CN105574231A - Storage battery surplus capacity detection method - Google Patents

Abstract

The invention provides a storage battery surplus capacity detection method based on a theory of a genetic algorithm and a particle swarm optimization algorithm-least squares support vector machine. According to the method, the least squares support vector machine is taken as the basis; the particle swarm optimization algorithm is adopted for performing parameter optimization for a kernel function parameter and a penalty coefficient, so that simulation precision is improved; meanwhile, the global searching ability of the algorithm is enhanced by introducing the genetic algorithm to prevent from occurring local optimal solution; the storage battery surplus capacity can be detected by only detecting the open circuit voltage, the environment temperature and the internal resistance of the storage battery; finally, through comparison between experiment data and prediction data, the average error of the detection method can be controlled within 5%; and the prediction accuracy is greatly improved compared with the conventional least squares support vector machine method.

Description

A kind of remaining battery capacity detection method

Technical field

The present invention relates to technical field of electric power, particularly relate to a kind of remaining battery capacity detection algorithm based on GAPSO-LSSVM.

Background technology

Lead-acid accumulator is simple, easy to use, the cheap electrochmical power source of a kind of structure.It is widely used among all trades and professions such as electric power, communication.Lead-acid accumulator stable, the operation of reliably working to whole system is most important.Research experiment shows, ensure the normal operation of system, extend the serviceable life of accumulator, just must detect the residual capacity SOC of accumulator (i.e. state-of-charge StateofCharge), thus facilitate slip-stick artist can understand the duty of accumulator, to take control strategy in time.Because lead-acid accumulator is of a great variety, purposes and external environment condition different, add that the influence factor of remaining battery capacity is numerous, therefore its prediction adopt method also varied.

The modeling method that general remaining battery capacity detects roughly can be divided into two large classes: a class is Method of Physical Modeling; Another kind of, be the identification and parameter estimation modeling method of system.

Having of Method of Physical Modeling is used: 1, ampere-hour electric discharge, this method needs initial capacity and the efficiency for charge-discharge of determining accumulator in battery remaining power prediction.2, internal resistance detection method, internal resistance method can not at 0 ~ 100% FR interior detection remaining battery capacity.3, open-circuit voltage method, has Linearity preferably between open-circuit voltage and residual capacity, but cannot get rid of this important influence factor of temperature to the impact of the SOC of accumulator.And the open-circuit voltage measuring accumulator needs accumulator to leave standstill for a long time under off-line state; The another kind of SOC Forecasting Methodology about the identification of system and the estimation of parameter has following several: 1 Kalman filtering method, and the method will set up corresponding comparatively complicate mathematical model, and calculated amount is larger.2 artificial neural network methods, although neural network precision of prediction is higher, need a large amount of training samples for its study, the speed of study is comparatively slow, and process is complicated, if will complete on-line analysis, higher to the requirement of hardware processor.Therefore, need to find additive method and solve remaining battery capacity test problems.

Summary of the invention

The technical problem to be solved in the present invention how to detect this inside of remaining battery capacity not easily detection limit at the easy detection limit in outside by detecting accumulator by algorithm, and in original method, improve detection degree of accuracy.

In order to solve this computational problem, the present invention is based on genetic algorithm and particle swarm optimization algorithm-least square method supporting vector machine proposes a kind of remaining battery capacity detection method, it is characterized in that, comprise the steps:

S100: gather the open-circuit voltage of accumulator to be measured, environment temperature, internal resistance be some test sample books, theoretical according to least square method supporting vector machine, obtain remaining battery capacity and detect output model $y=\underset{i=1}{\overset{N}{\Σ}}{a}_{i}K(x,x_i)+b;$

S200: random initializtion is carried out to population in particle cluster algorithm;

S300: upgrade each particle rapidity and position with the renewal rule preset, detects a of output model to described remaining battery capacity _iwith K (x, x _i) in the parameter (C, σ) that comprises carry out optimizing;

