CN115792625A - Lithium battery SOC estimation method, system, equipment and medium based on neural network - Google Patents
Lithium battery SOC estimation method, system, equipment and medium based on neural network Download PDFInfo
- Publication number
- CN115792625A CN115792625A CN202211372630.2A CN202211372630A CN115792625A CN 115792625 A CN115792625 A CN 115792625A CN 202211372630 A CN202211372630 A CN 202211372630A CN 115792625 A CN115792625 A CN 115792625A
- Authority
- CN
- China
- Prior art keywords
- value
- neural network
- soc
- lithium battery
- variance matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/10—Energy storage using batteries
Landscapes
- Tests Of Electric Status Of Batteries (AREA)
Abstract
The invention belongs to the technical field of lithium battery SOC estimation, and particularly relates to a lithium battery SOC estimation method, a lithium battery SOC estimation system, lithium battery SOC estimation equipment and a lithium battery SOC estimation medium based on a neural network. In conclusion, on the basis of the UKF algorithm, the solution that the ABP corrects the error of the UKF algorithm for estimating the SOC is provided, and the method improves the training iterative process of compensation correction of the ABP neural network by taking the unscented Kalman as the reference on the basis of the SO-PEC model on the basis of fully considering the grouping work of the lithium ion battery, and realizes the establishment of the SOC estimation model of the lithium ion battery and the reliable operation of the mathematical iterative algorithm of the SOC value.
Description
Technical Field
The invention belongs to the technical field of lithium battery SOC estimation, and particularly relates to a lithium battery SOC estimation method, system, equipment and medium based on a neural network.
Background
Energy safety and environmental protection play an important role in development and planning of China, and the aim for new energy to replace the traditional fossil fuel has become the key point of attention of all countries. Lithium ion batteries (hereinafter referred to as lithium batteries) have the advantages of high energy density, long service life, large output power, high cost performance and the like, so that the lithium ion batteries are widely applied and developed in the field of new energy, and have an important position in the field of renewable energy with the advantages of environmental protection. Under the condition that the application of lithium batteries in the field of new energy is more and more extensive, the health state detection of the lithium batteries is also more and more emphasized. Whether the State of Charge (SOC) of the lithium battery can be accurately estimated has important significance in fully exerting the performance of the battery and realizing real-time State detection and safety control on the lithium battery.
In the whole life cycle of the lithium Battery pack, the monitoring and adjustment of a Battery Management System (BMS) on a core parameter SOC will affect the effect and safety of emergency power output; therefore, how to accurately monitor this parameter is a critical loop. Reliable SOC value estimation can help the BMS system to carry out energy management better, avoids causing the emergence of phenomena such as lithium cell thermal runaway, cycle life sudden decrease. The accurate estimation of the SOC value is the key to guarantee the working performance of the battery pack, prolong the cycle life and reduce the use cost. However, SOC is a quantity that cannot be directly measured, and can only be indirectly estimated by a direct measurement value such as current and voltage and a definitional formula, and this process is affected by various factors such as battery aging, current measurement accuracy, SOC estimation initial value, and so on, thereby bringing great difficulty to the accurate estimation of SOC value.
At present, because a grouped SOC estimation technology in a BMS is not mature, potential safety hazards existing in the use process severely restrict the development of a lithium battery pack; for lithium battery packs, reliable BMS management relies on accurate SOC values; under the condition that the value is known, not only is reliable energy management and safety control carried out on the lithium battery pack, but also the lithium battery pack is prevented from being damaged in advance, and the service life of the lithium battery pack is prolonged; therefore, the SOC value is accurately estimated, and the method is very important for guaranteeing the working performance, energy and safety management of the lithium battery pack; the SOC estimation model construction and accurate estimation value of the lithium battery pack are worth obtaining and become the core problem of energy and safety management of the lithium battery pack; the lithium battery pack is formed by combining lithium cobaltate battery monomers with high energy density and closed circuit voltage, and the safety of the lithium battery pack is influenced by the working state of the lithium battery pack; in addition, the charge-discharge process of the lithium battery pack comprises the links of complicated electric energy, chemical energy, heat energy conversion and the like, the phenomena of overcharge and overdischarge easily cause safety accidents, and the accurate SOC estimation plays an important role in preventing overcharge and overdischarge; in the application of the lithium battery pack, the safety of the lithium battery pack is still the most concerned, and the SOC estimation is the basis and the premise for safe use of the lithium battery pack; the lithium battery pack adopts a battery monomer cascade structure, and meets the capacity and voltage requirements in the energy supply process of the auxiliary power; however, due to unavoidable material and process differences, the phenomenon of inconsistency between monomers is objective and unavoidable; moreover, the phenomenon becomes more and more obvious along with the increase of the cycle number, so that the expression and correction of the inconsistency among the monomers become an important component of the estimation of the SOC group, and meanwhile, great challenges are brought to the accurate estimation of the SOC group.
