CN105738817A - Battery charge state estimation method based on AEKF and estimation system - Google Patents

Abstract

The invention relates to the battery electrical testing technology, especially relates to a battery charge state estimation method based on AEKF and an estimation system. Battery SOC can be estimated by adopting the self-adaption expansion Kalman filtering algorithm, and the parameter self-adaption adjusting way of the Kalman filtering algorithm can be changed by additionally providing the weighting coefficient based on the forgetting factor, and then the influence of the parameter initial value setting on the whole algorithm is small, and the phenomena of the inaccurate battery SOC initial value calculated by adopting the original ampere-hour integral method and the accumulated error can be overcome, and in addition, the battery SOC can be estimated accurately and reliable. The battery charge state estimation method and the estimation system are advantageous in that the convergence performance is good, the convergence speed is fast, the algorithm transplantability is good, and the use stable and reliable; the estimation method and the estimation system can be used for the electric vehicle battery management field, and can be used for the SOC estimation of the electric vehicle storage battery, and therefore the endurance mileage of the electric vehicle can be calculated accurately, the control of the driver over the vehicle can be facilitated; the estimation method and the estimation system are more suitable for the electric vehicle environment having the strong current fluctuation.

Description

A kind of battery charge state method of estimation based on AEKF and estimating system

Technical field

The present invention relates to cell electrical measuring technology, particularly relate to a kind of battery charge state method of estimation based on AEKF and estimating system.

Background technology

At present, dump energy (StateofCharge, the SOC) method of estimation of battery is broadly divided into two big classes: direct method and indirect method.Direct method refers to that equipment directly measures the dump energy of battery by experiment；Indirect method, mainly through the physicochemical characteristic of inside battery, needs high-precision equipment in estimation procedure, is therefore difficulty with in practice.Ampere-hour integration method, open-circuit voltage method, internal resistance method etc. belong to indirect method, ampere-hour integration method can produce cumulative error in calculating process, causing that calculated SOC increases with the increase error of discharge and recharge time, the accuracy of the SOC initial value of ampere-hour integration method calculating simultaneously is difficult to determine；Open-circuit voltage method needs to stand for a long time to make inside battery voltage stabilization, is difficult to during battery dump energy in monitoring car running process；Internal resistance method also exists the difficulty of estimation internal resistance, is also difficult on hardware.Additionally, estimate battery SOC also by the indirect method such as artificial neural network algorithm, Kalman filtering algorithm, but neural network algorithm and Kalman filtering algorithm arrange difficulty due to its system, and in battery management system, application cost is high, provides no advantage against.

Summary of the invention

The technical problem to be solved is, for quality and the suitability of distinct methods, it is provided that a kind of battery charge state method of estimation based on AEKF and estimating system, to improve the accuracy of the battery SOC estimated.The present invention is achieved in that

A kind of battery charge state method of estimation based on AEKF, comprises the steps:

Step 1: initialize t₀The x in moment₀、P₀、Q₀、R₀, subsequently into step 2；Wherein x₀For battery charge state initial value, P₀For error covariance initial value, Q₀For process noise initial value, R₀For observation noise initial value；

Step 2: estimate the battery charge state in k momentAnd the state prior estimate error covariance in k momentSubsequently into step 3；Wherein:

Wherein, A is the transfer matrix in the sampling interval,For the posterior estimate of the battery charge state in k-1 moment, B is input matrix, u_k-1The input quantity of etching system, noise during for k-1_k-1For the white noise that the k-1 moment adds；

Wherein, P_k-1For the state estimation posteriori error covariance in k-1 moment, A^TFor the transposed matrix of transfer matrix A, Q_k-1Process noise for the k-1 moment；

Step 3: update the difference e between actual voltage signal and the model voltage signal in k moment_kWith Kalman filtering gain H_k, subsequently into step 4；Wherein:

Wherein, y_kFor the actual voltage signal of the battery that the k moment collects,For the model voltage signal of the battery model in k moment,Battery charge state priori estimates for the k moment；

