CN116674571A - Real-time estimation method for automobile quality and gradient based on data confidence factor - Google Patents

Abstract

The invention belongs to the technical field of automobiles, and particularly relates to an automobile quality and gradient real-time estimation method based on a data confidence factor. Comprising the following steps: 1. acquiring vehicle running state data; 2. establishing a relation model of the whole vehicle mass, the road gradient and the vehicle running state data based on the vehicle dynamics model; 3. calculating working condition characteristic parameters by using the vehicle running state data and the vehicle inherent parameters; 4. establishing a quality estimation confidence factor model based on a neural network algorithm; 5. estimating a vehicle mass based on a least squares method of the confidence factor; 6. judging whether the quality estimation value is stable; 7. the road grade is estimated based on an extended kalman filter algorithm. The method breaks the limitation of the traditional dynamic model-based estimated mass on the working condition of the vehicle, and realizes decoupling of the vehicle mass and road gradient estimation according to the measurement principle of the vehicle dynamic model and the vehicle-mounted acceleration sensor, thereby improving the accuracy of the estimated values of the vehicle mass and the road gradient.

Description

Real-time estimation method for automobile quality and gradient based on data confidence factor

Technical Field

The invention belongs to the technical field of automobiles, and particularly relates to an automobile quality and gradient real-time estimation method based on a data confidence factor.

Background

With the rapid development of automobile intellectualization, researchers have developed more automobile electronic control systems, which are control logic in real time according to the running state of a vehicle and road information. The mass and the road gradient of the whole vehicle are taken as two basic parameters in a vehicle model, and are main sources of inertia force, rolling resistance and gradient resistance of the vehicle. For vehicles, due to the limitation of production and manufacturing costs, the mass and the gradient cannot be measured in real time directly through sensors, so the mass and the gradient are generally estimated in real time based on a vehicle dynamics model or a data driving model, which has important influences on vehicle speed control, transmission control, powertrain optimization and the like of the vehicles. Therefore, the exploration of the real-time estimation method for the whole vehicle quality and the road gradient has important significance for improving the safety and the economy of the vehicle.

At present, the estimation of the whole vehicle quality and the road surface gradient is mainly completed by combining a vehicle longitudinal dynamics model with a corresponding estimation algorithm. The method based on the dynamic model has small dependence on the vehicle sensor and strong universality, wherein the more common estimation methods are a recursive least square method (Recursive Least Squares, RLS for short) and a Kalman Filter. In general, the least squares method is more suitable for estimating a vehicle mass with a relatively small amount of change, while the kalman filtering is more suitable for estimating a continuously changing road gradient. But the applicability of the mass estimation method based on the longitudinal dynamics model is limited, and the mass estimation method based on the longitudinal dynamics model is only suitable for a relatively simple linear acceleration working condition, because the running state of the vehicle is quite in accordance with the assumption of the longitudinal dynamics model of the vehicle, and then the mass estimation method based on the dynamics model can estimate more accurately. However, in reality, the vehicle is a complex system, the actual running condition is very complex, when the vehicle is in the complex conditions of gear shifting, braking and the like, the vehicle system cannot be expressed by the vehicle dynamics model, and at this time, the mass estimation based on the longitudinal dynamics model has larger error. In addition, the coupling between the mass and the road gradient is strong in the dynamics model, and if the mass estimation value is large in error, the estimation accuracy of the road gradient is seriously affected. Therefore, decoupling of the mass-gradient estimation model is required, and modeling of the vehicle is performed in nonlinear states such as gear shifting and braking, so that applicability of the mass estimation method under various working conditions is improved.

Disclosure of Invention

The invention provides an automobile quality and gradient estimation method based on a data confidence factor, which utilizes the nonlinear advantage of a data driving model to realize modeling of an automobile in a nonlinear state, so as to correct a quality estimation algorithm based on an automobile dynamics model, improve the applicability of the estimation algorithm under various working conditions, input the quality estimation value into a gradient estimation model after the estimation value of the quality estimation model is stable, realize decoupling of quality and gradient estimation, reduce the influence of the quality estimation precision on the gradient estimation precision, thereby completing joint estimation of the automobile quality and road gradient, further improving the safety and economy of the automobile, and solving the problems existing in the estimation of the existing whole automobile quality and road gradient.

The technical scheme of the invention is as follows in combination with the accompanying drawings:

a vehicle quality and gradient estimation method based on a data confidence factor comprises the following steps:

step one, acquiring vehicle running state data;

step two, building a relation model of the whole vehicle mass, the road gradient and the vehicle running state data based on the vehicle dynamics model;

step three, calculating working condition characteristic parameters by utilizing vehicle running state data and vehicle inherent parameters;

establishing a quality estimation confidence factor model based on a neural network algorithm;

estimating the vehicle mass based on a least square method of the confidence factor;

step six, judging whether the quality estimation value is stable;

and step seven, estimating the road gradient based on an extended Kalman filtering algorithm.

