CN107741568B - Lithium battery SOC estimation method based on state transition optimization RBF neural network - Google Patents

Lithium battery SOC estimation method based on state transition optimization RBF neural network Download PDF

Info

Publication number
CN107741568B
CN107741568B CN201711093872.7A CN201711093872A CN107741568B CN 107741568 B CN107741568 B CN 107741568B CN 201711093872 A CN201711093872 A CN 201711093872A CN 107741568 B CN107741568 B CN 107741568B
Authority
CN
China
Prior art keywords
network
soc
value
neural network
population
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201711093872.7A
Other languages
Chinese (zh)
Other versions
CN107741568A (en
Inventor
陈宁
李学鹏
阳春华
吴奇
张志平
桂卫华
金浩文
陆国雄
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Guangdong Greenway Technology Co Ltd
Original Assignee
DONGGUAN GREENWAY BATTERY Co Ltd
Central South University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by DONGGUAN GREENWAY BATTERY Co Ltd, Central South University filed Critical DONGGUAN GREENWAY BATTERY Co Ltd
Priority to CN201711093872.7A priority Critical patent/CN107741568B/en
Publication of CN107741568A publication Critical patent/CN107741568A/en
Application granted granted Critical
Publication of CN107741568B publication Critical patent/CN107741568B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
    • G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Abstract

A lithium battery SOC estimation method based on state transition optimization RBF neural network. The invention discloses a lithium battery SOC estimation method based on state transition optimization RBF neural network, and relates to the technical field of electric vehicles. Wherein, the method comprises the following steps: (1) collecting off-line training sample data, and carrying out normalization processing on all training samples; (2) establishing a lithium battery SOC estimation model based on an RBF neural network; (3) optimizing the established RBF neural network model by adopting an STA optimization algorithm; (4) storing the trained RBF network structure and various parameter values, and using the trained RBF network for estimating the SOC of the lithium iron phosphate battery; the method can accurately estimate the SOC of the lithium battery, has the characteristics of high estimation precision, strong reliability, simple estimation model and the like, and can be widely applied to the technical field of power batteries of electric automobiles.

