Multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization

By using the chaotic particle swarm feedback optimization algorithm with adaptive inertia weight in lithium-ion battery packs to train the feedback neural network, the problem of insufficient error accumulation and real-time correction in the SOC estimation of lithium-ion battery packs is solved, and the problem of insufficient error accumulation and real-time correction in the SOC estimation of lithium-ion battery packs is realized in a multi-temperature environment High-precision SOC estimation improves the safety and service life of the battery pack.

CN118011221BActive Publication Date: 2024-12-06SOUTHWEAT UNIV OF SCI & TECH +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410411497.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-08
Publication Date
2024-12-06
Estimated Expiration
2044-04-08

AI Technical Summary

Technical Problem

The existing SOC estimation methods of lithium-ion battery packs have problems of insufficient error accumulation and real-time correction, especially in multi-temperature environments, which affects the safety and service life of the battery.

Method used

The SOC estimation method of multi-temperature lithium battery based on chaotic particle swarm feedback optimization is adopted. The feedback neural network is trained through the chaotic particle swarm algorithm with adaptive inertia weights, and combined with real-time voltage and current data, the SOC estimation model is optimized to improve estimation accuracy and stability.

Benefits of technology

It realizes high-precision iterative calculation of SOC values ​​of lithium-ion battery packs, reduces the impact of environmental factors on state of charge estimation, improves the applicability and stability of the estimation model, overcomes the problems of insufficient error accumulation and real-time correction of traditional methods .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118011221B_ABST
    Figure CN118011221B_ABST
Patent Text Reader

Abstract

The present invention discloses a multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization, including: using a chaotic particle swarm algorithm with adaptive inertia weight to train and optimize a preset feedback neural network to obtain a SOC estimation model for a lithium-ion battery; inputting real-time voltage and current into the SOC estimation model to obtain the SOC value of the lithium-ion battery. The present invention introduces adaptive inertia weights into the chaotic particle swarm algorithm, improves the global search capability and local convergence speed of particles, solves the nonlinear problem of the system being greatly affected by the environment, and considers the problem of neural network parameter prediction caused by changes in the test environment, inputs voltage and current into the network in real time, and outputs SOC, thereby realizing effective iterative calculation of the SOC value of a lithium-ion battery pack and increasing estimation accuracy and stability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy measurement and control, and in particular relates to a multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization. Background Art

[0002] During the entire life cycle of lithium-ion batteries, the monitoring and adjustment of the core parameter SOC by the Battery Management System (BMS) will affect the effectiveness and safety of emergency power output; therefore, it is very necessary to monitor the changes in this parameter in real time and ensure the working performance of lithium-ion batteries based on this; since the SOC estimation technology in BMS is still immature, the potential safety hazards during use have seriously restricted the development of lithium-ion battery packs; for lithium-ion batteries, reliable BMS management relies on accurate SOC values; when this value is known, it can not only perform reliable energy management and safety control, but also avoid premature damage to the lithium-ion battery pack and extend its service life; therefore, accurate estimation of SOC values ​​is very important to ensure the performance of lithium-ion batteries The working performance of the battery and its energy and safety management are crucial; the construction of the SOC estimation model and accurate estimation of the SOC of lithium-ion batteries are worth obtaining, which has become the core issue of their energy and safety management; SOC represents the remaining capacity of the lithium-ion battery and is the most basic and important key parameter for the battery management system; in addition, the charging and discharging process of lithium-ion batteries involves complex electrical energy, chemical energy and thermal energy conversion, and overcharging and over-discharging can easily lead to safety accidents. Accurate SOC estimation plays an important role in preventing overcharging and over-discharging; in the application of lithium-ion batteries, their safety is still the most concerned issue, and SOC estimation is the basis and prerequisite for their safe use.