Further, the specific method of the first step is as follows:

the vehicle-mounted sensor longitudinal acceleration signal, the longitudinal vehicle speed signal, the engine output torque signal, the vehicle gear signal and the master cylinder brake pressure signal are obtained through a vehicle CAN bus.

Further, the specific method of the second step is as follows:

21 Simplifying and analyzing the stress of the vehicle to obtain a longitudinal dynamics model of the vehicle; wherein, the vehicle longitudinal dynamics model is expressed as:

F _t ＝F _w +F _i +F _f +F _j

in the formula ,F_t Is the driving force of the vehicle, andF _w is air resistance, andF _i is gradient resistance, and F _i ＝MgsinαF _f Is rolling resistance, and F _f ＝Mgfcosα；F _j Is acceleration resistance, and->

wherein ,

sinα≈tanα＝i

cosα≈1

22 Combining the above to obtain a relation model of the whole vehicle mass, the road gradient and the vehicle running state, wherein the relation model comprises the following steps:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio of the gearbox; i.e ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Respectively representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobile; f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a represents a windward area; ρ represents the air density; v denotes the vehicle longitudinal speed; i represents a road gradient;

the total equivalent moment of inertia of the rotating mass is represented by the following formula:

I _{total (S)} ＝∑I _w +I _f i _g ² i ₀ ² η _t

in the formula ,I_w The moment of inertia of the wheels; i _f The flywheel rotational inertia;

the rolling resistance coefficient is represented by the following formula:

f＝(f ₀ +f ₁ v)

wherein v is the longitudinal speed of the vehicle and the unit is km/h; f (f) ₀ Is a constant term of the rolling resistance coefficient; f (f) ₁ Is the coefficient of rolling resistance of oneA minor term; knowing the rolling resistance coefficient as described above is related to the longitudinal speed of the vehicle;

the longitudinal acceleration may be obtained by a difference in vehicle speed over time, represented by:

where Δt is the sampling period of data acquisition.

Further, the specific method of the third step is as follows:

31 Main speed reduction ratio i of the vehicle intrinsic parameters ₀ Efficiency eta of mechanical transmission _t Front and rear brake performance factor k _bf k _br The effective rolling radius r of the wheel and the total equivalent moment of inertia I of the rotating mass _{Total (S)} Rolling resistance coefficient f, vehicle air resistance coefficient C _D The windward area A is obtained through corresponding experimental measurement;

32 After obtaining the intrinsic parameters of the vehicle through experiments and obtaining the running state data of the vehicle through the CAN signals of the vehicle, calculating the characteristic parameters capable of expressing the running working conditions of the vehicle;

the calculated resistances are defined as:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio, i ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobile; f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a represents a windward area; ρ represents the air density; v denotes the vehicle longitudinal speed.

33 Calculating working condition characteristic parameters, namely acceleration, as follows:

in the formula ,a theoretical longitudinal vehicle acceleration obtained by a vehicle speed versus time difference; i is road grade; g is gravity acceleration; f is the rolling resistance coefficient.

Further, the specific method of the fourth step is as follows:

41 Establishing a quality estimation confidence factor model by adopting an error back propagation neural network algorithm, namely a BP algorithm;

the definition of the BP algorithm is as follows:

for a given training set d= { (x) ₁ ,y ₁ ),(x ₂ ,y ₂ ),...,(x _m ,y _m )}，I.e. the input is described by d-dimensional attributes and output as an l-dimensional real value vector; the neural network is a multi-layer feedforward network with d input neurons, l output neurons and q hidden layer neurons, wherein the threshold value of the jth neuron of the output layer is represented by theta _j Representing the threshold value gamma of the h neuron in the hidden layer _h A representation;

the connection weight between the ith neuron in the input layer and the h neuron in the hidden layer is v _ih The method comprises the steps of carrying out a first treatment on the surface of the The connection weight between the h neuron in the hidden layer and the j neuron in the output layer is w _hj The method comprises the steps of carrying out a first treatment on the surface of the The input received by the h neuron in the hidden layer isThe input received by the j-th neuron in the output layer is +.> wherein b_h The output of the h neuron in the hidden layer;

for training (x _k ,y _k ) The output of the neural network isNamely:

the neural network is then in (x _k ,y _k ) The mean square error is:

there are (d+l+1) q+l parameters in the neural network to be determined: d×q weights from the input layer to the hidden layer, q×l weights from the hidden layer to the output layer, q hidden layer neuron thresholds, and l output layer neuron thresholds;

the BP neural network is an iterative learning algorithm, and the parameters are updated and estimated by adopting a generalized perceptron learning rule in each iteration round, and the update estimation formula of any parameter v is as follows:

v←v+Δv

BP algorithm adjusts parameters in the negative gradient direction of the target based on gradient descent strategy, and adjusts error E _k The given learning rate η is:

w _hj first affecting the input value beta of the jth output layer neuron _j And then influence the output valueThen influence E _k There is

