CN105808914A - Method and device for predicting life of satellite lithium ion battery - Google Patents

Abstract

The invention discloses a method and device for predicting the life of a satellite lithium ion battery. The method comprises the following steps: analyzing the characteristics of cycle life test data of the satellite lithium ion battery so as to obtain a fault evolution characteristic quantity; carrying out de-relax effect on capacity degradation data corresponding to the fault evolution characteristic quantity; carrying out least squares support vector machine LSSVM unsupervised learning training according to the capacity degradation data after the de-relax effect so as to complete the construction of an LSSVM model for life prediction; predicting the battery capacities of the battery in different periods through the LSSVM model, and carrying out battery failure threshold value extrapolation according to the battery capacities so as to realize the real-time prediction of the remaining useful life RUL of the lithium battery. According to the method provided by the invention, the LSSVM model for life prediction is constructed after de-relax effect is carried out on the capacity degradation data, and then life prediction is carried out through the model, so that the prediction result is correct and the problem that the process of measuring the life of the lithium ion battery is not correct is solved.

Description

The Forecasting Methodology in a kind of satellite lithium ion battery life-span and device

Technical field

The present invention relates to field of lithium ion battery, particularly relate to Forecasting Methodology and the device in a kind of satellite lithium ion battery life-span.

Background technology

Lithium ion battery is as novel storage battery, type accumulator has greater advantage than ever, to aerospace equipment such as energy storage power supply electrical property, occasion such as low earth-orbit satellite (LEO) that reliability requirement is higher, geo-synchronous orbit satellite (GEO), space stations, lithium-ions battery group will become first choice.

But, in prior art, there is more inaccurate part in the process measuring the lithium ion battery life-span, causes that the lithium ion life error obtained is relatively big, the impact accurate evaluation to the lithium ion battery life-span.

Summary of the invention

The present invention provides Forecasting Methodology and the device in a kind of satellite lithium ion battery life-span, in order to solve in prior art, there is more inaccurate part in the process measuring the lithium ion battery life-span, causes that the lithium ion life error obtained is relatively big, the problem affecting the accurate evaluation to the lithium ion battery life-span.

On the one hand, the present invention provides the Forecasting Methodology in a kind of satellite lithium ion battery life-span, including: analyze the feature of satellite cycle life of lithium ion battery test data, to obtain fault Characteristics of Evolution amount；Go to loosen effect to the degradation in capacity data that described fault Characteristics of Evolution amount is corresponding；According to the described degradation in capacity data gone after loosening effect, carry out the training of least square method supporting vector machine LSSVM unsupervised learning, to complete the LSSVM model construction of bimetry；Predict the battery battery capacity at different cycles by LSSVM model, carry out battery failure threshold values extrapolation according to battery capacity, it is achieved the real-time estimate to lithium battery residual life RUL.

Further, LSSVM model formation is:Wherein, k (x, x_i) it is kernel function, α_iFor Lagrange factor, b is bias term, and x is the vector of all battery capacities composition, x_iFor battery capacity original value.

Further, kernel function k (x, x_i) at least include one below: Polynomial kernel function k (x, x_i)=[1+ (x x_i)]^q；Sigmoid kernel function k (x, x_i)=tanh (ν_k(x·x_i)+c_k)；RBF k (x, x_i)=exp (-| x-x_i|²/2σ²)；Wherein, x is the vector of all battery capacities composition, x_iFor battery capacity original value, ν_k(x·x_i) for displacement parameter, work as v_k> 0 time, its be input data an amplitude adjusted parameter, c_kBeing a displacement parameter controlling to map threshold value, σ is the width of kernel function.

Further, after completing the LSSVM model construction of bimetry, also include: adopting mean square error MSE and squared correlation coefficient SCC to carry out model accuracy checking, checking formula is as follows:

$\mathrm{MSE}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-y_i)}^2,$ $\mathrm{SCC}={\left[\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2}}\right]}^2,$ Wherein, x_iFor battery capacity original value, y_iFor battery capacity prediction value, x is battery capacity meansigma methods, and y is battery capacity prediction meansigma methods, and N is natural number.

