CN106501721A - A kind of lithium battery SOC estimation method based on biological evolution - Google Patents

Abstract

The invention provides a lithium battery SOC estimation method based on biological evolution, which is characterized in that it includes BP network neural model, BP network neural algorithm, network sample data acquisition, sample data preprocessing, BP neural network structure design, network estimation SOC Test, GA genetic algorithm, optimized framework design, network test after optimization, utilize described GA genetic algorithm to optimize the weight and threshold of described BP neural network, its steps are: a, determine network topology, b, determine genetic Algorithm parameters and encoding, c, decoding to get the weight and threshold, d, calculating the output of the neural network and getting the fitness value, e, determining whether the termination condition is satisfied, if the termination condition is satisfied, it will end, if the termination condition is not satisfied, it will go through genetic Algorithm selection, mutation, and crossover generate new groups and then return to step c to continue training. This invention greatly reduces the error of lithium battery SOC estimation.

Description

A Lithium Battery SOC Estimation Method Based on Biological Evolution

technical field

The invention relates to a method for estimating the SOC of a lithium battery, in particular to a method for estimating the SOC of a lithium battery based on biological evolution.

Background technique

With the development of science and technology and industrial technology, the problem of energy crisis and air pollution is becoming more and more serious. According to the survey data of the Beijing Municipal Environmental Protection Bureau in 2013, motor vehicle emissions accounted for 31.1% of the total PM2.5 emissions. Considering the economy, technology and environmental protection, the advantages of electric vehicles are zero emission, low noise, and high efficiency. The development of electric vehicles will become one of the important ways to control air pollution and solve the energy crisis. At present, electric vehicles have problems such as short battery life and insufficient power performance, and the key technical issue in the development of electric vehicles lies in the development of power battery technology. Governments of various countries have proposed corresponding technical strategies. Japan implemented the power storage strategy in 2012, and proposed that the lithium battery industry should reach 50% of the world share by 2020; the United States also proposed the goal of having 14 million electric vehicles by 2020; France and Germany both proposed corresponding power battery technology strategies ; my country has also listed power battery technology as an important research direction for major projects of energy saving and new energy vehicles in the "Eleventh Five-Year Plan" and "Twelfth Five-Year Plan".

Lithium batteries have stable safety performance and high specific capacity, so they become the main component of electric vehicle power batteries. SOC is an important parameter reflecting the internal state of lithium batteries. Accurate estimation of SOC is the basic and key in lithium battery technology. one ring. Accurately estimating the SOC of lithium batteries has important practical significance: First, SOC directly reflects the working capacity of lithium batteries, which can be used as a reference for designing hybrid electric vehicle control systems, effectively improving the performance of hybrid electric vehicle energy management systems, and rationally distributing electric vehicle energy. use; secondly, accurate SOC estimation can effectively prevent overcharge and discharge, maximize battery life, monitor battery SOC and perform energy distribution according to the difference in SOC of each single battery in the battery pack to ensure the consistency of battery power, which can Extend battery life on a single charge.

Inaccurate estimation of SOC will reduce the safety of the battery and shorten the service life of the battery; at the same time, it will affect the accurate calculation of the battery life of the electric vehicle and reduce the efficiency of the battery. As an inherent characteristic of lithium batteries, SOC is difficult to be directly measured by sensors. It is affected by factors such as temperature, discharge current, self-discharge, and battery life during operation, which presents complex nonlinearities. Therefore, an accurate SOC estimation method is sought It is an important research topic in the current power battery industry.

SOC is an important parameter to describe the operating state of lithium batteries, but the relationship between SOC and other battery parameters, such as temperature, internal resistance, self-discharge, life and other influencing factors, is highly nonlinear, making it very difficult to estimate the SOC of lithium batteries. There are mainly the following strategies for battery SOC estimation: open circuit voltage method, impedance analysis method, ampere-hour integration method, and Kalman filter method.

The open circuit voltage method was proposed by the Japanese EV Project Department and DENSO Corporation. After the battery has been left standing for a long time, the voltage at both ends of the battery has a relatively fixed functional relationship with the SOC. Therefore, the SOC can be estimated based on the bipolar voltage of the lithium battery. The open-circuit voltage method is effective in the initial and final stages of battery operation, but it has obvious disadvantages. The open-circuit voltage measurement of the battery must leave the battery for a long time, and the measured voltage is valid when the voltage reaches a relatively stable state, and Impressed by factors such as temperature and battery life, the open circuit voltage method is more suitable for estimating SOC when the battery is left standing.

The impedance analysis method is that during the use of lithium batteries, the internal resistance of the battery includes AC internal resistance and DC internal resistance. Current AC impedance is the transfer function between battery voltage and current. For the AC impedance of the battery, it is greatly affected by the temperature, and there are still great differences in the detection status of the AC impedance; the DC resistance refers to the ratio of the battery voltage change to the current change, and the detection of the DC resistance is determined by the detection time, Influenced by working status, if the detection time is too short, only ohmic resistance can be detected; if the detection time is too long, the internal resistance model will become complicated; in the early stage of discharge, the DC resistance changes relatively, and in the late discharge period, the internal resistance is relatively stable. Therefore, it is difficult to estimate SOC using the internal resistance method, and it is generally not used for actual estimation.