The learning rate eta epsilon (0, 1) controls the updating step length in each iteration of the algorithm;

42 For each training sample, the BP algorithm performs the following operations: firstly, providing input data to neurons of an input layer, and then forwarding training signals layer by layer until a result of an output layer is generated; and then calculating the error of the output layer, back-propagating the error to the hidden layer neuron, and finally adjusting the connection weight and the threshold value according to the error of the hidden neuron. The iterative process is looped until certain training stop conditions are reached;

wherein the tag value confidence coefficient in the dataset is defined as:

wherein m is a mass value estimated by a least square method; m is m ₀ The vehicle quality true value; confidence coefficient c _f ∈(0,1)；

Wherein, the threshold value algorithm of the confidence factor is defined as:

in the formula ,c_f For confidence coefficient, C _f Is a confidence factor;

43 Confidence factor C calculated by a threshold value algorithm _f Representing the rationality of the current working condition for quality estimation, and controlling an updating strategy of the vehicle during quality estimation under different working conditions through a confidence factor;

when C _f When the vehicle is 1, the running condition of the vehicle is very in accordance with the assumed condition of the longitudinal dynamics model of the vehicle, the vehicle is in an ideal linear region, the dynamics model-based estimation algorithm is better in performance, and the quality estimation value is more accurate;

when C _f When the value is 0, the running condition of the vehicle is not consistent with the longitudinal movement of the vehicleUnder the assumption of the mechanical model, the vehicle is in a relatively complex nonlinear state, and the estimation effect based on the dynamic model is inaccurate at the moment, the updating of the quality estimation value should be stopped, and the estimation value which is relatively accurate at the last moment is used as the estimation value at the current moment.

Further, the specific method of the fifth step is as follows:

51 The working principle of the vehicle-mounted accelerometer is as follows:

wherein ,the vehicle theoretical longitudinal acceleration is obtained through vehicle speed difference; i is road grade; g is gravity acceleration;

52 According to the working principle of the vehicle-mounted accelerometer and the longitudinal dynamics model of the vehicle, the model decoupling of the vehicle mass and the road gradient is realized, and the decoupling model of the mass estimation is as follows:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio of the gearbox; i.e ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobile; f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a represents a windward area; ρ represents the air density; v denotes the vehicle longitudinal speed; a, a _{x_sensor} Measuring a vehicle acceleration for the vehicle accelerometer;

53 The output quantity y=θh of the recursive least squares method, the parameter θ=m to be identified, the vehicle mass M is estimated using the recursive least squares method with confidence factors according to the following equation:

in the formula ,C_f Is a confidence factor; λ is a time-varying forgetting factor; p is the recursive covariance matrix.

Further, the specific method in the sixth step is as follows:

judging whether the covariance matrix P in the recursive least square algorithm in the quality estimation model is smaller than a critical value P ₀ ；

If P is less than P ₀ The quality estimated value at the current moment tends to be converged, the quality estimated value is accurate, and is considered as an inherent parameter of the vehicle because the vehicle does not greatly change in the running process, and the estimated value is input into a gradient estimation model at the moment so as to perform gradient estimation;

if P is greater than or equal to P ₀ The quality estimated value at the current moment is rapidly updated, and the change of the quality estimated value is relatively large, so that if the quality estimated value is input into the gradient estimation model, the error of gradient estimation is relatively large, and the accuracy required by vehicle control cannot be achieved, and therefore the steps one to five should be continuously repeated until the quality estimated value tends to be stable and converged, and then the road gradient estimation is performed.

Further, the specific method of the step seven is as follows:

71 Obtaining a state space model of the system according to the vehicle dynamics model as follows:

and (3) making:

72 Obtaining a discrete state space equation of the system according to the state space model of the system as follows:

wherein v (k) is the longitudinal vehicle speed at the current moment; t (T) _tq (k) The torque is output for the engine at the current moment; p is p _mc (k) The master cylinder braking pressure at the current moment; i (k) is the road gradient at the current time;

73 The observation equation for the system is as follows:

74 Jacobian matrix for the system:

in the formula ,J_f Jacobian matrix as system state equation and deltat as discrete step length of system;

75 State transition matrix a=j of system _f ；

76 System observation matrix is

77 After the Jacobian matrix and the observation matrix of the system are obtained through calculation, a road gradient estimation model is established through a Kalman filtering algorithm, and the flow is as follows:

(1) Initializing state variablesAnd a posterior estimated bias covariance P _k-1 ；

(2) According to the formulaCalculating a priori estimated value of the state variable;

(3) According to the formulaCalculating a priori estimated deviation covariance;

(4) According to the formulaCalculating Kalman filtering gain;

(5) According to the formulaCalculating posterior estimation deviation covariance;

(6) According to the formulaA state variable posterior estimate is calculated.