[0003] In view of the necessity and urgency of SOC estimation, relevant research institutions and universities, as well as many domestic and foreign journals have set up highly targeted columns to display relevant research results; the SOC estimation methods for lithium-ion batteries are mainly divided into traditional methods, model-based methods and intelligent algorithms. Among them, traditional methods include the ampere-hour integration method (Ah) and the open circuit voltage method (OCV), the models include Kalman filtering (FK) and its extended algorithm (EKF), particle filtering method (PF) and observer method, and the intelligent algorithms include neural network method (NN), fuzzy logic control and machine learning, etc.; due to the influence of many factors such as charging and discharging current, temperature, internal resistance, self-discharge and aging, the performance changes of lithium-ion batteries will have a significant impact on the accuracy of SOC estimation, and there is no universal method to achieve accurate estimation of SOC values; coupled with the influence of consistency between cells during group work, lithium-ion battery packs still lack effective SOC estimation methods; the current practical SOC estimation is achieved through the basic ampere-hour integration method, However, the estimation error is large and is affected by many factors, resulting in a significant cumulative effect. For the SOC estimation research of lithium-ion battery packs, the above-mentioned related research provides a reference for ideas. On this basis, the SOC estimation method under aviation conditions is explored to achieve effective SOC estimation of lithium-ion battery packs. At the same time, for the application of multi-temperature lithium battery packs, it is necessary to consider the balance state of each battery cell in the group for SOC estimation, and then use BMS for effective energy management. Constructing an SOC estimation model with parameter correction and adjustment capabilities and using parameter estimation theory based on equivalent circuit models have become the development trend of SOC estimation, seeking the best balance between improving accuracy and reducing calculation amount, and continuously optimizing and improving the estimation method.

[0004] In the existing BMS applications of lithium-ion battery packs, the SOC estimation method based on neural network fails to accurately characterize the cumulative error in SOC estimation, and cannot perform parameter correction in combination with the current state. Through the analysis of existing SOC estimation methods, based on the research of particle swarm algorithm, the closed-circuit voltage and current are used as parameters of the real-time input feedback network, and the working condition information of the lithium-ion battery is considered in the SOC estimation process, which overcomes the shortcomings of large errors and gradual accumulation caused by insufficient real-time correction of traditional SOC estimation methods. At the same time, the inertia weight is introduced to interfere with the direction of particle motion, improve the global search ability and local convergence speed, thereby reducing the error caused by parameter changes caused by environmental factors to the state of charge estimation. For the multi-temperature SOC estimation problem of lithium-ion battery packs, combined with the advantage analysis of the iterative calculation process of the particle swarm algorithm, a particle feedback method based on inertia weight is proposed, and the construction and experimental verification of the SOC estimation model are realized. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of existing lithium-ion battery pack SOC estimation methods, provide a multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization, and solve the problem of accurate SOC value estimation in lithium-ion battery group applications.

[0006] To achieve the above object, the present invention provides a multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization, comprising:

[0007] The preset feedback neural network is trained and optimized using a chaotic particle swarm algorithm with adaptive inertia weight to obtain a SOC estimation model for lithium-ion batteries.

[0008] The real-time voltage and current are input into the SOC estimation model to obtain the SOC value of the lithium-ion battery.

[0009] Optionally, using a chaotic particle swarm algorithm with adaptive inertia weights to train and optimize a preset feedback neural network includes:

[0010] Taking the test voltage and current as input and the estimated SOC value as output, training the preset feedback neural network to obtain error feedback of the preset feedback neural network;

[0011] Using a chaotic particle swarm algorithm with adaptive inertia weights, the preset feedback neural network is updated in weight, and the weights are dynamically converged to correct the error feedback;

[0012] The modified preset feedback neural network is regarded as the SOC estimation model.

[0013] Optionally, the preset feedback neural network includes: an input layer, a hidden layer and an output layer;

[0014] The input layer is:

[0015]

[0016] in, is the input of the feedback neural network, is the connection weight of the ith neuron to the jth neuron in the mth layer, is the input of the ith neuron to the jth neuron in the mth layer, is the bias of the jth neuron in the mth layer, is a summary of the weight of the jth neuron, is the overall representation of the jth neuron input, is the overall representation of the bias of the jth neuron;

[0017] The hidden layer is:

[0018] in, is the number of neurons in the hidden layer, m and n represent the number of neurons in the input layer and output layer respectively. Represents a random number between the interval [1,9];

[0019] The output layer is:

[0020] in, is the output of the feedback neural network, is the activation function from hidden layer to output layer, is the weight of the ith neuron in the mth layer to the jth neuron, is the input of the i-th neuron in the m-th layer to the j-th neuron, For multiplication, is the expression of the weight and input of the jth neuron under the activation function.