Wherein, H_kFor the Kalman filtering gain matrix in k moment, R_kFor the observation noise in k moment, C is output matrix, C^TTransposition for output matrix C；

Step 4: update the weight coefficient based on forgetting factor, subsequently into step 5；d_k-1=(1-b) (1-b^k)^-1, wherein, b is forgetting factor, and d is the weight coefficient based on forgetting factor, d_k-1The weighting coefficient values based on forgetting factor for the k-1 moment；

Step 5: renewal process noise Q_kWith observation noise R_k, subsequently into step 6:

$Q_k=(1-d_{k-1})Q_{k-1}+d_{k-1}G(H_k{e}_k{e}_k^T{H}_k^T+P_k^{-}-{\mathrm{AP}}_k^{-}{A}^T)G^T;$

$R_k=(1-d_{k-1})R_{k-1}+d_{k-1}(e_k{e}_k^T-{\mathrm{CP}}_k^{-}{C}^T);$ Wherein:

G is white noise, Q_k-1For the process noise in k-1 moment, R_k-1For the observation noise in k-1 moment,For the difference e between actual voltage signal and the model voltage signal in k moment_kTransposition,Kalman filtering gain matrix H for the k moment_kTransposition；

Step 6: update the posterior estimate of the battery charge state in k momentBattery charge state Posterior estimator error covariance with the k momentSubsequently into step 7；I is unit matrix；

Step 7:k value increase by 1, and return step 1.

Further, the value of described b is 0.95.

A kind of battery charge state estimating system based on AEKF, including initialization module, estimate module, voltage difference and Kalman filtering gain more new module, based on the weight coefficient more new module of forgetting factor, process noise and observation noise more new module, state-of-charge posterior estimate and error covariance more new module, iteration module；Wherein:

Initialization module is used for initializing t₀The x in moment₀、P₀、Q₀、R₀, then branch to estimate module；Wherein x₀For battery charge state initial value, P₀For error covariance initial value, Q₀For process noise initial value, R₀For observation noise initial value；

Estimate module for estimating the battery charge state in k momentAnd the state prior estimate error covariance in k momentThen branch to voltage difference and Kalman filtering gain more new module；Wherein:

Wherein, A is the transfer matrix in the sampling interval,For the posterior estimate of the battery charge state in k-1 moment, B is input matrix, u_k-1The input quantity of etching system, noise during for k-1_k-1For the white noise that the k-1 moment adds；

Wherein, P_k-1For the state estimation posteriori error covariance in k-1 moment, A^TFor the transposed matrix of transfer matrix A, Q_k-1Process noise for the k-1 moment；

Voltage difference and Kalman filtering gain more new module are for updating the difference e between the actual voltage signal in k moment and model voltage signal_kWith Kalman filtering gain H_k, then branch to the more new module of the weight coefficient based on forgetting factor；Wherein:

Wherein, y_kFor the actual voltage signal of the battery that the k moment collects,For the model voltage signal of the battery model in k moment,Battery charge state priori estimates for the k moment；Wherein, H_kFor the Kalman filtering gain matrix in k moment, R_kFor the observation noise in k moment, C is output matrix, C^TTransposition for output matrix C；

Weight coefficient more new module based on forgetting factor is used for updating the weight coefficient based on forgetting factor, then branches to process noise and observation noise more new module；d_k-1=(1-b) (1-b^k)^-1, wherein, b is forgetting factor, and d is the weight coefficient based on forgetting factor, d_k-1The weighting coefficient values based on forgetting factor for the k-1 moment；

Process noise and observation noise more new module is used for renewal process noise Q_kWith observation noise R_k, then branch to state-of-charge posterior estimate and error covariance more new module thereof:

$Q_k=(1-d_{k-1})Q_{k-1}+d_{k-1}G(H_k{e}_k{e}_k^T{H}_k^T+P_k^{-}-{\mathrm{AP}}_k^{-}{A}^T)G^T;$