In view of the importance of obtaining the SOC value for battery management, many scholars have studied SOC estimation of lithium batteries at home and abroad. The lithium battery is used as a system with high nonlinear working characteristics, and the acquisition and modeling of internal time-varying parameters are important factors influencing the accurate working characteristic characterization and the accurate state of charge estimation of the lithium battery. To effectively establish the state space expression of the lithium battery, an equivalent model with high adaptability needs to be established. Meanwhile, to ensure accurate and real-time estimation of the state of charge of the lithium battery, accurate acquisition of state space model parameters of the lithium battery under complex working conditions must be ensured, which aggravates the difficulty of estimation of the SOC of the lithium battery. With respect to the necessity and urgent need of SOC estimation, a great deal of research and intensive research has been conducted on SOC estimation in relevant research institutes and universities, such as the massachusetts institute of technology, state university of bingzhou, southern card university of usa, litz university of uk, robert university of uk, national renewable energy house of usa, leideng energy company, germany english-flying-technologies company, qinghua university, beijing aerospace university, beijing university of rationale, beijing university of transportation, chongqing university, china university of scientific technology, and harbin university of industry; many periodicals at home and abroad, such as Journal of Power Sources, applied Energy, IEEE Transactions on Power Systems, power technology and the like, establish highly targeted columns for relevant research result display; aiming at the problem of SOC estimation of lithium batteries, relevant research workers at home and abroad make a great research progress at present; as described in Hu et al, there are currently mainly an Ampere-hour integration method (Ah), an Open Circuit Voltage method (OCV), kalman filtering and its extended algorithm, a Particle Filter method (PF), a Neural Network method (NN), and the like; due to the influence of various factors such as charging and discharging current, temperature, internal resistance, self-discharge, aging and the like, the performance change of the lithium battery can obviously influence the SOC estimation precision, and a universal method for realizing the accurate estimation of the SOC value is not available; in addition, the consistency among the monomers in the grouping working process is influenced, and the lithium battery pack still lacks an effective SOC estimation method; at present, SOC estimation in practical application is realized by a basic ampere-hour integration method, but the estimation error is large, and the accumulation effect is obvious under the influence of a plurality of factors; aiming at the SOC estimation research of the lithium battery pack, the related research provides thought reference; on the basis, SOC estimation methods under various working conditions are explored, and effective SOC estimation of the lithium battery pack is achieved; an SOC estimation model with parameter correction and regulation capabilities is built, a parameter estimation theory based on an equivalent circuit model is applied, the trend of SOC estimation development is formed, an optimal balance point is found between the improvement of precision and the reduction of calculated quantity, and an estimation method is continuously optimized and improved.
In the conventional BMS application of the lithium battery pack, an SOC estimation method based on ampere-hour integration and open-circuit voltage cannot accurately represent accumulated errors existing in SOC estimation and cannot be combined with the current state for parameter correction; meanwhile, the traditional filtering algorithm has considerable limitation on the estimation of the relevant state of the object of the nonlinear working characteristic of the lithium battery. In view of the above problems, the present invention provides a lithium battery SOC estimation model based on adaptive back propagation neural network-unscented kalman filter (ABP-UKF). The UKF algorithm is used as a nonlinear system estimation method, abandons the idea of nonlinear function approximate linearization, however, in the later stage of SOC estimation, the estimation result is easy to deviate from the reference value. Therefore, in consideration of the robust learning of the neural network algorithm on data processing, a proper Back Propagation (BP) neural network algorithm is selected for error compensation of the UKF algorithm to perform compensation correction. The BP neural network is a multilayer feedforward network trained according to an error back propagation algorithm, and although the BP neural network has strong self-learning capability when processing large-scale data, the obtained network has poor performance and unstable learning rate. In conclusion, on the basis of the UKF algorithm, a solution scheme that ABP corrects errors of the UKF algorithm for estimating the SOC is provided, and the construction and experimental verification of the SOC estimation model are realized.