S400: according to fitness function, calculates the fitness value of each particle, and described fitness function is described test sample book and detects according to described remaining battery capacity the root-mean-square error that output model calculates the predicted data obtained;

S500: the described fitness value of each particle is sorted, and 1/2nd particles selecting fitness value relatively high carry out the interlace operation of genetic algorithm;

S600: the mutation operation in genetic algorithm is carried out to particle;

S700: according to described fitness function, the fitness value of particle after recalculating genetic algorithm selection, crossover and mutation operation;

S800: according to particle fitness value in the process of current iteration optimizing, selects individual optimal value and colony's optimal value of particle, completes an iteration optimizing;

S900: compare the colony's optimal value in current iteration optimizing and the colony's optimal value retained, retain colony's optimal value that fitness value is low, judge whether to reach default iteration optimizing number of times;

If reached default iteration optimizing number of times, enter S1000 with regard to termination of iterations evolutionary operation;

If not yet reach default iteration optimizing number of times, just return S300, continue the initialization population of next iteration optimizing with selected colony's optimal value;

S1000: configure described remaining battery capacity according to the described colony optimal value retained last in optimizing result and detect output model, exports the residual capacity testing result of this accumulator to be measured according to the open-circuit voltage of accumulator to be measured, environment temperature, internal resistance.

As a kind of prioritization scheme, in described step S300, described default renewal rule is:

$V_{id}^{k+1}={\mathrm{\ωV}}_{id}^k+c_1{r}_1(P_{id}^k-X_{id}^k)+c_2{r}_2(P_{gd}^k-X_{id}^k)$

$X_{id}^{k+1}=X_{id}^k+V_{id}^{k+1}$

Wherein, the X in particle cluster algorithm _i=(x _i1, x _i2... x _in) represent the position vector of i-th particle, and X=(C, σ), wherein C is punishment parameter, and C is included in a that described remaining battery capacity detects output model _iin, σ is kernel functional parameter, and σ is included in K (x, the x that described remaining battery capacity detects output model _i) in, V _i=(v _i1, v _i2... v _in) represent that the pre-set velocity of i-th particle is vectorial, P _i=(p _i1, p _i2... p _in) be described individual optimal value, P _g=(p _g1, p _g2... p _gn) be the described colony optimal value of population, ω is inertia weight, c ₁, c ₂for acceleration factor, r ₁, r ₂for the random number between 0 and 1.

As a kind of prioritization scheme, in described step S500, described interlace operation is specially speed and the position interlace operation of random two two particles in selected population, and formula is as follows:

$V_{new1}^k=\frac{V_1^k+V_2^k}{\left|V_1^k\right|+\left|V_2^k\right|}\left|V_1^k\right|$

$V_{new2}^k=\frac{V_1^k+V_2^k}{\left|V_1^k\right|+\left|V_2^k\right|}\left|V_2^k\right|$

$X_{new1}^k=rand*X_1^k+(1-rand)*X_2^k$

$X_{new2}^k=rand*X_2^k+(1-rand)*X_1^k$

Wherein with the speed of particle and position before being respectively interlace operation, with the speed of new particle and position after being respectively interlace operation.

As a kind of prioritization scheme, in described step S100, least square method supporting vector machine calculates according to described test sample book and obtains remaining battery capacity detection output model process be specially:

For the sample set that the open-circuit voltage of accumulator to be measured gathered, environment temperature, the some test sample books of internal resistance are formed u _i∈ R ⁿ, r _i∈ R ⁿ, t _i∈ R ⁿ, input/output relation formula is

Wherein: for nuclear space mapping function; W is weight vector, has the dimension identical with nuclear space, and b is departure;

Be constructed as follows the optimized parameter optimized formula and determine in described input/output relation formula:

in formula, C is punishment parameter, the Langrange function of structure dual space:

A in formula _i>=0, a _ifor Langrange dual variable;

Can obtain after calculating according to the Langrange function of dual space and the condition of parameter optimization:

