A kind of lithium battery group SOC prediction technique based on artificial life model
Technical field
The present invention relates to cell energy management system fields, more particularly to a kind of lithium battery based on artificial life model
Group SOC prediction technique.
Background technique
Explanation of nouns:
Cellular automata: cellular automata is that a kind of time, space, state are all discrete, steric interaction
It is the grid kinetic model of part with time causality, there is simulation complication system temporal-spatial evolution ability, be a kind of typical case
Artificial life model.
SOC: state-of-charge is also remaining capacity, and representative is after battery is lain idle using a period of time or for a long time
The ratio of the capacity of residual capacity and its fully charged state, commonly using percentage indicates;
BP neural network: Back Propagation neural network, by the multilayer feedforword net of Back Propagation Algorithm training
Network.
QPSO algorithm: Quantum-behaved Particle Swarm Optimization, quantum particle group's algorithm,
It is a kind of improved Particle Swarm Optimization (PSO), it is limited and easily fall into local extremum not overcomes PSO algorithm search space
Foot, while the algorithm has many advantages, such as few parameter, Yi Shixian, fast convergence rate.
Haze continuously emerges in recent years, is because traditional fuel-engined vehicle is more and more, generated vehicle exhaust is also more
More, with the gradual depletion of fossil energy, environmental problem becomes increasingly conspicuous.The appearance of lithium battery can satisfy modern society just
To not only green but also the new energy demand of environmental protection, lithium battery are widely used in electric car because of its high-energy density and high-specific-power
In battery system, the performance of battery not only decides the safety of vehicle electric system, but also decide its reliability and
Efficiency.Battery management system (BMS) needs to provide accurate battery status information so that ginseng is examined, and the core of battery management system
The heart is then battery charge state (SOC), only accurate lithium battery SOC prediction just can with the residual capacity of Accurate Prediction battery,
Effective battery management strategy is determined simultaneously, to avoid the occurrence of the overcharge and over-discharge situation of battery, extends battery
Service life.
But SOC can not directly measure to obtain again, it is necessary to it is predicted by measurable voltage, electric current, SOC prediction
Accuracy has decisive influence to the service efficiency, service life and safety of battery.
At present, there are many research method of SOC, and common measurement or prediction technique have an open circuit voltage method, ampere-hour
Integration method, discharge test method, neural network algorithm, Kalman filtering etc..Analysis obtain existing method shortcoming, now enumerate as
Under: 1. open circuit voltage methods, by the corresponding relationship of open-circuit voltage and SOC, but the time for needing to stand is longer, is obtained by tabling look-up
The SOC of battery, this method is easy to operate, but battery pack will intermittently be stood, only applicable use for laboratory, is not able to satisfy electronic vapour
Vehicle requires online;2. current integration method, can on-line testing, using simple, but this method can cause to estimate with the accumulation of time
Error increases;3. discharge test method, this method can measure to obtain reliable SOC, but this method needs off-line test, and
Charge-discharge machine is needed, and the charge and discharge time is long, thus leads to increased costs;4. Neural Network Prediction, this method
Precision of prediction depends on a large amount of training data and suitable training method;5. Kalman filtering algorithm is suitble to electric vehicle violent
The operating condition of operation, but traditional Kalman filtering algorithm is only suitable for linear system, in operation at acutely nonlinear
Lithium battery is not suitable for.
Generally speaking, the just on-line measurement to electric vehicle lithium battery group SOC at present or prediction technique there are errors big, essence
The problems such as low, at high cost or modeling accuracy is low is spent, is difficult accurately to obtain the SOC of electric automobile lithium battery.
Summary of the invention
For the problems in above technical background, the purpose of the present invention is to propose to a kind of, and the lithium based on artificial life model is electric
Pond group SOC prediction technique.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of lithium battery SOC prediction technique based on artificial life model, comprising steps of
S1, basic artificial life body Model is established;
S2, the structure for improving basic artificial life body Model;
S3, the learning algorithm for improving artificial life body Model.
S4, the history charge and discharge data and corresponding SOC data for acquiring lithium battery group;
S5, the history charge and discharge data of lithium battery group and corresponding SOC data are normalized;
S6, the data that processing is completed are input in each individual cellular, and assign each cellular in cellular space
Itself a set of network weight coefficient, using the reality output of each cellular of weight network query function, if the average reality in cellular space
Border output reaches standard or completes train epochs, then terminates algorithm, otherwise continue.
S7, fitness is calculated.The reality output for comparing desired output Yu each cellular calculates the fitness of each cellular.