The ampere-hour integration method is to calculate the SOC of the battery by calculating the integral of the discharge current of the lithium battery for a certain period of time. Assuming that the initial SOC is SOC0, then the SOC at the current moment is CN is the rated capacity of the battery, I(τ) is the discharge current at time τ, and η is the Coulomb efficiency. The ampere-hour integration method never needs to establish a complex SOC model, but records the flow of battery energy, but this will also As a result, the estimation results are affected by state factors such as battery temperature, charge-discharge rate, and battery aging, and there are errors. At the same time, there is also a time-accumulated error in current measurement. In addition, the self-discharge phenomenon at the initial moment will also affect the estimation of SOC. Therefore, in In practical applications, the ampere-hour integration method is generally used as a reference value for SOC estimation, and it is used in combination with other methods to improve estimation accuracy.

The Kalman filter algorithm is based on the principle of minimum mean square error, uses SOC as the state quantity of the system, uses the state equation of the system to describe the state transition process, and describes the state correlation function between various moments through the transition characteristics of the state equation, so as to realize SOC estimation [7 ], the core of the Kalman filter algorithm is to establish the equivalent model of the battery. The common models include equivalent circuit and electrochemical model. Accurate SOC can only be obtained with a battery model, and the higher the accuracy requirement, the more complex the model will be; Kalman, the filtering algorithm needs to do a lot of matrix operations, and the calculation is very heavy.

Contents of the invention

Based on the insufficiency of the existing SOC estimation method for lithium batteries, the purpose of the invention is to provide a method to solve the problems existing due to the characteristics of the lithium battery itself. The genetic algorithm is used to optimize the weight threshold of the BP neural network, which greatly reduces the lithium battery. Error in battery SOC estimation.

Existing SOC estimation methods for lithium batteries: the open circuit voltage method is affected by factors such as temperature and battery life, so the open circuit voltage method is more suitable for estimating the SOC when the battery is static; the impedance analysis method is affected by the temperature for the AC impedance of the battery It is very large, and there is still a big difference in the detection state of AC impedance. The detection of DC resistance is affected by the detection time and working status. If the detection time is too short, only ohmic resistance can be detected. If the detection time is too long, the internal resistance model will change. Complicated; in the early stage of discharge, the DC resistance changes relatively greatly, and the internal resistance is relatively stable in the late stage of discharge. Therefore, it is difficult to estimate SOC using the internal resistance method, and it is generally not used for actual estimation; the Abe integral method will cause the estimation results to be affected by the battery temperature, Errors occur due to the influence of state factors such as charge-discharge rate and battery aging. At the same time, there is also a time-accumulated error in current measurement. In addition, the self-discharge phenomenon at the initial moment will also affect the estimation result of SOC; the Kalman filter algorithm is used for battery SOC estimation The main problem lies in the high requirements for the battery model, the establishment of an accurate battery model to obtain accurate SOC, and the higher the accuracy requirements, the complexity of the model will increase; Kalman, the filter algorithm needs to do a lot of matrix operations , the calculation is very heavy.

The purpose of the present invention is to: solve the problems existing due to the characteristics of the lithium battery itself, use the genetic algorithm to optimize the weight threshold of the BP neural network, and greatly reduce the error of the SOC estimation of the lithium battery.

In order to achieve the above object, the present invention provides a lithium battery SOC estimation method based on biological evolution, characterized in that: comprising BP network neural model, BP network neural algorithm, network sample data acquisition, preprocessing of sample data, BP neural network Structural design, network estimation SOC test, GA genetic algorithm, optimized framework design, optimized network test, utilize described GA genetic algorithm to optimize the weight and threshold of described BP neural network, its steps are: a, determine network topology Structure, b. Determine genetic algorithm parameters and encoding, c. Decode to get weight and threshold, d. Calculate the output of neural network and get fitness value, e. Determine whether the termination condition is met. If the termination condition is met, it will end. If not The termination condition is to go through genetic algorithm selection, mutation, and crossover to generate new groups, and then return to step c to continue training.