Further, battery failure threshold values extrapolation is carried out according to battery capacity, it is achieved after the real-time estimate to lithium battery residual life RUL, also include:

With actual measured value, the RUL of prediction is carried out error analysis, and error analysis RPE formula is as follows:

Wherein C_rFor actual measured value, C_pRUL for prediction.

On the other hand, the present invention also provides for the prediction unit in a kind of satellite lithium ion battery life-span, including: analysis module, for analyzing the feature of satellite cycle life of lithium ion battery test data, to obtain fault Characteristics of Evolution amount；Processing module, for going to loosen effect to the degradation in capacity data that described fault Characteristics of Evolution amount is corresponding；Build module, described in basis, go the degradation in capacity data after loosening effect, carry out the training of least square method supporting vector machine LSSVM unsupervised learning, to complete the LSSVM model construction of bimetry；Prediction module, for predicting the battery battery capacity at different cycles by LSSVM model, carries out battery failure threshold values extrapolation according to battery capacity, it is achieved the real-time estimate to lithium battery residual life RUL.

Further, the LSSVM model formation of described structure module construction is:Wherein, k (x, x_i) it is kernel function, α_iFor Lagrange factor, b is bias term, and x is the vector of all battery capacities composition, x_iFor battery capacity original value.

Further, described structure module build LSSVM model time, kernel function k (x, x_i) at least include one below: Polynomial kernel function k (x, x_i)=[1+ (x x_i)]^q；Sigmoid kernel function k (x, x_i)=tanh (ν_k(x·x_i)+c_k)；RBF k (x, x_i)=exp (-| x-x_i|²/2σ²)；Wherein, x is the vector of all battery capacities composition, x_iFor battery capacity original value, ν_k(x·x_i) for displacement parameter, work as v_k> 0 time, its be input data an amplitude adjusted parameter, c_kBeing a displacement parameter controlling to map threshold value, σ is the width of kernel function.

Further, also including: authentication module, be used for adopting mean square error MSE and squared correlation coefficient SCC to carry out model accuracy checking, checking formula is as follows:

$\mathrm{MSE}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-y_i)}^2,$ $\mathrm{SCC}={\left[\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2}}\right]}^2,$ Wherein, x_iFor battery capacity original value, y_iFor battery capacity prediction value, x is battery capacity meansigma methods, and y is battery capacity prediction meansigma methods, and N is natural number.

Further, also including: error analysis module, for the RUL of prediction is carried out error analysis with actual measured value, error analysis RPE formula is as follows:Wherein C_rFor actual measured value, C_pRUL for prediction.

Forecasting Methodology provided by the invention is after going to loosen effect to degradation in capacity data, build the LSSVM model of bimetry, the prediction in life-span is carried out again through this model, predict the outcome accurately, solve the process measuring the lithium ion battery life-span and there is more inaccurate part, cause that the lithium ion life error obtained is relatively big, the problem affecting the accurate evaluation to the lithium ion battery life-span.

Accompanying drawing explanation

Fig. 1 is the flow chart of the Forecasting Methodology in embodiment of the present invention Satellite lithium ion battery life-span；

Fig. 2 is the prediction unit structural representation in embodiment of the present invention Satellite lithium ion battery life-span；

Fig. 3 is the prediction unit preferred structure schematic diagram in embodiment of the present invention Satellite lithium ion battery life-span；

Fig. 4 is the flow chart of cycle life of lithium ion battery Forecasting Methodology in the preferred embodiment of the present invention；

Fig. 5 utilizes LSSVM to build Cycle life prediction model schematic in the preferred embodiment of the present invention；

Fig. 6 is lithium ion battery original battery capacity decline data and curves figure in the preferred embodiment of the present invention；

Fig. 7 removes to loosen battery capacity decline data and curves figure after effect in the preferred embodiment of the present invention；

Fig. 8 is for training No. 5 battery model checking curve charts one of LSSVM model in the preferred embodiment of the present invention；

Fig. 9 is for training No. 5 battery model checking curve charts two of LSSVM model in the preferred embodiment of the present invention；