Disclosure of Invention
In order to solve the technical problems in the prior art, the invention provides a lithium battery SOC estimation method, system, device and medium based on a neural network, aiming at improving the lithium battery SOC estimation precision.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a lithium battery SOC estimation method based on a neural network comprises the following steps:
step 1: acquiring sampling data at the moment of k +1, and inputting the acquired sampling data into a UKF algorithm to obtain a measurement variance matrix at the moment of k +1 and a variance matrix of a state and a measured value; obtaining the SOC value of the lithium battery based on a UKF algorithm;
step 2: updating a system state variable value and an error variance matrix based on the obtained measurement variance matrix and the variance matrix of the state and the measured value;
and step 3: taking a system state variable and an error variance matrix as the input of a trained BP neural network, and obtaining a correction value of an SOC value based on the BP neural network;
and 4, step 4: and adding the corrected value of the SOC value and the SOC value of the lithium battery obtained by the UKF algorithm to obtain an SOC estimated value.
On the basis of the traditional unscented Kalman filtering, the SOC value obtained by the unscented Kalman filtering is corrected through a BP neural network, so that the estimation precision of the SOC value is ensured.
Further, the specific steps of obtaining the measurement variance matrix and the variance matrix of the state and the measurement values are as follows:
A. predicting a system state variable and an error variance matrix at the k +1 moment based on a UKF algorithm, and simultaneously obtaining a prediction mean value of the system state variable, wherein the specific calculation mode is as follows:
in the formula: i =1,2,3.. 2n +1;
is the system state change at time k +1The amount of the sample is predicted,
is a sampling point one-step prediction value of the UKF algorithm,
is that
Corresponding weight, P xx,k+1|k Is the state of charge estimation error variance matrix at time k +1,
weight of the predicted value joint covariance of x, Q k+1 Is gaussian noise;
B. updating the average weight of the measured value at the moment of k +1 and the measured variance matrix to obtain the variance matrix of the state and the measured value at the moment of k + 1:
in the formula:
is a predicted value for each of the sampling points,
is that
And is an input variable, P yy,k+1 Is the average weight at time k +1 and the variance matrix of the measurements; p xy,k+1 Is the variance matrix of the state and measurements at time k +1; r k+1 Is the noise at time k + 1.
Further, the step 2 comprises the following steps:
step 2.1: obtaining a Kalman gain based on the measurement variance matrix at the moment of k +1 and the variance matrix of the state and the measurement value:
step 2.2: updating a system state variable value and an error variance matrix based on Kalman gain:
in the formula:
is a value of a system state variable at time K +1 based on Kalman gain, K k+1 Is the Kalman gain, y k+1 Is the measured value at the time k +1, P xx,k+1|k+1 Is a system state variable (state of charge estimation) error variance matrix at time k +1 based on Kalman gain;
further, the training step of the BP neural network is as follows:
A. acquiring historical sampling data, and constructing a training data set based on the historical sampling data;
B. calculating a training data set based on a UKF algorithm to obtain a Kalman filter gain matrix and a state of charge estimation matrix error;
C. and training the BP neural network based on the obtained Kalman filter gain matrix and the charge state estimation matrix error as an input layer of the BP neural network, and correcting the weight and the threshold of the BP neural network model based on the training result to finally obtain the trained BP neural network model.
Further, the threshold and the activation function of the current node of the BP neural network are as follows:
in the formula: omega ej Represents the weight between e and node j; b j Threshold value representing node j；f(S j ) Is an activation function; a is j Representing the output value of node j.