As an improvement of the biological evolution-based lithium battery SOC estimation method of the present invention, the BP neural network of the biological evolution-based lithium battery SOC estimation method of the present invention is a multi-layer neural network composed of an input layer, a hidden layer, and an output layer. The most basic constituent unit of each layer is exactly neuron, constructs described BP neural network model according to described neuron, draws that corresponding described output layer transformation function is linear function f(x)=x, and described hidden layer The transformation function is a unipolar Sigmoid function Or a bipolar sigmoid function

As an improvement of the biological evolution-based lithium battery SOC estimation method of the present invention, the BP neural network algorithm flow of the biological evolution-based lithium battery SOC estimation method of the present invention is: (1) initialization, (2) input sample pairs, and calculate each layer Error, (3) Calculate the network output error In practical applications, more root mean square error is used (4) Calculate the error number of each layer (5) Adjust weights and thresholds (6) Determine whether all samples are trained. If the sample count is less than the number of network training times, then the sample count and network training times are increased by 1, and return to step (2), otherwise go to step (7), and (7) determine whether it is satisfied Error accuracy requirements and termination conditions, if the root mean square error is less than the minimum error or the sample count is less than the maximum sample count, the training is complete, otherwise the error is set to 0, the sample count is set to 1, and return to (2) to continue training.

As an improvement of the biological evolution-based lithium battery SOC estimation method of the present invention, the network sample data acquisition of the biological evolution-based lithium battery SOC estimation method of the present invention includes network sample data collection and sample SOC calculation. The network sample data collection can obtain the voltage-SOC relationship. According to the sample SOC calculation, the SOC estimated by the ampere-hour integration method is feasible as the standard SOC. The calculation formula is: The sample SOC calculations take into account the effects of temperature and discharge current rate.

As an improvement of the biological evolution-based lithium battery SOC estimation method of the present invention, the preprocessing of the sample data in the biological evolution-based lithium battery SOC estimation method of the present invention includes normalization processing of data, random arrangement of data, and The normalization process normalizes the data to the interval [-1,1], using the mapminmax function that comes with the MATLAB toolbox. The random arrangement of the data is to randomly sort the training sample data, which can avoid similar samples from being too large. concentrated.

As an improvement of the lithium battery SOC estimation method based on biological evolution in the present invention, the BP neural network structure design of the lithium battery SOC estimation method based on biological evolution in the present invention includes input and output layer design, hidden layer design, and design of the BP network neural network. The number of layers of the structure obtains the number of hidden layers, and the design of the hidden layer is by using the heuristic method, the formula: j=log ²ⁿ , According to the above formula, determine the range of the number of hidden layer nodes, use the MATLAB toolbox to use the tansig function as the transformation function of the hidden layer nodes, and use the LM algorithm as the network learning algorithm to build a network for testing, and take the average MSE of multiple trainings as the judgment Standard, the relationship between network performance and hidden layer nodes can be obtained, and the number of hidden layer nodes can be set.

As an improvement of the biological evolution-based lithium battery SOC estimation method of the present invention, the biological evolution-based lithium battery SOC estimation method of the present invention is based on the number of layers of the BP network neural structure determined above, the number of hidden layers, the hidden Contains the number of nodes to perform the network estimation SOC test.

As an improvement of the lithium battery SOC estimation method based on biological evolution in the present invention, the genetic operator of the lithium battery SOC estimation method based on biological evolution in the present invention includes a selection operator, a crossover operator, and a mutation operator. The GA genetic algorithm is based on The fitness value of the individual is used to simulate the screening process in the natural environment, and new individuals are generated through operations such as selection, crossover, and mutation. This process will make the entire population develop in a direction that is conducive to the optimal approximate solution. The chromosome code uses is a floating-point encoding. The fitness function should satisfy conditions such as single value, non-negative, continuous, etc., and the fitness function should be as simple as possible to reduce the amount of calculation.

As an improvement of the lithium battery SOC estimation method based on biological evolution in the present invention, the selection operator of the lithium battery SOC estimation method based on biological evolution in the present invention is a fitness proportional method, and the selection operator also includes uniform sorting method, optimal Save strategies, random league selection, crowd-out selection, and more. The crossover operator is the arithmetic crossover of the floating-point number coding, and the floating-point number coding also includes discrete crossover, etc., and the crossover operator is also applicable to single-point crossover, multi-point crossover, uniform crossover, etc. of binary codes, so The above mutation operator is a non-uniform mutation method, and the commonly used mutation operators also include uniform mutation, boundary mutation, Gaussian mutation and so on.

As an improvement of the lithium battery SOC estimation method based on biological evolution in the present invention, the optimization framework design of the lithium battery SOC estimation method based on biological evolution in the present invention is that the genetic algorithm optimizes the BP neural network algorithm and the genetic algorithm is global The advantages of strong search ability and fast convergence speed are combined with the nonlinear fitting ability of neural network to achieve the purpose of overcoming the shortcomings of slow convergence speed of neural network and easy to fall into local error. The termination condition of the genetic algorithm is to reach the required precision or to reach the maximum number of evolutions. Finally, compare the estimation effect and error comparison between optimization and pre-optimization.

Compared with the prior art, the present invention has the following beneficial effects: it solves the problems existing due to the characteristics of the lithium battery itself, uses the genetic algorithm to optimize the weight threshold of the BP neural network, and greatly reduces the error of the SOC estimation of the lithium battery.

