Mass center slip angle fusion estimation method based on intelligent automobile

Abstract

The invention discloses a side slip angle fusion estimation method based on an intelligent automobile, relates to the technical field of state estimation and control of the intelligent automobile, and aims to solve the problem of how to ensure high precision and adapt to side slip angle estimation of different driving conditions when the intelligent automobile is under complex driving working conditions. The vehicle operation information acquisition module acquires motion state parameters of the vehicle in real time; the side slip angle estimation influence factor calculation module comprehensively considers multiple factors and quantifies various influence factors, so that the estimation model adapts to different driving working conditions; the side slip angle estimation execution module executes a plurality of estimation modes in parallel under different working conditions; the side slip angle fusion estimation module based on the genetic algorithm and the neural network optimizes the weight factor offline through the genetic algorithm, and the neural network predicts the weight factor of each estimation mode in real time according to the influence factor in the online reasoning stage. According to the method, accurate side slip angle estimation can be provided in real-time driving, and the driving stability of the intelligent automobile is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent vehicle state estimation and control, and in particular to a method for fusion estimation of center of mass sideslip angle based on an intelligent vehicle. Background Art

[0002] With the rapid development of intelligent and automated technologies, smart cars have become a core component of future transportation systems. In order to ensure the high safety and stability of smart cars, accurate estimation of the vehicle's center of mass slip angle is one of the key technologies to improve driving safety and control system responsiveness. The center of mass slip angle is an important parameter to measure the lateral slip of the vehicle. Its estimation accuracy directly affects the stability of the autonomous driving system, the vehicle's path tracking ability, and the precise control of the steering system. Especially in complex and dynamically changing driving environments, accurate center of mass slip angle estimation can effectively improve the vehicle's dynamic control capabilities, thereby ensuring driving stability and driving safety.

[0003] However, the existing methods for estimating the center of mass sideslip angle have certain limitations in practical applications. Many traditional methods rely on simplified linear models or assumed physical models, which makes them show low robustness and accuracy when dealing with high noise environments and variable driving behaviors. Although some nonlinear processing methods, such as unscented Kalman filtering, have solved the shortcomings of traditional methods to a certain extent, single estimation methods still have obvious problems when facing diversified driving conditions. Specifically, in specific driving scenarios such as complex road conditions or extreme weather conditions, a single estimation method may be difficult to fully adapt to environmental changes. Considering that different estimation methods have different adaptability to different working conditions, how to select the best estimation model according to different application scenarios is a key factor in improving the accuracy of center of mass sideslip angle estimation. Therefore, there is an urgent need for a center of mass sideslip angle fusion estimation method based on intelligent vehicles. By combining multiple estimation methods and adaptively adjusting the contribution weight of each estimation method according to different working conditions, each estimation method can give full play to its advantages under specific conditions, effectively improve the performance of center of mass sideslip angle estimation, and ensure the stable operation of intelligent vehicles in variable environments. Summary of the invention

[0004] In order to solve the problems faced by the above-mentioned background technology in the estimation of the center of mass sideslip angle of an intelligent vehicle, the present invention provides a center of mass sideslip angle fusion estimation method based on an intelligent vehicle.

[0005] In order to achieve the above object, the content of the present invention is as follows:

[0006] A method for fusing and estimating the centroid side slip angle based on an intelligent vehicle, which includes a vehicle operation information acquisition module, a centroid side slip angle estimation influence factor calculation module, a centroid side slip angle estimation execution module, and a centroid side slip angle fusion estimation module based on a genetic algorithm and a neural network;

[0007] The vehicle operation information acquisition module is used to obtain the longitudinal speed v x , longitudinal acceleration a x , wheel slip ratio λ, longitudinal tire force F x1 and F x2 , lateral speed v y , lateral acceleration a y , yaw rate r, tire lateral force F y , tire side slip angle α, front wheel steering angle δ f , body roll angle φ, body pitch angle θ, front wheel suspension stroke s f , rear wheel suspension stroke s r , front wheel normal force F z,f , rear wheel normal force F z,r , road surface friction coefficient μ, road curvature κ, crosswind speed v w ;

[0008] The centroid side slip angle estimation influence factor calculation module calculates the vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3; the vehicle dynamic characteristic factor influence factor E1 evaluates the influence of the vehicle's own dynamic characteristics on the centroid side slip angle estimation, so that the estimation method adapts to different driving states; the driver factor influence factor E2 measures the influence of the driver's driving behavior and personal characteristics on the centroid side slip angle estimation, and identifies the uncertainty brought by human operation; the environmental road condition factor influence factor E3 quantifies the influence of the external environment and road conditions on the centroid side slip angle estimation, so that the estimation method is applicable to different road conditions and meteorological conditions; the vehicle dynamic characteristic factor influence factor E1 depends on the longitudinal dynamic characteristic influence factor E 1x , lateral dynamic characteristic influence factor E 1y , vertical dynamic characteristic influence factor E 1z , where the longitudinal dynamic characteristic influence factor E 1x depends on the longitudinal speed v x , longitudinal acceleration a x , wheel slip ratio λ, driving force F d , braking force F b , longitudinal tire force F x1 and F x2 , the lateral dynamic characteristic influence factor E 1y depends on the vehicle lateral inertia factor G1, the tire lateral adhesion factor G2, and the driver lateral input factor G3, where the vehicle lateral inertia factor G1 depends on the lateral speed vy 、Lateral acceleration a y 、Yaw rate r, and the tire lateral adhesion factor G2 depends on the tire lateral force F y 、Tire side slip angle α, empirical parameter k1, empirical parameter k2, and the driver lateral input factor G3 depends on the front wheel steering angle δ f , and the vertical dynamic characteristic influence factor E 1z depends on the vehicle body roll angle φ, vehicle body pitch angle θ, front wheel suspension stroke s f 、Rear wheel suspension stroke s r 、Front wheel normal force F z,f 、Rear wheel normal force F z,r ; The driver factor influence factor E2 depends on the driver experience category C i and the driver experience category probability factor P(C i ), where the driver experience category probability factor P(C i ) depends on the driving age Y, total driving mileage S, driving style D, accident rate A; The environmental road condition factor influence factor E3 depends on the road adhesion factor R1, road gradient factor R2, environmental weather factor R3, road curvature factor R4, crosswind influence factor R5, where the road adhesion factor R1 depends on the current road surface friction coefficient μ, standard dry road surface friction coefficient μ0, and the road gradient factor R2 depends on the longitudinal gradient angle θ x 、Lateral gradient angle θ y , the environmental weather factor R3 depends on the solar altitude angle ξ, air temperature Tem, precipitation intensity Precip, air temperature adjustment coefficient γ1, precipitation intensity influence coefficient γ2, and the road curvature factor R4 depends on the road curvature The crosswind influence factor R5 depends on the crosswind speed v w 、Longitudinal speed v x ;

[0009] The calculation formula of the vehicle dynamic characteristic factor influence factor E1 is as follows:

[0010]

[0011] In the formula, E 1x is the longitudinal dynamic characteristic influence factor, E 1y is the lateral dynamic characteristic influence factor, E 1z is the vertical dynamic characteristic influence factor; m1 is the longitudinal dynamic characteristic weight coefficient, m2 is the lateral dynamic characteristic weight coefficient, m3 is the vertical dynamic characteristic weight coefficient, and their value ranges are all between 0 and 1; u is the adjustment parameter that controls the sensitivity and smoothness of the output of the calculation formula of the vehicle dynamic characteristic factor influence factor E1, and its value range is between 0.5 and 2;

[0012] The calculation formula of the driver factor influence factor E2 is as follows:

[0013]

[0014] Wherein, C i is the driver experience category; among which, C1 is a novice driver with a short driving experience, low driving mileage, aggressive driving style, and a relatively high accident rate, and the value of C1 is taken as 1 during calculation; C2 is an average driver with medium driving experience and mileage, mild driving style, and a relatively low accident rate, and the value of C2 is taken as 2 during calculation; C3 is a skilled driver with high driving experience and mileage, stable driving style, and a very low accident rate, and the value of C3 is taken as 3 during calculation; C4 is a professional driver with a long driving experience, long driving mileage, stable driving style, and almost no accidents, and the value of C4 is taken as 4 during calculation; P(C i ) is the driver experience category probability factor. To improve the accuracy of the driver factor influence factor E2, the calculation of P(C i ) uses a deep feedforward neural network structure and is obtained by a neural network classifier; among which, P(C1) is the novice driver probability factor, P(C2) is the average driver probability factor, P(C3) is the skilled driver probability factor, and P(C4) is the professional driver probability factor, and their value ranges are all between 0 and 1; the post-processing of E2 uses the weighted expectation method;