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

Wherein

y＝(y ₁,y ₂...y _N) ^T；I _v＝(1,1...1) ^Τ；a＝(a ₁,a ₂...a _N) ^Τ；

and K (x _i, x _j) for meeting the kernel function of mercer theorem;

Described input/output relation formula is finally converted to described remaining battery capacity and detects output model $y=\underset{i=1}{\overset{N}{\Σ}}{a}_{i}K(x,x_i)+b.$

The present invention proposes the remaining battery capacity detection method based on genetic algorithm and particle swarm optimization algorithm-least square method supporting vector machine in, the method is based on the most basic least square method supporting vector machine, particle cluster algorithm is to wherein kernel functional parameter and penalty coefficient carry out parameter optimization in addition, improve simulation accuracy, introduce the ability of searching optimum that genetic algorithm strengthens algorithm simultaneously, prevent from being absorbed in locally optimal solution.Only needing by detecting battery open-circuit voltage, environment temperature, accumulator internal resistance these three amounts, just can detect the residual capacity of accumulator.Finally by contrast experiment's data and predicted data, find that the average error of the detection of this method can control within 5%, more traditional least square method supporting vector machine method, substantially increases precision of prediction.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of a kind of remaining battery capacity detection method in optional embodiment of the present invention;

Fig. 2 is the simulation result figure in one embodiment of the invention;

The error analysis schematic diagram that Fig. 3 (a), (b), (c) are the present invention one test data.

Embodiment

The technical problem to be solved in the present invention how to detect this inside of remaining battery capacity not easily detection limit at the easy detection limit in outside by detecting accumulator by intelligent algorithm, and improve accuracy of detection.

In order to address this problem, the present invention is based on genetic algorithm and particle swarm optimization algorithm-least square method supporting vector machine proposes a kind of remaining battery capacity detection method, it is characterized in that, comprise the steps:

S100: gather the open-circuit voltage of accumulator to be measured, environment temperature, internal resistance be some test sample books, theoretical according to least square method supporting vector machine, obtain remaining battery capacity and detect output model $y=\underset{i=1}{\overset{N}{\Σ}}{a}_{i}K(x,x_i)+b;$

S200: random initializtion is carried out to population in particle cluster algorithm;

S300: upgrade each particle rapidity and position with the renewal rule preset, detects a of output model to described remaining battery capacity _iwith K (x, x _i) in the parameter (C, σ) that comprises carry out optimizing;

S400: according to fitness function, calculates the fitness value of each particle, and described fitness function is described test sample book and detects according to described remaining battery capacity the root-mean-square error that output model calculates the predicted data obtained;

S500: the described fitness value of each particle is sorted, and 1/2nd particles selecting fitness value relatively high carry out the interlace operation of genetic algorithm;

S600: the mutation operation in genetic algorithm is carried out to particle;

S700: according to described fitness function, the fitness value of particle after recalculating genetic algorithm selection, crossover and mutation operation;

S800: according to particle fitness value in the process of current iteration optimizing, selects individual optimal value and colony's optimal value of particle, completes an iteration optimizing;

S900: compare the colony's optimal value in current iteration optimizing and the colony's optimal value retained, retain colony's optimal value that fitness value is low, judge whether to reach default iteration optimizing number of times;

If reached default iteration optimizing number of times, enter S1000 with regard to termination of iterations evolutionary operation;

If not yet reach default iteration optimizing number of times, just return S300, continue the initialization population of next iteration optimizing with selected colony's optimal value;

S1000: configure described remaining battery capacity according to the described colony optimal value retained last in optimizing result and detect output model, exports the residual capacity testing result of this accumulator to be measured according to the open-circuit voltage of accumulator to be measured, environment temperature, internal resistance.

