CN110232432A - A kind of lithium battery group SOC prediction technique based on artificial life model - Google Patents

Abstract

The invention provides a method for predicting the SOC of a lithium battery pack based on an artificial life model. The method includes two parts: the establishment of the artificial life model and the evolution of the artificial life model. First, an artificial life model for lithium battery SOC prediction is established. Secondly, in order to make the artificial life body learn more quickly, the artificial neural network is introduced and the QPSO algorithm is used as the learning algorithm of the artificial life body. After that, the collected historical charge and discharge data of the lithium battery pack of the electric vehicle and the corresponding SOC data are provided to the artificial life body to make them evolve continuously. Thirdly, when the evolution is completed, the optimal individual in the optimal population is selected as the model for predicting the SOC of the lithium battery. Finally, the resulting model is evaluated on the test data and the accuracy of the model is calculated. The present invention can accurately obtain prediction results without complicated parameter configuration, has high prediction accuracy, small error and low prediction cost, and can be widely used in the control industry of electric vehicles.

Description

A SOC prediction method of lithium battery pack based on artificial life model

technical field

The invention relates to the field of battery energy management systems, in particular to a method for predicting the SOC of a lithium battery pack based on an artificial life model.

Background technique

Glossary:

Cellular automata: cellular automata, is a grid dynamics model with discrete time, space and state, spatial interaction and time causal relationship as local, with the ability to simulate the space-time evolution of complex systems, is a typical artificial life Model.

SOC: state of charge, also known as remaining power, represents the ratio of the remaining capacity of the battery after a period of use or long-term shelving to the capacity of the fully charged state, commonly expressed as a percentage;

BP neural network: Back Propagation neural network, a multi-layer feedforward network trained according to the error back propagation algorithm.

QPSO algorithm: Quantum-behaved Particle Swarm Optimization, quantum particle swarm optimization, is an improved particle swarm optimization algorithm (PSO), which overcomes the shortcomings of the limited search space and easy to fall into local extreme values of the PSO algorithm, and the algorithm has fewer parameters. , easy to implement, fast convergence speed and so on.

In recent years, smog has continued to appear because there are more and more traditional fuel vehicles and more vehicle exhaust. With the gradual depletion of fossil energy, environmental problems have become increasingly prominent. The emergence of lithium batteries can just meet the needs of modern society for new energy that is both green and environmentally friendly. Lithium batteries are widely used in electric vehicle battery systems because of their high energy density and high specific power. The performance of the battery not only determines the performance of the electric vehicle system. Safety, but also determines its reliability and efficiency. The battery management system (BMS) needs to provide accurate battery status information for human reference, and the core of the battery management system is the battery state of charge (SOC). Only accurate lithium battery SOC prediction can accurately predict the remaining capacity of the battery. At the same time, an effective battery management strategy is determined to avoid overcharging and overdischarging of the battery and prolong the service life of the battery.

However, SOC cannot be directly measured, and must be predicted by measurable voltage and current. The accuracy of SOC prediction has a decisive impact on the efficiency, service life and safety of the battery.

From the current point of view, there are many research methods for SOC. Common measurement or prediction methods include open circuit voltage method, ampere-hour integration method, discharge experiment method, neural network algorithm, Kalman filter, etc. The shortcomings of the existing methods are analyzed and listed as follows: 1. Open circuit voltage method, through the corresponding relationship between open circuit voltage and SOC, but it needs to stand for a long time, and the SOC of the battery is obtained by looking up the table, this method is simple to operate , but the battery pack should be left to stand intermittently, which is only suitable for laboratory use and cannot meet the online requirements of electric vehicles; 2. The ampere-hour integration method can be tested online and is simple to use, but this method will increase the estimation error with the accumulation of time. 3. Discharge experiment method, this method can measure reliable SOC, but this method requires off-line test, and requires charging and discharging machine, and the charging and discharging time is long, which leads to an increase in cost; 4. Neural network algorithm predicts, this method The prediction accuracy depends on a large amount of training data and suitable training methods; 5. The Kalman filter algorithm is suitable for the severe running conditions of electric vehicles, but the traditional Kalman filter algorithm is only suitable for linear systems, and it is not suitable for severe abnormal conditions during operation. Linear lithium batteries are not suitable.

In general, the current online measurement or prediction methods for the SOC of lithium battery packs for electric vehicles have problems such as large errors, low accuracy, high cost or low modeling accuracy, and it is difficult to accurately obtain the SOC of lithium batteries for electric vehicles.