Figure 10 is the #5 battery Cycle life prediction curve chart when 40 cycle in the preferred embodiment of the present invention；

Figure 11 is the #6 battery Cycle life prediction curve chart when 40 cycle in the preferred embodiment of the present invention；

Figure 12 is the #5 battery Cycle life prediction curve chart when 70 cycle in the preferred embodiment of the present invention；

Figure 13 is the #6 battery Cycle life prediction curve chart when 70 cycle in the preferred embodiment of the present invention；

Figure 14 is #5 and the #6 battery Cycle life prediction error analysis figure when 40 cycle in the preferred embodiment of the present invention；

Figure 15 is #5 and the #6 battery Cycle life prediction error analysis figure when 70 cycle in the preferred embodiment of the present invention.

Detailed description of the invention

In order to solve in prior art, there is more inaccurate part in the process measuring the lithium ion battery life-span, cause that the lithium ion life error obtained is bigger, the problem affecting the accurate evaluation to the lithium ion battery life-span, the invention provides the Forecasting Methodology in a kind of satellite lithium ion battery life-span and device, below in conjunction with accompanying drawing and embodiment, the present invention is further elaborated.Should be appreciated that specific embodiment described herein is only in order to explain the present invention, does not limit the present invention.

Embodiments providing the Forecasting Methodology in a kind of satellite lithium ion battery life-span, the flow process of the method is as it is shown in figure 1, include S101 to S104:

S101, analyzes the feature of satellite cycle life of lithium ion battery test data, to obtain fault Characteristics of Evolution amount；

The degradation in capacity data that fault Characteristics of Evolution amount is corresponding are gone to loosen effect by S102；

S103, according to going the degradation in capacity data after loosening effect, carries out the training of least square method supporting vector machine LSSVM unsupervised learning, to complete the LSSVM model construction of bimetry；

S104, predicts the battery battery capacity at different cycles by LSSVM model, carries out battery failure threshold values extrapolation according to battery capacity, it is achieved the real-time estimate to lithium battery residual life RUL.

The Forecasting Methodology that the embodiment of the present invention provides is after going to loosen effect to degradation in capacity data, build the LSSVM model of bimetry, the prediction in life-span is carried out again through this model, predict the outcome accurately, solve the process measuring the lithium ion battery life-span and there is more inaccurate part, cause that the lithium ion life error obtained is relatively big, the problem affecting the accurate evaluation to the lithium ion battery life-span.

In realizing process, LSSVM model formation is:Wherein, k (x, x_i) it is kernel function, α_iFor Lagrange factor, b is bias term, and x is the vector of all battery capacities composition, x_iFor battery capacity original value.Wherein, kernel function k (x, x_i) at least include one below:

Polynomial kernel function k (x, x_i)=[1+ (x x_i)]^q；Sigmoid kernel function k (x, x_i)=tanh (ν_k(x·x_i)+c_k)；RBF k (x, x_i)=exp (-| x-x_i|²/2σ²)；Wherein, x is the vector of all battery capacities composition, x_iFor battery capacity original value, ν_k(x·x_i) for displacement parameter, work as v_k> 0 time, its be input data an amplitude adjusted parameter, c_kBeing a displacement parameter controlling to map threshold value, σ is the width of kernel function.

After the LSSVM model construction completing bimetry, it is also possible to the accuracy of checking model, then mean square error (MSE) and squared correlation coefficient (SCC) can being adopted to carry out model accuracy checking, checking formula is as follows:

$\mathrm{MSE}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-y_i)}^2,$ $\mathrm{SCC}={\left[\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2}}\right]}^2,$

Wherein, x_iFor battery capacity original value, y_iFor battery capacity prediction value,For battery capacity meansigma methods,For battery capacity prediction meansigma methods, N is natural number.

Further, battery failure threshold values extrapolation is being carried out according to battery capacity, it is achieved after the real-time estimate to lithium battery residual life RUL, it is also possible to actual measured value, the RUL of prediction is carried out error analysis, and error analysis RPE formula is as follows:Wherein C_rFor actual measured value, C_pRUL for prediction.

