the steering drive torque on both sides of the track, T ri is the steering resistance torque on both sides of the track, F N1 ,F N2 is the ground pressure on the left and right tracks, g is the acceleration due to gravity, and f r is the ground resistance coefficient;

[0028] The model state variable is x = [v x ,v y ,ω] T , the control quantity is u=[F x1 ,F x2 ,F y1 ,F y2 ,T r1 ,T r2 ,T d1 ,T d2 ] T , the quantity is measured as y = [a x ,a y ,ω] T , where a x ,a yare the longitudinal and lateral accelerations of the vehicle respectively, and ω is the yaw angular velocity;

[0029] The state transfer function is:

[0030]

[0031] According to the tracked vehicle dynamics model, the measurement function is obtained as follows:

[0032]

[0033] In a second aspect, the present application provides a method for predicting the driving state of a tracked vehicle, the method comprising:

[0034] Obtaining the state vector data, real-time measurement vector data and real-time measurable control vector data of the tracked vehicle to be predicted at the previous moment;

[0035] The state vector data of the previous moment, the real-time measurement vector data and the real-time measurable control vector data are input into the tracked vehicle driving state prediction model to obtain the real-time longitudinal speed, lateral speed and yaw angular velocity of the tracked vehicle to be predicted; the tracked vehicle driving state prediction model is trained by the tracked vehicle driving state prediction model training method described in any of the above items.

[0036] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned tracked vehicle driving state prediction model training methods or tracked vehicle driving state prediction methods.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned tracked vehicle driving state prediction model training methods or tracked vehicle driving state prediction methods.

[0038] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned tracked vehicle driving state prediction model training methods or tracked vehicle driving state prediction methods.

[0039] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0040] The present application provides a method for training a tracked vehicle driving state prediction model, a prediction method and a product, the method comprising: obtaining measurement vector data and measurable control vector data of the tracked vehicle; the measurement vector data comprising: longitudinal, lateral acceleration and yaw angular velocity of the tracked vehicle; the measurable control vector data comprising: left and right drive motor torques, left and right driving wheel speeds and transmission gears; inputting the measurable control vector data into a trained transmission-track model to obtain actual control vector data; the actual control vector data comprising: lateral and longitudinal forces exerted on the track by the ground, steering drive torque and steering resistance torque; obtaining the state vector data of the tracked vehicle at the previous moment; the state vector data The method includes: predicting the longitudinal, lateral speed and yaw rate of a tracked vehicle; calculating the prior estimate of the state vector data at the current moment by using the Kalman recursive framework to obtain the prior state estimate; calculating the prior estimate of the measurement vector data by using the Kalman recursive framework to obtain the prior measurement estimate; converting the prior state estimate, the prior measurement estimate and the actual control vector data into a differential sequence; training a recurrent neural network by using the differential sequence to obtain a tracked vehicle driving state prediction model; the output of the tracked vehicle driving state estimator is the state vector prediction data; the state vector prediction data is the predicted real-time longitudinal, lateral speed and yaw rate of the tracked vehicle. This application combines a neural network with good nonlinear fitting characteristics into the Kalman filter algorithm to learn potential noise characteristics, covariance matrix and Kalman gain, avoiding dependence on the mechanism model and operations such as matrix inversion, and finally realizing effective state estimation of the strong nonlinear system, and integrating the Kalman filter state estimation principle, using the mechanism model to calculate the prior estimation of the state quantity and quantity measurement, and combining the control quantity to perform differential sequence operations, and then the internal neural network only needs to use these differential sequences to learn the Kalman gain, with clear tasks, and the learning behavior is highly interpretable, so that there are fewer parameters to be learned, and the training efficiency is improved. This application can realize effective state estimation of the strong nonlinear system, and the neural network has fewer parameters to be learned, which improves the training efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A schematic diagram of a flow chart of a method for training a tracked vehicle driving state prediction model provided in one embodiment of the present application;

[0043] Figure 2A schematic diagram of a track shoe force analysis provided in an embodiment of the present application;

[0044] Figure 3 A schematic diagram of a state estimation process based on a Kalman neural network provided in an embodiment of the present application;

[0045] Figure 4 A schematic diagram of the internal recurrent neural network framework design of a Kalman neural network provided in an embodiment of the present application;

[0046] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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.

[0048] The present application provides a method for predicting the state of a tracked vehicle based on a Kalman neural network, which includes the following steps: considering the development of relative sliding between the track and the ground, establishing a track force model; performing data-driven modeling on the transmission-track model, and online predicting the ground force and steering torque on the track; combining the Kalman filter based on the mechanism model with the recurrent neural network to construct a Kalman neural network with a recursive structure; training the constructed network to learn the Kalman gain from the data to update and correct the state of the tracked vehicle. The present application is an architecture that combines mechanism knowledge and data-driven, which solves the problems of model mismatch and inability to adapt to strong nonlinearity that occur when traditional Kalman filtering is applied, and improves the interpretability of the neural network, with the characteristics of fast training convergence and high state estimation accuracy.