[0015] The calculation formula of the environmental road condition factor influence factor E3 is as follows:

[0016]

[0017] Wherein, R1 is the road adhesion factor, R2 is the road slope factor, R3 is the environmental weather factor, R4 is the road curvature factor, and R5 is the crosswind influence factor; q1 is the road adhesion factor weight coefficient, q2 is the road slope factor weight coefficient, q3 is the environmental weather factor weight coefficient, q4 is the road curvature factor weight coefficient, and q5 is the crosswind influence factor weight coefficient, and their value ranges are all between 0 and 1;

[0018] The centroid side slip angle estimation execution module includes three estimation modes. The first estimation mode is the least squares estimation mode based on kinematics, the second estimation mode is the unscented Kalman filter estimation mode based on dynamics, and the third estimation mode is the cubature Kalman filter estimation mode based on dynamics; when the centroid side slip angle estimation execution module executes the first estimation mode, the vehicle is in a stable driving condition, and the least squares estimation mode based on kinematics can provide high-precision estimation and reduce the calculation amount. Design the adaptive weight factor at time t Perform dynamic adjustment; when the centroid sideslip angle estimation execution module executes the second estimation mode, the dynamic model can more accurately describe the vehicle state. Use the unscented Kalman filter estimation mode based on dynamics for estimation, recursively estimate the state variables, make full use of the observation data for non-linear state estimation, and improve the tracking accuracy of the vehicle centroid sideslip angle; when the centroid sideslip angle estimation execution module executes the third estimation mode, use the cubature Kalman filter estimation mode based on dynamics for estimation, perform multi-point sampling on the state variables through cubature transformation, and combine with the non-linear observation model to improve the estimation accuracy;

[0019] The specific calculation formula for defining the error minimization objective function J(w1, w2, w3) of the genetic algorithm in the centroid sideslip angle fusion estimation module based on genetic algorithm and neural network is as follows:

[0020]

[0021] In the formula, w1 is the weight factor of the least squares estimation mode based on kinematics, w2 is the weight factor of the unscented Kalman filter estimation mode based on dynamics, w3 is the weight factor of the cubature Kalman filter estimation mode based on dynamics. The value ranges of the three weight factors are all between 0 and 1; j is the sample index, representing the j-th data point; N is the total number of samples, representing the number of data points used for optimization, and the value range is from 10000 to 50000; β 1,t is the centroid sideslip angle estimated at time t by the least squares estimation mode based on kinematics, β 2,t is the centroid sideslip angle estimated at time t by the unscented Kalman filter estimation mode based on dynamics, β 3,t is the centroid sideslip angle estimated at time t by the cubature Kalman filter estimation mode based on dynamics; β meas,t is the value of the centroid sideslip angle measured by the sensor at time t; f1(E1, E2, E3) is the dynamic adjustment factor of the least squares estimation mode based on kinematics, f2(E1, E2, E3) is the dynamic adjustment factor of the unscented Kalman filter estimation mode based on dynamics, f3(E1, E2, E3) is the dynamic adjustment factor of the cubature Kalman filter estimation mode based on dynamics. The value ranges of the three dynamic adjustment factors are all between 0 and 1; the calculation formulas of the three dynamic adjustment factors are as follows:

[0022]

[0023] In the formula, E1 is the influence factor of vehicle dynamic characteristics, E2 is the influence factor of driver factors, and E3 is the influence factor of environmental road conditions;

[0024] The constraint function is as follows:

[0025]

[0026] In the formula, w1 is the weight factor of the least squares estimation mode based on kinematics, w2 is the weight factor of the unscented Kalman filter estimation mode based on dynamics, and w3 is the weight factor of the cubature Kalman filter estimation mode based on dynamics. The value ranges of the three weight factors are all between 0 and 1; E1 is the influencing factor of vehicle dynamic characteristics, E2 is the influencing factor of driver factors, and E3 is the influencing factor of environmental road conditions; f1(E1, E2, E3) is the dynamic adjustment factor of the least squares estimation mode based on kinematics, f2(E1, E2, E3) is the dynamic adjustment factor of the unscented Kalman filter estimation mode based on dynamics, and f3(E1, E2, E3) is the dynamic adjustment factor of the cubature Kalman filter estimation mode based on dynamics. The value ranges of the three dynamic adjustment factors are all between 0 and 1.

[0027] The centroid side slip angle estimation influencing factor calculation module calculates the longitudinal dynamic characteristic influencing factor E according to the following formula 1x :

[0028]

[0029] In the formula, v x is the longitudinal speed; a x is the longitudinal acceleration; λ is the wheel slip ratio; F d is the driving force; F b is the braking force; F x1 and F x2 are the longitudinal tire forces;

[0030] The lateral dynamic characteristic influencing factor E 1y is calculated according to the following formula:

[0031]

[0032] In the formula, G1 is the vehicle lateral inertia factor, which reflects the influence of the vehicle's lateral speed, lateral acceleration, and yaw rate on lateral dynamics; G2 is the tire lateral adhesion factor, which reflects the influence of the tire side slip angle and tire lateral force on lateral stability. Considering the nonlinear characteristics of the tire, the magic formula tire model is used in combination; G3 is the driver's lateral input factor, which reflects the influence of steering input on lateral dynamics through the vehicle steering angle;

[0033] The vehicle lateral inertia factor G1 is calculated according to the following formula:

[0034]

[0035] In the formula, v y is the lateral speed; a y is the lateral acceleration; r is the yaw rate;

[0036] The tire lateral adhesion factor G2 is calculated according to the following formula:

[0037]

[0038] where F y is the tire lateral force; α is the tire sideslip angle; k1 is an empirical parameter for adjusting the growth rate of the non-linear region of the tire force, with a value range between 0.1 and 2; k2 is an empirical parameter for controlling the influence of the sideslip angle on lateral adhesion, with a value range between 0.5 and 5;

[0039] The driver's lateral input factor G3 is calculated according to the following formula:

[0040]

[0041] where δ f is the front wheel steering angle;

[0042] The vertical dynamic characteristic influence factor E 1z is calculated according to the following formula:

[0043]

[0044] where φ is the body roll angle; θ is the body pitch angle; s f is the front wheel suspension stroke, s r is the rear wheel suspension stroke; F z,f is the front wheel normal force, F z,r is the rear wheel normal force.

[0045] The centroid sideslip angle estimation influence factor calculation module calculates the driver experience category probability factor P(C i ) according to the following formula:

[0046]

[0047] where Y is the driving age, normalized by Min-Max, with a value range between 0 and 1; S is the total driving mileage, normalized by Min-Max, with a value range between 0 and 1; D is the driving style score, with a value range between 0 and 1; A is the accident rate, and the influence of extreme values is reduced through logarithmic transformation, with a value range between 0 and 1; is the output of the first hidden layer, is the output of the second hidden layer, is the output of the third hidden layer; ReLU is the activation function of the first hidden layer, extracting basic driving features; Swish is the activation function Swish of the second hidden layer, enhancing the non-linear mapping ability; ReLU is the activation function of the third hidden layer, optimizing feature expression and improving classification ability; is the weight matrix from the input layer to the first hidden layer, is the weight matrix from the first hidden layer to the second hidden layer, is the weight matrix from the second hidden layer to the third hidden layer, is the weight matrix from the third hidden layer to the output layer, all of which are determined through training; is the bias term of the first hidden layer, is the bias term of the second hidden layer, is the bias term of the third hidden layer, is the bias term of the output layer, all of which are determined through training; U E2,i is the original score of the unnormalized output layer; represents applying an exponential transformation to map U E2,i ; the driver experience category probability factor P(C i ) represents the probability that the driver belongs to the i-th experience category. Softmax normalization makes P(C1)+P(C2)+P(C3)+P(C4)=1; P(C1) is the probability factor for novice drivers, P(C2) is the probability factor for average drivers, P(C3) is the probability factor for skilled drivers, P(C4) is the probability factor for professional drivers, and their value ranges are all between 0 and 1; C i is the driver experience category; where C1 is a novice driver with a short driving age, low driving mileage, an aggressive driving style, and a relatively high accident rate, and C1 takes the value of 1 during calculation; C2 is an average driver with medium driving age and mileage, a relatively mild driving style, and a relatively low accident rate, and C2 takes the value of 2 during calculation; C3 is a skilled driver with a high driving age and mileage, a stable driving style, and a very low accident rate, and C3 takes the value of 3 during calculation; C4 is a professional driver with a long driving age, far driving mileage, a stable driving style, and almost no accidents, and C4 takes the value of 4 during calculation;

[0048] The input layer contains 4 nodes, corresponding to the driving age Y, total driving mileage S, driving style D, and accident rate A respectively; the hidden layer adopts a three-layer structure, including the first hidden layer, the second hidden layer, and the third hidden layer; the first hidden layer contains 64 neurons, the second hidden layer contains 32 neurons, and the third hidden layer contains 16 neurons; the output layer contains 4 nodes, which are the probability factor P(C1) for novice drivers, the probability factor P(C2) for average drivers, the probability factor P(C3) for skilled drivers, and the probability factor P(C4) for professional drivers.