Further, the error function of the BP neural network is established based on a gradient descent method, which is specifically as follows:
in the formula: e (ω, b) refers to the error function of the BP neural network with respect to the weights and the threshold;
is to make a partial derivative of it, d j Representing the results of the output layer; y is j The compensation correction value of the UKF algorithm is represented; f (S) j ) Is an activation function;
obtaining a weight between the hidden layer and the output layer according to an error function:
in the formula: omega ej And representing the weight between e and the node j, namely obtaining the update of the weight between the hidden layer and the output layer, and l represents the learning rate.
Further, the weight optimization method of the BP neural network is as follows:
in the formula: omega ej (k + 1) is a weight optimization expression of the BP neural network at the next time, ω ej (k) For the weight optimization expression at this time, mu a Is the learning rate factor rise factor, mu b To learn the rate factor reduction coefficient, E (k + 1) and E (k) are custom comparison values.
A lithium battery SOC estimation system based on a neural network comprises:
the data acquisition module is used for acquiring parameters of the equivalent circuit;
the first calculation module calls a UKF algorithm stored in a database, and substitutes the obtained equivalent circuit parameters into the UKF algorithm for calculation to obtain a measurement variance matrix, a variance matrix of a state measurement value and an SOC value of the lithium battery;
the second calculation module is used for calling the trained BP neural network model stored in the database, taking the measurement variance matrix and the variance matrix of the state measurement value calculated by the first calculation module as the input layer of the BP neural network model, and calculating through the BP neural network model to obtain the correction value of the SOC value;
and the third calculation module is used for adding the SOC value of the lithium battery calculated by the first calculation module and the corrected value of the SOC calculated by the second calculation module to obtain an SOC estimated value.
An electronic device comprising a memory and a processor, the memory having stored therein a computer program for execution on the processor, the computer program, when executed on the processor, performing the method of the invention.
A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method of the invention.
The beneficial effects of the invention include:
1. on the basis of the traditional unscented Kalman filtering, the SOC value obtained by the unscented Kalman filtering is corrected through a BP neural network, so that the estimation precision of the SOC value is ensured.
2. The method is mainly used for realizing the accurate real-time monitoring of the battery management system and solving the SOC estimation of the lithium battery pack; the problem that the estimated value deviates from the mean value when the UKF algorithm estimates the SOC can be solved by adjusting the adaptive learning rate through the algorithm; and the effective iterative calculation of the SOC value of the lithium battery pack is realized.
3. The method abandons the traditional filtering thought, designs a high-precision SOC estimation model by utilizing the thought of the traditional filtering and neural network algorithm, adjusts the connection weight and the threshold value between each layer of the network structure by utilizing the error back propagation in the network training process according to the error data of the previous iteration until the error converges below a set value, and performs self-adaptive processing on the learning rate at the same time, thereby increasing the practicability of the method.
4. The method can provide method reference for establishing lithium battery SOC estimation models and calculating SOC values under different application scenes aiming at the experimental analysis of the power application requirements and the working characteristics of the lithium battery, and has the advantages of simplicity in calculation, good adaptability and high precision.
Drawings
Fig. 1 is a schematic structural diagram of an SOC estimation model of a lithium battery pack according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. The components of the embodiments of the present application, as generally described and illustrated in the figures herein, could be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, as presented in the figures, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present application without making any creative effort, shall fall within the protection scope of the present application.
The invention is described in further detail below with reference to the accompanying figure 1:
a lithium battery SOC estimation method based on a neural network comprises the following steps:
accurate lithium ion battery modeling is a premise for accurately estimating the SOC, and the working condition of the lithium ion battery can be well represented by adopting a second-order polarization equivalent circuit (SO-PEC) model; the training network is the key of SOC estimation, and the estimation error of the algorithm is corrected by using the ABP-UKF algorithm and the related training of the UKF algorithm estimation result, so that the estimation efficiency of the SOC is improved; the final SOC estimation is a verification of the proposed algorithm; the iteration and training of the data are the important core logic and content of the invention.