Description of drawings

Figure 1 is a single neuron model of a preferred embodiment of the lithium battery SOC estimation method based on biological evolution in the present invention

Fig. 2 is a schematic diagram of a three-layer neural network structure of a preferred embodiment of the lithium battery SOC estimation method based on biological evolution in the present invention

Figure 3 is a flow chart of the transfer algorithm of the preferred embodiment of the lithium battery SOC estimation method based on biological evolution in the present invention

Fig. 4 is a flow chart of the genetic algorithm optimization BP neural network in the preferred embodiment of the lithium battery SOC estimation method based on biological evolution of the present invention

detailed description

The method for estimating the SOC of a lithium battery based on biological evolution of the present invention will be described below with reference to the accompanying drawings 1-4.

Firstly, the BP network neural algorithm and GA genetic algorithm are explained as follows.

The BP network neural algorithm, also known as the error back propagation algorithm, is the most widely used neural network so far. The BP neural network has a simple structure and strong scalability, and is widely used in function approximation, pattern recognition, classification, data compression and other fields. BP neural network is a multi-layer neural network composed of input layer, hidden layer and output layer. The most basic unit of each layer is neuron. The basic neuron model is shown in Figure 1, where X=(x ₁ ,x ₂ ,…) is the input of the neuron, y is the output of the neuron, W=(w ₁ ,w ₂ ,…) is the adjustable input weight, B=b is the threshold of the neuron, f(net) is the neuron Element activation function. The input signal enters the neuron through the input weight connection (weighted summation), and the output y is obtained through the activation function.

GA genetic algorithm is an optimization algorithm developed based on Darvin's evolution theory and Mendel's genetic inheritance principle. The genetic algorithm expresses the problem to be solved as a "chromosome". The initial population is the individuals within the solution set range of the problem to be solved. The population is composed of a certain number of encoded individuals. A better approximate solution. The genetic algorithm simulates the screening process in the natural environment according to the fitness value of the individual, and generates new individuals through operations such as selection, crossover, and mutation. This process will make the entire population develop in a direction that is conducive to the optimal approximate solution.

The invention provides a lithium battery SOC estimation method based on biological evolution, which is characterized in that it includes BP network neural model, BP network neural algorithm, network sample data acquisition, sample data preprocessing, BP neural network structure design, network Estimate SOC test, GA genetic algorithm, optimized framework design, optimized network test, utilize described GA genetic algorithm to optimize the weight and threshold of described BP neural network, its steps are: a, determine network topology, b, Determine the genetic algorithm parameters and encoding, c, decode to get the weight and threshold, d, calculate the output of the neural network and get the fitness value, e, determine whether the termination condition is satisfied, if the termination condition is satisfied, it will end, if the termination condition is not satisfied, the After genetic algorithm selection, mutation, and crossover to generate new populations, return to step c to continue training.

In this embodiment, the BP neural network of the lithium battery SOC estimation method based on biological evolution in the present invention is a multi-layer neural network composed of an input layer, a hidden layer, and an output layer, and the most basic unit of each layer is Be exactly neuron, construct described BP neural network model according to described neuron, draw corresponding described output layer transformation function is linear function f(x)=x, and described hidden layer transformation function is unipolar Sigmoid function Or a bipolar sigmoid function .

In this embodiment, the BP neural network algorithm flow of the lithium battery SOC estimation method based on biological evolution of the present invention is: (1) initialization, (2) input sample pairs, calculate the error of each layer, (3) calculate the network output error In practical applications, more root mean square error is used (4) Calculate the error signal of each layer (5) Adjust weights and thresholds (6) Determine whether all samples are trained. If the sample count is less than the number of network training times, then the sample count and network training times are increased by 1, and return to step (2), otherwise go to step (7), and (7) determine whether it is satisfied Error accuracy requirements and termination conditions, if the root mean square error is less than the minimum error or the sample count is less than the maximum sample count, the training is complete, otherwise the error is set to 0, the sample count is set to 1, and return to (2) to continue training.

In this embodiment, the network sample data acquisition of the lithium battery SOC estimation method based on biological evolution in the present invention includes network sample data collection and sample SOC calculation. The network sample data collection can obtain the voltage-SOC relationship. According to the sample SOC calculation, the SOC estimated by the ampere-hour integration method is feasible as the standard SOC. The calculation formula is: The sample SOC calculations take into account the effects of temperature and discharge current rate.

In this embodiment, the preprocessing of the sample data of the lithium battery SOC estimation method based on biological evolution in the present invention includes normalization processing of data and random arrangement of data, and the normalization processing of data normalizes the data To the interval [-1,1], the mapminmax function that comes with the MATLAB toolbox is used. The random arrangement of the data is to randomly sort the training sample data, which can avoid the excessive concentration of similar samples.