[0049] The centroid sideslip angle estimation influence factor calculation module calculates the road adhesion factor R1 according to the following formula:

[0050]

[0051] Wherein, μ is the current road surface friction coefficient; μ0 is the standard dry road surface friction coefficient, and its value range is between 0.8 and 1.0;

[0052] The road slope factor R2 is calculated according to the following formula:

[0053]

[0054] Wherein, θ x is the longitudinal slope angle, and θ y is the lateral slope angle;

[0055] The environmental weather factor R3 is calculated according to the following formula:

[0056]

[0057] Wherein, ξ is the solar altitude angle; Tem is the air temperature; Precip is the precipitation intensity; γ1 is the air temperature adjustment coefficient, and its value range is between 0.01 and 0.1; γ2 is the precipitation intensity influence coefficient, and its value range is between 0.01 and 0.5;

[0058] The road curvature factor R4 is calculated according to the following formula:

[0059]

[0060] Wherein, is the road curvature, which depends on the road type;

[0061] The crosswind influence factor R5 is calculated according to the following formula:

[0062]

[0063] Wherein, v w is the crosswind speed; v x is the longitudinal speed.

[0064] The centroid side slip angle estimation execution module includes executing the first estimation mode, the second estimation mode, and the third estimation mode;

[0065] S1. When the centroid side slip angle estimation execution module executes the first estimation mode, that is, the least squares estimation mode based on kinematics, design the adaptive weight factor at time t The estimated value β of the centroid side slip angle at time t 1,t The final calculation formula is as follows:

[0066]

[0067] Wherein, is the centroid side slip angle estimated by the least squares method at time t-1; a y,t is the lateral acceleration of the vehicle at time t; is the set critical lateral acceleration, with a value range of 0.8 to 1.5 m / s 2 ; ε is an adjustment factor for controlling the Sigmoid transition rate, with a value range of 3 to 8; is the error covariance matrix at time t-1 of the least squares method; is the observation matrix at time t of the least squares method, is the transpose of the observation matrix at time t of the least squares method; is the measurement noise covariance matrix at time t of the least squares method; is the measurement data at time t of the least squares method;

[0068] S2. When the centroid side slip angle estimation execution module executes the second estimation mode, that is, the unscented Kalman filter estimation mode based on dynamics, the estimated value of the centroid side slip angle β at time t 2,t has the following final calculation formula:

[0069]

[0070] In the formula, is the lateral velocity estimated by the unscented Kalman filter at time t, is the longitudinal velocity estimated by the unscented Kalman filter at time t; is the longitudinal velocity at time t calculated by prediction from time t-1 in the unscented Kalman filter estimation mode, is the lateral velocity at time t calculated by prediction from time t-1 in the unscented Kalman filter estimation mode; is the measurement data of the unscented Kalman filter at time t; is the measurement data at time t calculated by prediction from time t-1 in the unscented Kalman filter estimation mode; is the Kalman gain in the longitudinal velocity direction of the unscented Kalman filter at time t, is the Kalman gain in the lateral velocity direction of the unscented Kalman filter at time t;

[0071] S3. When the centroid side slip angle estimation execution module executes the third estimation mode, that is, the cubature Kalman filter estimation mode based on dynamics, the estimated value of the centroid side slip angle β at time t 3,t has the following final calculation formula:

[0072]

[0073] In the formula, is the lateral velocity estimated by the cubature Kalman filter at time t, is the longitudinal velocity estimated by the cubature Kalman filter at time t; is the longitudinal velocity at time t calculated from the prediction at time t-1 in the cubature Kalman filter estimation mode, is the lateral velocity at time t calculated from the prediction at time t-1 in the cubature Kalman filter estimation mode; is the measurement data at time t of the cubature Kalman filter; is the measurement data at time t calculated from the prediction at time t-1 in the cubature Kalman filter estimation mode; is the Kalman gain in the longitudinal velocity direction at time t of the cubature Kalman filter, is the Kalman gain in the lateral velocity direction at time t of the cubature Kalman filter.

[0074] The centroid sideslip angle fusion estimation module based on genetic algorithm and neural network first initializes the population, sets the population size as M = 100, randomly generates the weight factors and conforms to the Dirichlet(1,1,1) distribution, so that the sum of weights is 1 and possible solutions can be explored evenly; calculates the objective function J(w1,w2,w3) to minimize the centroid sideslip angle estimation error; the fitness function Υ=(1 + J(w1,w2,w3)) -1 The closer to 1 within the value range from 0 to 1 indicates that the individual is better; the selection strategy adopts tournament selection, selects the optimal individuals from the population for replication and inheritance; adopts two-point crossover, so that the information of different individuals can be effectively combined, and the crossover probability P c is set to 0.8; the mutation adopts non-uniform mutation so that the population can finely adjust the parameters in the later stage, and the mutation rate P m is set to 0.05 to keep the optimization search diverse but not overly disturb the optimization process; the termination condition is that the fitness function J(w1,w2,w3) converges to a range less than the threshold 0.000001 or reaches the maximum iteration number 200;

[0075] Considering that the actual working conditions are constantly changing and the vehicle needs millisecond-level calculation response, the online operation of the genetic algorithm has too high calculation cost and calculation time lag, and is not suitable for vehicle real-time control; in the training stage, the vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3 are used as the data set to train the neural network; in the online inference stage, the neural network can directly predict the weight factor w1 of the kinematics-based least squares estimation mode optimized by the genetic algorithm, the weight factor w2 of the unscented Kalman filter estimation mode based on dynamics, and the weight factor w3 of the cubature Kalman filter estimation mode based on dynamics according to the input vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3; the genetic algorithm learns offline to improve the calculation efficiency; the neural network architecture is as follows:

[0076]

[0077] In the formula, is the output of the first hidden layer, is the output of the second hidden layer; Swish is the activation function of the first hidden layer, and ReLU is the activation function of the second hidden layer; is the weight matrix from the input layer to the first hidden layer, is the weight matrix from the first hidden layer to the second hidden layer, is the weight matrix from the second hidden layer to the output layer, all of which are determined through training; is the bias term of the first hidden layer, is the bias term of the second hidden layer, is the bias term of the output layer, all of which are determined through training; U w is the output vector [w1, w2, w3] T ;

[0078] The input layer contains 3 nodes, and the inputs are the vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3; the hidden layer adopts a double-layer structure, including the first hidden layer and the second hidden layer; the first hidden layer contains 32 neurons, and the second hidden layer contains 16 neurons; the output layer contains 3 nodes, and the outputs are the weight factor w1 of the least squares estimation mode based on kinematics, the weight factor w2 of the unscented Kalman filter estimation mode based on dynamics, and the weight factor w3 of the cubature Kalman filter estimation mode based on dynamics.