The beneficial effects of the invention are as follows:

1) The method breaks the limitation of the traditional dynamic model-based estimation quality on the working condition of the vehicle;

2) According to the vehicle dynamics model and the measurement principle of the vehicle-mounted acceleration sensor, decoupling of vehicle mass and road gradient estimation is realized, so that accuracy of estimated values of the vehicle mass and the road gradient estimation is improved

3) According to the invention, modeling of the vehicle in a nonlinear state is realized by utilizing the nonlinear advantage of the data-driven model, the quality estimation confidence coefficient factor is obtained through the data-driven model, and the quality estimation algorithm of the dynamic model is corrected according to the confidence coefficient factor, so that the accuracy of the quality estimation algorithm is higher, and the robustness is better;

4) The invention realizes accurate estimation of the vehicle mass and the road gradient, and is beneficial to improvement of the safety and the economy of the vehicle.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic flow chart of the present invention;

FIG. 2 is a schematic illustration of the force applied by a vehicle accelerating uphill;

FIG. 3 is a schematic diagram of a neural network based quality estimation confidence factor model;

FIG. 4 is a schematic illustration of an algorithm for estimating road grade based on Kalman filtering.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Referring to fig. 1, a method for estimating the quality and gradient of an automobile based on a data confidence factor includes the following steps:

step one, acquiring vehicle running state data, wherein the specific method comprises the following steps:

the vehicle-mounted sensor longitudinal acceleration signal, the longitudinal vehicle speed signal, the engine output torque signal, the vehicle gear signal and the master cylinder brake pressure signal are obtained through a vehicle CAN bus.

Step two, a relation model of the whole vehicle mass, the road gradient and the vehicle running state data is established based on the vehicle dynamics model, and the specific method is as follows:

21 The stress of the vehicle when accelerating up a slope is shown in figure 2, wherein G is the gravity of the automobile; alpha is the road slope angle; h is a _g Is high in mass center of the automobile; f (F) _w Is air resistance; f (F) _z1 、F _z2 Is the ground normal reaction force acting on the front wheel and the rear wheel; f (F) _X1 、F _X2 To act on the front and rear wheelsTangential reaction force on the ground; l is the wheelbase of the automobile; a. b is the distance from the mass center of the automobile to the front and rear axles respectively;

the stress of the vehicle is simplified and analyzed to obtain a longitudinal dynamics model of the vehicle; wherein, the vehicle longitudinal dynamics model is expressed as:

F _t ＝F _w +F _i +F _f +F _j

in the formula ,F_t Is the driving force of the vehicle, andF _w is air resistance, andF _i is gradient resistance, and F _i ＝MgsinαF _f Is rolling resistance, and F _f ＝Mgfcosα；F _j Is acceleration resistance, and->

According to the road design specifications of China, the gradient of a general road is smaller in practice, and the road can be considered as follows:

sinα≈tanα＝i

cosα≈1

22 Combining the above to obtain a relation model of the whole vehicle mass, the road gradient and the vehicle running state, wherein the relation model comprises the following steps:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio of the gearbox; i.e ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Respectively representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobileThe method comprises the steps of carrying out a first treatment on the surface of the f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a represents a windward area; ρ represents the air density; v denotes the vehicle longitudinal speed; i represents a road gradient;

the total equivalent moment of inertia of the rotating mass is represented by the following formula:

I _{total (S)} ＝∑I _w+ I _f i _g ² i ₀ ² η _t

in the formula ,I_w The moment of inertia of the wheels; i _f The flywheel rotational inertia;

the rolling resistance coefficient is represented by the following formula:

f＝(f ₀ +f ₁ v)

wherein v is the longitudinal speed of the vehicle and the unit is km/h; f (f) ₀ Is a constant term of the rolling resistance coefficient; f (f) ₁ A rolling resistance coefficient primary term; knowing the rolling resistance coefficient as described above is related to the longitudinal speed of the vehicle;

the longitudinal acceleration may be obtained by a difference in vehicle speed over time, represented by:

where Δt is the sampling period of data acquisition.

Calculating working condition characteristic parameters by using vehicle running state data and vehicle inherent parameters, wherein the specific method comprises the following steps:

31 Main speed reduction ratio i of the vehicle intrinsic parameters ₀ Efficiency eta of mechanical transmission _t Front and rear brake performance factor k _bf k _br The effective rolling radius r of the wheel and the total equivalent moment of inertia I of the rotating mass _{Total (S)} Rolling resistance coefficient f, vehicle air resistance coefficient C _D The windward area A is obtained through corresponding experimental measurement;

32 After obtaining the intrinsic parameters of the vehicle through experiments and obtaining the running state data of the vehicle through the CAN signals of the vehicle, calculating the characteristic parameters capable of expressing the running working conditions of the vehicle;

the calculated resistances are defined as:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio, i ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobile; f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a represents a windward area; ρ represents the air density; v denotes the vehicle longitudinal speed.

33 Calculating working condition characteristic parameters, namely acceleration, as follows:

in the formula ,a theoretical longitudinal vehicle acceleration obtained by a vehicle speed versus time difference; i is road grade; g is gravity acceleration; f is the rolling resistance coefficient.