Description

Lithium battery SOC estimation method based on state transition optimization RBF neural network
Technical Field
The invention relates to the technical field of power batteries of electric vehicles, in particular to a lithium battery SOC estimation method based on state transition optimization RBF neural network.
Background
With the deterioration of global environment and the serious problem of resource shortage, especially the worsening of domestic PM2.5 pollution in recent years, the traditional automobile industry at home and abroad has changed greatly, and the development of new energy automobiles becomes the focus of increasing attention of people; among the numerous alternative energy sources of automobiles, electric energy has the characteristics of safety, high efficiency and cleanness, so that new energy automobiles using power batteries as main power sources or auxiliary power sources become the main object of research of people; the Battery Management System (BMS) is a control system based on optimized management and power battery protection, and is also a management system for evaluating the working state and performance of the vehicle-mounted battery of the electric vehicle, so that the safe driving of the vehicle is guaranteed, and the efficient utilization and stability of the battery are maintained; the core part Of the battery management system is the accurate estimation Of the battery SOC (State Of Charge); the battery is used as a complex nonlinear system, and the mathematical relationship between the SOC of the battery and parameters such as voltage, current, temperature, internal resistance and the like is difficult to find; according to the characteristic that the SOC of the battery and parameters such as current, voltage, temperature, aging degree and the like of the battery present nonlinearity, how to utilize measurable parameter data of the battery to realize accurate estimation of the current remaining capacity of the battery is always a core problem of a storage battery management system and a technical problem which needs to be solved urgently, and is also an important and challenging task.
Currently, the commonly used SOC estimation methods include: open circuit voltage method, impedance analysis method, ampere-hour measurement method, neural network method, kalman filter method, and the like; the open-circuit voltage method has the obvious defects that the battery needs to be kept still for a long time during measurement, usually several to more than ten hours, and the method is only suitable for testing the SOC of the electric automobile in the parking state; the impedance analysis method is to estimate the SOC of the battery by researching the relation between the battery resistance and the SOC; however, the difficulty of estimating the SOC by adopting the battery impedance is very high, and the estimation precision cannot be ensured; the ampere-hour metering method is an SOC estimation method which is used in practice, and has some problems in application; errors in current measurement can cause SOC to generate calculation errors, the errors are accumulated continuously and are larger and larger, and the errors are larger under the conditions of high temperature and current fluctuation; when a Kalman filtering method (KF, Kalman Filter) is adopted to estimate the SOC of the battery, the battery is taken as a power system to be researched, and the SOC is taken as an internal state of the system; the kalman filtering algorithm for SOC estimation mainly has the following two problems: firstly, the dependence on a battery model is strong, a more accurate battery model needs to be established to obtain an accurate SOC, and the accuracy and the complexity of the battery model are in direct proportion; secondly, the Kalman algorithm comprises a large number of matrix operations, and the calculated amount is large.
Disclosure of Invention
The invention aims to provide a lithium battery SOC estimation method based on state transition optimization RBF neural network; the technical scheme adopted by the invention for solving the specific technical problems is as follows:
a lithium battery SOC estimation method based on state transition optimization RBF neural network comprises the following steps:
(A) acquiring off-line training sample data, wherein the sample data comprises a monomer terminal voltage, a charge-discharge current, a tab temperature, a cycle life parameter and corresponding SOC data of a lithium battery under the conditions that a charge-discharge multiplying power interval is 0.2C and a temperature interval is 5 ℃, the monomer terminal voltage, the charge-discharge current, the tab temperature and the cycle life parameter are used as an input layer vector of a network, and the SOC is used as an output layer vector of the network; all training samples were normalized according to the following equation:
wherein R is the true value of the actual sample, R is the normalized data, RmaxIs the maximum value of the corresponding type of data sample, RminIs the minimum value of the corresponding type data sample.
(B) Establishing a lithium battery SOC estimation model based on an RBF neural network:
firstly, a gaussian function is selected as a basis function of hidden layer nodes of the RBF neural network, which is expressed as follows:
wherein x is a 4-dimensional input vector which respectively corresponds to the single terminal voltage, the charge-discharge current, the tab temperature and the cycle life parameter of the lithium battery, and ciIs the central vector of the ith neuron node of the hidden layer, the dimension is the same as the input vector x, sigmai 2The width of the center of the ith neuron node.
Secondly, according to the training sample, taking the charge and discharge data of the lithium battery at any moment as the input data of the RBF neural network, and taking the SOC data corresponding to the moment as the output data of the RBF neural network, wherein the output expression corresponding to the model is as follows:
whereinFor the network output, i.e. SOC estimation, wiAnd k is the number of nodes of the hidden layer of the network.
(C) Optimizing the established RBF neural network model by adopting an STA optimization algorithm:
firstly, classifying RBF network input sample sets by using a K-means clustering algorithm, reducing the number of hidden layer nodes, and determining the classified number K by adopting a distance cost principle:
lk1is defined as the average minimum inter-cluster distance from the center of all clusters to the cluster nearest to the centerAverage of the sum of the distances from the heart, ciIs the center of the ith cluster, cjIs the center of the jth cluster;
lk2is the average intra-class distance, defined as the average of the sum of the intra-cluster distances of all clusters, n is the total number of samples, miTotal number of samples for ith cluster, xijThe jth sample inside the cluster.
lk＝lk1-lk2 (6)
lkFor the distance cost when the classification number is k, the optimal classification number k should be the maximum classification number kmaxInner, lkTo the maximum, i.e.:
lk＝max(li)，i＝1,2,......kmax (7)
meanwhile, for the determined classification number K, when the K-means algorithm is used for clustering, in order to overcome the defect that the K-means algorithm is easy to fall into local optimum when clustering is carried out, the STA algorithm is adopted to optimize the selection of the central point of each iteration of the K-means algorithm in the clustering process, and the corresponding optimization problem can be equivalent to the following equation:
wherein C isk(i+1)Represents a state, in particular k center point positions,/k(i+1)Is corresponding to Ck(i+1)The distance cost under the state is specifically calculated according to the formulas (4) to (6); a. theiIs a state transition matrix and can be regarded as an operation operator of an optimization algorithm; and after the optimization of the STA algorithm, the optimal classification scheme under the condition that the classification number is k can be obtained.
Secondly, determining a central point c of the RBF network by using an STA algorithmiExtended width σi 2And wi(ii) a The problem of optimizing RBF network parameters using STAs can be expressed as the following relationship:
wherein S iskRepresenting a state, corresponding to the hub ciExtended width σi 2And the connection weight wiA group of solutions ofkIs a state transition matrix, which is an operation operator, SOC, for optimizing the algorithmiIs the SOC value of the ith sample,is an estimate of the ith sample, n is the total number of samples, err (x)k+1) An objective function is defined as the mean square error of the SOC true value and the SOC estimated value, and the specific steps of training the RBF network by using the STA are as follows:
(a) initializing population individual number SE equal to 30, and randomly and uniformly initializing c in feasible domaini、σi 2And wiGenerating an initial population and an initial feasible solution of an SE group by three variables;
(b) selecting a group c of the current population which enables the objective function f to reach the minimum valuei、σi 2And w value, denoted as best, corresponding to a cost of fbest, copy best into a population with an individual number of SE, denoted as S (k), and perform scaling transformation according to equation (10) to obtain a new population:
S(k+1)＝S(k)+γReS(k) (10)
wherein gamma is a normal number called a scaling factor and takes the value of 1, Re∈Rn×nIs a random diagonal matrix, and x (k +1) is S (k) which is a new population after scaling transformation.
And if the gbest is smaller than fbest, performing translational transformation on the individual newbest according to a formula (11), updating the begt and fbest after the translational transformation, and otherwise, not performing the translational transformation.
Wherein beta is a normal number called translation factor and takes the value of 1, Rm∈Rn×nIs a random variable, the value of which is
In the range of [0,1], S (k-1) is a value before the newbest individual is subjected to scaling transformation;
(c) copying best into a population with the individual number of SE, then carrying out rotation transformation according to a formula (12) to obtain a new population, selecting the optimal individual newbest in the population after the rotation transformation, wherein the corresponding cost is gbest;
wherein alpha is a normal number, called twiddle factor, and takes the value of 1, and R belongs to Rn×nThe elements of the random matrix take on the values of [ -1,1]Scope, | | · | luminance2Is the 2 norm of the vector, In is the identity matrix;
if gbest is smaller than fbest, performing translation transformation according to the formula (11), and updating best and fbest after translation transformation, otherwise, not performing translation transformation;
(d) copying best into a population with the individual number of SE, then carrying out axis transformation according to a formula (13) and judging, selecting the optimal individual in the transformed population as newbest, and taking the corresponding cost as gbest;
Sk+1＝Sk+δRaSk (13)
wherein, delta is an axial factor and takes the value of 1, RaIs a random diagonal matrix;
if gbest is smaller than fbest, performing translation transformation according to the formula (11), and updating best and fbest after translation transformation, otherwise, not performing translation transformation;
(e) repeating the steps b) to d) until the fitness meets the minimum requirement or the iteration times are reached;
(f) saving kernel center c obtained by STA optimizationiExtended width σi 2And the connection weight wi(ii) a (D) And storing the trained RBF network structure and various parameter values, and using the trained RBF network for estimating the SOC of the lithium iron phosphate battery.