[0079] The beneficial effects of the present invention are as follows:

[0080] By comprehensively considering various factors such as vehicle dynamics characteristics, driver behavior, and environmental road conditions, the present invention quantifies various influence factors, enabling the estimation model to adapt to different driving conditions; combining three estimation modes, namely the least squares estimation mode based on kinematics, the unscented Kalman filter estimation mode based on dynamics, and the cubature Kalman filter estimation mode based on dynamics, and executing the three estimation modes in parallel under different conditions to enhance the estimation accuracy and robustness; optimizing the weight factors in the estimation model offline through a genetic algorithm, training the neural network based on the training data set of the influence factors, and in the online inference stage, the neural network predicts the weight factors of each estimation mode in real time according to the input influence factors, further improving the estimation accuracy and calculation efficiency; the present invention can provide accurate yaw angle estimation during real-time driving and enhance the driving stability of intelligent vehicles. Description of the Drawings

[0081] The present invention will be further described below with reference to the accompanying drawings:

[0082] Figure 1It is a framework diagram of a method for fusing and estimating the centroid side slip angle based on an intelligent vehicle proposed by the present invention. Specific implementation mode

[0083] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0084] Refer to Figure 1 , a method for fusing and estimating the centroid side slip angle based on an intelligent vehicle described in the present invention includes a vehicle operation information acquisition module, a centroid side slip angle estimation influence factor calculation module, a centroid side slip angle estimation execution module, and a centroid side slip angle fusion estimation module based on genetic algorithm and neural network;

[0085] The vehicle operation information acquisition module is used to obtain the longitudinal speed v x , longitudinal acceleration a x , wheel slip ratio λ, longitudinal tire force F x1 and F x2 , lateral speed v y , lateral acceleration a y , yaw rate r, tire lateral force F y , tire side slip angle α, front wheel steering angle δ f , body roll angle φ, body pitch angle θ, front wheel suspension stroke s f , rear wheel suspension stroke s r , front wheel normal force F z,f , rear wheel normal force F z,r , road surface friction coefficient μ, road curvature κ, crosswind speed v w ;

[0086] The centroid side slip angle estimation influence factor calculation module calculates the vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3; the vehicle dynamic characteristic factor influence factor E1 evaluates the influence of the vehicle's own dynamic characteristics on the centroid side slip angle estimation, so that the estimation method adapts to different driving states; the driver factor influence factor E2 measures the influence of the driver's driving behavior and personal characteristics on the centroid side slip angle estimation, and identifies the uncertainty brought by human operation; the environmental road condition factor influence factor E3 quantifies the influence of the external environment and road conditions on the centroid side slip angle estimation, so that the estimation method is applicable to different road conditions and meteorological conditions; the vehicle dynamic characteristic factor influence factor E1 depends on the longitudinal dynamic characteristic influence factor E 1x , lateral dynamic characteristic influence factor E 1y , vertical dynamic characteristic influence factor E 1z , where the longitudinal dynamic characteristic influence factor E1x depending on the longitudinal speed v x and the longitudinal acceleration a x the wheel slip ratio λ, the driving force F d the braking force F b the longitudinal tire force F x1 and F x2 the lateral dynamic characteristic influence factor E 1y depends on the vehicle lateral inertia factor G1, the tire lateral adhesion factor G2, and the driver lateral input factor G3, where the vehicle lateral inertia factor G1 depends on the lateral speed v y and the lateral acceleration a y and the yaw rate r, the tire lateral adhesion factor G2 depends on the tire lateral force F y the tire sideslip angle α, the empirical parameter k1, the empirical parameter k2, and the driver lateral input factor G3 depends on the front wheel steering angle δ f the vertical dynamic characteristic influence factor E 1z depends on the vehicle body roll angle φ, the vehicle body pitch angle θ, the front wheel suspension stroke s f and the rear wheel suspension stroke s r the front wheel normal force F z,f and the rear wheel normal force F z,r ; the driver factor influence factor E2 depends on the driver experience category C i and the driver experience category probability factor P(C i ), where the driver experience category probability factor P(C i ) depends on the driving age Y, the total driving mileage S, the driving style D, and the accident rate A; the environmental road condition factor influence factor E3 depends on the road adhesion factor R1, the road slope factor R2, the environmental weather factor R3, the road curvature factor R4, and the crosswind influence factor R5, where the road adhesion factor R1 depends on the current road surface friction coefficient μ and the standard dry road surface friction coefficient μ0, and the road slope factor R2 depends on the longitudinal slope angle θ x and the lateral slope angle θ y , the environmental weather factor R3 depends on the solar altitude angle ξ, the air temperature Tem, the precipitation intensity Precip, the air temperature adjustment coefficient γ1, and the precipitation intensity influence coefficient γ2, the road curvature factor R4 depends on the road curvature the crosswind influence factor R5 depends on the crosswind speed v w and the longitudinal speed v x ;

[0087] The calculation formula for the vehicle dynamic characteristic factor influence factor E1 is as follows:

[0088]

[0089] In the formula, E 1xis the longitudinal dynamic characteristic influence factor, E 1y is the lateral dynamic characteristic influence factor, E 1z is the vertical dynamic characteristic influence factor; m1 is the longitudinal dynamic characteristic weight coefficient, m2 is the lateral dynamic characteristic weight coefficient, m3 is the vertical dynamic characteristic weight coefficient, and their value ranges are all between 0 and 1; u is the adjustment parameter for controlling the sensitivity and smoothness output by the calculation formula of the vehicle dynamic characteristic factor E1, and its value range is between 0.5 and 2;

[0090] The calculation formula of the driver factor influence factor E2 is as follows:

[0091]

[0092] In the formula, C i is the driver experience category; among them, C1 is a novice driver with short driving experience, low driving mileage, aggressive driving style, and relatively high accident rate, and C1 takes the value of 1 during calculation; C2 is an average driver with medium driving experience and mileage, mild driving style, and relatively low accident rate, and C2 takes the value of 2 during calculation; C3 is a skilled driver with high driving experience and mileage, stable driving style, and very low accident rate, and C3 takes the value of 3 during calculation; C4 is a professional driver with long driving experience, far driving mileage, stable driving style, and almost no accidents, and C4 takes the value of 4 during calculation; P(C i ) is the driver experience category probability factor. To improve the accuracy of the driver factor influence factor E2, the calculation of P(C i ) uses a deep feedforward neural network structure and is obtained by a neural network classifier; among them, P(C1) is the novice driver probability factor, P(C2) is the average driver probability factor, P(C3) is the skilled driver probability factor, P(C4) is the professional driver probability factor, and their value ranges are all between 0 and 1; the post-processing of E2 uses the weighted expectation method;

[0093] The calculation formula of the environmental road condition factor influence factor E3 is as follows:

[0094]

[0095] In the formula, R1 is the road adhesion factor, R2 is the road gradient factor, R3 is the environmental weather factor, R4 is the road curvature factor, R5 is the crosswind influence factor; q1 is the road adhesion factor weight coefficient, q2 is the road gradient factor weight coefficient, q3 is the environmental weather factor weight coefficient, q4 is the road curvature factor weight coefficient, q5 is the crosswind influence factor weight coefficient, and their value ranges are all between 0 and 1;

[0096] The centroid sideslip angle estimation execution module includes three estimation modes. The first estimation mode is the least squares estimation mode based on kinematics, the second estimation mode is the unscented Kalman filter estimation mode based on dynamics, and the third estimation mode is the cubature Kalman filter estimation mode based on dynamics. When the centroid sideslip angle estimation execution module executes the first estimation mode, the vehicle is in a stable driving condition, and the least squares estimation mode based on kinematics can provide high-precision estimation and reduce the computational load. An adaptive weight factor at time t is designed for dynamic adjustment. When the centroid sideslip angle estimation execution module executes the second estimation mode, the dynamic model can more accurately describe the vehicle state. The unscented Kalman filter estimation mode based on dynamics is used for estimation, and the state variables are recursively estimated. The observed data is fully utilized for non-linear state estimation to improve the tracking accuracy of the vehicle centroid sideslip angle. When the centroid sideslip angle estimation execution module executes the third estimation mode, the cubature Kalman filter estimation mode based on dynamics is used for estimation. The state variables are multi-point sampled through cubature transformation, and the estimation accuracy is improved by combining with the non-linear observation model;

[0097] The specific calculation formula for defining the error minimization objective function J(w1, w2, w3) of the genetic algorithm and neural network-based centroid sideslip angle fusion estimation module is as follows:

[0098]

[0099] In the formula, w1 is the weight factor of the least squares estimation mode based on kinematics, w2 is the weight factor of the unscented Kalman filter estimation mode based on dynamics, and w3 is the weight factor of the cubature Kalman filter estimation mode based on dynamics. The value ranges of the three weight factors are all between 0 and 1; j is the sample index, representing the jth data point; N is the total number of samples, representing the number of data points used for optimization, and the value range is from 10000 to 50000; β 1,t is the centroid sideslip angle estimated at time t by the least squares estimation mode based on kinematics, β 2,t is the centroid sideslip angle estimated at time t by the unscented Kalman filter estimation mode based on dynamics, β 3,t is the centroid sideslip angle estimated at time t by the cubature Kalman filter estimation mode based on dynamics; β meas,tis the value of the centroid side slip angle actually measured by the sensor at time t; f1(E1, E2, E3) is the dynamic adjustment factor of the least squares estimation mode based on kinematics, f2(E1, E2, E3) is the dynamic adjustment factor of the unscented Kalman filter estimation mode based on dynamics, f3(E1, E2, E3) is the dynamic adjustment factor of the cubature Kalman filter estimation mode based on dynamics, and the value ranges of the three dynamic adjustment factors are all between 0 and 1; the calculation formulas of the three dynamic adjustment factors are as follows:

[0100]

[0101] In the formula, E1 is the influence factor of vehicle dynamic characteristics, E2 is the influence factor of driver factors, and E3 is the influence factor of environmental road conditions;

[0102] The constraint function is as follows:

[0103]

[0104] In the formula, w1 is the weight factor of the least squares estimation mode based on kinematics, w2 is the weight factor of the unscented Kalman filter estimation mode based on dynamics, w3 is the weight factor of the cubature Kalman filter estimation mode based on dynamics, and the value ranges of the three weight factors are all between 0 and 1; E1 is the influence factor of vehicle dynamic characteristics, E2 is the influence factor of driver factors, and E3 is the influence factor of environmental road conditions; f1(E1, E2, E3) is the dynamic adjustment factor of the least squares estimation mode based on kinematics, f2(E1, E2, E3) is the dynamic adjustment factor of the unscented Kalman filter estimation mode based on dynamics, f3(E1, E2, E3) is the dynamic adjustment factor of the cubature Kalman filter estimation mode based on dynamics, and the value ranges of the three dynamic adjustment factors are all between 0 and 1.