In this embodiment, the BP neural network structure design of the lithium battery SOC estimation method based on biological evolution in the present invention includes input and output layer design, hidden layer design, and the number of layers of the BP network neural structure is designed to obtain hidden Contains the number of layers, the hidden layer is designed by using the heuristic method, the formula: j=log ²ⁿ , According to the above formula, determine the range of the number of hidden layer nodes, use the MATLAB toolbox to use the tansig function as the transformation function of the hidden layer nodes, and use the LM algorithm as the network learning algorithm to build a network for testing, and take the average MSE of multiple trainings as the judgment Standard, the relationship between network performance and hidden layer nodes can be obtained, and the number of hidden layer nodes can be set.

In this embodiment, the present invention is based on the lithium battery SOC estimation method based on biological evolution, according to the number of layers of the neural structure of the BP network determined above, the number of hidden layers, and the number of hidden nodes to perform the network Estimated SOC test.

The genetic operator of the lithium battery SOC estimation method based on biological evolution in the present invention includes a selection operator, a crossover operator, and a mutation operator. The GA genetic algorithm simulates the screening process in the natural environment according to the fitness value of the individual, and generates new individuals through operations such as selection, crossover, and mutation. This process will make the entire population develop in a direction that is conducive to the optimal approximate solution . The chromosomal encoding adopts floating-point encoding. The fitness function should satisfy conditions such as single value, non-negative, continuous, etc., and the fitness function should be as simple as possible to reduce the amount of calculation.

In this embodiment, the selection operator of the lithium battery SOC estimation method based on biological evolution of the present invention is a fitness ratio method. The selection operator also includes uniform sorting method, optimal preservation strategy, random league selection, exclusion selection and so on. The crossover operator is the arithmetic crossover of the floating-point number code, and the floating-point number code also includes discrete crossover, etc., and the crossover operator is also applicable to single-point crossover, multi-point crossover, uniform crossover, etc. of binary code. The mutation operator is a non-uniform mutation method, and commonly used mutation operators include uniform mutation, boundary mutation, Gaussian mutation, and the like.

In this embodiment, the optimization framework design of the lithium battery SOC estimation method based on biological evolution in the present invention is that the genetic algorithm optimizes the BP neural network algorithm, and the genetic algorithm has strong global search ability and fast convergence speed. The combination of the advantages and the nonlinear fitting ability of the neural network achieves the purpose of overcoming the shortcomings of the slow convergence speed of the neural network and the tendency to fall into the local error minimum. The termination condition of the genetic algorithm is to reach the required precision or to reach the maximum number of evolutions. Finally, compare the estimation effect and error comparison between optimization and pre-optimization.

In this embodiment, the BP network neural algorithm of the lithium battery SOC estimation method based on biological evolution of the present invention first constructs the BP neural network model, the structure diagram of the three-layer BP neural network is shown in Figure 2, and the three-layer BP neural network includes all The input layer, the hidden layer, and the output layer, X=(x ₁ ,x ₂ ,...x _n ) is the input matrix of the network, x _n is the input feature vector, W=(w ₀ ,w ₁ , …w _n ) is the connection weight matrix between the input layer and the hidden layer, w _n is the weight vector, where w ₀ is the threshold vector, V=(v ₀ ,v ₁ ,…v _n ) is the hidden layer and the connection weight between the output layer, v _n is the weight vector, where v ₀ is the threshold vector, y is the output vector, for the output layer, the following relationship:

y _k = f(net _k ) k = 1, 2, 3, . . . l (3-1)

For the hidden layer, there are the following relations:

y _j = f(net _j ) j = 1, 2, 3, ... m (3-3)

In the above relationship, the output layer transformation function is a linear function:

f(x)=x (3-5)

The hidden layer transformation function is a unipolar Sigmoid function:

Or a bipolar sigmoid function:

In this embodiment, the BP network neural algorithm flow of the lithium battery SOC estimation method based on biological evolution in the present invention is as follows:

(1) Initialization

Assign initial values to the weight vectors W and V, initialize the sample count p and network training times q to 1, set the error E to 0, set the learning rate η to a decimal between 0 and 1, and set the network expected precision E _min to a positive decimal .

(2) Input the sample pair and calculate the error of each layer

Input the current sample X ^p , d ^p , and calculate the output values Y and O of each layer according to formula (3-5) and formula (3-7);

(3) Calculate the network output error

For P pairs of training samples, the error of the network for sample p The geometric average of the output errors of all samples can be calculated as the total error. In practical applications, the root mean square error is more commonly used.

(4) Calculate the error signal of each layer

Calculation error with

Adjust weights and thresholds

Adjust the next weight W, V according to the error:

(5) Determine whether all samples are trained

If p<P, p, q increase by 1, return to step (2), otherwise go to step (7).

(6) Judging whether the error accuracy requirements and termination conditions are met

If E _RME <E _min or p<P _max is satisfied, the training is completed, otherwise E is set to 0, p is set to 1, and returns to step (2) to continue training.

In this embodiment, the network sample data acquisition of the lithium battery SOC estimation method based on biological evolution in the present invention:

1.1.1 Network sample data collection

Network training samples need to cover the entire parameter range well.