S8, selection and duplication.Each cellular selects the maximum individual of fitness as learning object in neighbor scope,
Each cellular carries out duplication operation to the excellent individual in the neighbours selected, updates the network weight coefficient of itself, then returns
Return step S6
Further, the step S1, specifically:
The grid Moore type cellular automata of n × n is established, wherein n takes 20, the artificial life model as this patent.
Further, the step S2, specifically:
Artificial neural network is introduced into each cellular, the neural network for needing training, mind are housed in each cellular
It is three-decker through network, is input layer, hidden layer, output layer respectively.
Further, the step S3, specifically:
Using QPSO algorithm as the learning algorithm of cellular, steps are as follows:
S31, the number of iterations n=0 is set.Initialization population, the number including particle, search space, random initializtion are every
A particle initial position;
S32, in first time iteration, the initial position of each particle is the best position of individual.The position of each particle
The parameter being set in neural network calculates the reality output of all input group data neural networks, according to reality output and expectation
Output obtains target function value corresponding to the particle, and the target function value of all particles compares and can find one and have
The particle of minimum target functional value, the position of the particle is global best position, while saving knot corresponding to the particle
By parameter matrix;
S33, the average desired positions that individual is calculated by the individual desired positions of all particles, to the position of each particle
It is updated, calculates the target function value of each particle, and more new individual desired positions.Finally, the individual of more all particles
Desired positions obtain global desired positions, while saving with parameter matrix corresponding to global desired positions particle;
S34, when reaching iteration termination condition, training terminate, global desired positions are neural network to be determined
Parameter value, the consequent parameter matrix of preservation are the consequent parameter to be determined.When iteration termination condition is not up to, if n=n+
1, go to step S33.
Further, the step S4, specifically:
The charge and discharge data include charging and discharging currents, the monomer voltage of each single battery of lithium battery group.
Further, the step S5, specifically:
According to the following formula, the history charge and discharge data of electric automobile lithium battery group are normalized, while to SOC number
According to progress anti-normalization processing:
In above formula, X indicates the data after normalized, xkIndicate the history charge and discharge data actually entered, k is nature
Number indicates serial number, xmaxIndicate xkMaximum value in the data sequence at place, xminIndicate xkMinimum in the data sequence at place
Value.
Further, the step S6, comprising:
S61, history charge and discharge data and corresponding SOC data according to the electric automobile lithium battery group of acquisition will appoint
Output data of the SOC data at one moment as BP neural network, while by the history charge and discharge in preset time before the moment
Input data of the electric data as BP neural network, and then for outputting and inputting node in the network structure of determining first cell space
Number;
S62, itself a set of network weight coefficient of each cellular in cellular space is assigned
S63, the identical processed data of step S5 are provided for each cellular, utilize each cellular of weight network query function
Reality output, if the average reality output in cellular space reach standard or complete train epochs, terminate algorithm, otherwise after
It is continuous.
Further, the step S7, specifically:
Each cellular compares the gap of itself reality output and desired output, calculates fitness, and mode is as follows:
In above formula, fitness is fitness, outaIt is cellular reality output, outqIt is desired output.
Further, the step S8, specifically:
Each cellular selects fitness fit in neighbor scope in (including itself)star=max { fitj, j ∈ Ω } make
For learning object, each cellular is replicated to the maximum cellular individual of fitness is selected, and updates the network weight of itself
Coefficient:
WI, t'=WStar, t
VI, t'=VStar, t
Then return step S6.
The beneficial effects of the present invention are: the present invention is the lithium battery group SOC prediction technique based on artificial life model, should
Invention is accurately modeled to the SOC value of electric vehicle lithium battery group by artificial life model and is obtained the on-line prediction of SOC
Value, and there is no the accumulation of error, precision of prediction is high, error is small, forecast cost is low, convenient for being widely used.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples
Fig. 1 is a kind of lithium battery group SOC prediction technique flow chart based on artificial life model of the invention;
Fig. 2 is life model --- --- Moore type cellular automata established by the present invention
Fig. 3 is improved cellular automata --- --- the cellular genetic neural network structure of the present invention
Specific embodiment
Referring to Fig.1, the lithium battery group SOC prediction technique based on artificial life model that the present invention provides a kind of, including step
It is rapid:
S1, basic artificial life body Model is established;
S2, the structure for improving basic artificial life body Model;
S3, the learning algorithm for improving artificial life body Model.