[0105] The centroid side slip angle estimation influence factor calculation module calculates the longitudinal dynamic characteristic influence factor E according to the following formula 1x :

[0106]

[0107] In the formula, v x is the longitudinal speed; a x is the longitudinal acceleration; λ is the wheel slip ratio; F d is the driving force; F b is the braking force; F x1 and F x2 are the longitudinal tire forces;

[0108] The lateral dynamic characteristic influence factor E 1y is calculated according to the following formula:

[0109]

[0110] In the formula, G1 is the vehicle lateral inertia factor, which reflects the influence of the vehicle's lateral speed, lateral acceleration, and yaw rate on lateral dynamics; G2 is the tire lateral adhesion factor, which reflects the influence of the tire slip angle and tire lateral force on lateral stability. Considering the nonlinear characteristics of the tire, the Magic Formula tire model is used in combination; G3 is the driver's lateral input factor, which reflects the influence of steering input on lateral dynamics through the vehicle steering angle;

[0111] The vehicle lateral inertia factor G1 is calculated according to the following formula:

[0112]

[0113] In the formula, v y is the lateral speed; a y is the lateral acceleration; r is the yaw rate;

[0114] The tire lateral adhesion factor G2 is calculated according to the following formula:

[0115]

[0116] In the formula, F y is the tire lateral force; α is the tire slip angle; k1 is an empirical parameter for adjusting the growth rate of the nonlinear region of the tire force, and its value range is between 0.1 and 2; k2 is an empirical parameter for controlling the influence of the slip angle on lateral adhesion, and its value range is between 0.5 and 5;

[0117] The driver's lateral input factor G3 is calculated according to the following formula:

[0118]

[0119] In the formula, δ f is the front wheel steering angle;

[0120] The vertical dynamic characteristic influence factor E 1z is calculated according to the following formula:

[0121]

[0122] In the formula, φ is the body roll angle; θ is the body pitch angle; s f is the front wheel suspension stroke, s r is the rear wheel suspension stroke; F z,f is the front wheel normal force, F z,r is the rear wheel normal force.

[0123] The centroidal slip angle estimation influence factor calculation module calculates the driver experience category probability factor P(C i):

[0124]

[0125] Wherein, Y is the driving age, normalized by Min - Max, with a value range between 0 and 1; S is the total driving mileage, normalized by Min - Max, with a value range between 0 and 1; D is the driving style score, with a value range between 0 and 1; A is the accident rate, and the influence of extreme values is reduced through logarithmic transformation, with a value range between 0 and 1; is the output of the first hidden layer, is the output of the second hidden layer, is the output of the third hidden layer; ReLU is the activation function of the first hidden layer, extracting basic driving features; Swish is the activation function Swish of the second hidden layer, enhancing the non - linear mapping ability; ReLU is the activation function of the third hidden layer, optimizing feature expression and improving classification ability; is the weight matrix from the input layer to the first hidden layer, is the weight matrix from the first hidden layer to the second hidden layer, is the weight matrix from the second hidden layer to the third hidden layer, is the weight matrix from the third hidden layer to the output layer, all determined through training; is the bias term of the first hidden layer, is the bias term of the second hidden layer, is the bias term of the third hidden layer, is the bias term of the output layer, all determined through training; U E2,i is the original score of the un - normalized output layer; represents using exponential transformation to map U E2,i ; The driver experience category probability factor P(C i ) represents the probability that the driver belongs to the i - th experience category. Softmax normalization makes P(C1)+P(C2)+P(C3)+P(C4) = 1; P(C1) is the probability factor for novice drivers, P(C2) is the probability factor for average drivers, P(C3) is the probability factor for proficient drivers, P(C4) is the probability factor for professional drivers, and their value ranges are all between 0 and 1; C i is the driver experience category; among them, C1 is a novice driver with a short driving age, low driving mileage, aggressive driving style, and a relatively high accident rate, and C1 takes the value of 1 during calculation; C2 is an average driver with medium driving age and mileage, a relatively mild driving style, and a relatively low accident rate, and C2 takes the value of 2 during calculation; C3 is a proficient driver with a high driving age and mileage, a stable driving style, and a very low accident rate, and C3 takes the value of 3 during calculation; C4 is a professional driver with a long driving age, long driving mileage, a stable driving style, and almost no accidents, and C4 takes the value of 4 during calculation;

[0126] The input layer contains 4 nodes, corresponding to driving age Y, total driving mileage S, driving style D, and accident rate A respectively; the hidden layer adopts a three-layer structure, including the first hidden layer, the second hidden layer, and the third hidden layer; the first hidden layer contains 64 neurons, the second hidden layer contains 32 neurons, and the third hidden layer contains 16 neurons; the output layer contains 4 nodes, namely the probability factor of novice drivers P(C1), the probability factor of general drivers P(C2), the probability factor of proficient drivers P(C3), and the probability factor of professional drivers P(C4).

[0127] The centroid sideslip angle estimation influence factor calculation module calculates the road adhesion factor R1 according to the following formula:

[0128]

[0129] In the formula, μ is the current road surface friction coefficient; μ0 is the standard dry road surface friction coefficient, and its value range is between 0.8 and 1.0;

[0130] The road slope factor R2 is calculated according to the following formula:

[0131]

[0132] In the formula, θ x is the longitudinal slope angle, and θ y is the lateral slope angle;

[0133] The environmental weather factor R3 is calculated according to the following formula:

[0134]

[0135] In the formula, ξ is the solar altitude angle; Tem is the air temperature; Precip is the precipitation intensity; γ1 is the air temperature adjustment coefficient, and its value range is between 0.01 and 0.1; γ2 is the precipitation intensity influence coefficient, and its value range is between 0.01 and 0.5;

[0136] The road curvature factor R4 is calculated according to the following formula:

[0137]

[0138] In the formula, is the road curvature, which depends on the road type;

[0139] The crosswind influence factor R5 is calculated according to the following formula:

[0140]

[0141] In the formula, v w is the crosswind speed; v x is the longitudinal speed.

[0142] The centroid sideslip angle estimation execution module includes executing the first estimation mode, the second estimation mode, and the third estimation mode;

[0143] S1. When the centroid sideslip angle estimation execution module executes the first estimation mode, that is, the least squares estimation mode based on kinematics, an adaptive weight factor at time t is designed The estimated value of the centroid sideslip angle β at time t 1,t The final calculation formula is as follows:

[0144]

[0145] In the formula, is the centroid sideslip angle estimated by the least squares method at time t - 1; a y,t is the lateral acceleration of the vehicle at time t; is the set critical lateral acceleration, and the value range is 0.8 to 1.5 m / s 2 ; ε is the adjustment factor for controlling the Sigmoid transition rate, and the value range is 3 to 8; is the error covariance matrix of the least squares method at time t - 1; is the observation matrix of the least squares method at time t, is the transpose of the observation matrix of the least squares method at time t; is the measurement noise covariance matrix of the least squares method at time t; is the measurement data of the least squares method at time t;

[0146] S2. When the centroid sideslip angle estimation execution module executes the second estimation mode, that is, the unscented Kalman filter estimation mode based on dynamics, the estimated value of the centroid sideslip angle β at time t 2,t The final calculation formula is as follows:

[0147]