[0049] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0050] In an exemplary embodiment, Figure 1 As shown, a method for training a tracked vehicle driving state prediction model is provided, comprising:

[0051] S1. Obtaining measurement vector data and measurable control vector data of a tracked vehicle; the measurement vector data includes: longitudinal acceleration, lateral acceleration and yaw angular velocity of the tracked vehicle; the measurable control vector data includes: left and right drive motor torques, left and right driving wheel speeds and transmission gear.

[0052] S2. Input the measurable control vector data into the trained transmission-track model to obtain actual control vector data; the actual control vector data includes: the lateral and longitudinal forces exerted on the track by the ground, the steering drive torque and the steering resistance torque.

[0053] The training method of the transmission-track model specifically includes:

[0054] Acquire driving data of the tracked vehicle; the driving data includes: left and right driving motor torque training data, left and right driving wheel speed training data and transmission gear training data.

[0055] A force analysis is performed on the tracked vehicle to obtain vehicle force data; the vehicle force data includes: training data of lateral and longitudinal forces exerted on the track of the tracked vehicle, training data of steering drive torque and training data of steering resistance torque.

[0056] The driving data is used as input and the vehicle force data is used as output to train a neural network to obtain a transmission-track model. The neural network is a long short-term memory neural network.

[0057] S3. Obtain the state vector data of the tracked vehicle at the previous moment; the state vector data includes: the predicted values ​​of the longitudinal velocity, lateral velocity and yaw angular velocity of the tracked vehicle. The state vector data in this embodiment cannot be directly obtained, and each state vector data is obtained by prediction, wherein the first state vector data (i.e., the initial value) is obtained by initialization. S4. Use the Kalman recursive framework to calculate the prior estimate of the state vector data at the current moment to obtain the prior state estimate.

[0058] The Kalman recursive framework is a Kalman recursive framework based on a mechanism model;

[0059] The processing of the Kalman recursive framework includes:

[0060] Build a tracked vehicle dynamics model:

[0061]

[0062] Where m is the vehicle mass, I z is the turning inertia of the vehicle, B is the track center distance, v x is the longitudinal velocity, v y is the lateral speed, ω is the steering angular velocity, i=1,2 represents the left and right tracks respectively, F xi is the longitudinal driving force of the ground acting on the tracks on both sides, F yi is the lateral force of the ground acting on the tracks on both sides, T di is the steering drive torque on both sides of the track, T ri is the steering resistance torque on both sides of the track, FN1 ,F N2 is the ground pressure on the left and right tracks, g is the acceleration due to gravity, and f r is the ground resistance coefficient;

[0063] The model state variable is x = [v x ,v y ,ω] T , the control quantity is u=[F x1 ,F x2 ,F y1 ,F y2 ,T r1 ,T r2 ,T d1 ,T d2 ] T , the quantity is measured as y = [a x ,a y ,ω] T , where a x ,a y are the longitudinal and lateral accelerations of the vehicle respectively, and ω is the yaw angular velocity;

[0064] The state transfer function is:

[0065]

[0066] According to the vehicle dynamics model, the measurement function is obtained as follows:

[0067]

[0068] S5. Calculate the prior estimate of the measurement vector data using a Kalman recursive framework to obtain a priori measurement estimate.

[0069] S6. Convert the a priori state estimate, the a priori measurement estimate and the actual control vector data into a differential sequence.

[0070] S7. Use the differential sequence to train a recurrent neural network to obtain a tracked vehicle driving state prediction model; the output of the tracked vehicle driving state estimator is state vector prediction data; the state vector prediction data is the predicted real-time longitudinal speed, lateral speed and yaw angular velocity of the tracked vehicle.

[0071] The loss function of the recurrent neural network is: in, represents the loss function, Δy k Indicates the measurement update difference.

[0072] The recurrent neural network is a long short-term memory neural network; the long short-term memory neural network includes: three independent LSTM units and several FC layers; the three independent LSTM units are used for learning process noise covariance matrix, state covariance matrix and measurement covariance matrix respectively.

[0073] In this embodiment, the above process is implemented through the following four steps:

[0074] Step A1: Consider the relative sliding development process between the track and the ground, establish the track force model, first analyze the force of each track shoe, and obtain the normal force p of the ground on each track shoe. ik , longitudinal force ε xik and the lateral force ε yik , and then synthesize the distributed forces into concentrated forces, that is, the forces and moments acting on the tracks on both sides by the ground. Figure 2 , which describes the lateral and longitudinal sliding and ground forces on each track shoe when the tracked vehicle turns. Specifically, the contact force between each track shoe and the ground is decomposed into the normal force p ik , longitudinal force ε xik and the lateral force ε yik , where the subscript ik represents the kth track shoe from front to back on the track on side i.