The embodiment of the present invention additionally provides the prediction unit in a kind of satellite lithium ion battery life-span, its structural representation is as shown in Figure 2, including what couple successively: analysis module 10, for analyzing the feature of satellite cycle life of lithium ion battery test data, to obtain fault Characteristics of Evolution amount；Processing module 20, for going to loosen effect to the degradation in capacity data that described fault Characteristics of Evolution amount is corresponding；Build module 30, described in basis, go the degradation in capacity data after loosening effect, carry out the training of least square method supporting vector machine LSSVM unsupervised learning, to complete the LSSVM model construction of bimetry；Prediction module 40, for predicting the battery battery capacity at different cycles by LSSVM model, carries out battery failure threshold values extrapolation according to battery capacity, it is achieved the real-time estimate to lithium battery residual life RUL.

Wherein, the LSSVM model formation of described structure module construction is:Wherein, k (x, x_i) it is kernel function, α_iFor Lagrange factor, b is bias term, and x is the vector of all battery capacities composition, x_iFor battery capacity original value.The kernel function of LSSVM is chosen, and the Nonlinear Modeling of LSSVM depends on feature space method and kernel function skill.Kernel function thought essentially consists in and makes derivation carry out in the input space, rather than carries out at feature space, ensures that the two has the corresponding relation determined in theory simultaneously.Selecting different kernel functions, it will form different algorithms, data can be mapped to different spaces, it is meant that takes different standards that similarity core similarity degree is evaluated.

In LSSVM algorithm, different kernel functions will form different algorithms, and that uses relatively more mainly has three classes: Polynomial kernel function k (x, x_i)=[1+ (x x_i)]^q, for q rank Polynomial kernel function；Sigmoid kernel function k (x, x_i)=tanh (ν_k(x·x_i)+c_k), LSSVM algorithm at this moment contains the multilayer perceptron of a hidden layer, hidden layer node is automatically determined by algorithm, and the problem that algorithm is absent from perplexing the local minimum point of neutral net；RBF k (x, x_i)=exp (-| x-x_i|²/2σ²), wherein owing to Radial basis kernel function characteristic of correspondence space is infinite dimensional, limited sample is linear separability certainly in this feature space, therefore adopts the kernel function of radially base core；Wherein, x is the vector of all battery capacities composition, x_iFor battery capacity original value, ν_k(x·x_i) for displacement parameter, work as v_k> 0 time, its be input data an amplitude adjusted parameter, c_kBeing a displacement parameter controlling to map threshold value, σ is the width of kernel function.

The parameter of LSSVM model is chosen in process, and two in LSSVM are positive parameter σ²Being unknown with γ, σ is the width of kernel function, and value is crossed senior general and made model Premature Convergence, does not reach the purpose of prediction, and γ is mistake penalty factor, and value can make more greatly the better of number of training evidence and test result matching, namely improves measuring accuracy.Nuclear parameter σ²Determined by k mono-folding cross validation method.K mono-folding cross-validation method concretely comprises the following steps: 1. by (γ, σ²) training sample set be randomly assigned to k mutually disjoint subset, being substantially equal to the magnitudes of each folding.2. utilize k-1 training subset, one group of given parameter is set up regression model, utilize the mean square deviation MSE of last subset remaining to assess the performance of parameter.Repeating k time according to above procedure, therefore each subset has the opportunity to test, and 3. estimates to expect extensive error according to the meansigma methods of the mean square deviation obtained after k iteration, finally selects parameter one group optimum.

Said apparatus can also as it is shown on figure 3, include: authentication module 50, couples with structure module 30 and prediction module 40, is used for adopting mean square error MSE and squared correlation coefficient SCC to carry out model accuracy checking, and checking formula is as follows:

$\mathrm{MSE}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-y_i)}^2,$ $\mathrm{SCC}={\left[\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2}}\right]}^2;$

Error analysis module 60, couples with prediction module 40, and for the RUL of prediction is carried out error analysis with actual measured value, error analysis RPE formula is as follows:Wherein, x_iFor battery capacity original value, y_iFor battery capacity prediction value,For battery capacity meansigma methods,For battery capacity prediction meansigma methods, N is natural number, C_rFor actual measured value, C_pRUL for prediction.