[0148] In the formula, is the lateral velocity estimated by the unscented Kalman filter at time t, is the longitudinal velocity estimated by the unscented Kalman filter at time t; is the longitudinal velocity at time t calculated from the prediction at time t - 1 in the unscented Kalman filter estimation mode, is the lateral velocity at time t calculated from the prediction at time t - 1 in the unscented Kalman filter estimation mode; is the measurement data of the unscented Kalman filter at time t; is the measurement data at time t calculated from the prediction at time t - 1 in the unscented Kalman filter estimation mode; is the Kalman gain in the longitudinal velocity direction at time t of the unscented Kalman filter, is the Kalman gain in the lateral velocity direction at time t of the unscented Kalman filter;

[0149] S3. When the centroid side slip angle estimation execution module executes the third estimation mode, that is, the volume Kalman filter estimation mode based on dynamics, the estimated value β of the centroid side slip angle at time t 3,t has the following final calculation formula:

[0150]

[0151] In the formula, is the estimated lateral velocity at time t of the volume Kalman filter, is the estimated longitudinal velocity at time t of the volume Kalman filter; is the longitudinal velocity at time t calculated from the prediction at time t-1 in the volume Kalman filter estimation mode, is the lateral velocity at time t calculated from the prediction at time t-1 in the volume Kalman filter estimation mode; is the measurement data at time t of the volume Kalman filter; is the measurement data at time t calculated from the prediction at time t-1 in the volume Kalman filter estimation mode; is the Kalman gain in the longitudinal velocity direction at time t of the volume Kalman filter, is the Kalman gain in the lateral velocity direction at time t of the volume Kalman filter.

[0152] The centroid side slip angle fusion estimation module based on genetic algorithm and neural network first initializes the population, sets the population size to M = 100, randomly generates the weight factors and conforms to the Dirichlet(1,1,1) distribution, so that the sum of the weights is 1 and the possible solutions can be explored evenly; calculates the objective function J(w1,w2,w3) to minimize the centroid side slip angle estimation error; the fitness function Υ=(1 + J(w1,w2,w3)) -1 The closer to 1 in the value range from 0 to 1 indicates that the individual is better; the selection strategy adopts tournament selection, and the optimal individuals are selected from the population for replication and inheritance; two-point crossover is adopted so that the information of different individuals can be effectively combined, and the crossover probability P c is set to 0.8; mutation adopts non-uniform mutation so that the population can finely adjust the parameters in the later stage, and the mutation rate P m is set to 0.05 so that the optimization search maintains diversity but does not overly disturb the optimization process; the termination condition is that the fitness function J(w1,w2,w3) converges to a range less than the threshold 0.000001 or reaches the maximum number of iterations 200;

[0153] Considering that the actual working conditions are constantly changing and the vehicle requires millisecond-level computing response, the online operation of the genetic algorithm has too high computing costs and a time lag in computing, which is not suitable for vehicle real-time control; in the training stage, the vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3 are used as the data set to train the neural network; in the online inference stage, the neural network can directly predict the weight factor w1 of the least squares estimation mode based on kinematics optimized by the genetic algorithm, the weight factor w2 of the unscented Kalman filter estimation mode based on dynamics, and the weight factor w3 of the cubature Kalman filter estimation mode based on dynamics according to the input vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3; the genetic algorithm performs offline learning to improve the computing efficiency; the neural network architecture is as follows:

[0154]

[0155] In the formula, is the output of the first hidden layer, is the output of the second hidden layer; Swish is the activation function of the first hidden layer, and ReLU is the activation function of the second hidden layer; is the weight matrix from the input layer to the first hidden layer, is the weight matrix from the first hidden layer to the second hidden layer, is the weight matrix from the second hidden layer to the output layer, all of which are determined through training; is the bias term of the first hidden layer, is the bias term of the second hidden layer, is the bias term of the output layer, all of which are determined through training; U w is the output vector [w1, w2, w3] T ;

[0156] The input layer contains 3 nodes, and the inputs are the vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3; the hidden layer adopts a double-layer structure, including the first hidden layer and the second hidden layer; the first hidden layer contains 32 neurons, and the second hidden layer contains 16 neurons; the output layer contains 3 nodes, and the outputs are the weight factor w1 of the least squares estimation mode based on kinematics, the weight factor w2 of the unscented Kalman filter estimation mode based on dynamics, and the weight factor w3 of the cubature Kalman filter estimation mode based on dynamics.