According to the method, on the basis of fully considering various complex working conditions of the lithium ion battery, the establishment of an equivalent model of the lithium ion battery is accurately represented, the mathematical expression of a state space equation is obtained, and a lithium battery SOC estimation model of a self-adaptive back propagation neural network-unscented Kalman filtering is constructed. As will be well known to those skilled in the art, the adaptive back propagation neural network-unscented Kalman filter SOC estimation of various lithium ion battery packs can be realized according to the technical idea of the invention;
the invention relates to a model establishing thought for performing ABP neural network reprocessing on an estimation result by taking a UKF algorithm (unscented Kalman filtering) as a reference. Among them, the key of the UKF algorithm is the problem of unscented transformation. The basic principle is that according to the statistical characteristics of state variables, a fixed sampling method (usually symmetric sampling) is adopted to obtain a certain number of sampling points, the sampling points need to have the same mean value and covariance with the original state variables, the point set uses a state equation to carry out nonlinear transfer, the transformed mean value and covariance can be obtained according to weight distribution, and the value is shown as the following formula:
in the formula:
the transformed mean and covariance, α, β, λ are scaling parameters; kappa is a secondary scale factor; four parameters are typically defined as: beta is more than or equal to 0,0.2 is more than or equal to alpha is less than or equal to 1, kappa =3-n, and lambda = alpha 2 (n + κ) -n. Because the nonlinear mapping of the BP neural network has strong learning capacity, the filtering precision of the SOC estimation of the lithium ion battery can be improved by adding the BP neural network, and the problem of overlarge estimation error of the UKF algorithm is solved. In the BP neural network algorithm, the learning rate eta is a fixed value, but in the whole lithium ion battery state estimation process, the target of compensating errors needs to be formedThe filtering precision can be effectively improved only if the filtering precision is constantly changed; the central idea of the self-adaptive BP neural network algorithm for compensating errors in the invention is as follows: changing the value of the learning rate eta into a variable for selecting a learning factor according to the learning rate; the central idea of the adaptive BP neural network algorithm for correcting errors is to use the value of the learning rate as a variable of the learning factor, as shown in the following formula:
in the formula: μ is the learning rate; μ (k + 1) is the value of the adaptive learning rate, hereinafter in μ a Mu as the learning rate factor rise coefficient b Is a learning rate factor reduction factor; k is a radical of a Is an ascending learning rate parameter; k is a radical of b Is a falling learning rate parameter; μ (n) is the value of the original learning rate.
And integrating the advantages of the two algorithms to establish an SOC estimation model of the lithium ion battery. The SOC estimation model of the lithium ion battery is established by taking the SOC as a variable in a state equation of the SOC and outputting closed-circuit voltage as a variable of an observation equation, and the SOC estimation model is specifically as follows:
step 1: acquiring sampling data at the moment of k +1, and inputting the acquired sampling data into a UKF algorithm to obtain a measurement variance matrix at the moment of k +1 and a variance matrix of a state and a measured value; obtaining the SOC value of the lithium battery based on a UKF algorithm;
the specific steps for obtaining the measurement variance matrix and the variance matrix of the state and measurement values are as follows:
A. predicting a system state variable and an error variance matrix at the k +1 moment based on a UKF algorithm, and simultaneously obtaining a prediction mean value of the system state variable, wherein the specific calculation mode is as follows:
in the formula: i =1,2,3.. 2n +1;
is the predicted value of the system state variable at the moment k +1,
is a sampling point one-step prediction value of the UKF algorithm,
is that
Corresponding weight, P xx,k+1|k Is the state of charge estimation error variance matrix at time k +1,
weight of the predicted value joint covariance of x, Q k+1 Is gaussian noise;
B. updating the average weight of the measured value at the moment k +1 and the measured variance matrix to obtain the state at the moment k +1 and the variance matrix of the measured value:
in the formula:
is a predicted value for each of the sampling points,
is that
And is an input variable, P yy,k+1 Is the mean weight at time k +1 and the variance matrix of the measurements; p is xy,k+1 Is the variance matrix of the state and measurements at time k +1; r k+1 Is the noise at time k + 1.