[0021] Optionally, obtaining a preset error feedback of the feedback neural network includes:

[0022] Obtaining the outputs of the preset hidden layer and input layer of the feedback neural network;

[0023] Based on the outputs of the hidden layer and the input layer, obtaining an error value between the output of the output layer and the expected value;

[0024] Based on the error value, error functions of the hidden layer and the input layer are obtained.

[0025] Optionally, the method for updating the weight of the preset feedback neural network is:

[0026]

[0027] in, is the update of the m+1th connection weight of the kth neuron to the jth neuron in the hidden layer, is the mth connection weight update of the kth neuron to the jth neuron in the hidden layer, is the m-th update amount of the j-th neuron to the k-th neuron in the hidden layer, is the error of the kth neuron with error coefficient, is the output of the k-th neuron output layer, is the hidden layer output of the jth neuron, is the m+1th connection weight update of the i-th neuron in the input layer to the j-th neuron, Update the mth connection weight of the i-th neuron in the input layer to the j-th neuron. is the weight update amount of the i-th neuron in the input layer to the j-th neuron, is the mth connection weight update of the i-th neuron in the hidden layer to the j-th neuron, To distinguish it from the error coefficient of the hidden layer, is the output of the jth neuron in the hidden layer, is the connection weight of the jth neuron in the hidden layer to the kth neuron, is the input summary representation of the i-th neuron.

[0028] Optionally, the expression of the adaptive inertia weight is:

[0029]

[0030] in, represents the weight of the directional parameter σ, parameter r i,n-1 k is a random variable that follows a uniform distribution, with parameter n i,n-1 k is a random variable that follows a normal distribution, is the adaptive inertia weight, is the previous weight coefficient of the ith particle in the kth iteration, is the minimum value within the weight correction range of the random component, The maximum value within the weight correction range in the random component.

[0031] Optionally, when training the preset feedback neural network, it also includes:

[0032] A validation set is preset, and cross validation is performed while inputting voltage and current.

[0033] Optionally, when training the preset feedback neural network, it also includes:

[0034] Build a constant temperature control system, and use the BMS battery management system to control the temperature of the lithium battery test by the preset feedback neural network under the chaotic particle algorithm;

[0035] A charge and discharge test management system is built to input the voltage and current under different working conditions into the optimized feedback neural network in real time.

[0036] The present invention has the following beneficial effects:

[0037] The present invention is mainly used to estimate the SOC of lithium-ion battery packs in a high-stability battery system of a petroleum base station. By introducing adaptive inertia weights into a chaotic particle swarm algorithm, the global search capability and local convergence speed of particles are improved, the nonlinear problem of the system being greatly affected by the environment is solved, and the problem of neural network parameter prediction due to changes in the test environment is considered. The voltage and current are input into the network in real time, and the SOC is the output value, thereby realizing effective iterative calculation of the SOC value of the lithium-ion battery pack and increasing the estimation accuracy and stability.

[0038] The present invention is based on the experimental analysis of the application requirements and working characteristics of petroleum base station energy storage, combined with the research ideas of modern control theory, and proposes a multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization, which has strong applicability; for the accurate estimation of the SOC value of the lithium-ion battery pack, the present invention establishes a three-layer feedback neural network to overcome the error caused by the polarization effect, and solves the problem that the environmental factors have a great influence on the charge state estimation through the real-time input of voltage and current characterization data, and adds interval and directional inertia weights to achieve high-speed convergence of group SOC estimation and improve network stability; the present invention can provide a method reference for the establishment of SOC estimation models and SOC value estimation of lithium-ion battery packs in different application scenarios, and has the advantages of simple calculation, good adaptability and high precision. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0040] Figure 1 A schematic diagram of multi-temperature system control according to an embodiment of the present invention;