Claims

1. A method for fusing and estimating the centroid side slip angle based on an intelligent vehicle, characterized in that It includes the following contents: It includes a vehicle operation information acquisition module, a centroid side slip angle estimation influence factor calculation module, a centroid side slip angle estimation execution module, and a centroid side slip angle fusion estimation module based on genetic algorithm and neural network; The vehicle operation information acquisition module is used to obtain the longitudinal speed v x , longitudinal acceleration a x , wheel slip ratio λ, longitudinal tire force F x1 and F x2 , lateral speed v y , lateral acceleration a y , yaw angular velocity r, tire side force F y , tire sideslip angle α, front wheel steering angle δ f , body roll angle φ, body pitch angle θ, front wheel suspension stroke s f , rear wheel suspension stroke s r , front wheel normal force F z,f , rear wheel normal force F z,r , road surface friction coefficient μ, road curvature κ, crosswind speed v w ; The centroid side slip angle estimation influence factor calculation module calculates the vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3; the vehicle dynamic characteristic factor influence factor E1 evaluates the influence of the vehicle's own dynamic characteristics on the centroid side slip angle estimation, so that the estimation method adapts to different driving states; the driver factor influence factor E2 measures the influence of the driver's driving behavior and personal characteristics on the centroid side slip angle estimation, and identifies the uncertainty brought by human operation; the environmental road condition factor influence factor E3 quantifies the influence of the external environment and road conditions on the centroid side slip angle estimation, so that the estimation method is applicable to different road conditions and meteorological conditions; the vehicle dynamic characteristic factor influence factor E1 depends on the longitudinal dynamic characteristic influence factor E 1x , the lateral dynamic characteristic influence factor E 1y , and the vertical dynamic characteristic influence factor E 1z , where the longitudinal dynamic characteristic influence factor E 1x depends on the longitudinal speed v x , the longitudinal acceleration a x , the wheel slip ratio λ, the driving force F d , the braking force F b , the longitudinal tire force F x1 and F x2 , the lateral dynamic characteristic influence factor E 1y depends on the vehicle lateral inertia factor G1, the tire lateral adhesion factor G2, and the driver lateral input factor G3. Among them, the vehicle lateral inertia factor G1 depends on the lateral speed v y , the lateral acceleration a y , the yaw rate r, the tire lateral adhesion factor G2 depends on the tire lateral force F y , the tire side slip angle α, the empirical parameter k1, and the empirical parameter k2. The driver lateral input factor G3 depends on the front wheel steering angle δ f , the vertical dynamic characteristic influence factor E 1z depends on the body roll angle φ, the body pitch angle θ, the front wheel suspension stroke s f , the rear wheel suspension stroke s r , the front wheel normal force F z,f , the rear wheel normal force F z,r ; the driver factor influence factor E2 depends on the driver experience category C i and the driver experience category probability factor P(C i ), where the driver experience category probability factor P(C i )It depends on the driving experience Y, the total driving mileage S, the driving style D, and the accident rate A; the environmental road condition factor influence factor E3 depends on the road adhesion factor R1, the road slope factor R2, the environmental weather factor R3, the road curvature factor R4, and the crosswind influence factor R5. Among them, the road adhesion factor R1 depends on the current road surface friction coefficient μ and the standard dry road surface friction coefficient μ0, and the road slope factor R2 depends on the longitudinal slope angle θ x and the lateral slope angle θ y . The environmental weather factor R3 depends on the solar altitude angle ξ, the air temperature Tem, the precipitation intensity Precip, the air temperature adjustment coefficient γ1, and the precipitation intensity influence coefficient γ2. The road curvature factor R4 depends on the road curvature . The crosswind influence factor R5 depends on the crosswind speed v w and the longitudinal speed v x ; The calculation formula of the vehicle dynamic characteristic factor influence factor E1 is as follows: Where, E 1x is the longitudinal dynamic characteristic influence factor, E 1y is the lateral dynamic characteristic influence factor, E 1z is the vertical dynamic characteristic influence factor; m1 is the longitudinal dynamic characteristic weight coefficient, m2 is the lateral dynamic characteristic weight coefficient, m3 is the vertical dynamic characteristic weight coefficient, and their value ranges are all between 0 and 1; u is the adjustment parameter for controlling the sensitivity and smoothness output by the calculation formula of the vehicle dynamic characteristic factor E1, and its value range is between 0.5 and 2; The calculation formula of the driver factor influence factor E2 is as follows: Where C i is the driver experience category; among them, C1 is a novice driver with a short driving experience, a low driving mileage, an aggressive driving style, and a relatively high accident rate. When calculating, C1 takes the value of 1; C2 is an average driver with a medium driving experience and mileage, a relatively mild driving style, and a relatively low accident rate. When calculating, C2 takes the value of 2; C3 is a skilled driver with a high driving experience and mileage, a stable driving style, and a very low accident rate. When calculating, C3 takes the value of 3; C4 is a professional driver with a long driving experience, a long driving mileage, a stable driving style, and almost no accidents. When calculating, C4 takes the value of 4; P(C i ) is the driver experience category probability factor. To improve the accuracy of the driver factor influence factor E2, the calculation of P(C i ) uses a deep feedforward neural network structure and is obtained by a neural network classifier; among them, P(C1) is the novice driver probability factor, P(C2) is the average driver probability factor, P(C3) is the skilled driver probability factor, and P(C4) is the professional driver probability factor. Their value ranges are all between 0 and 1; the post-processing of E2 uses the weighted expectation method. The calculation formula of the environmental road condition factor influence factor E3 is as follows: In the formula, R1 is the road adhesion factor, R2 is the road slope factor, R3 is the environmental weather factor, R4 is the road curvature factor, R5 is the crosswind influence factor; q1 is the road adhesion factor weight coefficient, q2 is the road slope factor weight coefficient, q3 is the environmental weather factor weight coefficient, q4 is the road curvature factor weight coefficient, q5 is the crosswind influence factor weight coefficient, and their value ranges are all between 0 and 1; The centroid side slip angle estimation execution module includes three estimation modes. The first estimation mode is the least squares estimation mode based on kinematics, the second estimation mode is the unscented Kalman filter estimation mode based on dynamics, and the third estimation mode is the cubature Kalman filter estimation mode based on dynamics; When the centroid side slip angle estimation execution module executes the first estimation mode, the vehicle is in a stable driving condition, and the least squares estimation mode based on kinematics can provide high-precision estimation and reduce the calculation amount. The adaptive weight factor at time t is designed for dynamic adjustment; When the centroid side slip angle estimation execution module executes the second estimation mode, the dynamic model can more accurately describe the vehicle state. The unscented Kalman filter estimation mode based on dynamics is used for estimation, and the state variables are recursively estimated, making full use of the observation data for non-linear state estimation to improve the tracking accuracy of the vehicle centroid side slip angle; When the centroid side slip angle estimation execution module executes the third estimation mode, the cubature Kalman filter estimation mode based on dynamics is used for estimation. The state variables are multi-point sampled through cubature transformation, and the estimation accuracy is improved by combining with the non-linear observation model; The specific calculation formula for defining the error minimization objective function J(w1, w2, w3) of the genetic algorithm in the centroid side slip angle fusion estimation module based on genetic algorithm and neural network is as follows: where w1 is the weight factor of the least squares estimation mode based on kinematics, w2 is the weight factor of the unscented Kalman filter estimation mode based on dynamics, w3 is the weight factor of the cubature Kalman filter estimation mode based on dynamics, and the value ranges of the three weight factors are all between 0 and 1; j is the sample index, representing the j-th data point; N is the total number of samples, representing the number of data points used for optimization, and the value range is from 10,000 to 50,000; β 1,t is the estimated sideslip angle of the center of mass at time t in the least squares estimation mode based on kinematics, β 2,t is the estimated sideslip angle of the center of mass at time t in the unscented Kalman filter estimation mode based on dynamics, β 3,t is the estimated sideslip angle of the center of mass at time t in the cubature Kalman filter estimation mode based on dynamics; β meas,t is the true measured value of the sideslip angle of the center of mass at time t; f1(E1, E2, E3) is the dynamic adjustment factor of the least squares estimation mode based on kinematics, f2(E1, E2, E3) is the dynamic adjustment factor of the unscented Kalman filter estimation mode based on dynamics, f3(E1, E2, E3) is the dynamic adjustment factor of the cubature Kalman filter estimation mode based on dynamics, and the value ranges of the three dynamic adjustment factors are all between 0 and 1; the calculation formulas of the three dynamic adjustment factors are as follows: In the formula, E1 is the vehicle dynamic characteristic factor influence factor, E2 is the driver factor influence factor, and E3 is the environmental road condition factor influence factor; The constraint function is as follows: In the formula, w1 is the weight factor of the least squares estimation mode based on kinematics, w2 is the weight factor of the unscented Kalman filter estimation mode based on dynamics, w3 is the weight factor of the cubature Kalman filter estimation mode based on dynamics, and the value ranges of the three weight factors are all between 0 and 1; E1 is the vehicle dynamic characteristic factor influence factor, E2 is the driver factor influence factor, E3 is the environmental road condition factor influence factor; f1(E1, E2, E3) is the dynamic adjustment factor of the least squares estimation mode based on kinematics, f2(E1, E2, E3) is the dynamic adjustment factor of the unscented Kalman filter estimation mode based on dynamics, f3(E1, E2, E3) is the dynamic adjustment factor of the cubature Kalman filter estimation mode based on dynamics, and the value ranges of the three dynamic adjustment factors are all between 0 and 1.

2. The centroid side slip angle fusion estimation method based on an intelligent vehicle according to claim 1, characterized in that The centroid sideslip angle estimation influence factor calculation module calculates the longitudinal dynamic characteristic influence factor E according to the following formula 1x : where \(v\) x is the longitudinal speed; \(a\) x is the longitudinal acceleration; \(\lambda\) is the wheel slip ratio; \(F\) d is the driving force; \(F\) b is the braking force; \(F\) x1 and \(F\) x2 are the longitudinal tire forces; Lateral dynamic characteristic influence factor E 1y Calculated according to the following formula: In the formula, G1 is the vehicle lateral inertia factor, which reflects the influence of the vehicle's lateral speed, lateral acceleration, and yaw rate on lateral dynamics; G2 is the tire lateral adhesion factor, which reflects the influence of the tire side slip angle and tire lateral force on lateral stability. Considering the nonlinear characteristics of the tire, the Magic Formula tire model is used in combination; G3 is the driver lateral input factor, which reflects the influence of steering input on lateral dynamics through the vehicle steering angle. The vehicle lateral inertia factor G1 is calculated according to the following formula: where v y is the lateral velocity; a y is the lateral acceleration; r is the yaw rate; The tire lateral adhesion factor G2 is calculated according to the following formula: Where, F y is the lateral force of the tire; α is the slip angle of the tire; k1 is an empirical parameter for adjusting the growth rate of the non-linear region of the tire force, and its value range is between 0.1 and 2; k2 is an empirical parameter for controlling the influence of the slip angle on the lateral adhesion, and its value range is between 0.5 and 5; The driver lateral input factor G3 is calculated according to the following formula: where δ f is the front wheel steering angle; Vertical dynamic characteristic influence factor E 1z Calculated according to the following formula: Where φ is the body roll angle; θ is the body pitch angle; s f is the front wheel suspension stroke, s r is the rear wheel suspension stroke; F z,f is the front wheel normal force, F z,r is the rear wheel normal force.