Hair brushEstimating the SOC of the lithium battery of the electric automobile based on the angle of data, and establishing a RBF neural network model from the external characteristics of the lithium battery to accurately estimate the SOC of the lithium battery; when the external characteristics of the lithium battery are considered, the influence of the voltage, the current and the temperature of the battery on the SOC of the battery is considered, the cycle life parameter of the battery is introduced, the error caused by the aging of the battery on the SOC of the battery is eliminated, and meanwhile, in order to enable a training sample set to have high representativeness, the charging and discharging multiplying power interval is set to be 0.2C, and the temperature interval is set to be 5 ℃; determining the number of network hidden layers by using a distance cost principle, introducing an average minimum inter-class distance and an average intra-class distance, and for a clustered central point set, the more discrete the more the clustering effect is better, namely the larger the distance between a certain central point and the central point closest to the certain central point is, the better the clustering effect is, and the average minimum inter-class distance is defined as the average value of the sum of the distances from all the clustering centers to the clustering center closest to the certain central point; for a specific class, the closer the distance between a sample point in the class and the class center is, the better the clustering effect is, and the average value of the sum of the distances between the sample point in the class and the interior of all clustering clusters is averaged; the introduction of the two indexes can improve the accuracy of classification; meanwhile, when the sample set is clustered, in order to overcome the defect that the K-means clustering algorithm is easy to fall into local optimization, a state transition global optimization algorithm is introduced to carry out global optimization on the K-means; meanwhile, optimizing network parameters by adopting a state transition algorithm to obtain an optimal central point ciExtended width σi 2And wi。
Drawings
FIG. 1 is a flow chart of a lithium battery SOC estimation method based on state transition optimization RBF neural network according to the present invention;
FIG. 2 is a model of SOC estimation based on RBF neural network according to the present invention;
FIG. 3 is a comparison graph of SOC estimation results at 25 ℃ after 1C discharge of the lithium battery SOC estimation method based on state transition optimization RBF neural network of the present invention;
FIG. 4 is a comparison graph of SOC estimation errors at 25 ℃ after 1C discharge of the lithium battery SOC estimation method based on state transition optimization RBF neural network.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and the detailed description; it is to be understood that the scope of the invention is not to be limited by the specific embodiments;
in the embodiment, hundred million latitude lithium energy LF56K-56AH are adopted as objects, and according to the embodiment of the invention, a method for estimating the SOC of a lithium battery in real time based on a neural network is provided, fig. 1 is a flow chart of the method, and the specific steps are as follows:
(A) collecting off-line training sample data, wherein the sample data comprises single terminal voltage, charging and discharging current, tab temperature and cycle life parameters of the lithium battery at the charging and discharging multiplying power interval of 0.2C and the temperature interval of 5 ℃ and corresponding SOC data; in order to ensure that the selected training samples have high representativeness, discharge tests are carried out under 35 conditions of 5 discharge multiplying factors (set to be 0.2C to 1C) and 7 temperatures (set to be 0 ℃ to 30 ℃) to obtain a training sample set, the size of each sample is 100, and then the total size of the training samples is 3500; the training samples were normalized according to the following equation:
wherein R is the true value of the actual sample, R is the normalized data, RmaxIs the maximum value of the corresponding type of data sample, RminIs the minimum value of the corresponding type data sample;
(B) referring to fig. 2, establishing a lithium battery SOC estimation model based on an RBF neural network, taking a single terminal voltage (V), a current (I), an ambient temperature (T) and a cycle life parameter (Cn) of a power battery as input vectors of the RBF network, and taking a state of charge (SOC) as an output vector; the method comprises the following steps:
firstly, a gaussian function is selected as a basis function of hidden layer nodes of the RBF neural network, which is expressed as follows:
wherein x is a 4-dimensional input vector which respectively corresponds to the single terminal voltage, the charge-discharge current, the tab temperature and the cycle life parameter of the lithium battery, and ciIs the central vector of the ith neuron node of the hidden layer, the dimension is the same as the input vector x, sigmai 2The width of the center of the ith neuron node.
Secondly, according to the training sample, taking the charge and discharge data of the lithium battery at any moment as the input data of the RBF neural network, and taking the SOC data corresponding to the moment as the output data of the RBF neural network, wherein the output expression corresponding to the model is as follows:
whereinFor the network output, i.e. SOC estimation, k is the number of hidden nodes in the network, wiRepresenting the connection weight of the ith neuron node to the output node.
(C) Optimizing the established RBF neural network model by adopting an STA optimization algorithm:
firstly, determining a network hidden layer node by using a K-means algorithm; classifying the RBF network input sample set by using a K-means clustering algorithm, reducing the number of nodes of a hidden layer, and determining the classified number K by adopting a distance cost minimum principle:
lk1is the minimum inter-cluster mean distance, defined as the mean of the sum of the distances of all cluster centers to their nearest cluster center, ciIs the center of the ith cluster, cjIs the center of the jth cluster;