[0041] Figure 2 Schematic diagram of algorithm flow optimization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0043] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0044] This embodiment aims at the SOC estimation problem of the energy storage lithium-ion battery of the PetroChina base station in the neural network, and proposes a multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization. The real-time input voltage and current are used to train the typical three-layer feedback neural network model to reduce the interference of environmental changes on the SOC prediction; the chaotic particle swarm algorithm is used to optimize the initial weights and thresholds of the feedback neural network to achieve high-precision estimation of the base station lithium battery; based on the deep learning feedback neural network, inertia weights are added to the chaotic particles to achieve interval directional interference on the particles, improve the global search ability and local convergence speed of the particles, and reduce the errors caused by duplication and omission in the estimation; set a validation set, cross-validate the data set, improve data utilization and particle population size, thereby improving the test matching degree and stability of the neural network; build a temperature control system and feedback the inertia weight chaotic particles through the BMS battery management system Neural network lithium battery test temperature control; build a charge and discharge test management system to input the voltage and current under different working conditions into the feedforward neural network in real time under the optimization of the chaotic particle feedback algorithm, so as to realize the accurate prediction of the SOC of multi-temperature lithium battery; this method fully considers the working of lithium-ion batteries in groups at different temperatures, combines the neural network to establish a model for SOC estimation, and realizes the mathematical expression of error prediction and control of lithium-ion battery packs; in order to better reflect this embodiment, in this embodiment, only the lithium-ion battery pack is used as an example for explanation, but those skilled in the art should be aware that according to the technical idea of ​​this embodiment, a multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization can be realized for a variety of lithium-ion batteries; the following describes in detail the implementation steps of the chaotic particle swarm feedback optimization algorithm with adaptive inertia weight for accurate SOC estimation of lithium batteries in high-stability battery systems of petroleum base stations.

[0045] Figure 1 The figure shows the schematic diagram of the control system for the multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization. In order to make the estimation process have better stability and higher accuracy, temperature control and real-time monitoring are integrated into the parameter identification and data collation process, and the problems of battery SOC estimation stability and calculation time under neural network are considered.

[0046] A three-layer feedback neural network is constructed, and the test voltage and current are transmitted to the network in real time as the input layer for training. The estimated SOC value is the output layer of the network, and the hidden layer will be determined by formula (1).

[0047] (1)

[0048] Among them, m and n represent the number of input layers and output layers respectively. represents a random number between the interval [1,9]. In this embodiment, the traversal method is used to finally determine it to be 3. The network can meet the test requirements and maintain high stability. The feedback network consists of two parts, including feedforward and error feedback. Set any neuron N j [m] The corresponding input matrix is ​​(x1,x2,…x n ) T , get the linear input z j [m] =w1x1+w2x2+w3x3+…+w n x n +b j [m] , where W1 represents the weight connected to the first input X1, and so on, W1, W2, ... W n Represents the transition from the first input X1 to X n The corresponding weight, m ​​represents the layer. In this design, there are only input layer, hidden layer and output layer, so the value of m is only 1 and 2. It is used to represent the weight between the input layer and the hidden layer (m=1), and between the hidden layer and the output layer (m=2). Output y j [m] =f(*), where f is the activation function, is the input of the feedback neural network, is the connection weight of the ith neuron to the jth neuron in the mth layer, is the input of the ith neuron to the jth neuron in the mth layer, is the bias of the jth neuron in the mth layer, is a summary of the weight of the jth neuron, is the overall representation of the jth neuron input, is the overall representation of the jth neuron bias. The input of this system can be determined as:

[0049] (2)

[0050] The corresponding output can be expressed as:

[0051] (3)

[0052] in, is the output of the feedback neural network, is the activation function from hidden layer to output layer, is the weight of the ith neuron in the mth layer to the jth neuron, is the input of the i-th neuron in the m-th layer to the j-th neuron, For multiplication, is the expression of the weight and input of the jth neuron under the activation function.

[0053] For the error feedback process, the important idea is reflected in the gradient descent method, which transmits the error back to the hidden layer and input layer layer by layer, and sets the neurons of the input layer, hidden layer and output layer to be X i (n),H j (n),Y k (n), the connection weight of the i-th neuron in the j-th layer can be expressed as W ji 1 .It can be divided into the following steps:

[0054] Step 1: Determine the output of the hidden layer and the input layer according to formula (4). i It represents input, indicating that any input from the first input to the nth input is represented by i. Represents the bias of each neuron in the hidden layer, and any bias from the first to the nth is represented by j. The W in represents the weight between the hidden layer and the output layer. The k and j in represent the correction from the jth neuron in the hidden layer to the kth neuron in the output layer, and 2 represents the second layer between the hidden layer and the output layer. correspond, Represents the bias of each neuron in the output layer, and any bias from the first to the nth is represented by k. Represents the output of the hidden layer neurons, where H represents the hidden layer neurons, j represents the jth, and n represents the nth update. represents the output of the output layer, Y represents the output layer, k represents the kth neuron in the output layer, represents the connection weight W from the i-th neuron in the input layer to the j-th neuron in the hidden layer.