$R_k=(1-d_{k-1})R_{k-1}+d_{k-1}(e_k{e}_k^T-{\mathrm{CP}}_k^{-}{C}^T);$ Wherein:

G is white noise, Q_k-1For the process noise in k-1 moment, R_k-1For the observation noise in k-1 moment,For the difference e between actual voltage signal and the model voltage signal in k moment_kTransposition,Kalman filtering gain matrix H for the k moment_kTransposition；

State-of-charge posterior estimate and error covariance thereof more new module is for updating the posterior estimate of the battery charge state in k momentBattery charge state Posterior estimator error covariance with the k momentThen branch to iteration module；I is unit matrix；

Iteration module is for increasing by 1 by k value, and returns to initialization module.

Further, the value of described b is 0.95.

Compared with prior art, the present invention adopts adaptive extended kalman filtering algorithm to estimate battery SOC, and add the weight coefficient based on forgetting factor change Kalman filtering algorithm parameter adaptive adjust mode, promote the impact that whole algorithm is arranged by initial parameter value only small, overcome original ampere-hour integration method and calculate that battery SOC initial value is inaccurate and the phenomenon of cumulative errors, it is possible to more accurately and reliably estimate battery SOC.Meanwhile, convergence of the present invention is good, fast convergence rate, and algorithm transplantability is good, reliable and stable.Present invention can apply to batteries of electric automobile management domain, the SOC applying the present invention to accumulator of electric car estimates, the course continuation mileage of electric automobile can be calculated exactly, it is simple to driver's control to vehicle, be more suitable for the electric automobile environment that current fluctuation is violent.

Accompanying drawing explanation

Fig. 1: the battery charge state method of estimation schematic flow sheet based on AEKF that the embodiment of the present invention provides；

Fig. 2: the battery charge state estimating system structural representation based on AEKF that the embodiment of the present invention provides.

Detailed description of the invention

In order to make the purpose of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is further elaborated.

The invention provides a kind of battery charge state (StateofCharge, SOC) method of estimation based on AEKF (adaptive extended kalman filtering algorithm).The overall thought of this method is, first each estimation parameter is carried out initialization process, mainly includes t₀The Initialize installation of the SOC state in moment, covariance and noise matrix (process noise, observation noise), then process variable is updated, advance according to above-mentioned Kalman filtering algorithm stepping type, then the weight coefficient based on forgetting factor is determined, and then determine its forgetting factor, parameter in update algorithm, finally obtains SOC estimation.It is iterated by whole process repeatedly, constantly updates the SOC estimation obtaining optimum.

On the basis of above-mentioned overall thought, the battery charge state method of estimation based on AEKF provided by the invention comprises the steps:

Step 1: initialize t₀The x in moment₀、P₀、Q₀、R₀, subsequently into step 2；Wherein x₀For battery charge state initial value, P₀For error covariance initial value, Q₀For process noise initial value, R₀For observation noise initial value；

Step 2: estimate the battery charge state in k momentAnd the state prior estimate error covariance in k momentSubsequently into step 3；Wherein:

Wherein, A is the transfer matrix in the sampling interval,For the posterior estimate of the battery charge state in k-1 moment, B is input matrix, u_k-1The input quantity of etching system, noise during for k-1_k-1For the white noise that the k-1 moment adds；

Wherein, P_k-1For the state estimation posteriori error covariance in k-1 moment, A^TFor the transposed matrix of transfer matrix A, Q_k-1Process noise for the k-1 moment；

Step 3: update the difference e between actual voltage signal and the model voltage signal in k moment_kWith Kalman filtering gain H_k, subsequently into step 4；Wherein:

Wherein, y_kFor the actual voltage signal of the battery that the k moment collects,For the model voltage signal of the battery model in k moment,Battery charge state priori estimates for the k moment；

Wherein, H_kFor the Kalman filtering gain matrix in k moment, R_kFor the observation noise in k moment, C is output matrix, C^TTransposition for output matrix C；