lk2is the average intra-class distance, defined as the average of the sum of the intra-cluster distances of all clusters, n is the total number of samples, miTotal number of samples for ith cluster, xijThe jth sample inside the cluster;
lk＝lk1-lk2 (6)
lkfor the distance cost when the classification number is k, the optimal classification number k should be the maximum classification number kmaxInner, lkTo the maximum, i.e.:
lk＝max(li)，i＝1,2,......kmax (7)
meanwhile, for the determined classification number K, when the K-means algorithm is used for clustering, in order to overcome the defect that the K-means algorithm is easy to fall into local optimum when clustering is carried out, the STA algorithm is adopted to optimize the selection of the central point of each iteration of the K-means algorithm in the clustering process, and the corresponding optimization problem can be equivalent to the following equation:
wherein C isk(i+1)Represents a state, in particular k center point positions,/k(i+1)Is corresponding to Ck(i+1)The distance cost under the state is specifically calculated according to the formulas (4) to (6); a. theiIs a state transition matrix and can be regarded as an operation operator of an optimization algorithm; after the optimization by the STA algorithm, an optimal classification scheme under the condition that the number of classifications is k can be obtained, and in this example, the number of the determined optimal central points is 31.
Secondly, determining a central point c of the RBF network by using an STA algorithmiExtended width σi 2And wi(ii) a The problem of optimizing RBF network parameters using STAs can be expressed as the following relationship:
wherein Sk represents a state corresponding to the hubciExtended width σi 2And the connection weight wiA group of solutions ofkIs a state transition matrix, which is an operation operator, SOC, for optimizing the algorithmiIs the SOC value of the ith sample,is an estimate of the ith sample, n is the total number of samples, err (x)k+1) An objective function is defined as the mean square error of the SOC true value and the SOC estimated value, and the specific steps of training the RBF network by using the STA are as follows:
(a) initializing population individual number SE equal to 30, and randomly and uniformly initializing c in feasible domaini、σi 2And wiGenerating an initial population and an initial feasible solution of an SE group by three variables;
(b) selecting a group c of the current population which enables the objective function f to reach the minimum valuei、σi 2And w value, denoted as best, corresponding to a cost of fbest, copy best into a population with an individual number of SE, denoted as S (k), and perform scaling transformation according to equation (10) to obtain a new population:
S(k+1)＝S(k)+γReS(k) (10)
wherein gamma is a normal number called a scaling factor and takes the value of 1, Re∈Rn×nIs a random diagonal matrix, x (k +1) is S (k) and a new population after expansion and contraction transformation;
the optimal individual in the population after the telescopic transformation is newbest, the corresponding cost is gbest, if the gbest is smaller than fbest, the newbest of the individual is subjected to translational transformation according to a formula (11), and the begt and fbest after the translational transformation are updated, otherwise, the translational transformation is not performed;
wherein beta is a normal number called translation factor and takes the value of 1, Rm∈Rn×nIs a random variable with a value of [0,1]]In the range, S (k-1) is a value before the newbest individual is subjected to scaling transformation;
(c) copying best into a population with the individual number of SE, then carrying out rotation transformation according to a formula (12) to obtain a new population, selecting the optimal individual newbest in the population after the rotation transformation, wherein the corresponding cost is gbest;
wherein alpha is a normal number, called twiddle factor, and takes the value of 1, and R belongs to Rn×nThe elements of the random matrix take on the values of [ -1,1]Scope, | | · | luminance2Is a 2 norm of a vector, InIs an identity matrix;
if gbest is smaller than fbest, performing translation transformation according to the formula (11), and updating best and fbest after translation transformation, otherwise, not performing translation transformation;
(d) copying best into a population with the individual number of SE, then carrying out axis transformation according to a formula (13) and judging, selecting the optimal individual in the transformed population as newbest, and taking the corresponding cost as gbest;
Sk+1＝Sk+δRaSk (13)
wherein, delta is an axial factor and takes the value of 1, RaIs a random diagonal matrix;
if gbest is smaller than fbest, performing translation transformation according to the formula (11), and updating best and fbest after translation transformation, otherwise, not performing translation transformation;
(e) repeating the steps b) to d) until the fitness meets the minimum requirement or the iteration times are reached;
(f) saving kernel center c obtained by STA optimizationiExtended width σi 2And the connection weight wi(ii) a (D) The RBF network structure trained in the step (3) and various parameter values are stored, the method, the BP neural network method and the ampere-hour metering method are respectively adopted to estimate the state of charge of the power battery under the condition that the 1C discharge multiplying power is 25 ℃ at normal temperature, and the estimation result is shown in figure 3; the SOC estimation error is shown in fig. 4; as can be seen from fig. 3, the estimation results of the algorithm proposed herein are closer to the true values; as can be derived from FIG. 4, the STA-K' M-RBF neural network algorithmThe estimation error is about 2%, the estimation error of the BP neural network algorithm is about 5%, and the estimation error of the ampere-hour metering method is about 7%.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (1)