[0075] The track force model of step A1 is specifically as follows:

[0076] First, the track winding speed is calculated by the driving wheel speed output by the electromechanical hybrid transmission system, and then the force of each track shoe is calculated according to the movement speed of each track shoe, including the normal ground pressure and the tangential horizontal force. The track winding movement is driven by the torque output by the driving wheel, and the dynamic equation of the track travel device is expressed as follows:

[0077]

[0078] Among them, i = 1, 2, representing the left and right sides respectively, J tr is the equivalent moment of inertia of the single-side crawler travel device at the driving wheel, is the angular acceleration of the driving wheel on one side, T spi is the driving torque output by the driving wheel, which is obtained by the torque of the driving motors on both sides through the front transmission mechanism, dual motor coupling mechanism, transmission and side transmission mechanism of the electromechanical hybrid transmission system. z F is the radius of the driving wheel, xi is the longitudinal force of the ground acting on one side of the crawler, ignoring the influence of the rotational inertia and mechanical efficiency of the transmission mechanism, the transmission system model representing the relationship between the driving wheel torque and the driving motor torque is established as follows, where i q ,i b ,ic They are front transmission ratio, second gear transmission ratio, and side transmission ratio, respectively. x is the planetary gear parameter of the coupling mechanism, T m1 and T m2 They are the torques output by the drive motors on the left and right sides respectively.

[0079]

[0080] In the track force model, the horizontal force between the track and the ground is composed of concentrated force (longitudinal force F xi , lateral force F yi ) and concentrated moment (steering resistance moment T ri , Steering drive torque T di ) is expressed. In order to more accurately simulate the dynamic behavior of the tracked vehicle and the complex contact situation with the ground, the force on each track shoe is calculated separately, and the ground force on each track shoe is regarded as a concentrated force. The concentrated contact force is decomposed into the normal force p ik , longitudinal force ε xik and the lateral force ε yik , where the subscript ik represents the kth track shoe from the front to the back of the track on the i side. When the tracked vehicle does not roll significantly, the influence of the steering centrifugal force on the track ground pressure distribution is considered, and the track ground pressure is equivalent to a trapezoidal distribution. The ground pressure F on the two tracks is calculated. Ni As follows, where m is the mass of the tracked vehicle, g is the acceleration due to gravity, and h c is the height of the center of mass, B is the center distance of the track, ω is the yaw angular velocity, v x is the longitudinal speed, is the lateral acceleration.

[0081]

[0082] The number of track shoes on one side of the track that touch the ground is recorded as n. Assuming that the ground pressure of the track is distributed in a trapezoidal shape, due to the ground pressure F on one side, Ni The normal force p of each track shoe on this side ik The normal force (ground pressure) of each track shoe is calculated by integrating the track ground length as follows: ik is the x-coordinate of the track shoe, L is the ground contact length of the track, v y is the lateral speed, is the longitudinal acceleration.

[0083]

[0084] The horizontal friction force acting on each track shoe can be calculated by the following formula, where the friction coefficient μ and the ground shear modulus K are the contact parameters between the track and the ground, ε ikis the actual friction force, ε maxik is the maximum friction force, j ik is the sliding amount of the track shoe, that is, the shear amount of the ground here, j ik It can be calculated based on the sliding speed of each track shoe, p ik is the normal force of the track shoe, calculated by the above formula.

[0085]

[0086] The steering of the tracked vehicle is driven by the differential speed of the driving wheels on both sides. Assuming that the extension and contraction of the track can be ignored, the speed of the track relative to the vehicle body is v spi =ω spi r z , where ω spi and r z are the angular velocity and radius of the driving wheel respectively. In the vehicle coordinate system V(x,y), the sliding speed of the track shoe relative to the ground is calculated by the following formula.

[0087]

[0088] In the formula, v jxik is the longitudinal sliding speed of the track shoe relative to the ground, v jyik is the lateral sliding speed of the track shoe relative to the ground, R s is the turning radius, L is the track ground length, v spi is the speed of the track relative to the vehicle body, and the x coordinate of the track shoe x ik The calculation formula is:

[0089]

[0090] Where, t 0ik is the time when the track shoe starts to contact the ground, t is the current time, and in the definite integral, τ is the independent variable (time), and its upper and lower limits are t and t 0ik .

[0091] According to the direction angle φ of the tracked vehicle, the sliding speed v of the track shoe is obtained by transforming it into the inertial coordinate system. jXik 、v jYik , and then by integrating the sliding velocity, we get the shear amount j ik and longitudinal and transverse shear j Xik 、j Yik As follows:

[0092]

[0093] According to the relationship between friction and shear, the ground friction acting on each track shoe is obtained. Assuming that the friction on the track shoe is opposite to its sliding speed, the friction on each track shoe can be decomposed into the x and y directions by the following formula:

[0094]

[0095] In the formula, δ ik is the direction angle of the track shoe sliding velocity. ik Satisfies the following formula:

[0096]

[0097] According to the shear displacement relationship between the vehicle track and the ground, the normal force and tangential force of each track shoe are calculated, and finally the distributed force is converted into a concentrated force. The force and moment acting on the track by the ground are obtained as follows:

[0098]

[0099] In the above formula, F xi and F yi are the longitudinal and lateral concentrated forces acting on the track from the horizontal ground force, ε xik and ε yik T is the longitudinal and transverse contact force of the kth track shoe under the horizontal action of the ground. ri is the steering resistance torque caused by the distributed lateral contact force, T di is the steering drive torque caused by the distributed longitudinal contact force.