Step 4: update the weight coefficient based on forgetting factor, subsequently into step 5；d_k-1=(1-b) (1-b^k)^-1, wherein, b is forgetting factor, and d is the weight coefficient based on forgetting factor, d_k-1The weighting coefficient values based on forgetting factor for the k-1 moment；

Step 5: renewal process noise Q_kWith observation noise R_k, subsequently into step 6:

$Q_k=(1-d_{k-1})Q_{k-1}+d_{k-1}G(H_k{e}_k{e}_k^T{H}_k^T+P_k^{-}-{\mathrm{AP}}_k^{-}{A}^T)G^T;$

$R_k=(1-d_{k-1})R_{k-1}+d_{k-1}(e_k{e}_k^T-{\mathrm{CP}}_k^{-}{C}^T);$ Wherein:

G is white noise, Q_k-1For the process noise in k-1 moment, R_k-1For the observation noise in k-1 moment,For the difference e between actual voltage signal and the model voltage signal in k moment_kTransposition,Kalman filtering gain matrix H for the k moment_kTransposition；

Step 6: update the posterior estimate of the battery charge state in k momentBattery charge state Posterior estimator error covariance with the k momentSubsequently into step 7；I is unit matrix；

Step 7:k value increase by 1, and return step 1.

Fig. 1 is the simple flow schematic diagram of the present embodiment method of estimation.In the present embodiment step 4, the value of b is set to 0.95.Whole SOC estimation procedure utilizes the weight coefficient based on forgetting factor that Kalman filtering algorithm stepping type is updated, and continues to optimize the SOC value of battery of more new estimation.Step 4 is the critically important step that the present invention is different from prior art.By increasing the weight coefficient based on forgetting factor, the parameter adaptive changing Kalman filtering algorithm adjusts mode, promote the impact that whole algorithm is arranged by initial parameter value less, can the SOC value of battery of optimal estimating better, comparing and utilize existing ampere-hour integration method to estimate battery SOC, the present invention is more reliable.Present invention can apply in the battery management of electric automobile (including pure electric automobile and hybrid vehicle), it is simple to driver accurately grasps the course continuation mileage of electric automobile in real time.

Based on above-mentioned battery charge state method of estimation, present invention also offers a kind of battery charge state estimating system based on AEKF.As in figure 2 it is shown, this system includes:

Initialization module 1, estimate module 2, voltage difference and Kalman filtering gain more new module 3, based on weight coefficient more new module 4, process noise and the observation noise more new module 5 of forgetting factor, state-of-charge posterior estimate and error covariance more new module 6, iteration module 7；Wherein:

Initialization module 1 is used for initializing t₀The x in moment₀、P₀、Q₀、R₀, then branch to estimate module 2；Wherein x₀For battery charge state initial value, P₀For error covariance initial value, Q₀For process noise initial value, R₀For observation noise initial value；

Estimate module 2 for estimating the battery charge state in k momentAnd the state prior estimate error covariance in k momentThen branch to voltage difference and Kalman filtering gain more new module 3；Wherein:

Wherein, A is the transfer matrix in the sampling interval,For the posterior estimate of the battery charge state in k-1 moment, B is input matrix, u_k-1The input quantity of etching system, noise during for k-1_k-1For the white noise that the k-1 moment adds；

Wherein, P_k-1For the state estimation posteriori error covariance in k-1 moment, A^TFor the transposed matrix of transfer matrix A, Q_k-1Process noise for the k-1 moment；

Voltage difference and Kalman filtering gain more new module 3 is for updating the difference e between the actual voltage signal in k moment and model voltage signal_kWith Kalman filtering gain H_k, then branch to more new module 4 of the weight coefficient based on forgetting factor；Wherein:

Wherein, y_kFor the actual voltage signal of the battery that the k moment collects,For the model voltage signal of the battery model in k moment,Battery charge state priori estimates for the k moment；

Wherein, H_kFor the Kalman filtering gain matrix in k moment, R_kFor the observation noise in k moment, C is output matrix, C^TTransposition for output matrix C；