SUMMARY OF THE INVENTION

In view of the problems in the above technical background, the purpose of the present invention is to propose a method for predicting the SOC of a lithium battery pack based on an artificial life model.

The technical scheme adopted by the present invention to solve its technical problems is:

A lithium battery SOC prediction method based on an artificial life model, comprising the steps of:

S1. Establish a basic artificial life model;

S2. Improve the structure of the basic artificial life model;

S3, improving the learning algorithm of the artificial life model.

S4. Collect historical charge and discharge data and corresponding SOC data of the lithium battery pack;

S5, normalize the historical charge and discharge data of the lithium battery pack and the corresponding SOC data;

S6. Input the processed data into each individual cell, and assign each cell in the cell space a set of its own network weight coefficients, and use the weight network to calculate the actual output of each cell. If the average actual output of the cell space reaches the standard or the number of training steps is completed, the algorithm ends, otherwise it continues.

S7. Calculate the fitness. Compare the expected output with the actual output of each cell, and calculate the fitness of each cell.

S8. Select and copy. Each cell selects the individual with the largest fitness as the learning object within the neighbor range, and each cell replicates the outstanding individuals in the selected neighbors, updates its own network weight coefficient, and then returns to step S6

Further, the step S1 is specifically:

Establish an n×n grid Moore-type cellular automaton, where n is 20, as the artificial life model of this patent.

Further, the step S2 is specifically:

The artificial neural network is introduced into each cell, and each cell stores the neural network that needs to be trained. The neural network has a three-layer structure, namely the input layer, the hidden layer, and the output layer.

Further, the step S3 is specifically:

Using the QPSO algorithm as the learning algorithm of the cell, the steps are as follows:

S31. Set the number of iterations n=0. Initialize the particle swarm, including the number of particles, search space, and randomly initialize the initial position of each particle;

S32. In the first iteration, the initial position of each particle is the best position of the individual. The position of each particle is a parameter in the neural network. Calculate the actual output of the neural network for all input data groups, and obtain the objective function value corresponding to the particle according to the actual output and expected output. The objective function values of all particles can be compared. Find a particle with the smallest objective function value, the position of the particle is the best position in the world, and save the conclusion parameter matrix corresponding to the particle;

S33: Calculate the average best position of the individual from the individual best positions of all the particles, update the position of each particle, calculate the objective function value of each particle, and update the individual best position. Finally, compare the individual best positions of all particles to obtain the global best positions, and save the parameter matrix corresponding to the particles with the global best positions;

S34. When the iteration end condition is reached, the training ends, the global best position is the neural network parameter value to be determined, and the saved conclusion parameter matrix is the conclusion parameter to be determined. When the iteration end condition is not reached, set n=n+1, and go to step S33.

Further, the step S4 is specifically:

The charge-discharge data includes charge-discharge current and single-cell voltage of each single cell of the lithium battery pack.

Further, the step S5 is specifically:

According to the following formula, the historical charge and discharge data of the electric vehicle lithium battery pack is normalized, and the SOC data is de-normalized at the same time:

In the above formula, X represents the normalized data, x _k represents the actual input historical charge and discharge data, k is a natural number, representing the serial number, x _max represents the maximum value in the data sequence where x _k is located, and x _min represents x The minimum value in the data sequence in which _k is located.

Further, the step S6 includes:

S61. According to the collected historical charge and discharge data of the lithium battery pack of the electric vehicle and the corresponding SOC data, use the SOC data at any time as the output data of the BP neural network, and at the same time use the historical charge and discharge data in a preset time before the time As the input data of the BP neural network, and then determine the number of input and output nodes in the network structure of the cell body;

S62. Give each cell in the cell space a set of its own network weight coefficients

S63. Provide each cell with the same data processed in step S5, and use the weight network to calculate the actual output of each cell. If the average actual output of the cell space reaches the standard or the number of training steps is completed, the algorithm ends, otherwise continue.

Further, the step S7 is specifically:

Each cell compares the gap between its actual output and the expected output, and calculates the fitness as follows:

In the above formula, fitness is the fitness, out _a is the actual output of the cell, and out _q is the expected output.

Further, the step S8 is specifically:

Each cell selects fitness fit _star =max{fit _j , j∈Ω} as the learning object in the range of its neighbors (including itself), and each cell replicates the selected cell individual with the largest fitness, And update its own network weight coefficient:

Wi _,t '= _Wstar,t

Vi _,t '= _Vstar,t

Then it returns to step S6.