Step four, establishing a quality estimation confidence factor model based on a neural network algorithm, wherein the specific method comprises the following steps:

41 Establishing a quality estimation confidence factor model by adopting an error back propagation neural network algorithm, namely a BP algorithm; as shown in fig. 3, the inputs of the mass estimation confidence factor model based on the neural network are calculated resistance, calculated acceleration, engine output torque, brake master cylinder pressure, and output is a mass estimation confidence factor;

the definition of the BP algorithm is as follows:

for a given training set d= { (x) ₁ ,y ₁ ),(x ₂ ,y ₂ ),...,(x _m ,y _m )}，I.e. the input is described by d-dimensional attributes and output as an l-dimensional real value vector; the neural network is a multi-layer feedforward network with d input neurons, l output neurons and q hidden layer neurons, wherein the threshold value of the jth neuron of the output layer is represented by theta _j Representing the threshold value gamma of the h neuron in the hidden layer _h A representation;

the connection weight between the ith neuron in the input layer and the h neuron in the hidden layer is v _ih The method comprises the steps of carrying out a first treatment on the surface of the The connection weight between the h neuron in the hidden layer and the j neuron in the output layer is w _hj The method comprises the steps of carrying out a first treatment on the surface of the The input received by the h neuron in the hidden layer isThe input received by the j-th neuron in the output layer is +.> wherein b_h The output of the h neuron in the hidden layer;

for training (x _k ,y _k ) The output of the neural network isNamely:

the neural network is then in (x _k ,y _k ) The mean square error is:

as shown in fig. 3, there are (d+l+1) q+l parameters in the neural network to be determined: d×q weights from the input layer to the hidden layer, q×l weights from the hidden layer to the output layer, q hidden layer neuron thresholds, and l output layer neuron thresholds;

the BP neural network is an iterative learning algorithm, and the parameters are updated and estimated by adopting a generalized perceptron learning rule in each iteration round, and the update estimation formula of any parameter v is as follows:

v←v+Δv

the BP algorithm adjusts parameters in the negative gradient direction of the target based on gradient descent (gradient descent) strategy, and adjusts the error E _k The given learning rate η is:

w _hj first affecting the input value beta of the jth output layer neuron _j And then influence the output valueThen influence E _k There is

The learning rate eta epsilon (0, 1) controls the updating step length in each iteration of the algorithm;

42 For each training sample, the BP algorithm performs the following operations: firstly, providing input data to neurons of an input layer, and then forwarding training signals layer by layer until a result of an output layer is generated; then calculating the error of the output layer, back-propagating the error to the hidden layer neuron, and finally adjusting the connection weight and the threshold value according to the error of the hidden neuron; the iterative process is looped until certain training stop conditions are reached;

wherein the tag value confidence coefficient in the dataset is defined as:

wherein m is a mass value estimated by a least square method; m is m ₀ The vehicle quality true value; confidence coefficient c _f ∈(0,1)；

The neural network algorithm is a learning algorithm based on data driving, and a confidence coefficient calculation model can be obtained after the algorithm is learned through multiple training rounds. When the model is used for actually predicting the confidence coefficient, the confidence coefficient needs to be limited due to the defects of mislearning, strong divergence and the like of the neural network, so that the confidence coefficient which can finally express whether the quality estimation working condition is applicable or not is obtained.

Wherein, the threshold value algorithm of the confidence factor is defined as:

in the formula ,c_f For confidence coefficient, C _f Is a confidence factor;

43 Confidence factor C calculated by a threshold value algorithm _f Representing the rationality of the current working condition for quality estimation, and controlling an updating strategy of the vehicle during quality estimation under different working conditions through a confidence factor;

when C _f When the vehicle is 1, the running condition of the vehicle is very in accordance with the assumed condition of the longitudinal dynamics model of the vehicle, the vehicle is in an ideal linear region, the dynamics model-based estimation algorithm is better in performance, and the quality estimation value is more accurate;

when C _f When the estimated value is 0, the running condition of the vehicle is not in accordance with the assumed condition of the longitudinal dynamics model of the vehicle, the vehicle is in a complex nonlinear state, the estimated effect based on the dynamics model is inaccurate, the updating of the quality estimated value should be stopped, and the estimated value which is accurate at the last moment is used as the estimated value at the current moment.

Estimating the vehicle mass based on a least square method of a confidence factor, wherein the method comprises the following steps of:

51 The working principle of the vehicle-mounted accelerometer is as follows:

wherein ,the vehicle theoretical longitudinal acceleration is obtained through vehicle speed difference; i is road grade; g is gravity acceleration;

52 According to the working principle of the vehicle-mounted accelerometer and the longitudinal dynamics model of the vehicle, the model decoupling of the vehicle mass and the road gradient is realized, and the decoupling model of the mass estimation is as follows:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio of the gearbox; i.e ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobile; f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a represents a windward area; ρ represents the air density; v denotes the vehicle longitudinal speed; a, a _{x_sensor} Measuring a vehicle acceleration for the vehicle accelerometer;

53 The output quantity y=θh of the recursive least squares method, the parameter θ=m to be identified, the vehicle mass M is estimated using the recursive least squares method with confidence factors according to the following equation:

in the formula ,C_f Is a confidence factor; λ is a time-varying forgetting factor; p is the recursive covariance matrix.