S4, the history charge and discharge data and corresponding SOC data for acquiring lithium battery group;
S5, the history charge and discharge data of lithium battery group and corresponding SOC data are normalized;
S6, the data that processing is completed are input in each individual cellular, and assign each cellular in cellular space
Itself a set of network weight coefficient, using the reality output of each cellular of weight network query function, if the average reality in cellular space
Border output reaches standard or completes train epochs, then terminates algorithm, otherwise continue.
S7, fitness is calculated.The reality output for comparing desired output Yu each cellular calculates the fitness of each cellular.
S8, selection and duplication.Each cellular selects the maximum individual of fitness as learning object in neighbor scope,
Each cellular carries out duplication operation to the excellent individual in the neighbours selected, updates the network weight coefficient of itself, then returns
Return step S6.
It is further used as preferred embodiment, the step S1, comprising:
Referring to Fig. 2, the artificial life model of the Moore type cellular automata of n × n is established, wherein n takes 20.
It is further used as preferred embodiment, the step S2, comprising:
Referring to Fig. 3, artificial neural network is introduced into each cellular, the nerve net for needing training is housed in each cellular
Network, neural network are three-decker, are input layer, hidden layer, output layer respectively.
It is further used as preferred embodiment, the step S3, comprising:
Using QPSO algorithm as the learning algorithm of cellular, the specific steps are as follows:
S31, the number of iterations n=0 is set.Initialization population, the number including particle, search space, random initializtion are every
A particle initial position;
S32, in first time iteration, the initial position of each particle is the best position of individual.The position of each particle
The parameter being set in neural network calculates the reality output of all input group data neural networks, according to reality output and expectation
Output obtains target function value corresponding to the particle, and the target function value of all particles compares and can find one and have
The particle of minimum target functional value, the position of the particle is global best position, while saving knot corresponding to the particle
By parameter matrix;
S33, the average desired positions that individual is calculated by the individual desired positions of all particles, to the position of each particle
It is updated, calculates the target function value of each particle, and more new individual desired positions.Finally, the individual of more all particles
Desired positions obtain global desired positions, while saving with parameter matrix corresponding to global desired positions particle;
S34, when reaching iteration termination condition, training terminate, global desired positions are neural network to be determined
Parameter value, the consequent parameter matrix of preservation are the consequent parameter to be determined.When iteration termination condition is not up to, if n=n+
1, go to step S33.
It is further used as preferred embodiment, the step S4, comprising:
The charge and discharge data include charging and discharging currents, the monomer voltage of each single battery of lithium battery group.
It is further used as preferred embodiment, the step S5, comprising:
According to the following formula, the history charge and discharge data of electric automobile lithium battery group are normalized, while to SOC number
According to progress anti-normalization processing:
In above formula, X indicates the data after normalized, xkIndicate the history charge and discharge data actually entered, k is nature
Number indicates serial number, xmaxIndicate xkMaximum value in the data sequence at place, xminIndicate xkMinimum in the data sequence at place
Value.
It is further used as preferred embodiment, the step S6, comprising:
S61, history charge and discharge data and corresponding SOC data according to the electric automobile lithium battery group of acquisition will appoint
Output data of the SOC data at one moment as BP neural network, while by the history charge and discharge in preset time before the moment
Input data of the electric data as BP neural network, and then for outputting and inputting node in the network structure of determining first cell space
Number;
S62, itself a set of network weight coefficient of each cellular in cellular space is assigned
S63, the identical processed data of step S5 are provided for each cellular, utilize each cellular of weight network query function
Reality output, if the average reality output in cellular space reach standard or complete train epochs, terminate algorithm, otherwise after
It is continuous.
It is further used as preferred embodiment, the step S7, comprising:
Each cellular compares the gap of itself reality output and desired output, calculates fitness, and mode is as follows:
In above formula, fitness is fitness, outaIt is cellular reality output, outqIt is desired output.
It is further used as preferred embodiment, the step S8, comprising:
Each cellular selects fitness fit in neighbor scope in (including itself)star=max { fitj, j ∈ Ω } make
For learning object, each cellular is replicated to the maximum cellular individual of fitness is selected, and updates the network weight of itself
Coefficient:
WI, t'=WStar, t
VI, t'=VStar, t
Then return step S6.
It is to be illustrated to preferable implementation of the invention, but the invention is not limited to the implementation above
Example, those skilled in the art can also make various equivalent variations on the premise of without prejudice to spirit of the invention or replace
It changes, these equivalent deformations or replacement are all included in the scope defined by the claims of the present application.