[0100] After the transmission system model and track force model are established as above, the relationship between the ground force exerted on the tracks on both sides and the steering torque and the input torque of the driving motors on both sides can be described under a certain transmission gear.

[0101] Step A2: Use a neural network to perform data-driven modeling on the transmission system model and the track force model to obtain a transmission-track model to reduce the computational complexity. The model is used to online predict the lateral and longitudinal forces and steering torque exerted on the track under given torque inputs from the drive motors on both sides, so that they can be input as control quantities into the state transfer equation to calculate the prior state estimate.

[0102] The data-driven transmission-track model in step A2 is specifically as follows:

[0103] A2.1. First, use the transmission system model and track force model to simulate and generate a training data set, including:

[0104] Given a series of different driving conditions with target vehicle speed and target yaw rate, tracked vehicle tracking control is performed using different gears, and then the input torque T of the drive motors on both sides in the running data is used. m1 ,T m2 , driving wheel speed ω sp1 ,ω sp2 The transmission gear position z is used as the input of the data-driven model, and the ground lateral and longitudinal forces F calculated by the electromechanical hybrid transmission system and the track force model are xi ,F yi and steering torque T ri ,T di As the output of the data-driven model, the input and output are collected to build a training database.

[0105] A2.2. Then select the appropriate neural network to build a data-driven transmission-track model, including:

[0106] The neural network type is selected according to the type of input and output. Since the driving motor torque and the ground force and torque on the track are time series data, there is a dependency between the measurement and the historical input. Therefore, the long short-term memory neural network in the recurrent neural network is selected to predict the output according to the input. The long short-term memory neural network includes: input layer, LSTM layer, and output layer. The input layer model at any k moment is:

[0107]

[0108] In the formula, represents the input torque of the left and right drive motors, represents the speed of the left and right driving wheels, z is the gear position of the transmission, and at any k moment, each LSTM layer consists of 64 cells (hidden layer), and the model of each cell is:

[0109]

[0110] Where σ(·) is the sigmoid activation function; i k , is the input threshold layer; f k is the forget threshold layer; o k ,h k is the output threshold layer; c k is the cell state at the current moment; W i ,W c ,W f ,W o and b i ,b c ,b f ,b oare the weight matrices and bias terms to be learned for the input threshold layer, forget threshold layer, and output threshold layer of each cell unit; the subscripts k and k-1 indicate that the variable is the variable at the current moment and the variable at the previous moment respectively; each cell unit has an output h k ; The output layer converts the output h of 64 cells k Add them together to solve the ground force and moment at the current moment The superscript k indicates that the variable is the variable at the current moment; the output expression is as follows:

[0111]

[0112] After the data-driven transmission-track model is established as above, the lateral and longitudinal forces, steering drive torque and steering resistance torque exerted on the track by the ground can be predicted online according to the torque input of the drive motors on both sides under a certain transmission gear, so as to be input into the state transfer equation as control quantities to calculate the prior state estimation and prior measurement estimation.

[0113] Step A3, combining the Kalman filter based on the mechanism model with the data-driven recurrent neural network to construct a Kalman neural network with a recursive structure. Figure 3 , Figure 3 It is the state estimation flow chart based on Kalman neural network described in step A3 of the present application, including an external Kalman filter state estimation framework and an internal recurrent neural network framework, which describes the principle process of using Kalman neural network for differential and recursive operations to finally realize state estimation.

[0114] The Kalman neural network principle and design of step A3 are as follows:

[0115] The principle and design of the Kalman neural network include Kalman filtering theory, state estimation principle based on Kalman neural network and Kalman neural network design based on the principle.

[0116] A3.1, the Kalman filter theory, analyzes the principle of Kalman filter for state estimation of nonlinear systems, which is the theoretical basis for the subsequent design of the internal and external recursive framework of the Kalman neural network, specifically:

[0117] A time-varying nonlinear system is established, and the system noise has a Gaussian distribution. The state equation after discretization is as follows:

[0118]

[0119] where x k ,y k ,u k are the state value, measured value and controlled value at time k, respectively, k-1represents the state value at time k-1, X k-1 represents the state value at the previous moment (k-1 moment); N represents the Gaussian distribution; w k ,v k are process noise and measurement noise, respectively, which conform to Gaussian distribution with mean 0. Their covariance matrices are Q and R, respectively. f(·) and h(·) are the state transfer function and measurement function of the nonlinear system, respectively.