The beneficial effects of the present invention are as follows: the present invention is a method for predicting the SOC of a lithium battery pack based on an artificial life model, and the present invention accurately models the SOC value of the lithium battery pack of an electric vehicle through the artificial life model and obtains the online predicted value of the SOC, Moreover, there is no accumulation of errors, the prediction accuracy is high, the error is small, and the prediction cost is low, which is convenient for wide use.

Description of drawings

The present invention will be further described below in conjunction with the accompanying drawings and embodiments

1 is a flowchart of a method for predicting the SOC of a lithium battery pack based on an artificial life model of the present invention;

Fig. 2 is the life model established by the present invention---Moore cellular automata

Fig. 3 is the improved cellular automata of the present invention ------ cellular genetic neural network structure

Detailed ways

1, the present invention provides a method for predicting the SOC of a lithium battery pack based on an artificial life model, including the steps:

S1. Establish a basic artificial life model;

S2. Improve the structure of the basic artificial life model;

S3, improving the learning algorithm of the artificial life model.

S4. Collect historical charge and discharge data and corresponding SOC data of the lithium battery pack;

S5, normalize the historical charge and discharge data of the lithium battery pack and the corresponding SOC data;

S6. Input the processed data into each individual cell, and assign each cell in the cell space a set of its own network weight coefficients, and use the weight network to calculate the actual output of each cell. If the average actual output of the cell space reaches the standard or the number of training steps is completed, the algorithm ends, otherwise it continues.

S7. Calculate the fitness. Compare the expected output with the actual output of each cell, and calculate the fitness of each cell.

S8. Select and copy. Each cell selects the individual with the greatest fitness as the learning object within the neighbor range, and each cell replicates the outstanding individuals in the selected neighbors, updates its own network weight coefficient, and then returns to step S6.

Further as a preferred embodiment, the step S1 includes:

Referring to FIG. 2 , an artificial life model of n×n Moore-type cellular automata is established, where n is 20.

Further as a preferred embodiment, the step S2 includes:

Referring to Figure 3, an artificial neural network is introduced into each cell, and each cell stores a neural network that needs to be trained. The neural network has a three-layer structure, which are an input layer, a hidden layer, and an output layer.

Further as a preferred embodiment, the step S3 includes:

Using the QPSO algorithm as the learning algorithm of the cell, the specific steps are as follows:

S31. Set the number of iterations n=0. Initialize the particle swarm, including the number of particles, search space, and randomly initialize the initial position of each particle;

S32. In the first iteration, the initial position of each particle is the best position of the individual. The position of each particle is a parameter in the neural network. Calculate the actual output of the neural network for all input data groups, and obtain the objective function value corresponding to the particle according to the actual output and expected output. The objective function values of all particles can be compared. Find a particle with the smallest objective function value, the position of the particle is the best position in the world, and save the conclusion parameter matrix corresponding to the particle;

S33: Calculate the average best position of the individual from the individual best positions of all the particles, update the position of each particle, calculate the objective function value of each particle, and update the individual best position. Finally, compare the individual best positions of all particles to obtain the global best positions, and save the parameter matrix corresponding to the particles with the global best positions;

S34. When the iteration end condition is reached, the training ends, the global best position is the neural network parameter value to be determined, and the saved conclusion parameter matrix is the conclusion parameter to be determined. When the iteration end condition is not reached, set n=n+1, and go to step S33.

Further as a preferred embodiment, the step S4 includes:

The charge-discharge data includes charge-discharge current and single-cell voltage of each single cell of the lithium battery pack.

Further as a preferred embodiment, the step S5 includes:

According to the following formula, the historical charge and discharge data of the electric vehicle lithium battery pack is normalized, and the SOC data is de-normalized at the same time:

In the above formula, X represents the normalized data, x _k represents the actual input historical charge and discharge data, k is a natural number, representing the serial number, x _max represents the maximum value in the data sequence where x _k is located, and x _min represents x The minimum value in the data sequence in which _k is located.

Further as a preferred embodiment, the step S6 includes:

S61. According to the collected historical charge and discharge data of the lithium battery pack of the electric vehicle and the corresponding SOC data, use the SOC data at any time as the output data of the BP neural network, and at the same time use the historical charge and discharge data in a preset time before the time As the input data of the BP neural network, and then determine the number of input and output nodes in the network structure of the cell body;

S62. Give each cell in the cell space a set of its own network weight coefficients

S63. Provide each cell with the same data processed in step S5, and use the weight network to calculate the actual output of each cell. If the average actual output of the cell space reaches the standard or the number of training steps is completed, the algorithm ends, otherwise continue.