1.1.2 Sample SOC Calculation

As mentioned above, the ampere-hour integration method is simple and reliable, and is suitable for all batteries to estimate the SOC. It is generally used as the SOC standard and used in combination with other methods. However, the ampere-time integration method has the following problems in practical application:

(1) Easily affected by battery temperature and current fluctuations;

(2) Errors in battery measurement will gradually accumulate;

(3) A large amount of sample data is required;

The test scheme adopted in the present invention is a constant current discharge under constant temperature conditions, so the fluctuation of temperature and current is very small and can be ignored; the current measurement accuracy of the battery test system is very high, and the cumulative error is very small; the interval sampling time of the battery is 30s, The total amount of data reaches more than 6,500, which meets the requirements of the number of sample data. Therefore, it is feasible to use the SOC estimated by the ampere-hour integration method as the standard SOC. The calculation formula is:

Among them, I is the current discharge current, η charge and discharge efficiency, C _{I, T} is the total power that the battery can discharge under the current current and temperature, because the lithium battery can discharge all the power under different conditions of temperature and discharge current rate are different, so the influence of temperature and discharge current rate is taken into account here.

In this embodiment, the preprocessing of the sample data of the lithium battery SOC estimation method based on biological evolution of the present invention includes two steps:

(1) Normalization of data

The sample data has different dimensions. Usually, the training samples should be normalized before network training. The usual method is to normalize the data to the interval [-1,1], using the MATLAB toolbox. mapminmax function.

(2) Random sorting of data

Since the sample data is measured by constant current discharge at a certain temperature, there will be a situation where the voltage changes, but the current and temperature are constant. If the network training is performed according to the original order of the sample data, the current and temperature are too concentrated. It will cause the network training and learning process to oscillate, making the convergence time longer or non-convergent. Therefore, it is necessary to randomly sort the training sample data to avoid excessive concentration of similar samples.

In this embodiment, the BP neural network structure design of the lithium battery SOC estimation method based on biological evolution in the present invention is as follows:

Input and output layer design

The input parameters are selected as voltage, discharge current, and temperature, so the input layer has 3 nodes; the output parameter is SOC, so the output layer has 1 node, and the transformation function of the output node is purelin (pure linear function).

hidden layer design

The designed BP neural network structure is 3 layers, so the number of hidden layers is 1. The determination of the number of nodes in the hidden layer is due to the lack of perfect research on the network structure theory, and there is no specific theory to guide, so the heuristic method is used according to the actual situation. Before testing, the approximate range of the number of hidden layer nodes is obtained according to some empirical formulas:

j = log ²ⁿ (3-13)

n is the number of input layer nodes.

m is the number of nodes in the output layer, n is the number of nodes in the input layer, and α is an integer between 0 and 10. The present invention adopts formula (3-14) to determine the range of the number of nodes in the hidden layer, and utilize the MATLAB toolbox to adopt the tansig function As the transformation function of the hidden layer nodes, the Leverberg-Marquardt (LM) algorithm is used as the network learning algorithm to build a network for testing, and the average MSE (minimum mean square error) of multiple trainings is taken as the judgment standard, and the network performance and hidden Layer node relationship.

In this embodiment, the present invention is based on the network estimation SOC test of the lithium battery SOC estimation method based on biological evolution:

According to the above analysis, get the corresponding data and use the MATLAB programming environment to test the network estimation effect. The estimation effect of BP neural network is better, and the actual value is basically consistent with the estimated value. In order to analyze the error more intuitively and accurately, the actual value and the estimated value are used as absolute errors.

In this embodiment, the above of the lithium battery SOC estimation method based on biological evolution of the present invention mainly expounds the relevant theory of BP neural network, and analyzes the model of BP neural network for SOC estimation, and briefly introduces the establishment of the entire BP neural network. Finally, the effect of estimating SOC by BP neural network is analyzed.

In this embodiment, the GA genetic algorithm of the lithium battery SOC estimation method based on biological evolution of the present invention first constructs a genetic algorithm flow chart, as shown in Figure 3, the present invention uses floating-point number encoding 1 to generate an initial population 2, and calculates individual fitness 3. Whether the end condition 4 is met, if the condition is met, end 9, if the condition is not met, select 5, crossover 6, mutation 7, get the next generation population 8, and then return to the individual fitness calculation 3 to continue the calculation.

In this embodiment, the chromosomal code of the lithium battery SOC estimation method based on biological evolution in the present invention selects floating-point code 1. The length is equal to the number of variables. What this article adopts is the method of floating-point number encoding 1.

In this embodiment, the fitness function of the lithium battery SOC estimation method based on biological evolution of the present invention is used to simulate the adaptability of the population individual to the entire natural environment, corresponding to the optimal position of the individual in the solution set range in the genetic algorithm. close to the optimal solution. The fitness function is a criterion for evaluating the adaptability of an individual in the environment. The fitness function should be non-negative. Sometimes it seeks the maximum value, and sometimes it seeks the minimum value. The design fitness function should meet the conditions of single value, non-negative, continuous and so on; it also needs to be as simple as possible to achieve the purpose of reducing the calculation workload; the degree of complexity depends on the specific problem.