Further, in described step S300, described default renewal rule is:

$V_{id}^{k+1}={\mathrm{\ωV}}_{id}^k+c_1{r}_1(P_{id}^k-X_{id}^k)+c_2{r}_2(P_{gd}^k-X_{id}^k)$

$X_{id}^{k+1}=X_{id}^k+V_{id}^{k+1}$

Wherein, the X in particle cluster algorithm _i=(x _i1, x _i2... x _in) represent the position vector of i-th particle, and X=(C, σ), wherein C is punishment parameter, and C is included in a that described remaining battery capacity detects output model _iin, σ is kernel functional parameter, and σ is included in K (x, the x that described remaining battery capacity detects output model _i) in, V _i=(v _i1, v _i2... v _in) represent that the pre-set velocity of i-th particle is vectorial, P _i=(p _i1, p _i2... p _in) be described individual optimal value, P _g=(p _g1, p _g2... p _gn) be the described colony optimal value of population, ω is inertia weight, c ₁, c ₂for acceleration factor, r ₁, r ₂for the random number between 0 and 1.

Further, in described step S500, described interlace operation is specially speed and the position interlace operation of random two two particles in selected population, and formula is as follows:

$V_{new1}^k=\frac{V_1^k+V_2^k}{\left|V_1^k\right|+\left|V_2^k\right|}\left|V_1^k\right|$

$V_{new2}^k=\frac{V_1^k+V_2^k}{\left|V_1^k\right|+\left|V_2^k\right|}\left|V_2^k\right|$

$X_{new1}^k=rand*X_1^k+(1-rand)*X_2^k$

$X_{new2}^k=rand*X_2^k+(1-rand)*X_1^k$

Wherein with the speed of particle and position before being respectively interlace operation, with the speed of new particle and position after being respectively interlace operation.

Further, in described step S100, least square method supporting vector machine calculates according to described test sample book and obtains remaining battery capacity detection output model process be specially:

For the sample set that the open-circuit voltage of accumulator to be measured gathered, environment temperature, the some test sample books of internal resistance are formed u _i∈ R ⁿ, r _i∈ R ⁿ, t _i∈ R ⁿ, input/output relation formula is

Wherein: for nuclear space mapping function; W is weight vector, has the dimension identical with nuclear space, and b is departure;

Be constructed as follows the optimized parameter optimized formula and determine in described input/output relation formula:

$\underset{w,b,e}{min}J(w,e)=\frac{1}{2}{w}^{T}w+\frac{1}{2}C\underset{i=1}{\overset{N}{\Σ}}{e_i}^2$

In formula, C is punishment parameter, the Langrange function of structure dual space:

A in formula _i>=0, a _ifor Langrange dual variable;

Can obtain after calculating according to the Langrange function of dual space and the condition of parameter optimization:

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

Wherein

y＝(y ₁,y ₂...y _N) ^T；I _v＝(1,1...1) ^Τ；a＝(a ₁,a ₂...a _N) ^Τ；

and K (x _i, x _j) for meeting the kernel function of mercer theorem;

Described input/output relation formula is finally converted to described remaining battery capacity and detects output model $y=\underset{i=1}{\overset{N}{\Σ}}{a}_{i}K(x,x_i)+b.$

Below describe to have done and expand explanation more specifically, and the process of calculating has been brought in the flow process of method.

Above-mentioned steps specifically comprises the steps:

S1: the detection limit choosing remaining battery capacity.Because accumulator is as a closed system, residual capacity is measured as inside, can not directly be obtained by measurement means, needs indirectly to be calculated by outside relevant characterization amount to obtain.

Battery open-circuit voltage and remaining battery capacity have Linearity preferably; Environment temperature also has residual capacity comparatively significantly to be affected; Accumulator internal resistance and residual capacity in certain residual capacity scope internal linear better, therefore select battery open-circuit voltage, environment temperature and internal resistance three as detection limit.

S2: adopt least square method supporting vector machine to carry out Residual capacity prediction for battery, only need experimentally measured data to carry out the training of model, just can carry out the prediction of residual capacity according to the model trained.