[0120] The model-based Kalman filter algorithm is a linear recursive estimation algorithm that uses the posterior state estimation of the previous moment. To predict the prior state estimate at the current moment At the same time, use the current measurement value y k To correct the prior estimate to obtain the posterior state estimate at the current moment The posterior state estimate is also called the optimal state estimate, and the cycle continues until the iteration condition is met. The application of this algorithm in nonlinear systems is the extended Kalman filter algorithm, which uses the first-order Taylor expansion to locally linearize the nonlinear problem and improves the estimation effect of the nonlinear system. The extended Kalman filter algorithm can be described by a two-step process: state update and measurement correction, in which each step calculates the estimated mean and covariance matrix. Specifically, it includes:

[0121] 1) In the state update step, the prior state estimate is predicted based on the posterior state estimate at the previous moment and the control amount at the current moment and prior measurement estimates The specific prediction formula is as follows:

[0122]

[0123] When f(·) and h(·) are differentiable, the extended Kalman filter algorithm linearizes them by calculating the Jacobian matrix by taking partial derivatives, and and The value at which the state transfer Jacobian matrix A is taken k and the measurement Jacobian matrix H k The calculation of is as follows:

[0124]

[0125] The prediction uncertainty is given by the state covariance matrix P k|k-1 and the measurement covariance matrix S k|k-1 Description, the specific calculation is as follows:

[0126]

[0127] 2): In the measurement correction step, the a posteriori estimate is calculated based on the prior estimate. The key to this step is to calculate the Kalman gain Kk , calculated through the state covariance matrix, the measurement covariance matrix and the measurement matrix hidden in the measurement covariance matrix:

[0128]

[0129] The current measurement value y k Compare with the prior measurement estimate to get the measurement update difference Δy k , combined with the Kalman gain, the posterior estimate of the state is updated and the state covariance matrix P k , the specific calculation is as follows:

[0130]

[0131] A3.2, the state estimation principle based on Kalman neural network is as follows:

[0132] Similar to the extended Kalman filter algorithm, at each time t = k, the Kalman neural network estimates the state in two steps: state update and measurement correction.

[0133] The state update step is the same as the extended Kalman filter algorithm, only predicting the prior estimates of the state and measurement and Specifically, the Kalman neural network first uses the posterior state estimate of the previous moment The control quantity u at the current moment k and the state transfer function f(·) to calculate the current prior state estimate Then, using the current prior state estimate The control quantity u at the current moment k and the measurement function h(·) to compute the current prior measurement estimate

[0134] The measurement correction steps are the same as the extended Kalman filter algorithm, using the current measurement value y k and prior measurement estimates To calculate the updated difference Δy k , combined with the Kalman gain K k To calibrate Get the posterior state estimate at the current moment Unlike the extended Kalman filter algorithm, the Kalman gain in the Kalman neural network does not need to calculate the covariance matrix P first. k|k-1 and S k|k-1 Instead of being solved explicitly, it is learned from the data by a recurrent neural network without knowing the noise statistics.

[0135] A3.3, the principle-based Kalman neural network design, according to the principle described in A3.2, the Kalman neural network is designed to estimate the state of the tracked vehicle, including an external Kalman recursive framework based on the mechanism model and an internal recurrent neural network framework based on data drive. Figure 4 , Figure 4 It is the design of the internal recurrent neural network framework described in step A3.3 of this application, which describes the process of using the recurrent neural network architecture to track the covariance matrices in the Kalman filter principle from the data and finally obtain the Kalman gain.

[0136] The external Kalman recursive framework design based on the mechanism model in A3.3 is specifically as follows:

[0137] In order to carry out the state update step described in A3.2, a tracked vehicle dynamics model is established to describe the movement in the longitudinal, lateral and yaw directions on the horizontal ground, as follows:

[0138]

[0139] Where m is the vehicle mass, I z is the turning inertia of the vehicle, B is the track center distance, v x is the longitudinal velocity, v y is the lateral speed, ω is the steering angular velocity, i=1,2 represents the left and right tracks respectively, F xi is the longitudinal driving force of the ground acting on the tracks on both sides, F yi is the lateral force of the ground acting on the tracks on both sides, T ri is the steering resistance torque on both sides of the track, T di is the steering drive torque on both sides of the track, F N1 ,F N2 is the ground pressure on the left and right tracks, g is the acceleration due to gravity, and f r is the ground resistance coefficient.

[0140] The model state variable is x = [v x ,v y ,ω] T , the control quantity is u=[F x1 ,F x2 ,F y1 ,F y2 ,T r1 ,T r2 ,T d1 ,T d2 ] T , the quantity is measured as y = [a x ,a y ,ω] T , where a x ,ay are the longitudinal and lateral accelerations of the vehicle, respectively, measured by the accelerometer, and the yaw rate ω is measured by the yaw rate sensor. The control quantity u is calculated online by the data-driven transmission-track model at a certain transmission gear according to the torque input of the drive motors on both sides.