In this embodiment, the genetic operator of the lithium battery SOC estimation method based on biological evolution of the present invention includes:

(1) Select 5

Selection 5 is the process of selecting individuals with strong adaptability in the group to participate in the evolution of the new group. This process is also called replication. The selection 5 operation is the basis of crossover and mutation, and its main purpose is to avoid loss of adaptive genes and improve global convergence and computational efficiency. The method used in this embodiment is the fitness ratio method. In other embodiments, methods such as the fitness ratio method (roulette method), uniform sorting method, optimal preservation strategy, random league selection, and exclusion selection are also used.

The operation process of the fitness proportional method is assuming an individual i, whose fitness is f _i , and the population size is N, then the probability of the individual being selected is

(2) Cross 6

The crossover 6 operation in the genetic algorithm is to imitate the process of two chromosomes forming new chromosomes through recombination in the process of natural evolution of organisms. The crossover 6 operation can continuously generate new individuals, increase population diversity, and expand the scope of optimization. The solution space plays an important role. What this embodiment adopts is floating-point code 1, so this embodiment adopts arithmetic intersection 6, and in other embodiments also adopts the single-point intersection of binary code, multi-point intersection, uniform intersection etc., the discrete intersection of floating-point number 1 Wait.

Arithmetic crossover 6 is the process of selecting individual genes for linear combination to generate new chromosomes. Assuming that two individuals X _A and X _B are selected, the process of performing crossover 6 operations on two individuals is:

X′ _A ＝αX _B +(1-α)X _A (4-2)

X' _B =αX _A +(1-a)X _B (4-3)

If α is constant, it is uniform arithmetic crossing6, and if α is variable, it is called non-uniform arithmetic crossing.

(3) Variation 7

Mutation 7 operation is the process of simulating the mutation of genes in the natural evolution process to generate new chromosomes. The mutation 7 operation is to replace some genes in the individual chromosome code string with a certain mutation 7 probability, which can effectively expand the local search range of the genetic algorithm and prevent the algorithm from prematurely converging. This embodiment mainly uses the non-uniform variation method, and in other embodiments, uniform variation, boundary variation, non-uniform variation, Gaussian variation, etc. are also used:

The process of non-uniform mutation operator is:

Where γ is a random 0 or 1, v _k is a gene selected according to a certain probability, v′ _k is a newly generated gene, and the specific formula of the function Δ(t, y) can be:

Where r is a random number in [0,1], T is the maximum number of generations, t is the current number of generations, and b is the system parameter of non-uniformity, which generally takes a value of 2 to 5.

In this embodiment, the optimization framework design of the lithium battery SOC estimation method based on biological evolution of the present invention is as follows:

Genetic algorithm optimization of BP neural network algorithm is to combine the advantages of strong global search ability and fast convergence speed of genetic algorithm with the nonlinear fitting ability of neural network, so as to overcome the shortcomings of slow convergence speed of neural network and easy to fall into local error minimization The purpose, as shown in Figure 4, the process is: determine the network topology 10, determine the genetic algorithm parameters and encoding 11, decode to obtain the weight and threshold 12, calculate the output of the neural network and obtain the fitness value 13, whether to meet the condition 14, if Satisfy the process and end 16, if it is not satisfied, the genetic algorithm will select, cross, and mutate to generate a new group, and return to decoding to obtain weights and thresholds 12 to continue the process.

The fitness function uses the maximum absolute error between the actual SOC value and the SOC value estimated by the network, namely:

E＝max(Y′ _out -Y _out ) (4-6)

Among them, Y' _out is the SOC value estimated by the network, Y _out is the actual SOC value, and the termination condition of the algorithm is to achieve the required accuracy or reach the maximum number of evolutions.

In this embodiment, the lithium battery SOC estimation method based on biological evolution of the present invention is finally optimized for a network test, and the estimation effects and error comparisons are compared between optimized and pre-optimized.

Compared with the prior art, the present invention has the following beneficial effects: it solves the problems existing due to the characteristics of the lithium battery itself, uses the genetic algorithm to optimize the weight threshold of the BP neural network, and greatly reduces the error of SOC estimation.

Although the present invention has been described with reference to specific embodiments, those skilled in the art will be able to make obvious modifications and modifications to the present invention after reading the above description without departing from the intent and essence of the present invention. Modifications and modifications are included within the scope of the claims.