LSSVM (least square method supporting vector machine) algorithm is as follows:

For sample set u _i∈ R ⁿ, R _i∈ R ⁿ, T _i∈ R ⁿ, wherein, U is the open-circuit voltage of accumulator to be measured, and T is environment temperature, and R is the internal resistance of accumulator to be measured.According to the basic theories of least square method supporting vector machine, the nonlinear relationship between input and output can be described as

Wherein: for nuclear space mapping function; W is weight vector, has the dimension identical with nuclear space; B is departure.For determining the optimized parameter in input/output relation formula, be constructed as follows optimization problem:

In formula, C is punishment parameter.For solving above-mentioned optimization problem, the Langrange function of structure dual space:

A in formula _i>=0, a _ibe called Langrange dual variable.

Condition according to parameter optimization can obtain:

Cancellation above formula w, e, can obtain system of equations:

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

Wherein

y＝(y ₁,y ₂...y _N) ^T；I _v＝(1,1...1) ^Τ；a＝(a ₁,a ₂...a _N) ^Τ；

and K (x _i, x _j) for meeting the kernel function of mercer theorem.

The relation that the introducing of kernel function makes formula (1) describe is converted to

$y=\underset{i=1}{\overset{N}{\Σ}}{a}_{i}K(x,x_i)+b---\left(7\right)$

Now only need solve linear equations formula (6), just can obtain the parameter of the described model of formula (7).And usually can being solved by conjugate gradient iteration of system of linear equations.

When LSSVM is used for Residual capacity prediction for battery, after comprehensively analyzing the remaining various influence factor of accumulator, select battery open-circuit voltage, temperature and internal resistance value as input quantity, remaining battery capacity is as output quantity.So just, can detect by the residual capacity that LSSVM realizes accumulator.

Here it should be noted that, the least square method supporting vector machine function simlssvm in MATLAB can achieve the most basic sample training and forecast function, therefore directly can use the simlssvm function in MATLAB when calculating.

S3: the theoretical and simulation analysis according to support vector machine, kernel functional parameter and punishment parameter affect very large on simulation accuracy.Traditional grid-search algorithms not only length consuming time and there is certain subjective blindness.Herein adopt particle swarm optimization algorithm to punishment parameter and kernel functional parameter carry out parameter optimization, improve to a certain extent simulation accuracy decrease simultaneously emulation consuming time.

Particle swarm optimization algorithm (ParticleSwarmOptimization, PSO) is the Swarm Intelligence Algorithm that a kind of structure is simple, parameter is few.X in particle cluster algorithm _i=(x _i1, x _i2... x _in) represent the position vector of i-th particle, V _i=(v _i1, v _i2... v _in) represent the velocity vector of i-th particle, P _i=(p _i1, p _i2... p _in) be individual optimal value, P _g=(p _g1, p _g2... p _gn) be the global optimum of population.Particle upgrades speed and the position of self by individual optimal value and global optimum, and more new formula is as follows:

$\begin{array}{c}{V}_{id}^{k+1}={\mathrm{\ωV}}_{id}^k+c_1{r}_1(P_{id}^k-X_{id}^k)+c_2{r}_2(P_{gd}^k-X_{id}^k)\\ X_{id}^{k+1}=X_{id}^k+V_{id}^{k+1}\end{array}---\left(8\right),\left(9\right)$

In formula, ω is inertia weight, c ₁, c ₂for nonnegative constant, be called acceleration factor; r ₁, r ₂for the random number between 0 and 1, X=(C, σ), wherein C is punishment parameter, and C is included in a that described remaining battery capacity detects output model _iin, σ is kernel functional parameter, and σ is included in K (x, the x that described remaining battery capacity detects output model _i) in.

When being optimized for least square method supporting vector machine, basic thought finds the parameter (C, σ) of one group of optimum to make objective function minimum by iteration optimizing, and wherein C is punishment parameter, and σ is kernel functional parameter.Position and the velocity vector of getting particle are argument sequence (C, σ), with the root-mean-square error of simulation and prediction data and actual test data for fitness function.Be finally that optimized parameter carries out model and forecast to LSSVM model with the argument sequence making fitness function minimum.