[0055] (4)

[0056] Step 2: Determine the error value according to formula (5). The error value E is defined as half of the sum of squared errors. The calculation process is as follows. After obtaining the error value, the error value is used to find the partial derivative of the corresponding weight, and the error between the output layer and the expected value and the partial derivative of the weight are obtained by simplification and parameter substitution, which is converted into a functional relationship with the weight output of each layer. and represent the true value and the estimated value respectively. It represents the square of the error generated by the input from the jth neuron in the hidden layer to the output layer and the output from the output layer. E represents error. It represents the calculation of 1 / 2 of the sum of the errors output by the output layer after the j-th neuron in the hidden layer acts on each neuron in the output layer, and m represents the m-th layer.

[0057] (5)

[0058] Step 3: Calculate the error function of the hidden layer and the input layer, as shown in formula (6). The error can be directly obtained through the arithmetic relationship between the system expected value and the actual output of the output layer, but the error function is obtained by taking the partial derivative of the error and its corresponding weight, and through substitution, it is transformed into the relationship expression obtained by the error and the output of the input layer and the hidden layer, which becomes the error function. Represents the error of the kth neuron in the output layer. * represents multiplication. Represents the output of the kth neuron after weight update. represents the output of the jth neuron in the hidden layer.

[0059] (6)

[0060] Step 4: Combined with the learning step size, update the weights according to formula (7).

[0061] (7)

[0062] in, is the update of the m+1th connection weight of the kth neuron to the jth neuron in the hidden layer, is the mth connection weight update of the kth neuron to the jth neuron in the hidden layer, is the m-th update amount of the j-th neuron to the k-th neuron in the hidden layer, is the error of the kth neuron with error coefficient, is the output of the k-th neuron output layer, is the hidden layer output of the jth neuron, is the m+1th connection weight update of the i-th neuron in the input layer to the j-th neuron, Update the mth connection weight of the i-th neuron in the input layer to the j-th neuron. is the weight update amount of the i-th neuron in the input layer to the j-th neuron, is the mth connection weight update of the i-th neuron in the hidden layer to the j-th neuron, To distinguish it from the error coefficient of the hidden layer, is the output of the jth neuron in the hidden layer, is the connection weight of the jth neuron in the hidden layer to the kth neuron, is the input summary representation of the i-th neuron.

[0063] For the chaotic particle swarm algorithm, the position and direction of the i-th particle after the m-th iteration can be expressed as X i m =(X i 1 X i 2 …X i N ) and V i m =(V i 1 V i 2 …V i N ), while the position component and velocity component can be expressed as:

[0064] (8)

[0065] Represents the inertia weight with adaptive force, p is used to distinguish it from the connection weight. In the equation, n represents the nth time that the adaptive inertia weight is introduced to achieve dynamic convergence of the weight, i represents the i-th particle, j represents the j-th dimension, and V represents the velocity component. Represents a random number, with a value range between (0, 1). and Represents a constant that satisfies uniform distribution, with a value between (0, 1) and a subscript corresponding to that in V. Where G represents the best particle in the particle swarm. Where p represents the individual optimality, n represents the nth time, j represents the jth dimension, Where X represents the position, n represents the nth time, i represents the i-th particle, and j represents the j-th dimension. represents uniform distribution, the interval is (0, 1), and the parameter σ can control the randomization of particles to prevent particles from missing the global optimum during movement. The definition of particle motion convergence is different from that of random optimization algorithms, because the process of updating from a single particle is random. When all particles transmit information about position and velocity to each other, they need to correspond to the individual optimum and the global optimum. In the chaotic particle swarm algorithm, the velocity contains random components and drift components, and the weight part is weakened. Its expression with paranoid components is shown in formula (9).

[0066] (9)

[0067] In the equation, V represents the velocity component of the ith particle after the n-1th update in the Kth dimension. , Represents two random constants after the (n-1)th update of the ith particle in the kth dimension, with values ​​between (0, 1). The subscripts are the same as in V, and X represents the position component. Where ibest and gbest represent individual optimum and global optimum respectively. The adaptive inertia weight can help complete the weakened part, improve the convergence time of particles when approaching individual optimum, and improve the search ability and search range of particles in finding global optimum. The expression is shown in formula (10).