Preferred embodiment

The embodiment of the present invention is the satellite lithium ion battery Forecasting Methodology of research NCA/C battery system, lays the foundation for realizing satellite novel third generation high energy energy storage power supply efficient application in-orbit in future, accurately management.

The embodiment of the present invention proposes a kind of satellite lithium ion battery life-span (cycle life) Forecasting Methodology based on degraded data and least square method supporting vector machine (LSSVM), purpose is to solve in existing cycle life of lithium ion battery prediction process, the accurate Forecasting Methodology of neither one, causes that true problem do not calculated accurately by coulant meter.The embodiment of the present invention is achieved by the following scheme:

1) feature of cycle life of lithium ion battery test data is analyzed, it is thus achieved that fault Characteristics of Evolution amount.

2) owing to the measurement (EIS) of battery capacity is to carry out (this time is called relaxation time) after placement a period of time after charging process, the battery capacity that can cause next cycle is uprushed, and the uncertain prediction that also can affect remaining battery life of the time placed.Thus, for satellite accumulator failure Characteristics of Evolution amount capacity, it is necessary to go to loosen effect to the degradation in capacity data of battery.

3) the degradation in capacity data after loosening effect are gone in utilization, carry out LSSVM unsupervised learning training, and complete model construction.

4) utilize LSSVM to build model, it was predicted that battery battery capacity distribution after different cycles, realize the real-time estimate to lithium battery residual life (RUL) according to battery failure threshold values Extrapolating model.

5) RUL of prediction is carried out error analysis with the actual measurement life-span.

The least square method supporting vector machine method that the present invention is above-mentioned, is by the inequality constraints condition of support vector classification, changes under the constraints of equation, then passes through the solution that solution is a series of system of linear equations solving dual problem class.So, least square method supporting vector machine has only to solve a system of linear equations just can obtain classifying face, simply many compared with the support vector machine demand quadratic programming problem of standard, and amount of storage is reduced, and training speed is improved.The advantage of the least square method supporting vector machine (LSSVM) adopted is in that small-sample learning ability and generalization ability are strong, and its derivation is as follows:

LSSVM adds error sum of squares item in the object function of standard SVM.For nonlinear system, it is considered to nonlinear solshing:

F (x)=ω^T·φ(x_k)+b(1)

Wherein: x ∈ Rⁿ, y ∈ R, ω represent weight vector, nonlinear functionThe input space is mapped as high-dimensional feature space, and then LSSVM can be defined as follows optimization problem, and γ is regularization parameter:

$\underset{\mathrm{\ω},b,e}{\min }J(\mathrm{\ω},\mathrm{\ξ})=\frac{1}{2}{\mathrm{\ω}}^T\mathrm{\ω}+\frac{1}{2}\mathrm{\γ}\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\ξ}}_k^2---\left(2\right)$

s.t.y_k=ω^T·φ(x_k)+b+e_k, k=1 ..., N (3)

It is re-introduced into Lagrange function, constrained optimization problems is converted into unconstrained optimization:

$L(\mathrm{\ω},b,\mathrm{\ξ},\mathrm{\α})=J(\mathrm{\ω},\mathrm{\ξ})-\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_k[{\mathrm{\ω}}^T\·\mathrm{\φ}\left(x_k\right)+b+\mathrm{\ξ}-y_k]---\left(4\right)$

Further according to KKT condition, formula (4) is sought local derviation:

$\left\{\begin{array}{c}\frac{\∂L}{\∂\mathrm{\ω}}=0\→\mathrm{\ω}=\underset{K=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_K\mathrm{\φ}\left(x_k\right)\\ \frac{\∂L}{\∂b}=0\→\underset{K=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_K=0\\ \frac{\∂L}{\∂\mathrm{\ξ}}=0\→{\mathrm{\α}}_K=\mathrm{\γ}{\mathrm{\ξ}}_k\\ \frac{\∂L}{\∂{\mathrm{\α}}_k}=0\→{\mathrm{\ω}}^T\·\mathrm{\φ}\left(x_k\right)+b+{\mathrm{\ξ}}_k-y_k\end{array}\right.---\left(5\right)$