Step six, judging whether the quality estimation value is stable, wherein the specific method is as follows:

judging whether the covariance matrix P in the recursive least square algorithm in the quality estimation model is smaller than a critical value P ₀ ；

If P is less than P ₀ The quality estimated value at the current moment tends to be converged, the quality estimated value is accurate, and is considered as an inherent parameter of the vehicle because the vehicle does not greatly change in the running process, and the estimated value is input into a gradient estimation model at the moment so as to perform gradient estimation;

if P is greater than or equal to P ₀ The quality estimated value at the current moment is rapidly updated, and the change of the quality estimated value is relatively large, and at the moment, if the quality estimated value is input into the gradient estimation model, the error of gradient estimation is relatively large, and the accuracy required by vehicle control cannot be achieved, so that the steps one to five should be continuously repeated until the quality estimated value tends to be stable and converged, and then the road gradient estimation can be performed.

Step seven, estimating the road gradient based on an extended Kalman filtering algorithm, wherein the specific method is as follows:

71 Obtaining a state space model of the system according to the vehicle dynamics model as follows:

and (3) making:

72 Obtaining a discrete state space equation of the system according to the state space model of the system as follows:

wherein v (k) is the currentLongitudinal vehicle speed at moment; t (T) _tq (k) The torque is output for the engine at the current moment; p is p _mc (k) The master cylinder braking pressure at the current moment; i (k) is the road gradient at the current time;

73 The observation equation for the system is as follows:

74 Since the system state equation involved is highly nonlinear, a further derivation of the jacobian matrix that expresses the vehicle's nonlinearity is needed, the jacobian matrix of the system is:

in the formula ,J_f Jacobian matrix as system state equation and deltat as discrete step length of system;

75 For this system, the state transition matrix of the system a=j _f ；

76 System observation matrix is

77 After the Jacobian matrix and the observation matrix of the system are calculated, a road gradient estimation model is established through a Kalman filtering algorithm, as shown in fig. 4, and the flow is as follows:

(1) Initializing state variablesAnd a posterior estimated bias covariance P _k-1 ；

(2) According to the formulaCalculating a priori estimated value of the state variable;

(3) According to the formulaCalculating a priori estimated deviation covariance;

(4) According to the formulaCalculating Kalman filtering gain;

(5) According to the formulaCalculating posterior estimation deviation covariance;

(6) According to the formulaA state variable posterior estimate is calculated.

In summary, the method breaks the limitation of the traditional dynamic model-based estimated mass on the working condition of the vehicle, and realizes decoupling of the vehicle mass and road gradient estimation according to the measuring principle of the vehicle dynamic model and the vehicle-mounted acceleration sensor, thereby improving the accuracy of the estimated values of the vehicle mass and the road gradient.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (8)

1. The automobile quality and gradient estimation method based on the data confidence factor is characterized by comprising the following steps of:

step one, acquiring vehicle running state data;

step two, building a relation model of the whole vehicle mass, the road gradient and the vehicle running state data based on the vehicle dynamics model;

step three, calculating working condition characteristic parameters by utilizing vehicle running state data and vehicle inherent parameters;

establishing a quality estimation confidence factor model based on a neural network algorithm;

estimating the vehicle mass based on a least square method of the confidence factor;

step six, judging whether the quality estimation value is stable;

and step seven, estimating the road gradient based on an extended Kalman filtering algorithm.

2. The method for estimating the quality and gradient of the vehicle based on the data confidence factor according to claim 1, wherein the specific method of the first step is as follows:

the vehicle-mounted sensor longitudinal acceleration signal, the longitudinal vehicle speed signal, the engine output torque signal, the vehicle gear signal and the master cylinder brake pressure signal are obtained through a vehicle CAN bus.