[0068] (10)

[0069] in, Represents the weight of the directional parameter σ, σ is adjusted in the interval [0.2, 0.5], * represents multiplication, W represents weight, r represents component, and is distinguished from connection weight. is a random variable that follows a uniform distribution, with parameter n i,n-1 k is a random variable that follows a normal distribution. It represents the difference between the maximum and minimum weight of the i-th particle in the n-1th update of the k-th dimension. Respectively represent the maximum and minimum values ​​of the weight with directional parameters, r represents the component, max represents the maximum, and min represents the minimum. Therefore, the particle velocity expression can be updated as shown in formula (11).

[0070] (11)

[0071] represents the velocity component of particle i in the k-th dimension of the n-1th order, V represents the velocity, Where D represents the chaotic component, i represents the i-th particle, n-1 represents the n-1-th update, k represents the k-th dimension, and σ is a parameter with direction guidance. is the weight, r is the component, and W is the weight. The particle swarm algorithm cannot generate new populations, and the utilization of data is not improved. To address this problem, the data is preprocessed and a validation set is set. While inputting voltage and current in real time, cross-validation is performed to complete the adjustment of parameters before the final output result is generated, thereby improving efficiency. In order to further improve the system's matching of different data, a constant temperature system is built, and a charge and discharge test platform is constructed at the same time to complete the training of parameters under different temperatures and working conditions, further improving the stability and matching of the network architecture.

[0072] In the multi-temperature lithium-ion battery pack SOC estimation process, the algorithm flow optimization process of the lithium battery SOC estimation method based on the chaotic particle swarm feedback optimization algorithm with adaptive inertia weight is as follows Figure 2As shown in the figure: the weights and thresholds in the feedback network are optimized by using the chaotic particle swarm algorithm. The details are as follows: input data, initialize the particles, and use the mean square error of the particle swarm as the fitness value of the neural network to establish a mapping relationship between the network and the particle algorithm; the particles search for the optimal value, which is divided into individual optimal and group optimal. The optimal value is continuously updated by updating the learning step size and fitness. Finally, it is judged whether the optimization conditions are met. If so, the optimal value is passed to the neural network, and the best quality is determined as the weight and threshold of the neural network. The error is calculated and updated. After the conditions are met, the optimized result is output. If the conditions are not met, the step size is continued to be updated, and the particles are constantly approaching the best quality to find the optimal value and make a new round of judgment.

[0073] The feedback neural network under the weighted particle swarm algorithm is determined through the above series of formulas to obtain the optimal estimated SOC value of the system state variable; this method is based on the multi-temperature lithium battery SOC estimation process of the chaotic particle swarm feedback optimization algorithm with adaptive inertia weight, and combines grid traversal and random search to optimize the hyperparameters of data and genetic parameters, thereby realizing the construction of the SOC estimation model for lithium-ion batteries; the accuracy of the SOC estimation value is effectively improved.

[0074] In summary, this embodiment aims at the goal of accurate SOC estimation of lithium-ion batteries in a high-stability battery system of a petroleum base station, comprehensively considers the estimation accuracy, computational complexity and stability of the algorithm, and proposes a multi-temperature lithium battery SOC estimation method based on a chaotic particle swarm feedback optimization algorithm with adaptive inertia weights. On the basis of fully considering the estimation of lithium-ion battery data in a complex environment in the neural network framework, combined with the establishment of an SOC estimation model, high-precision optimization of the SOC estimation of lithium-ion batteries is achieved, providing ideas for the SOC estimation of lithium-ion battery packs and the construction of more neural network algorithms and models. The above embodiments of this embodiment only take the lithium-ion battery in the high-stability battery system of a petroleum base station as an example to illustrate the chaotic particle swarm feedback optimization algorithm based on adaptive inertia weights for accurate SOC estimation of multi-temperature lithium batteries.