1. A lithium battery SOC estimation method based on state transition optimization RBF neural network is characterized by comprising the following steps:
(A) acquiring off-line training sample data, wherein the sample data comprises a monomer terminal voltage, a charge-discharge current, a tab temperature, a cycle life parameter and corresponding SOC data of a lithium battery under the conditions that a charge-discharge multiplying power interval is 0.2C and a temperature interval is 5 ℃, the monomer terminal voltage, the charge-discharge current, the tab temperature and the cycle life parameter are used as an input layer vector of a network, and the SOC is used as an output layer vector of the network; all training samples were normalized according to the following equation:
wherein R is the true value of the actual sample, R is the normalized data, RmaxIs the maximum value of the corresponding type of data sample, RminIs the minimum value of the corresponding type data sample;
(B) establishing a lithium battery SOC estimation model based on an RBF neural network:
firstly, a gaussian function is selected as a basis function of hidden layer nodes of the RBF neural network, which is expressed as follows:
wherein, x is a 4-dimensional input vector which respectively corresponds to the single terminal voltage, the charge-discharge current and the electrode lug of the lithium batteryTemperature and cycle life parameters, ciIs the central vector of the ith neuron node of the hidden layer, the dimension is the same as the input vector x, sigmai 2Is the center width of the ith neuron node;
secondly, according to the training sample, taking the charge and discharge data of the lithium battery at any moment as the input data of the RBF neural network, and taking the SOC data corresponding to the moment as the output data of the RBF neural network, wherein the output expression corresponding to the model is as follows:
whereinFor the network output, i.e. SOC estimation, wiRepresenting the connection weight from the ith neuron node to the output node, wherein k is the number of nodes of the hidden layer of the network;
(C) optimizing the established RBF neural network model by adopting an STA optimization algorithm:
firstly, classifying RBF network input sample sets by using a K-means clustering algorithm, reducing the number of hidden layer nodes, and determining the number K of the network hidden layer nodes by adopting a distance cost principle:
lk1is the average minimum inter-cluster distance, defined as the average of the sum of the distances of all cluster centers to their nearest cluster center, ciIs the center of the ith cluster, cjIs the center of the jth cluster;
lk2is the average intra-class distance, defined as the average of the sum of the intra-cluster distances of all clusters, n is the total number of samples, miIs the ith polyTotal number of samples of class, xijThe jth sample in the ith cluster is taken as the ith sample;
lk＝lk1-lk2 (6)
lkfor the distance cost when the number of the hidden layer nodes of the network is k, the optimal number k of the hidden layer nodes of the network should be the maximum classification number kmaxInner, lkTo the maximum, i.e.:
lk＝max(li)，i＝1,2,......kmax (7)
meanwhile, for the determined number K of nodes of the network hidden layer, when the K-means algorithm is used for clustering, in order to overcome the defect that the K-means algorithm is easy to fall into local optimum when clustering is carried out, the STA algorithm is adopted to optimize the selection of the central point of each iteration in the clustering process of the K-means algorithm, and the corresponding optimization problem can be equivalent to the following equation:
wherein C isk(i+1)Represents a state, in particular k center point positions,/k(i+1)Is corresponding to Ck(i+1)The distance cost under the state is specifically calculated according to the formulas (4) to (6); a. theiIs a state transition matrix and can be regarded as an operation operator of an optimization algorithm; after STA algorithm optimization, an optimal classification scheme under the condition that the number of nodes of the network hidden layer is k is obtained;
secondly, determining the center c of the ith cluster of the RBF network by using an STA algorithmiCenter width σi 2And the connection weight w from the ith neuron node to the output nodei(ii) a The problem of optimizing RBF network parameters by adopting STA is expressed as the following relation:
wherein S iskRepresenting a state corresponding to the center c of the ith cluster of the networkiCenter width σi 2And the firstConnection weight w from i neuron nodes to output nodeiA group of solutions ofkIs a state transition matrix, which is an operation operator, SOC, for optimizing the algorithmiIs the SOC value of the ith sample,is an estimate of the ith sample, n is the total number of samples, err (x)k+1) For an objective function, defined as the mean square error of the true value and the estimated value of the SOC, the specific steps of training the RBF network using the STA are as follows:
(a) initializing population individual number SE equal to 30, and randomly and uniformly initializing c in feasible domaini、σi 2And wiGenerating an initial population and an initial feasible solution of an SE group by three variables;
(b) selecting a group c of the current population which enables the objective function f to reach the minimum valuei、σi 2And w value, denoted as best, corresponding to a cost of fbest, copy best into a population with an individual number of SE, denoted as S (k), and perform scaling transformation according to equation (10) to obtain a new population:
S(k+1)＝S(k)+γReS(k) (10)
wherein gamma is a normal number called a scaling factor and takes the value of 1, Re∈Rn×nIs a random diagonal matrix, and S (k +1) is a new population after S (k) is subjected to scaling transformation;
the optimal individual in the population after the telescopic transformation is newbest, the corresponding cost is gbest, if the gbest is smaller than fbest, the newbest of the individual is subjected to translational transformation according to a formula (11), and the begt and fbest after the translational transformation are updated, otherwise, the translational transformation is not performed;
wherein beta is a normal number called translation factor and takes the value of 1, Rm∈Rn×nIs a random variable with a value of [0,1]]In the range, S (k-1) is a value before the newbest individual is subjected to scaling transformation;
(c) copying best into a population with the individual number of SE, then carrying out rotation transformation according to a formula (12) to obtain a new population, selecting the optimal individual newbest in the population after the rotation transformation, wherein the corresponding cost is gbest;
wherein alpha is a normal number, called twiddle factor, and takes the value of 1, and R belongs to Rn×nThe elements of the random matrix take on the values of [ -1,1]Scope, | | · | luminance2Is a 2 norm of a vector, InIs an identity matrix;
if gbest is smaller than fbest, performing translation transformation according to the formula (11), and updating best and fbest after translation transformation, otherwise, not performing translation transformation;
(d) copying best into a population with the individual number of SE, then carrying out axis transformation according to a formula (13) and judging, selecting the optimal individual in the transformed population as newbest, and taking the corresponding cost as gbest;
Sk+1＝Sk+δRaSk (13)
wherein, delta is an axial factor and takes the value of 1, RaIs a random diagonal matrix;
if gbest is smaller than fbest, performing translation transformation according to the formula (11), and updating best and fbest after translation transformation, otherwise, not performing translation transformation;
(e) repeating the steps b) to d) until the fitness meets the minimum requirement or the iteration times are reached;
(f) saving the center c of the ith cluster obtained by STA optimizationiCenter width σi 2And the connection weight w from the ith neuron node to the output nodei；
(D) And storing the trained RBF network structure and various parameter values, and using the trained RBF network for estimating the SOC of the lithium iron phosphate battery.
CN201711093872.7A 2017-11-08 2017-11-08 Lithium battery SOC estimation method based on state transition optimization RBF neural network Active CN107741568B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711093872.7A CN107741568B (en) 2017-11-08 2017-11-08 Lithium battery SOC estimation method based on state transition optimization RBF neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711093872.7A CN107741568B (en) 2017-11-08 2017-11-08 Lithium battery SOC estimation method based on state transition optimization RBF neural network