S4: particle cluster algorithm has parameter few, algorithm is simple, the advantages such as versatility is stronger, and also there is later stage iteration efficiency low, the problems such as Premature Convergence, particularly when running into the optimization problem that there is complicated solution space, are easily absorbed in locally optimal solution simultaneously.Compared with particle swarm optimization algorithm, genetic algorithm (GeneticAlgorithm-GA) has make a variation more widely ability and search capability compared with particle swarm optimization algorithm, particularly in ability of searching optimum, has high efficiency.The present invention proposes and introduce genetic algorithm in particle cluster algorithm, use for reference the operations such as the intersection of genetic algorithm, selection and variation, both are had complementary advantages in the overall situation and local search ability, and simulation analysis shows, GAPSO algorithm simulating precision proposed by the invention is significantly improved.

Genetic algorithm is the computation model of the simulation natural selection of Darwinian evolutionism and a kind of biological evolution process of Genetic Mechanisms, is a kind of optimized algorithm of simulating nature evolutionary process.Genetic algorithm is from arbitrary initial population, use for reference the selection of occurring in nature, crossover and mutation operation, produce a new population, new population is to the adaptability of environment has had a better fitness, along with constantly the carrying out of population colonization, Evolution of Population is to the better optimum population of environmental adaptation degree, has namely tried to achieve the optimum solution of problem.

Implementation procedure for the operatings of genetic algorithm operator of PSO-LSSVM is as follows:

A. the realization of selection opertor

Selection operation is carried out to particle, according to the particle fitness of PSO algorithm select fitness value lower two/tap into into operation of future generation always, so just maintain the improved seeds of particle, more easily obtain globally optimal solution.

B. the realization of mutation operator

Use for reference biological natural hereditary variation principle, crossover operator simulates the gene mutation in biological evolution process, performs variation, obtain new individuality with certain 1/2nd probability to original some individual gene.

C. the realization of crossover operator

Two individualities are selected, with certain crossover probability P in the population that the fitness selected after selection opertor operation is poor _cthe gene of the random identical position to them is carried out and is exchanged.Information exchanging process in crossing operation operant response biological evolution process, its object is exactly to produce new gene and individuality.

And new individual gene is intersected by original two individualities participating in variation and forms.Concrete for speed individual in PSO and location updating formula is:

$V_{new1}^k=\frac{V_1^k+V_2^k}{\left|V_1^k\right|+\left|V_2^k\right|}\left|V_1^k\right|$

$V_{new2}^k=\frac{V_1^k+V_2^k}{\left|V_1^k\right|+\left|V_2^k\right|}\left|V_2^k\right|$

$X_{new1}^k=rand*X_1^k+(1-rand)*X_2^k$

$X_{new2}^k=rand*X_2^k+(1-rand)*X_1^k$

Wherein with the speed of particle and position before being respectively interlace operation, with the speed of new particle and position after being respectively interlace operation.

S5: according to the model obtained after parameter optimization, can carry out the detection of remaining battery capacity.

Detection algorithm is by MATLAB programming realization.

In order to verify the validity of particle cluster algorithm and the algorithm based on GAPSO-LSSVM, the present invention builds simulated program and verifies in MATLAB.With the input that battery open-circuit voltage, environment temperature and accumulator internal resistance are simulated program in experimental data, the residual capacity of accumulator is for exporting.Data are totally 124 groups of data, and wherein 100 groups for training built model, remaining 24 groups as test data, detect precision of prediction.

With the root-mean-square error (root-mean-squareerror, RMSE) of test data and measured data for fitness function.