Weight coefficient more new module 4 based on forgetting factor, for updating the weight coefficient based on forgetting factor, then branches to process noise and observation noise more new module 5；d_k-1=(1-b) (1-b^k)^-1, wherein, b is forgetting factor, and d is the weight coefficient based on forgetting factor, d_k-1The weighting coefficient values based on forgetting factor for the k-1 moment；

Process noise and observation noise more new module 5 is for renewal process noise Q_kWith observation noise R_k, then branch to state-of-charge posterior estimate and error covariance more new module 6 thereof:

$Q_k=(1-d_{k-1})Q_{k-1}+d_{k-1}G(H_k{e}_k{e}_k^T{H}_k^T+P_k^{-}-{\mathrm{AP}}_k^{-}{A}^T)G^T;$

$R_k=(1-d_{k-1})R_{k-1}+d_{k-1}(e_k{e}_k^T-{\mathrm{CP}}_k^{-}{C}^T);$ Wherein:

G is white noise, Q_k-1For the process noise in k-1 moment, R_k-1For the observation noise in k-1 moment,For the difference e between actual voltage signal and the model voltage signal in k moment_kTransposition,Kalman filtering gain matrix H for the k moment_kTransposition；

State-of-charge posterior estimate and error covariance thereof more new module 6 is for updating the posterior estimate of the battery charge state in k momentBattery charge state Posterior estimator error covariance with the k momentThen branch to iteration module 7；I is unit matrix；

Iteration module 7 is for increasing by 1 by k value, and returns to initialization module 1.

Wherein, the value of b is 0.95.

Each module in this estimating system and each step one_to_one corresponding in above-mentioned method of estimation, be referred to each step in above-mentioned method of estimation, no longer repeat one by one in this function to each module and operation principle.

The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all any amendment, equivalent replacement and improvement etc. made within the spirit and principles in the present invention, should be included within protection scope of the present invention.

Claims (4)

1. the battery charge state method of estimation based on AEKF, it is characterised in that comprise the steps:

Step 1: initialize t₀The x in moment₀、P₀、Q₀、R₀, subsequently into step 2；Wherein x₀For battery charge state initial value, P₀For error covariance initial value, Q₀For process noise initial value, R₀For observation noise initial value；

Step 2: estimate the battery charge state in k momentAnd the state prior estimate error covariance in k momentSubsequently into step 3；Wherein:

Wherein, A is the transfer matrix in the sampling interval,For the posterior estimate of the battery charge state in k-1 moment, B is input matrix, u_k-1The input quantity of etching system, noise during for k-1_k-1For the white noise that the k-1 moment adds；

Wherein, P_k-1For the state estimation posteriori error covariance in k-1 moment, A^TFor the transposed matrix of transfer matrix A, Q_k-1Process noise for the k-1 moment；

Step 3: update the difference e between actual voltage signal and the model voltage signal in k moment_kWith Kalman filtering gain H_k, subsequently into step 4；Wherein:

Wherein, y_kFor the actual voltage signal of the battery that the k moment collects,For the model voltage signal of the battery model in k moment,Battery charge state priori estimates for the k moment；

Wherein, H_kFor the Kalman filtering gain matrix in k moment, R_kFor the observation noise in k moment, C is output matrix, C^TTransposition for output matrix C；

Step 4: update the weight coefficient based on forgetting factor, subsequently into step 5；d_k-1=(1-b) (1-b^k)^-1, wherein, b is forgetting factor, and d is the weight coefficient based on forgetting factor, d_k-1The weighting coefficient values based on forgetting factor for the k-1 moment；

Step 5: renewal process noise Q_kWith observation noise R_k, subsequently into step 6:

$Q_k=(1-d_{k-1})Q_{k-1}+d_{k-1}G(H_k{e}_k{e}_k^T{H}_k^T+P_k^{-}-{\mathrm{AP}}_k^{-}{A}^T)G^T;$

$R_k=(1-d_{k-1})R_{k-1}+d_{k-1}(e_k{e}_k^T-{\mathrm{CP}}_k^{-}{C}^T);$ Wherein:

G is white noise, Q_k-1For the process noise in k-1 moment, R_k-1For the observation noise in k-1 moment,For the difference e between actual voltage signal and the model voltage signal in k moment_kTransposition,Kalman filtering gain matrix H for the k moment_kTransposition；

Step 6: update the posterior estimate of the battery charge state in k momentBattery charge state Posterior estimator error covariance with the k momentSubsequently into step 7；I is unit matrix；

Step 7:k value increase by 1, and return step 1.