Eliminate ω and ξ, solving equation (5):

$\left[\begin{array}{cc}0& 1^T\\ 1& K+{\mathrm{\γ}}^{-1}I\end{array}\right]\left[\begin{array}{c}b\\ a\end{array}\right]=\left[\begin{array}{c}0\\ Y\end{array}\right]---\left(6\right)$

With least-squares calculation a and b, obtaining LSSVM model is:

$f\left(x\right)=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{\α}}_{k}k(x,x_k)+b---\left(7\right)$

K in above-mentioned formula (7) is equal to i, simply there is difference on form of presentation.

The flow process of the cycle life of lithium ion battery Forecasting Methodology based on degraded data and LSSVM of the embodiment of the present invention is as shown in Figure 4.First, go to loosen effect to cell degradation data, next degraded data sample being utilized respectively different cycles carries out LSSVM unsupervised learning training and sets up corresponding model, then battery battery capacity (distribution) after different cycles is predicted respectively, foundation battery failure threshold values Extrapolating model realizes the real-time estimate to lithium battery residual life (RUL), and the RUL finally prediction obtained carries out error analysis with the actual measurement life-span and compares with standard SVM.

Compared with prior art, the present invention has the beneficial effect that:

The present invention provides a kind of cycle life of lithium ion battery Forecasting Methodology based on degraded data and LSSVM, degraded data goes loosen effect process, then LSSVM is adopted to train cell degradation data, follow the tracks of degradation trend, advantage is in that small-sample learning ability and generalization ability are strong, when not providing lithium battery capacity degenerate distribution function, LSSVM is utilized to train degraded data to set up model.Realize RUL real-time estimate according to failure threshold, and carry out error analysis and the SVM with standard compares.Below in conjunction with example, said process is described in detail.

Embodiment 1

Embodiments provide and how to utilize LSSVM to build Cycle life prediction model, see Fig. 5, implement step as follows.

The first step: select suitable degraded data as training sample, based on RBF superior performance in LSSVM regression forecasting is applied, chooses the RBF LSSVM kernel function as this programme.

Second step: initialization model, namely utilizes the core width cs initializing Radial basis kernel function and penalty factor γ.

3rd step: utilize LSSVM cross validation function tunelssvm Optimization Solution core width cs and penalty factor γ.

4th step: core width cs and penalty coefficient γ are substituted into LSSVM training pattern, after utilizing trainlssvm order solving model parameter and b value, model is set up in training.

5th step: checking model.The accuracy of Definition Model is carried out with mean square error (MSE) and squared correlation coefficient (SCC).Wherein, MSE is more little, and SCC is closer to 1, and model-fitting degree is more high.

$\mathrm{MSE}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-y_i)}^2---\left(8\right)$

$\mathrm{SCC}={\left[\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2}}\right]}^2---\left(9\right)$

Wherein, x_iFor battery capacity original value, y_iFor battery capacity prediction value.

6th step: carry out RUL prediction according to inefficacy threshold values, and carry out corresponding error analysis.

Embodiment 2

The embodiment of the present invention is to the described checking setting up cycle life of lithium ion battery forecast model based on degraded data and LSSVM.In the present embodiment, adopt the lithium ion battery test data in NASAPCoE research center to verify, and verification experimental verification result has been carried out relative analysis.Choose #05, #06 and the #07 battery capacity decline test data that test data is concentrated, and using 70% battery capacity (dropping to about 1.4Ah from 2Ah) as the threshold value in lithium ion battery life-span.

First, the measurement (EIS) of battery capacity can cause that capacity occurs loosening effect, and curve spike as shown in Figure 6 is loose effect, and capacity data is gone to loosen effect by the mode being taken based on Mathematical Morphology, as shown in Figure 7.