3. The method for estimating the quality and the gradient of the automobile based on the data confidence factor according to claim 1, wherein the specific method of the second step is as follows:

21 Simplifying and analyzing the stress of the vehicle to obtain a longitudinal dynamics model of the vehicle; wherein, the vehicle longitudinal dynamics model is expressed as:

F _t ＝F _w +F _i +F _f +F _j

in the formula ,F_t Is the driving force of the vehicle, andF _w is air resistance and->F _i Is gradient resistance, and F _i ＝MgsinαF _f Is rolling resistance, and F _f ＝Mgfcosα；F _j Is acceleration resistance, and

wherein sin α≡tanα=i

cosα≈1

22 Combining the above to obtain a relation model of the whole vehicle mass, the road gradient and the vehicle running state, wherein the relation model comprises the following steps:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio of the gearbox; i.e ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Respectively representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobile; f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a represents a windward area; ρ represents the air density; v denotes the vehicle longitudinal speed; i represents a road gradient;

the total equivalent moment of inertia of the rotating mass is represented by the following formula:

I _{total (S)} ＝∑I _w +I _f i _g ² i ₀ ² η _t

in the formula ,I_w The moment of inertia of the wheels; i _f The flywheel rotational inertia;

the rolling resistance coefficient is represented by the following formula:

f＝(f ₀ +f ₁ v)

wherein v is the longitudinal speed of the vehicle and the unit is km/h; f (f) ₀ Is a constant term of the rolling resistance coefficient; f (f) ₁ A rolling resistance coefficient primary term; knowing the rolling resistance coefficient as described above is related to the longitudinal speed of the vehicle;

the longitudinal acceleration may be obtained by a difference in vehicle speed over time, represented by:

where Δt is the sampling period of data acquisition.

4. The method for estimating the quality and gradient of the automobile based on the data confidence factor according to claim 1, wherein the specific method in the third step is as follows:

31 Main speed reduction ratio i of the vehicle intrinsic parameters ₀ Efficiency eta of mechanical transmission _t Front and rear brake performance factor k _bf k _br The effective rolling radius r of the wheel and the total equivalent moment of inertia I of the rotating mass _{Total (S)} Rolling resistance coefficient f, vehicle air resistance coefficient C _D The windward area A is obtained through corresponding experimental measurement;

32 After obtaining the intrinsic parameters of the vehicle through experiments and obtaining the running state data of the vehicle through the CAN signals of the vehicle, calculating the characteristic parameters capable of expressing the running working conditions of the vehicle;

the calculated resistances are defined as:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio, i ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobile; f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a represents a windward area; ρ represents the air density; v denotes the vehicle longitudinal speed.

33 Calculating working condition characteristic parameters, namely acceleration, as follows:

in the formula ,a theoretical longitudinal vehicle acceleration obtained by a vehicle speed versus time difference; i is road grade; g is gravity acceleration; f is the rolling resistance coefficient.

5. The method for estimating the quality and the gradient of the automobile based on the data confidence factor according to claim 1, wherein the specific method of the fourth step is as follows:

41 Establishing a quality estimation confidence factor model by adopting an error back propagation neural network algorithm, namely a BP algorithm;

the definition of the BP algorithm is as follows:

for a given training set d= { (x) ₁ ,y ₁ ),(x ₂ ,y ₂ ),...,(x _m ,y _m )}，I.e. the input is described by d-dimensional attributes and output as an l-dimensional real value vector; the neural network is a multi-layer feedforward network with d input neurons, l output neurons and q hidden layer neurons, wherein the threshold value of the jth neuron of the output layer is represented by theta _j Representing the threshold value gamma of the h neuron in the hidden layer _h A representation;

the connection weight between the ith neuron in the input layer and the h neuron in the hidden layer is v _ih The method comprises the steps of carrying out a first treatment on the surface of the The connection weight between the h neuron in the hidden layer and the j neuron in the output layer is w _hj The method comprises the steps of carrying out a first treatment on the surface of the The input received by the h neuron in the hidden layer isThe input received by the j-th neuron in the output layer is +.> wherein b_h The output of the h neuron in the hidden layer;

for training (x _k ,y _k ) The output of the neural network isNamely:

the neural network is then in (x _k ,y _k ) The mean square error is:

there are (d+l+1) q+l parameters in the neural network to be determined: d×q weights from the input layer to the hidden layer, q×l weights from the hidden layer to the output layer, q hidden layer neuron thresholds, and l output layer neuron thresholds;

the BP neural network is an iterative learning algorithm, and the parameters are updated and estimated by adopting a generalized perceptron learning rule in each iteration round, and the update estimation formula of any parameter v is as follows:

v←v+Δv

BP algorithm adjusts parameters in the negative gradient direction of the target based on gradient descent strategy, and adjusts error E _k The given learning rate η is:

w _hj first affecting the input value beta of the jth output layer neuron _j And then influence the output valueThen influence E _k There is

The learning rate eta epsilon (0, 1) controls the updating step length in each iteration of the algorithm;

42 For each training sample, the BP algorithm performs the following operations: firstly, providing input data to neurons of an input layer, and then forwarding training signals layer by layer until a result of an output layer is generated; and then calculating the error of the output layer, back-propagating the error to the hidden layer neuron, and finally adjusting the connection weight and the threshold value according to the error of the hidden neuron. The iterative process is looped until certain training stop conditions are reached;

wherein the tag value confidence coefficient in the dataset is defined as:

wherein m is a mass value estimated by a least square method; m is m ₀ The vehicle quality true value; confidence coefficient c _f ∈(0,1)；