Further as a preferred embodiment, the step S7 includes:

Each cell compares the gap between its actual output and the expected output, and calculates the fitness as follows:

In the above formula, fitness is the fitness, out _a is the actual output of the cell, and out _q is the expected output.

Further as a preferred embodiment, the step S8 includes:

Each cell selects fitness fit _star =max{fit _j , j∈Ω} as the learning object in the range of its neighbors (including itself), and each cell replicates the selected cell individual with the largest fitness, And update its own network weight coefficient:

Wi _,t '= _Wstar,t

Vi _,t '= _Vstar,t

Then it returns to step S6.

The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments, and those skilled in the art can also make various equivalent deformations or replacements on the premise that does not violate the spirit of the present invention , these equivalent modifications or substitutions are all included within the scope defined by the claims of the present application.

Claims (9)

1. a kind of lithium battery group SOC prediction technique based on artificial life model, which is characterized in that including specific steps:

S1, basic artificial life body Model is established；

S2, the structure for improving basic artificial life body Model；

S3, the learning algorithm for improving artificial life body Model.

S4, the history charge and discharge data and corresponding SOC data for acquiring lithium battery group；

S5, the history charge and discharge data of lithium battery group and corresponding SOC data are normalized；

S6, the data that processing is completed are input in each individual cellular, and it is a set of to assign each cellular in cellular space The network weight coefficient of itself, using the reality output of each cellular of weight network query function, if the average reality in cellular space is defeated Reach standard out or complete train epochs, then terminates algorithm, otherwise continue.

S7, fitness is calculated.The reality output for comparing desired output Yu each cellular calculates the fitness of each cellular.

S8, selection and duplication.Each cellular selects the maximum individual of fitness as learning object in neighbor scope, each Cellular carries out duplication operation to the excellent individual in the neighbours selected, updates the network weight coefficient of itself, then returns to step Rapid S6.

2. a kind of lithium battery group SOC prediction technique based on artificial life model according to claim 1, feature exist In establishing the grid cellular Automation Model for establishing n × n that basic artificial life model refers in the step S1.

3. a kind of lithium battery group SOC prediction technique based on artificial life model according to claim 1, feature exist In: in the step S2, storage in each individual cellular is needed to the artificial neural network of training.The artificial mind stored herein It is the network model of three-decker through network model, is made of respectively input layer, hidden layer, output layer.

4. a kind of lithium battery group SOC prediction technique based on artificial life model according to claim 1, feature exist In, the step S3, specifically: after replacing the original learning algorithm of artificial life using QPSO algorithm, pass through tune Whole its parameter beta realizes learning process, so that more convenient adjust model.

5. a kind of lithium battery group SOC prediction technique based on artificial life model according to claim 1, feature exist In, the step S4, specifically: the acquisition data are charging and discharging currents, the list of each single battery of lithium battery group Bulk voltage.

6. a kind of lithium battery group SOC prediction technique based on artificial life model according to claim 1, feature exist In, the step S5, specifically: the acquisition data fill the history of lithium battery group according to following formula by normalized Discharge data is normalized:

In above formula, X indicates the data after normalized, x_kIndicate the history charge and discharge data actually entered, k is natural number, table Show serial number, x_maxIndicate x_kMaximum value in the data sequence at place, x_minIndicate x_kMinimum value in the data sequence at place.

7. a kind of lithium battery group SOC prediction technique based on artificial life model according to claim 1, feature exist In the step S6, specifically: artificial life body Model that step S3 is established by the data handled based on step S5 into Row study, the model can constantly evolve, until completing to evolve.

8. a kind of lithium battery group SOC prediction technique based on artificial life model according to claim 1, feature exist In, the step S7, specifically: each cellular compares the gap of itself reality output and desired output, calculates fitness, side Formula is as follows:

In above formula, fitness is fitness, out_aIt is cellular reality output, out_qIt is desired output.

9. a kind of lithium battery group SOC prediction technique based on artificial life model according to claim 1, feature exist In, the step S8, specifically: each cellular selects fitness fit in neighbor scope in (including itself)_star= max{fit_j, j ∈ Ω } and it is used as learning object, each cellular is replicated to the maximum cellular individual of fitness is selected, and more Itself new network weight coefficient:

W_{I, t}'=W_{Star, t}

V_{I, t}'=V_{Star, t}

Then return step S6.