Then, the kernel function of Rational choice LSSVM model and model parameter.The expression formula of LSSVM model is:Wherein, k (x, x_i) it is kernel function, α_iFor Lagrange factor, b is bias term.

Set up the crucial optimum option first choosing suitable kernel function and kernel functional parameter of accurate regression model and include core width cs and penalty factor γ, then pass through the α in the accurate calculating formula of LSSVM training algorithm (10)_iWith b value, set up corresponding model.Using 50 cycles before #5 and before #5 the degraded data in 100 cycles as LSSVM training sample and carry out regression forecasting, result is as shown in Figure 8 and Figure 9.

Calculating mean-square value MSE now and squared correlation coefficient SCC as shown in table 1, table 1 represents MSE and the SCC of the model of #5.

Table 1

	MSE	SCC
			Front 50 cycles	6.649921e-006	0.994767
Front 100 cycles	8.664163e-006	0.999164

Can being drawn by table 1, now the MSE of model is only small, and SCC trends towards 1, good by the fitting effect of the visible model of Fig. 8 and Fig. 9, so utilizing LSSVM can set up more accurate degradation model.

Embodiment 3

40 cycles before being utilized respectively #5 and #6 described in the embodiment of the present invention and the degraded data in front 70 cycles are set up on the basis of model 1 and model 2, predict that 60 cycles and model 2 predicted for 30 cycles backward backward at corresponding model 1 respectively, then carry out biometry according to the threshold values (70%) of battery capacity.

The SIMLSSVM order in LSSVM algorithm is utilized to carry out RUL prediction.Gap between bimetry and actual life is exactly forecast error, same with next by mean-square value (MSE) and squared correlation coefficient (SCC), also increase relative error (RPE) simultaneously and carry out linear expression error, as shown in formula (11).

$\mathrm{RPE}=\left|\frac{C_r-C_p}{C_p}\right|\×100\%---\left(11\right)$

Wherein, C_rFor actual measured value, C_pFor predictive value.

When 40 cycle, making prediction result is as shown in Figure 10 and Figure 11, and when 70 cycle, making prediction result is as shown in Figures 12 and 13, and when 40 cycle, the relative error of making prediction is as shown in figure 14, and the relative error of making prediction is as shown in figure 15 when 70 cycle for #6.

MSE and SCC and relative error (RPE) that RUL predicts the outcome are as shown in table 2, and table 2 represents #5 and the #6 error analysis predicted the outcome.

Table 2

Be can be seen that the relative error of prediction was less than for 40 cycles when 70 cycle by Figure 14 and Figure 15, and along with the continuous increase of predetermined period number, relative error can be gradually increased.Can being found out that the MSE of prediction was less than for 40 cycles when 70 cycle quantitatively by table 1 and table 2, SCC is also the closer in 1.Because relative to 40 cycles, having more training sample during 70 cycle, energy is matching degradation trend better, it was predicted that precision increases.

Although being example purpose, having been disclosed for the preferred embodiments of the present invention, it is also possible for those skilled in the art will recognize various improvement, increase and replacement, and therefore, the scope of the present invention should be not limited to above-described embodiment.

Claims (10)

1. the Forecasting Methodology in a satellite lithium ion battery life-span, it is characterised in that including:

Analyze the feature of satellite cycle life of lithium ion battery test data, to obtain fault Characteristics of Evolution amount；

Go to loosen effect to the degradation in capacity data that described fault Characteristics of Evolution amount is corresponding；

According to the described degradation in capacity data gone after loosening effect, carry out the training of least square method supporting vector machine LSSVM unsupervised learning, to complete the LSSVM model construction of bimetry；

Predict the battery battery capacity at different cycles by LSSVM model, carry out battery failure threshold values extrapolation according to battery capacity, it is achieved the real-time estimate to lithium battery residual life RUL.

2. the method for claim 1, it is characterised in that LSSVM model formation is:

Wherein, k (x, x_i) it is kernel function, α_iFor Lagrange factor, b is bias term, and x is the vector of all battery capacities composition, x_iFor battery capacity original value.