Publications (2)

Publication Number Publication Date
CN107741568A CN107741568A (en) 2018-02-27
CN107741568B true CN107741568B (en) 2019-12-31

Family

ID=61234245

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711093872.7A Active CN107741568B (en) 2017-11-08 2017-11-08 Lithium battery SOC estimation method based on state transition optimization RBF neural network

Country Status (1)

Country Link
CN (1) CN107741568B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110336336A (en) * 2018-03-29 2019-10-15 中国科学院微电子研究所 A kind of remaining battery capacity calculation method and device
CN108519556A (en) * 2018-04-13 2018-09-11 重庆邮电大学 A kind of lithium ion battery SOC prediction techniques based on Recognition with Recurrent Neural Network
CN108647423B (en) * 2018-05-04 2020-12-08 中南大学 Optimization method for size of orifice of flash evaporation system in alumina dissolution process
CN109031152A (en) * 2018-10-15 2018-12-18 重庆科技学院 Lead-acid accumulator residual capacity flexible measurement method based on neural network transparence
CN112363058B (en) * 2020-10-30 2022-06-24 哈尔滨理工大学 Lithium ion battery safety degree estimation method and device based on impedance spectrum and Markov characteristic
CN112462267A (en) * 2020-11-25 2021-03-09 珠海格力电器股份有限公司 Lithium battery SOC estimation method, lithium battery module and power utilization equipment
CN113655385B (en) * 2021-10-19 2022-02-08 深圳市德兰明海科技有限公司 Lithium battery SOC estimation method and device and computer readable storage medium
CN114895206B (en) * 2022-04-26 2023-04-28 合肥工业大学 Lithium ion battery SOH estimation method based on RBF neural network of improved gray wolf optimization algorithm
CN114740376B (en) * 2022-05-05 2023-01-13 杭州科工电子科技有限公司 On-line diagnosis method for battery state of charge
CN116627708B (en) * 2023-07-24 2023-09-19 湖南惟储信息技术有限公司 Storage fault analysis system and method thereof