The errors table of table 1 simulation algorithm:

The errors table of table 1 simulation algorithm

Simulation algorithm	Largest percentage error	Mean percent ratio error	Root-mean-square error
				LSSVM	22.7％	5.6％	0.0384
PSO-LSSVM	14.4％	3.2％	0.0200
				GAPSO-LSSVM	9.9％	2.7％	0.0147

The theory that the present invention is based on genetic algorithm and particle swarm optimization algorithm-least square method supporting vector machine proposes a kind of remaining battery capacity detection method, the method is based on the most basic least square method supporting vector machine, particle cluster algorithm is to wherein kernel functional parameter and penalty coefficient carry out parameter optimization in addition, improve simulation accuracy, introduce the ability of searching optimum that genetic algorithm strengthens algorithm simultaneously, prevent from being absorbed in locally optimal solution.Only needing by detecting battery open-circuit voltage, environment temperature, accumulator internal resistance these three amounts, just can detect the residual capacity of accumulator.Finally by contrast experiment's data and predicted data, the average error of the detection of this method can control within 5%, and more traditional least square method supporting vector machine method, substantially increases precision of prediction.Fig. 2 is the simulation result figure in one embodiment of the invention, and the curve formed with actual measured value that predicts the outcome as seen overlaps substantially completely, and the present invention has higher degree of accuracy.Fig. 3 (a), (b), c () is the error analysis schematic diagram of the present invention one test data, by Fig. 3 (a), (b), (c) known LSSVM, PSO-LSSVM, three's model prediction data of GAPSO-LSSVM algorithm and the relative error analysis situation of experimental data, the root-mean-square error of associative list 1 simulation algorithm is known simultaneously, contrast PSO-LSSVM and LSSVM algorithm, PSO-LSSVM graph of errors fluctuation after particle cluster algorithm optimization obviously diminishes, largest percentage error drops to 14.4% by 22.7%, mean percent ratio error drops to 3.2% by 5.6%, root-mean-square error drops to 0.0200 by 0.0384, effect of optimization is obvious, simulation accuracy significantly improves, after further introducing genetic algorithm, the every error criterion of error obviously reduces further.Relatively GAPSO-LSSVM algorithm and traditional LSSVM algorithm, largest percentage error is dropped to 9.9% by initial 22.7% by proposed GAPSO-LSSVM algorithm; Mean percent ratio error drops to 2.7% by 5.6%, and root-mean-square error is reduced to 0.0147 by 0.0384, and every error criterion is all reduced over half, and algorithm improvement is quite obvious.

The present invention emulates case and just sets forth the present invention for helping.Emulation case does not have all details of detailed descriptionthe, does not limit the embodiment that this invention is only described yet.Obviously, according to the content of this instructions, can make many modifications and variations.This instructions is chosen and is specifically described these embodiments, is to explain principle of the present invention and practical application better, thus makes art technician can utilize the present invention well.The present invention is only subject to the restriction of claims and four corner and equivalent.

Claims (4)

1. a remaining battery capacity detection method, is characterized in that, comprises the steps:

S100: gather the open-circuit voltage of accumulator to be measured, environment temperature, internal resistance be some test sample books, theoretical according to least square method supporting vector machine, obtain remaining battery capacity and detect output model $y=\underset{i=1}{\overset{N}{\Σ}}{a}_{i}K(x,x_i)+b;$

S200: random initializtion is carried out to population in particle cluster algorithm;

S300: upgrade each particle rapidity and position with the renewal rule preset, detects a of output model to described remaining battery capacity _iwith K (x, x _i) in the parameter (C, σ) that comprises carry out optimizing;

S400: according to fitness function, calculates the fitness value of each particle, and described fitness function is described test sample book and detects according to described remaining battery capacity the root-mean-square error that output model calculates the predicted data obtained;

S500: the described fitness value of each particle is sorted, and 1/2nd particles selecting fitness value relatively high carry out the interlace operation of genetic algorithm; S600: the mutation operation in genetic algorithm is carried out to particle;

S700: according to described fitness function, the fitness value of particle after recalculating genetic algorithm selection, crossover and mutation operation;

S800: according to particle fitness value in the process of current iteration optimizing, selects individual optimal value and colony's optimal value of particle, completes an iteration optimizing;

S900: compare the colony's optimal value in current iteration optimizing and the colony's optimal value retained, retain colony's optimal value that fitness value is low, judge whether to reach default iteration optimizing number of times;

If reached default iteration optimizing number of times, enter S1000 with regard to termination of iterations evolutionary operation;

If not yet reach default iteration optimizing number of times, just return S300, continue the initialization population of next iteration optimizing with selected colony's optimal value;

S1000: configure described remaining battery capacity according to the described colony optimal value retained last in optimizing result and detect output model, exports the residual capacity testing result of this accumulator to be measured according to the open-circuit voltage of accumulator to be measured, environment temperature, internal resistance.