[0141] Therefore, the state transfer function is:

[0142]

[0143] According to the vehicle dynamics model, the measurement function is obtained as follows:

[0144]

[0145] In order to carry out the measurement correction step described in A3.2, the state quantity, quantity measurement and control quantity are processed to obtain the input quantity of the internal recurrent neural network framework so that the recurrent neural network can learn the Kalman gain to correct the prior state estimate, as follows:

[0146] A3.1 The Kalman filter theory It shows that in order to obtain the Kalman gain, it is necessary to first learn the measurement matrix information H from the state transition information and noise statistical characteristics. k , covariance matrix information P k|k-1 and S k|k-1 , therefore, the input quantities of the internal recurrent neural network framework are obtained by processing the state quantity, quantity measurement and control quantity as follows:

[0147] Measure the evolution difference F1:

[0148] Measurement update difference F2:

[0149] State evolution difference F3:

[0150] Status update difference F4:

[0151] Control value difference F5: Δu k =u k -u k-1 ;

[0152] Among them, y k-1 is the measured value at the previous moment, is the posterior state estimate of the previous moment, u k-1 is the control quantity at the previous moment. The input features are all differential sequences. F1, F3 and F5 represent the measurement evolution difference, state evolution difference and control quantity difference of state transfer information. F2 and F4 include the measurement update difference and state update difference representing the statistical characteristics of noise.

[0153] Using the differential sequence as the network input can better learn the noise characteristics, optimize the error caused by the noise, avoid the calculation error caused by directly predicting the state quantity, and speed up the training convergence. In order to make the recurrent neural network fully learn the Kalman gain, a comprehensive differential sequence {F1, F2, F3, F4, F5} is selected as the input feature sequence. Among them, in the current time step k, The Kalman neural network is waiting for calculation, so the available F3 and F4 features are and

[0154] The external Kalman recursive framework processes each variable to obtain After that, the input feature difference sequence is used for the training and calculation of the internal recurrent neural network, and the Kalman gain K of the recurrent neural network output is k It is then used to correct the prior state estimate.

[0155] The A3.3 internal recurrent neural network framework design is specifically as follows:

[0156] The long short-term memory neural network in the recurrent neural network is selected to learn the covariance matrix in the Kalman filter principle in the data. This recurrent neural network structure with long-term memory and long-term dependence can control the flow of information through gating units and solve the gradient disappearance and explosion problems in traditional RNN.

[0157] The RNN architecture uses a separate LSTM unit to track the learning of the covariance matrix for each bottom layer. It consists of three independent LSTM units (LSTM 1, LSTM 2 and LSTM 3) and some FC layers. When tracking the state covariance matrix P k|k-1 Before that, we need to learn about process noise Q k Statistical characteristics, so the design of LSTM 1 from From learning and tracking the process noise Q, the predicted Q k With input features Δu k Input into LSTM 2 to learn and track the state covariance matrix P k|k-1 Similarly, the P predicted by LSTM 2 k|k-1 With the input feature Δy k , Δu k Input the learning measurement matrix H into LSTM 3 together k The measurement covariance matrix S is calculated by combining the measurement noise R k|k-1 Finally, combined with P k|k-1 and S k|k-1 To learn the Kalman gain K at the current momentk .

[0158] The hidden layer dimensions of each LSTM module and the input and output layer dimensions of the FC module in the recurrent neural network structure are designed specifically as follows: first, the state quantity dimension is regarded as m, m=3, the quantity measurement dimension is regarded as n, n=3, and the control quantity dimension is regarded as r, r=8; then the dimensions of each neural network module are designed based on these dimensions.

[0159] LSTM 1 hidden layer dimension of tracking process noise Q is m 2 , tracking state covariance matrix P k|k-1 The LSTM 2 hidden layer dimension is m 2 , tracking measurement covariance matrix S k|k-1 The LSTM 3 hidden layer dimension is n 2 ; There are 7 fully connected layer modules in total, the input layer dimension of FC 1 and the hidden layer dimension of LSTM2 are m 2 Same, the output layer dimension is the same as the LSTM 3 hidden layer dimension n 2 Same; the input layer dimension of FC 2 is the hidden layer dimension m of LSTM 2 2 and the hidden layer dimension n of LSTM 3 2 The sum of the output layer dimension and the Kalman gain K k The dimension of FC 3’s input layer is the same as the dimension of LSTM 3’s hidden layer, n. 2 The sum of the output layer dimension m·n of FC 2, the output layer dimension and the hidden layer dimension m of LSTM 2 2 Same; the input layer dimension of FC 4 is the hidden layer dimension m of LSTM 2 2 and the output layer dimension of FC 3 is m 2 The sum of the output layer dimension and the LSTM 2 hidden layer dimension m 2 Same; the input layer dimension of FC 5 is the state update difference The dimension of the output layer is an integer multiple of the dimension of the input layer; the dimension of the input layer of FC 6 is the state evolution difference Dimension m and control value difference Δu k The sum of the dimensions r, the output layer dimension is an integer multiple of the input layer dimension; the input layer dimension of FC 7 is the measured evolution difference Δy k Dimension n, measurement update difference Dimension n and control value difference Δu k The sum of the dimensions r, the output layer dimension is an integer multiple of the input layer dimension.

[0160] Step A4, training the network described in step A3 to learn the Kalman gain from the data, using the transmission-track model described in step S2 to obtain the actual control quantity, and updating and measuring the tracked vehicle state based on the Kalman filter principle, and finally obtaining the tracked vehicle state estimation value based on the Kalman neural network.