3. The centroid side slip angle fusion estimation method based on an intelligent vehicle according to claim 1, wherein The centroid side slip angle estimation influence factor calculation module calculates the driver experience category probability factor P(C i ) according to the following formula: In the formula, Y is the driving age, which is normalized by Min-Max and the value range is between 0 and 1. S is the total driving mileage, which is normalized using Min-Max and has a value range between 0 and 1; D is the driving style score, with a value range between 0 and 1; A is the accident rate, and the extreme value impact is reduced through logarithmic transformation, with a value range between 0 and 1; is the output of the first hidden layer, is the output of the second hidden layer, is the output of the third hidden layer; ReLU is the activation function of the first hidden layer, which extracts basic driving features; The activation function of the second hidden layer is Swish, which enhances the non-linear mapping ability; the activation function of the third hidden layer is ReLU, which optimizes the feature expression and improves the classification ability; is the weight matrix from the input layer to the first hidden layer, is the weight matrix from the first hidden layer to the second hidden layer, is the weight matrix from the second hidden layer to the third hidden layer, and is the weight matrix from the third hidden layer to the output layer, all of which are determined through training; is the bias term of the first hidden layer, is the bias term of the second hidden layer, is the bias term of the third hidden layer, and is the bias term of the output layer, all of which are determined through training; U E2,i is the raw score of the unnormalized output layer; represents the exponential transformation mapping of U E2,i ; the driver experience category probability factor P(C i ) represents the probability that the driver belongs to the i-th experience category. Softmax normalization makes P(C1)+P(C2)+P(C3)+P(C4)=1; P(C1) is the probability factor for novice drivers, P(C2) is the probability factor for average drivers, P(C3) is the probability factor for experienced drivers, and P(C4) is the probability factor for professional drivers, and their value ranges are all between 0 and 1; C i is the driver experience category; where C1 is a novice driver with a short driving age, low driving mileage, an aggressive driving style, and a relatively high accident rate, and C1 takes the value of 1 during calculation; C2 is an average driver with a medium driving age and mileage, a relatively mild driving style, and a relatively low accident rate, and C2 takes the value of 2 during calculation; C3 is an experienced driver with a high driving age and mileage, a stable driving style, and a very low accident rate, and C3 takes the value of 3 during calculation; C4 is a professional driver with a long driving age, a long driving mileage, a stable driving style, and almost no accidents, and C4 takes the value of 4 during calculation; The input layer contains 4 nodes, corresponding to the driving age Y, total driving mileage S, driving style D, and accident rate A respectively; the hidden layer adopts a three-layer structure, including the first hidden layer, the second hidden layer, and the third hidden layer; the first hidden layer contains 64 neurons, the second hidden layer contains 32 neurons, and the third hidden layer contains 16 neurons; the output layer contains 4 nodes, which are the novice driver probability factor P(C1), the average driver probability factor P(C2), the proficient driver probability factor P(C3), and the professional driver probability factor P(C4).

4. A centroid side slip angle fusion estimation method based on an intelligent vehicle according to claim 1, characterized in that The centroid side slip angle estimation influence factor calculation module calculates the road adhesion factor R1 according to the following formula: In the formula, μ is the current road surface friction coefficient; μ0 is the standard dry road surface friction coefficient, and the value range is between 0.8 and 1.

0. The road gradient factor R2 is calculated according to the following formula: In the formula, θ x is the longitudinal slope angle, and θ y is the transverse slope angle; The environmental weather factor R3 is calculated according to the following formula: In the formula, ξ is the solar altitude angle; Tem is the air temperature. Precip is the precipitation intensity; γ1 is the air temperature adjustment coefficient, and the value range is between 0.01 and 0.1; γ2 is the precipitation intensity influence coefficient, and the value range is between 0.01 and 0.

5. The road curvature factor R4 is calculated according to the following formula: In the formula, is the road curvature, which depends on the road type; The crosswind influence factor R5 is calculated according to the following formula: where v w is the crosswind speed; v x is the longitudinal speed.

5. A method for fusing and estimating the sideslip angle of the centroid based on an intelligent vehicle according to claim 1, characterized in that The centroid side slip angle estimation execution module includes executing the first estimation mode, the second estimation mode, and the third estimation mode. S1. When the centroid sideslip angle estimation execution module executes the first estimation mode, that is, the least squares estimation mode based on kinematics, an adaptive weight factor at time t is designed. The estimated value of the centroid sideslip angle β at time t 1,t The final calculation formula is as follows: In the formula, is the centroid sideslip angle estimated by the least squares method at time t-1; a y,t is the lateral acceleration of the vehicle at time t; is the set critical lateral acceleration, and the value range is 0.8 to 1.5 m / s 2 ; ε is the adjustment factor for controlling the Sigmoid transition rate, and the value range is 3 to 8; is the error covariance matrix of the least squares method at time t-1; is the observation matrix of the least squares method at time t, is the transpose of the observation matrix of the least squares method at time t; is the measurement noise covariance matrix of the least squares method at time t; is the measurement data of the least squares method at time t; S2. When the centroid sideslip angle estimation execution module executes the second estimation mode, i.e., the unscented Kalman filter estimation mode based on dynamics, the estimated value β of the centroid sideslip angle at time t 2,t has the following final calculation formula: In the formula, is the lateral velocity estimated by the unscented Kalman filter at time t, is the longitudinal velocity estimated by the unscented Kalman filter at time t; is the longitudinal velocity at time t calculated from the prediction at time t - 1 in the unscented Kalman filter estimation mode, is the lateral velocity at time t calculated from the prediction at time t - 1 in the unscented Kalman filter estimation mode; is the measurement data of the unscented Kalman filter at time t; is the measurement data at time t calculated from the prediction at time t - 1 in the unscented Kalman filter estimation mode; is the Kalman gain in the longitudinal velocity direction of the unscented Kalman filter at time t, is the Kalman gain in the lateral velocity direction of the unscented Kalman filter at time t; S3. When the centroid sideslip angle estimation execution module executes the third estimation mode, i.e., the volume Kalman filter estimation mode based on dynamics, the estimated value β of the centroid sideslip angle at time t 3,t has the following final calculation formula: Wherein, is the lateral velocity estimated by the cubature Kalman filter at time t, is the longitudinal velocity estimated by the cubature Kalman filter at time t; is the longitudinal velocity at time t calculated from the prediction at time t-1 in the cubature Kalman filter estimation mode, is the lateral velocity at time t calculated from the prediction at time t-1 in the cubature Kalman filter estimation mode; is the measurement data of the cubature Kalman filter at time t; is the measurement data at time t calculated from the prediction at time t-1 in the cubature Kalman filter estimation mode; is the Kalman gain in the longitudinal velocity direction of the cubature Kalman filter at time t, is the Kalman gain in the lateral velocity direction of the cubature Kalman filter at time t.

6. A centroidal side slip angle fusion estimation method based on an intelligent vehicle according to claim 1, characterized in that, The centroid sideslip angle fusion estimation module based on genetic algorithm and neural network first initializes the population, sets the population size as M = 100, randomly generates the weight factors which conform to the Dirichlet(1, 1, 1) distribution, so that the sum of weights is 1 and possible solutions can be explored evenly; calculates the objective function J(w1, w2, w3) to minimize the centroid sideslip angle estimation error; the fitness function Υ = (1 + J(w1, w2, w3)) -1 In the value range from 0 to 1, the closer to 1 indicates that the individual is better; The selection strategy adopts tournament selection to pick the optimal individuals from the population for replication and inheritance; two-point crossover is adopted to effectively combine the information of different individuals, and the crossover probability P c is set to 0.8; non-uniform mutation is used for mutation so that the population can finely adjust the parameters in the later stage, and the mutation rate P m is set to 0.05 to keep the optimization search diverse but not overly disturb the optimization process; the termination condition is that the fitness function J(w1, w2, w3) converges to a range less than the threshold 0.000001 or reaches the maximum number of iterations 200; Considering that the actual working conditions are constantly changing and the vehicle requires a millisecond-level calculation response, the online operation of the genetic algorithm has too high a calculation cost and a calculation time lag, which is not suitable for vehicle real-time control; in the training stage, the vehicle dynamic characteristic factor influence factor E1, the driver factor influence factor E2, and the environmental road condition factor influence factor E3 are used as the data set to train the neural network; in the online inference stage, the neural network can directly predict the weight factor w1 of the kinematics-based least squares estimation mode optimized by the genetic algorithm, the weight factor w2 of the dynamics-based unscented Kalman filter estimation mode, and the weight factor w3 of the dynamics-based cubature Kalman filter estimation mode according to the input vehicle dynamic characteristic factor influence factor E1, driver factor influence factor E2, and environmental road condition factor influence factor E3; the genetic algorithm learns offline to improve the calculation efficiency; the neural network architecture is as follows: Wherein, is the output of the first hidden layer, is the output of the second hidden layer; Swish is the activation function of the first hidden layer, and ReLU is the activation function of the second hidden layer; is the weight matrix from the input layer to the first hidden layer, is the weight matrix from the first hidden layer to the second hidden layer, is the weight matrix from the second hidden layer to the output layer, and all are determined through training; is the bias term of the first hidden layer, is the bias term of the second hidden layer, is the bias term of the output layer, and all are determined through training; U w is the output vector [w1, w2, w3] T ; The input layer contains 3 nodes, and the inputs are the influencing factor E1 of vehicle dynamic characteristics, the influencing factor E2 of driver factors, and the influencing factor E3 of environmental road conditions; the hidden layer adopts a double-layer structure, including a first hidden layer and a second hidden layer; the first hidden layer contains 32 neurons, and the second hidden layer contains 16 neurons; the output layer contains 3 nodes, and the outputs are the weight factor w1 of the least squares estimation mode based on kinematics, the weight factor w2 of the unscented Kalman filter estimation mode based on dynamics, and the weight factor w3 of the cubature Kalman filter estimation mode based on dynamics.

Images