Claims (10)

1. A lithium battery SOC estimation method based on biological evolution, characterized in that: it includes BP network neural model, BP network neural algorithm, network sample data acquisition, preprocessing of sample data, BP neural network structure design, network estimation SOC Test, GA genetic algorithm, optimized framework design, network test after optimization, utilize described GA genetic algorithm to optimize the weight and threshold of described BP neural network, its steps are: a, determine network topology, b, determine genetic Algorithm parameters and encoding, c, decoding to get the weight and threshold, d, calculating the output of the neural network and getting the fitness value, e, determining whether the termination condition is satisfied, if the termination condition is satisfied, it will end, if the termination condition is not satisfied, it will go through genetic Algorithm selection, mutation, and crossover generate new populations, and then return to step c to continue training. 2. The lithium battery SOC estimation method based on biological evolution according to claim 1 is characterized in that: the BP neural network is a multilayer neural network composed of an input layer, a hidden layer, and an output layer, and each layer is the most Basic constituent unit is exactly neuron, builds described BP neural network model according to described neuron, draws that corresponding described output layer transformation function is linear function f(x)=x, and described hidden layer transformation function is unit Polar Sigmoid function Or a bipolar sigmoid function 3. The lithium battery SOC estimation method based on biological evolution according to claim 2, characterized in that: the BP neural network algorithm process is: (1) initialization, (2) input sample pair, calculate the error of each layer, (3 ) to calculate the network output error In practical applications, more root mean square error is used (4) Calculate the error signal of each layer (5) Adjust weights and thresholds (6) Determine whether all samples are trained. If the sample count is less than the number of network training times, then the sample count and network training times are increased by 1, and return to step (2), otherwise go to step (7), and (7) determine whether it is satisfied Error accuracy requirements and termination conditions, if the root mean square error is less than the minimum error or the sample count is less than the maximum sample count, the training is complete, otherwise the error is set to 0, the sample count is set to 1, and return to (2) to continue training. 4. The lithium battery SOC estimation method based on biological evolution according to claim 3, wherein the network sample data acquisition includes network sample data collection and sample SOC calculation. The network sample data collection can obtain the voltage-SOC relationship. According to the sample SOC calculation, the SOC estimated by the ampere-hour integration method is feasible as the standard SOC. The calculation formula is: The sample SOC calculations take into account the effects of temperature and discharge current rate. 5. The lithium battery SOC estimation method based on biological evolution according to claim 4, characterized in that: the preprocessing of the sample data includes normalization processing of data, random arrangement of data, and the normalization processing of data To normalize the data to the interval [-1,1], the mapminmax function that comes with the MATLAB toolbox is used. The random arrangement of the data is to randomly sort the training sample data, which can avoid the excessive concentration of similar samples. 6. The lithium battery SOC estimation method based on biological evolution according to claim 5, characterized in that: the BP neural network structure design includes input and output layer design, hidden layer design, and the number of layers of the BP network neural structure is designed , to obtain the number of hidden layers, the hidden layer design is by using the heuristic method, the formula: j=log ²ⁿ , According to the above formula, determine the range of the number of hidden layer nodes, use the MATLAB toolbox to use the tansig function as the transformation function of the hidden layer nodes, and use the LM algorithm as the network learning algorithm to build a network for testing, and take the average MSE of multiple trainings as the judgment Standard, the relationship between network performance and hidden layer nodes can be obtained, and the number of hidden layer nodes can be set. 7. The lithium battery SOC estimation method based on biological evolution according to claim 6, characterized in that: according to the number of layers of the BP network neural structure determined above, the number of hidden layers, and the number of hidden nodes Perform the network estimation SOC test. 8. The method for estimating lithium battery SOC based on biological evolution according to claim 7, wherein the genetic operator includes a selection operator, a crossover operator, and a mutation operator. The GA genetic algorithm simulates the screening process in the natural environment according to the fitness value of the individual, and generates new individuals through operations such as selection, crossover, and mutation. This process will make the entire population develop in a direction that is conducive to the optimal approximate solution . The chromosomal encoding adopts floating-point encoding. The fitness function should satisfy conditions such as single value, non-negative, continuous, etc., and the fitness function should be as simple as possible to reduce the amount of calculation. 9. The biological evolution-based lithium battery SOC estimation method according to claim 8, characterized in that: the selection operator is a fitness proportional method. The selection operator also includes uniform sorting method, optimal preservation strategy, random league selection, exclusion selection and so on. The crossover operator is the arithmetic crossover of the floating-point number code, and the floating-point number code also includes discrete crossover, etc., and the crossover operator is also applicable to single-point crossover, multi-point crossover, uniform crossover, etc. of binary code. The mutation operator is a non-uniform mutation method, and commonly used mutation operators include uniform mutation, boundary mutation, Gaussian mutation, and the like. 10. The lithium battery SOC estimation method based on biological evolution according to claim 9, characterized in that: the optimization framework design is that the genetic algorithm optimizes the BP neural network algorithm by combining the genetic algorithm with strong global search ability and The advantages of fast convergence speed and the nonlinear fitting ability of neural network are combined to achieve the purpose of overcoming the shortcoming of slow convergence speed of neural network and easy to fall into the minimum local error. The termination condition of the genetic algorithm is to achieve the required accuracy or Reach the maximum number of evolutions, and finally compare the estimation effect and error comparison between optimization and optimization.