[0161] The training of the tracked vehicle state estimator based on the Kalman neural network in step S4 includes loss function design, learning basis and training process, specifically:

[0162] The loss function of the training is designed according to unsupervised learning, so that the Kalman neural network can train itself online to learn the Kalman gain and perform state estimation, as follows:

[0163] Although the training dataset lacks K k The network cannot directly calculate its loss to update the network weights. k and K k-1 There is a connection between them, and the difference Δy can be updated by calculating the measurement in an end-to-end manner k The loss is used to train the Kalman neural network, which indirectly affects K k For learning, use the squared error loss function The formula is as follows:

[0164]

[0165] The basis for learning the Kalman gain through this loss function is as follows:

[0166]

[0167] Among them, K k-1 is the Kalman gain at time k-1, Δy k-1 is the measurement update difference at time k-1, It is a custom difference value, defined as: About K k-1 The expression of is as follows, where is the prior state estimate of the previous moment:

[0168]

[0169] The training process of the tracked vehicle driving state estimator based on the Kalman neural network is as follows:

[0170] Step 1: Initialize the state vector of the tracked vehicle, and collect the measured vector data and measurable control vector data of the tracked vehicle in real time. The measured vector data include: the longitudinal, lateral acceleration and yaw angular velocity of the tracked vehicle, and the measurable control vector data include: the left and right drive motor torques, the left and right driving wheel speeds and the current transmission gear selection.

[0171] Step 2: Input the measurable control vector data into the data-driven transmission-track model described in claim 3 to obtain the lateral and longitudinal forces, steering drive torque and steering resistance torque exerted on the current track, and use them as actual control vector data.

[0172] Step 3: Input the measurement vector data and the actual control vector data into the Kalman neural network. After the prior estimation calculation and differential operation of the external Kalman filter framework, input the differential sequence into the internal recurrent neural network framework. The forward propagation obtains the Kalman gain, which is then used by the external Kalman filter framework to perform measurement correction to obtain the predicted state vector, and calculate the loss according to the loss function. Then, back propagation is performed to obtain the gradient value of the internal neural network parameters. Finally, the Adam optimizer updates the neural network parameters according to the calculated gradient value. After multiple time steps of update iteration to the preset loss threshold, a tracked vehicle driving state estimator based on the Kalman neural network is obtained.

[0173] At present, state estimation methods based on mechanism models, such as Kalman filtering and its variants, require the establishment of a kinematic or dynamic model of the tracked vehicle before combining sensor information for state estimation. They are suitable for linear or weakly nonlinear systems that can be modeled. However, when a tracked vehicle is traveling, there is dynamic coupling between the directions, the coupling of state variables in each direction is obvious, and the motion nonlinearity is strong. Since the track force model is based on the ground shear displacement, the integration of time and position is usually used, and the nonlinearity is strong. At the same time, the drastic changes in ground conditions and random disturbances caused by the tracked vehicle will lead to mismatch between the vehicle nominal model and the actual controlled object. Strong nonlinearity and model mismatch lead to poor results of state estimation methods based on mechanism models. Data-driven state estimation methods do not need to establish complex high-precision models, but learn the relationship between system input and output through data. They have good applicability, but lack interpretability. At the same time, the black box characteristics will lead to many parameters to be learned and long training time. In order to overcome the limitations of the above two types of state estimation methods, it is a trend to combine mechanism knowledge and data-driven state estimation methods.

[0174] This application combines a neural network with good nonlinear fitting characteristics into the Kalman filter algorithm to learn potential noise characteristics, covariance matrix and Kalman gain, avoiding dependence on the mechanism model and operations such as matrix inversion, and finally realizing effective state estimation of the strong nonlinear system. This application incorporates the Kalman filter state estimation principle. First, the mechanism model is used to calculate the prior estimate of the state quantity and quantity measurement, and the differential sequence operation is performed in combination with the control quantity. Then the internal neural network only needs to use these differential sequences to learn the Kalman gain, which has a clear task, and the neural network learns the process noise covariance matrix, the state covariance matrix and the measurement covariance matrix respectively by designing three LSTM modules. The learning behavior is highly interpretable, so that there are fewer parameters to be learned and the training efficiency is improved.

[0175] In another exemplary embodiment, a method for predicting a running state of a tracked vehicle is provided, the method comprising:

[0176] Obtaining the state vector data, real-time measurement vector data and real-time measurable control vector data of the tracked vehicle to be predicted at the previous moment;

[0177] The state vector data of the previous moment, the real-time measurement vector data and the real-time measurable control vector data are input into the tracked vehicle driving state prediction model to obtain the real-time longitudinal speed, lateral speed and yaw angular velocity of the tracked vehicle to be predicted; the tracked vehicle driving state prediction model is trained by the above-mentioned tracked vehicle driving state prediction model training method.

[0178] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for training a tracked vehicle driving state prediction model or a method for predicting a tracked vehicle driving state is implemented.