Wherein, the threshold value algorithm of the confidence factor is defined as:

in the formula ,c_f For confidence coefficient, C _f Is a confidence factor;

43 Confidence factor C calculated by a threshold value algorithm _f Representing the rationality of the current working condition for quality estimation, and controlling an updating strategy of the vehicle during quality estimation under different working conditions through a confidence factor;

when C _f When the vehicle is 1, the running condition of the vehicle is very in accordance with the assumed condition of the longitudinal dynamics model of the vehicle, the vehicle is in an ideal linear region, the dynamics model-based estimation algorithm is better in performance, and the quality estimation value is more accurate;

when C _f When the estimated value is 0, the running condition of the vehicle is not in accordance with the assumed condition of the longitudinal dynamics model of the vehicle, the vehicle is in a complex nonlinear state, the estimated effect based on the dynamics model is inaccurate, the updating of the quality estimated value should be stopped, and the estimated value which is accurate at the last moment is used as the estimated value at the current moment.

6. The method for estimating the quality and gradient of the automobile based on the data confidence factor according to claim 1, wherein the specific method in the fifth step is as follows:

51 The working principle of the vehicle-mounted accelerometer is as follows:

wherein ,the vehicle theoretical longitudinal acceleration is obtained through vehicle speed difference; i is road grade; g is gravity acceleration;

52 According to the working principle of the vehicle-mounted accelerometer and the longitudinal dynamics model of the vehicle, the model decoupling of the vehicle mass and the road gradient is realized, and the decoupling model of the mass estimation is as follows:

in the formula ,T_tq The torque is output for the engine of the automobile; i.e _g Representing the transmission ratio of the gearbox; i.e ₀ Representing a final drive ratio; η (eta) _t Representing mechanical transmission efficiency; p is p _mc Indicating master cylinder pressure; k (k) _bf 、k _br Representing front and rear brake performance factors; r represents the effective rolling radius of the wheel; i _{Total (S)} Representing the total equivalent moment of inertia of the rotating mass; m represents the mass of the automobile; f represents a rolling resistance coefficient; c (C) _D Representing the air resistance coefficient of the vehicle; a tableShowing the windward area; ρ represents the air density; v denotes the vehicle longitudinal speed; a, a _{x_sensor} Measuring a vehicle acceleration for the vehicle accelerometer;

53 The output quantity y=θh of the recursive least squares method, the parameter θ=m to be identified, the vehicle mass M is estimated using the recursive least squares method with confidence factors according to the following equation:

in the formula ,C_f Is a confidence factor; λ is a time-varying forgetting factor; p is the recursive covariance matrix.

7. The method for estimating the quality and gradient of the automobile based on the data confidence factor according to claim 1, wherein the specific method in the sixth step is as follows:

judging whether the covariance matrix P in the recursive least square algorithm in the quality estimation model is smaller than a critical value P ₀ ；

If P is less than P ₀ The quality estimated value at the current moment tends to be converged, the quality estimated value is accurate, and is considered as an inherent parameter of the vehicle because the vehicle does not greatly change in the running process, and the estimated value is input into a gradient estimation model at the moment so as to perform gradient estimation;

if P is greater than or equal to P ₀ The quality estimated value at the current moment is rapidly updated, and the change of the quality estimated value is relatively large, so that if the quality estimated value is input into the gradient estimation model, the error of gradient estimation is relatively large, and the accuracy required by vehicle control cannot be achieved, and therefore the steps one to five should be continuously repeated until the quality estimated value tends to be stable and converged, and then the road gradient estimation is performed.

8. The method for estimating the quality and gradient of the automobile based on the data confidence factor according to claim 1, wherein the specific method in the step seven is as follows:

71 Obtaining a state space model of the system according to the vehicle dynamics model as follows:

and (3) making:c _i ＝mg、c _f0 ＝mgf ₀ 、c _f1 ＝3.6Mgf、

72 Obtaining a discrete state space equation of the system according to the state space model of the system as follows:

wherein v (k) is the longitudinal vehicle speed at the current moment; t (T) _tq (k) The torque is output for the engine at the current moment; p is p _mc (k) The master cylinder braking pressure at the current moment; i (k) is the road gradient at the current time;

73 The observation equation for the system is as follows:

74 Jacobian matrix for the system:

in the formula ,J_f Jacobian matrix as system state equation and deltat as discrete step length of system;

75 State transition matrix a=j of system _f ；

76 System observation matrix is

77 After the Jacobian matrix and the observation matrix of the system are obtained through calculation, a road gradient estimation model is established through a Kalman filtering algorithm, and the flow is as follows:

(1) Initializing state variablesAnd a posterior estimated bias covariance P _k-1 ；

(2) According to the formulaCalculating a priori estimated value of the state variable;

(3) According to the formulaCalculating a priori estimated deviation covariance;

(4) According to the formulaCalculating Kalman filtering gain;

(5) According to the formulaCalculating posterior estimation deviation covariance;

(6) According to the formulaA state variable posterior estimate is calculated.