[0075] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization, characterized in that: include: The preset feedback neural network is trained and optimized using a chaotic particle swarm algorithm with adaptive inertia weight to obtain a SOC estimation model for lithium-ion batteries. The expression of the adaptive inertia weight is: in, represents the weight of the directional parameter σ, and the parameter r i,n-1 k is a random variable that follows a uniform distribution, with parameter n i,n-1 k is a random variable that follows a normal distribution, is the previous weight coefficient of the ith particle in the kth iteration, is the minimum value within the weight correction range in the random component, It is the maximum value within the weight correction range of the random component; Input the real-time voltage and current into the SOC estimation model to obtain the SOC value of the lithium-ion battery; The training and optimization of the preset feedback neural network using the chaotic particle swarm algorithm with adaptive inertia weights includes: Taking the test voltage and current as input and the estimated SOC value as output, training the preset feedback neural network to obtain error feedback of the preset feedback neural network; When training the preset feedback neural network, it also includes: Preset the validation set and perform cross validation while inputting voltage and current; Using a chaotic particle swarm algorithm with adaptive inertia weights, the preset feedback neural network is updated in weight, and the weights are dynamically converged to correct the error feedback; The method for updating the weight of the preset feedback neural network is: in, is the update of the m+1th connection weight of the kth neuron to the jth neuron in the hidden layer, is the mth connection weight update of the kth neuron to the jth neuron in the hidden layer, is the m-th update amount of the j-th neuron to the k-th neuron in the hidden layer, is the error of the kth neuron with error coefficient, is the output of the k-th neuron output layer, is the hidden layer output of the jth neuron, is the m+1th connection weight update of the i-th neuron in the input layer to the j-th neuron, Update the mth connection weight of the i-th neuron in the input layer to the j-th neuron. is the weight update amount of the i-th neuron in the input layer to the j-th neuron, is the mth connection weight update of the i-th neuron in the hidden layer to the j-th neuron, To distinguish it from the error coefficient of the hidden layer, is the output of the jth neuron in the hidden layer, is the connection weight of the jth neuron in the hidden layer to the kth neuron, is the input summary representation of the i-th neuron; The modified preset feedback neural network is regarded as the SOC estimation model; Obtaining the preset error feedback of the feedback neural network includes: Obtaining the outputs of the preset hidden layer and input layer of the feedback neural network; Based on the outputs of the hidden layer and the input layer, obtaining an error value between the output of the output layer and the expected value; Based on the error value, obtaining error functions of the hidden layer and the input layer; The velocity expression for this particle is updated as follows: in, represents the random constant of the ith particle after the n-1th update in the kth dimension, represents the velocity component of particle i in the k-th dimension of the n-1th order, V represents the velocity, Where D represents the chaotic component, i represents the i-th particle, n-1 represents the n-1-th update, and k represents the k-th dimension; In the SOC estimation process of multi-temperature lithium-ion battery packs, the chaotic particle swarm algorithm is used to optimize the weights and thresholds in the feedback network. The details are as follows: input data, initialize the particles, and use the mean square error of the particle swarm as the fitness value of the neural network to establish a mapping relationship between the network and the particle algorithm; the particles search for the optimal value, which is divided into individual optimal and group optimal. The optimal value is continuously updated by updating the learning step size and fitness. Finally, it is judged whether the optimization conditions are met. If so, the optimal value is passed to the neural network, and the best quality is determined as the weight and threshold of the neural network. The error is calculated and updated. After the conditions are met, the optimized result is output. If the conditions are not met, the step size is continued to be updated, and the particles are constantly approaching the best quality to find the best, and a new round of judgment is carried out.

2. The multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization according to claim 1 is characterized in that: The preset feedback neural network includes: an input layer, a hidden layer and an output layer; The input layer is: in, is the input of the feedback neural network, is the connection weight of the ith neuron to the jth neuron in the mth layer, is the input of the ith neuron to the jth neuron in the mth layer, is the bias of the jth neuron in the mth layer, is a summary of the weight of the jth neuron, is the overall representation of the jth neuron input, is the overall representation of the bias of the jth neuron; The hidden layer is: in, is the number of neurons in the hidden layer, m and n represent the number of neurons in the input layer and output layer respectively. Represents a random number between the interval [1,9]; The output layer is: in, is the output of the feedback neural network, is the activation function from hidden layer to output layer, is the weight of the ith neuron in the mth layer to the jth neuron, is the input of the i-th neuron in the m-th layer to the j-th neuron, For multiplication, is the expression of the weight and input of the jth neuron under the activation function.

3. The multi-temperature lithium battery SOC estimation method based on chaotic particle swarm feedback optimization according to claim 1 is characterized in that: When training the preset feedback neural network, it also includes: Build a constant temperature control system, and use the BMS battery management system to control the temperature of the lithium battery test by the preset feedback neural network under the chaotic particle algorithm; A charge and discharge test management system is built to input the voltage and current under different working conditions into the optimized feedback neural network in real time.