2. a kind of remaining battery capacity detection method as claimed in claim 1, is characterized in that, in described step S300, described default renewal rule is:

$V_{id}^{k+1}={\mathrm{\ωV}}_{id}^k+c_1{r}_1(P_{id}^k-X_{id}^k)+c_2{r}_2(P_{gd}^k-X_{id}^k)$

$X_{id}^{k+1}=X_{id}^k+V_{id}^{k+1}$

Wherein, the X in particle cluster algorithm _i=(x _i1, x _i2... x _in) represent the position vector of i-th particle, and X=(C, σ), wherein C is punishment parameter, and C is included in a that described remaining battery capacity detects output model _iin, σ is kernel functional parameter, and σ is included in K (x, the x that described remaining battery capacity detects output model _i) in, V _i=(v _i1, v _i2... v _in) represent that the pre-set velocity of i-th particle is vectorial, P _i=(p _i1, p _i2... p _in) be described individual optimal value, P _g=(p _g1, p _g2... p _gn) be the described colony optimal value of population, ω is inertia weight, c ₁, c ₂for acceleration factor, r ₁, r ₂for the random number between 0 and 1.

3. a kind of remaining battery capacity detection method as claimed in claim 1 or 2, is characterized in that, in described step S500, described interlace operation is specially speed and the position interlace operation of random two two particles in selected population, and formula is as follows:

$V_{new1}^k=\frac{V_1^k+V_2^k}{\left|V_1^k\right|+\left|V_2^k\right|}\left|V_1^k\right|$

$V_{new2}^k\frac{V_1^k+V_2^k}{\left|V_1^k\right|+\left|V_2^k\right|}\left|V_2^k\right|$

$X_{enw1}^k=rand*X_1^k+(1-rand)*X_2^k$

$X_{new2}^k=rand*X_2^k+(1-rand)*X_1^k$

Wherein with the speed of particle and position before being respectively interlace operation, with the speed of new particle and position after being respectively interlace operation.

4. a kind of remaining battery capacity detection method as claimed in claim 1, is characterized in that, in described step S100, least square method supporting vector machine calculates according to described test sample book and obtains remaining battery capacity detection output model process be specially:

For the sample set that the open-circuit voltage of accumulator to be measured gathered, environment temperature, the some test sample books of internal resistance are formed u _i∈ R ⁿ, r _i∈ R ⁿ, t _i∈ R ⁿ, input/output relation formula is

Wherein: for nuclear space mapping function; W is weight vector, has the dimension identical with nuclear space, and b is departure;

Be constructed as follows the optimized parameter optimized formula and determine in described input/output relation formula:

$\underset{w,b,e}{\min }J(w,e)=\frac{1}{2}{w}^{T}w+\frac{1}{2}C\underset{i=1}{\overset{N}{\Σ}}{e_i}^2$

In formula, C is punishment parameter, the Langrange function of structure dual space:

A in formula _i>=0, a _ifor Langrange dual variable;

Can obtain after calculating according to the Langrange function of dual space and the condition of parameter optimization:

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

Wherein

y＝(y ₁,y ₂...y _N) ^T；I _v＝(1,1...1) ^Τ；a＝(a ₁,a ₂...a _N) ^Τ；

and K (x _i, x _j) for meeting the kernel function of mercer theorem;

Described input/output relation formula is finally converted to described remaining battery capacity and detects output model $y=\underset{i=1}{\overset{N}{\Σ}}{a}_{i}K(x,x_i)+b.$