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103822960A (en) * 2014-02-17 2014-05-28 中南大学 Online multi-metal ion concentration detection method utilizing polarography
CN104834975A (en) * 2015-05-13 2015-08-12 国家电网公司 Power network load factor prediction method based on intelligent algorithm optimization combination
CN106443453A (en) * 2016-07-04 2017-02-22 陈逸涵 Lithium battery SOC estimation method based on BP neural network
US9633148B2 (en) * 2014-11-28 2017-04-25 International Business Machines Corporation Static timing analysis in circuit design

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103822960A (en) * 2014-02-17 2014-05-28 中南大学 Online multi-metal ion concentration detection method utilizing polarography
US9633148B2 (en) * 2014-11-28 2017-04-25 International Business Machines Corporation Static timing analysis in circuit design
CN104834975A (en) * 2015-05-13 2015-08-12 国家电网公司 Power network load factor prediction method based on intelligent algorithm optimization combination
CN106443453A (en) * 2016-07-04 2017-02-22 陈逸涵 Lithium battery SOC estimation method based on BP neural network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
State of charge estimation of Li-ion batteries in an electric vehicle based on a radial-basis-function neural network;Bi Jun等;《Chin.Phys.B》;20121231;第21卷(第11期);118801-1至11801-5 *
基于双重状态转移优化RBF神经网络的锂离子电池SOC估算方法;阳春华等;《第28届中国过程控制会议（CPCC2017）暨纪念中国过程控制会议30周年摘要集》;20170731;摘要 *

Also Published As

Publication number Publication date
CN107741568A (en) 2018-02-27

Similar Documents

Publication Publication Date Title
CN107741568B (en) Lithium battery SOC estimation method based on state transition optimization RBF neural network
CN111398833B (en) Battery health state assessment method
CN110146822B (en) Vehicle power battery capacity online estimation method based on constant-current charging process
Tian et al. Consistency evaluation and cluster analysis for lithium-ion battery pack in electric vehicles
Ouyang et al. Improved parameters identification and state of charge estimation for lithium-ion battery with real-time optimal forgetting factor
CN110346734B (en) Machine learning-based lithium ion power battery health state estimation method
CN111352032A (en) Lithium battery dynamic peak power prediction method
CN113484787B (en) Electrochemical impedance spectrum-based rapid sorting and recombining method for retired lithium ion battery
CN108549030B (en) Lithium battery online health state rapid prediction method based on voltage key characteristics
CN111458646A (en) Lithium battery SOC estimation method based on PSO-RBF neural network
CN109358293B (en) Lithium ion battery SOC estimation method based on IPF
Li et al. A novel state estimation approach based on adaptive unscented Kalman filter for electric vehicles
CN112986830A (en) Lithium battery capacity estimation method based on convolution time memory neural network
CN111208438B (en) Method for cooperatively estimating residual capacity of lithium-ion battery and sensor deviation based on neural network and unscented Kalman filter
CN111064253A (en) Battery health degree rapid evaluation method based on average discrete Frechet distance
CN109613446A (en) A kind of aging of lithium battery detection method based on time series analysis
CN114839536A (en) Lithium ion battery health state estimation method based on multiple health factors
CN115327415A (en) Lithium battery SOC estimation method based on limited memory recursive least square algorithm
CN115219918A (en) Lithium ion battery life prediction method based on capacity decline combined model
Lin et al. Lithium-ion batteries SOH estimation with multimodal multilinear feature fusion
CN114646888A (en) Assessment method and system for capacity attenuation of power battery
Wang et al. State-of-charge estimation of lithium-ion batteries based on multiple filters method
Wu et al. State of health estimation for lithium-ion batteries in real-world electric vehicles
Wu et al. A novel state of health estimation of lithium-ion battery energy storage system based on linear decreasing weight-particle swarm optimization algorithm and incremental capacity-differential voltage method
CN112698217A (en) Battery monomer capacity estimation method based on particle swarm optimization algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CP01 Change in the name or title of a patent holder

Address after: Yuelu District City, Hunan province 410083 Changsha Lushan Road No. 932

Co-patentee after: Guangdong Boliwei Technology Co., Ltd.

Patentee after: CENTRAL SOUTH University

Address before: Yuelu District City, Hunan province 410083 Changsha Lushan Road No. 932

Co-patentee before: DONGGUAN GREENWAY BATTERY Co.,Ltd.

Patentee before: CENTRAL SOUTH University

CP01 Change in the name or title of a patent holder