3. method as claimed in claim 2, it is characterised in that kernel function k (x, x_i) at least include one below:

Polynomial kernel function k (x, x_i)=[1+ (x x_i)]^q；

Sigmoid kernel function k (x, x_i)=tanh (ν_k(x·x_i)+c_k)；

RBF k (x, x_i)=exp (-| x-x_i|²/2σ²)；

Wherein, x is the vector of all battery capacities composition, x_iFor battery capacity original value, ν_k(x·x_i) for displacement parameter, work as v_k> 0 time, its be input data an amplitude adjusted parameter, c_kBeing a displacement parameter controlling to map threshold value, σ is the width of kernel function.

4. the method for claim 1, it is characterised in that after completing the LSSVM model construction of bimetry, also include:

Adopting mean square error MSE and squared correlation coefficient SCC to carry out model accuracy checking, checking formula is as follows:

$\mathrm{MSE}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-y_i)}^2,\mathrm{SCC}={\left[\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2}}\right]}^2,$

Wherein, x_iFor battery capacity original value, y_iFor battery capacity prediction value,For battery capacity meansigma methods,For battery capacity prediction meansigma methods, N is natural number.

5. the method as according to any one of Claims 1-4, it is characterised in that carry out battery failure threshold values extrapolation according to battery capacity, it is achieved after the real-time estimate to lithium battery residual life RUL, also include:

With actual measured value, the RUL of prediction is carried out error analysis, and error analysis RPE formula is as follows:

Wherein C_rFor actual measured value, C_pRUL for prediction.

6. the prediction unit in a satellite lithium ion battery life-span, it is characterised in that including:

Analysis module, for analyzing the feature of satellite cycle life of lithium ion battery test data, to obtain fault Characteristics of Evolution amount；

Processing module, for going to loosen effect to the degradation in capacity data that described fault Characteristics of Evolution amount is corresponding；

Build module, described in basis, go the degradation in capacity data after loosening effect, carry out the training of least square method supporting vector machine LSSVM unsupervised learning, to complete the LSSVM model construction of bimetry；

Prediction module, for predicting the battery battery capacity at different cycles by LSSVM model, carries out battery failure threshold values extrapolation according to battery capacity, it is achieved the real-time estimate to lithium battery residual life RUL.

7. device as claimed in claim 6, it is characterised in that the LSSVM model formation of described structure module construction is:

Wherein, k (x, x_i) it is kernel function, α_iFor Lagrange factor, b is bias term, and x is the vector of all battery capacities composition, x_iFor battery capacity original value.

8. device as claimed in claim 7, it is characterised in that described structure module when the LSSVM model built, kernel function k (x, x_i) at least include one below:

Polynomial kernel function k (x, x_i)=[1+ (x x_i)]^q；

Sigmoid kernel function k (x, x_i)=tanh (ν_k(x·x_i)+c_k)；

RBF k (x, x_i)=exp (-| x-x_i|²/2σ²)；

Wherein, x is the vector of all battery capacities composition, x_iFor battery capacity original value, ν_k(x·x_i) for displacement parameter, work as v_k> 0 time, its be input data an amplitude adjusted parameter, c_kBeing a displacement parameter controlling to map threshold value, σ is the width of kernel function.

9. device as claimed in claim 6, it is characterised in that also include:

Authentication module, is used for adopting mean square error MSE and squared correlation coefficient SCC to carry out model accuracy checking, and checking formula is as follows:

$\mathrm{MSE}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-y_i)}^2,\mathrm{SCC}={\left[\frac{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(x_i-\stackrel{\‾}{x})(y_i-\stackrel{\‾}{y})}{\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(x_i-\stackrel{\‾}{x})}^2}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2}}\right]}^2,$

Wherein, x_iFor battery capacity original value, y_iFor battery capacity prediction value,For battery capacity meansigma methods,For battery capacity prediction meansigma methods, N is natural number.

10. the device as according to any one of claim 6 to 9, it is characterised in that also include:

Error analysis module, for the RUL of prediction is carried out error analysis with actual measured value, error analysis RPE formula is as follows:

Wherein C_rFor actual measured value, C_pRUL for prediction.