[0179] Those skilled in the art will understand that Figure 5The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0180] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0181] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0182] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0183] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0184] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0185] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0186] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0187] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for training a tracked vehicle driving state prediction model, characterized in that: The tracked vehicle driving state prediction model training method comprises: Acquire measurement vector data and measurable control vector data of the tracked vehicle; the measurement vector data includes: longitudinal acceleration, lateral acceleration and yaw rate of the tracked vehicle; the measurable control vector data includes: left and right drive motor torques, left and right driving wheel speeds and transmission gear position; Input the measurable control vector data into the trained transmission-track model to obtain actual control vector data; the actual control vector data includes: the lateral and longitudinal forces exerted on the track by the ground, the steering drive torque and the steering resistance torque; Acquire the state vector data of the tracked vehicle at the last moment; the state vector data includes: the predicted values ​​of the longitudinal velocity, lateral velocity and yaw angular velocity of the tracked vehicle; The Kalman recursive framework is used to calculate the prior estimate of the state vector data at the current moment to obtain the prior state estimate; Calculating a priori estimates of the measurement vector data using a Kalman recursive framework to obtain a priori measurement estimate; Converting the a priori state estimate, the a priori measurement estimate and the actual control vector data into a differential sequence; The differential sequence is used to train a recurrent neural network to obtain a tracked vehicle driving state prediction model; the output of the tracked vehicle driving state estimator is state vector prediction data; the state vector prediction data is the predicted real-time longitudinal speed, lateral speed and yaw angular velocity of the tracked vehicle.

2. The method for training a tracked vehicle driving state prediction model according to claim 1, characterized in that: The training method of the transmission-track model specifically includes: Acquire driving data of the tracked vehicle; the driving data includes: left and right driving motor torque training data, left and right driving wheel speed training data and transmission gear position training data; Performing force analysis on the tracked vehicle to obtain vehicle force data; the vehicle force data includes: training data of lateral and longitudinal forces of the track of the tracked vehicle, training data of steering drive torque and training data of steering resistance torque; The driving data is used as input and the vehicle force data is used as output to train a neural network and obtain a transmission-track model.

3. The method for training a tracked vehicle driving state prediction model according to claim 2, characterized in that: The neural network is a long short-term memory neural network.

4. The method for training a tracked vehicle driving state prediction model according to claim 1, characterized in that: The loss function of the recurrent neural network is: in, represents the loss function, Δy k Indicates the measurement update difference.

5. The method for training a tracked vehicle driving state prediction model according to claim 1, characterized in that: The recurrent neural network includes: three independent LSTM units and several FC layers; Three independent LSTM units are used to learn the process noise covariance matrix, the state covariance matrix, and the measurement covariance matrix, respectively.

6. The method for training a tracked vehicle driving state prediction model according to claim 1, characterized in that: The Kalman recursive framework is a Kalman recursive framework based on a mechanism model; The processing of the Kalman recursive framework includes: Build a tracked vehicle dynamics model: In the formula, is the longitudinal acceleration, is the lateral acceleration, is the yaw angular acceleration, i=1,2 represents the left and right tracks respectively, m is the vehicle mass, I z is the turning inertia of the vehicle, B is the track center distance, v x is the longitudinal velocity, v y is the lateral velocity, ω is the steering angular velocity, F xi is the longitudinal driving force of the ground acting on the tracks on both sides, F yi is the lateral force of the ground acting on the tracks on both sides, T di is the steering drive torque on both sides of the track, T ri is the steering resistance torque on both sides of the track, F N1 ,F N2 is the ground pressure on the left and right tracks, g is the acceleration due to gravity, and f r is the ground resistance coefficient; The model state variable is x = [v x ,v y ,ω] T , the control quantity is u=[F x1 ,F x2 ,F y1 ,F y2 ,T r1 ,T r2 ,T d1 ,T d2 ] T , the quantity is measured as y = [a x ,a y ,ω] T , where a x ,a y are the longitudinal and lateral accelerations of the vehicle respectively, and ω is the yaw angular velocity; The state transfer function is: According to the vehicle dynamics model, the measurement function is obtained as follows:

7. A method for predicting the driving state of a tracked vehicle, characterized in that: The method for predicting the running state of a tracked vehicle comprises: Obtaining the state vector data, real-time measurement vector data and real-time measurable control vector data of the tracked vehicle to be predicted at the previous moment; The state vector data of the previous moment, the real-time measurement vector data and the real-time measurable control vector data are input into the tracked vehicle driving state prediction model to obtain the real-time longitudinal speed, lateral speed and yaw angular velocity of the tracked vehicle to be predicted; the tracked vehicle driving state prediction model is trained by the tracked vehicle driving state prediction model training method described in any one of claims 1-6.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the tracked vehicle driving state prediction model training method described in any one of claims 1 to 6 or the tracked vehicle driving state prediction method described in claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for training a tracked vehicle driving state prediction model according to any one of claims 1 to 6 or the method for predicting a tracked vehicle driving state according to claim 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for training a tracked vehicle driving state prediction model according to any one of claims 1 to 6 or the method for predicting a tracked vehicle driving state according to claim 7 is implemented.

Images