2. the battery charge state method of estimation based on AEKF as claimed in claim 1, it is characterised in that the value of described b is 0.95.

3. the battery charge state estimating system based on AEKF, it is characterized in that, including initialization module, estimate module, voltage difference and Kalman filtering gain more new module, based on the weight coefficient more new module of forgetting factor, process noise and observation noise more new module, state-of-charge posterior estimate and error covariance more new module, iteration module；Wherein:

Initialization module is used for initializing t₀The x in moment₀、P₀、Q₀、R₀, then branch to estimate module；Wherein x₀For battery charge state initial value, P₀For error covariance initial value, Q₀For process noise initial value, R₀For observation noise initial value；

Estimate module for estimating the battery charge state in k momentAnd the state prior estimate error covariance in k momentThen branch to voltage difference and Kalman filtering gain more new module；Wherein:

Wherein, A is the transfer matrix in the sampling interval,For the posterior estimate of the battery charge state in k-1 moment, B is input matrix, u_k-1The input quantity of etching system, noise during for k-1_k-1For the white noise that the k-1 moment adds；

Wherein, P_k-1For the state estimation posteriori error covariance in k-1 moment, A^TFor the transposed matrix of transfer matrix A, Q_k-1Process noise for the k-1 moment；

Voltage difference and Kalman filtering gain more new module are for updating the difference e between the actual voltage signal in k moment and model voltage signal_kWith Kalman filtering gain H_k, then branch to the more new module of the weight coefficient based on forgetting factor；Wherein:

Wherein, y_kFor the actual voltage signal of the battery that the k moment collects,For the model voltage signal of the battery model in k moment,Battery charge state priori estimates for the k moment；

Wherein, H_kFor the Kalman filtering gain matrix in k moment, R_kFor the observation noise in k moment, C is output matrix, C^TTransposition for output matrix C；

Weight coefficient more new module based on forgetting factor is used for updating the weight coefficient based on forgetting factor, then branches to process noise and observation noise more new module；d_k-1=(1-b) (1-b^k)^-1, wherein, b is forgetting factor, and d is the weight coefficient based on forgetting factor, d_k-1The weighting coefficient values based on forgetting factor for the k-1 moment；

Process noise and observation noise more new module is used for renewal process noise Q_kWith observation noise R_k, then branch to state-of-charge posterior estimate and error covariance more new module thereof:

$Q_k=(1-d_{k-1})Q_{k-1}+d_{k-1}G(H_k{e}_k{e}_k^T{H}_k^T+P_k^{-}-{\mathrm{AP}}_k^{-}{A}^T)G^T;$

$R_k=(1-d_{k-1})R_{k-1}+d_{k-1}(e_k{e}_k^T-{\mathrm{CP}}_k^{-}{C}^T);$ Wherein:

G is white noise, Q_k-1For the process noise in k-1 moment, R_k-1For the observation noise in k-1 moment,For the difference e between actual voltage signal and the model voltage signal in k moment_kTransposition,Kalman filtering gain matrix H for the k moment_kTransposition；

State-of-charge posterior estimate and error covariance thereof more new module is for updating the posterior estimate of the battery charge state in k momentBattery charge state Posterior estimator error covariance with the k momentThen branch to iteration module；I is unit matrix；

Iteration module is for increasing by 1 by k value, and returns to initialization module.

4. the battery charge state estimating system based on AEKF as claimed in claim 3, it is characterised in that the value of described b is 0.95.