And 2, step: updating a system state variable value and an error variance matrix based on the obtained measurement variance matrix and the variance matrix of the state and the measured value;
the step 2 comprises the following steps:
step 2.1: obtaining a Kalman gain based on the measurement variance matrix at the moment of k +1 and the variance matrix of the state and the measurement value:
step 2.2: updating a system state variable value and an error variance matrix based on Kalman gain:
in the formula:
is a value of a system state variable at time K +1 based on Kalman gain, K k+1 Is Kalman gain, y k+1 Is the measured value at time k +1, P xx,k+1|k+1 Is a system state variable (state of charge estimation) error variance matrix at time k +1 based on Kalman gain;
and step 3: taking the system state variable and the error variance matrix as the input of a trained BP neural network, and obtaining a corrected value of the SOC value based on the BP neural network;
the training steps of the BP neural network are as follows:
A. acquiring historical sampling data, and constructing a training data set based on the historical sampling data;
B. calculating a training data set based on a UKF algorithm to obtain a Kalman filter gain matrix and a charge state estimation matrix error;
C. and training the BP neural network based on the obtained Kalman filter gain matrix and the charge state estimation matrix error as an input layer of the BP neural network, and correcting the weight and the threshold of the BP neural network model based on the training result to finally obtain the trained BP neural network model.
The threshold and the activation function of the current node of the BP neural network are as follows:
in the formula: omega ej Represents the weight between e and node j; b j A threshold value representing node j; f (S) j ) Is an activation function; a is j Representing the output value of node j.
The error function of the BP neural network is established based on a gradient descent method, which is specifically as follows:
in the formula: e (ω, b) refers to the error function of the BP neural network with respect to the weights and the threshold;
is to make a partial derivation of it, d j Representing the results of the output layer; y is j The compensation correction value of the UKF algorithm is represented; f (S) j ) Is an activation function;
obtaining a weight between the hidden layer and the output layer according to an error function:
in the formula: omega ej And (3) representing the weight between e and the node j, namely, obtaining the weight update between the hidden layer and the output layer, and l represents the learning rate.
The weight optimization method of the BP neural network is as follows:
in the formula: omega ej (k + 1) is the weight of the BP neural network at the next time instantReoptimization expression, ω ej (k) For the weight optimization expression at this time, mu a Is the learning rate factor rise factor, mu b For learning rate factor reduction factor, E (k + 1) and E (k) are custom comparison values.
And 4, step 4: adding the corrected value of the SOC value and the SOC value of the lithium battery obtained by the UKF algorithm to obtain an SOC estimated value:
SOC ABP-UKF ＝SOC UKF +SOC Err ；
SOC ABP-UKF is the estimated value obtained, SOC UKF Is an estimate of the UKF algorithm; SOC Err Is an error value obtained after the ABP neural network training, also called a correction value. When the SOC of the lithium ion battery is estimated by the ABP-UKF algorithm, the SOC of a target estimator is used as a state variable, and the terminal voltage of the battery is used as an observation variable. In the later estimation process, the compensation effect of the ABP algorithm on the UKF algorithm has an obvious optimization effect on the final estimation result, and the estimation precision of the SOC of the lithium ion battery in the battery management system is improved to a considerable extent.
In the SOC estimation process of the lithium ion battery, the obtained SOC value can be obtained by iteration through the series of formulas. The method is based on the unscented Kalman algorithm as a reference, the compensation correction is carried out by the ABP algorithm, and the construction of the SOC estimation model of the lithium ion battery pack is realized through the iterative computation process.
In conclusion, the invention provides a lithium battery SOC estimation model of a self-adaptive back propagation neural network-unscented Kalman filtering aiming at an accurate SOC estimation target of a lithium battery pack and comprehensively considering estimation precision, calculation complexity and algorithm stability. Meanwhile, under the condition of fully considering different complex working conditions of the lithium ion battery, the iterative computation of the SOC estimation of the lithium ion battery pack is realized by combining the establishment of an SOC estimation model, and a basis is provided for the SOC estimation and the real-time monitoring of the working state of the lithium ion battery pack in a battery management system.
On the basis of the traditional unscented Kalman filtering, the SOC value obtained by the unscented Kalman filtering is corrected through a BP neural network, so that the estimation precision of the SOC value is ensured.
A lithium battery SOC estimation system based on a neural network comprises:
the data acquisition module is used for acquiring parameters of the second-order RC equivalent circuit;
the first calculation module calls a UKF algorithm stored in a database, and substitutes the obtained equivalent circuit parameters into the UKF algorithm for calculation to obtain a measurement variance matrix, a variance matrix of a state measurement value and an SOC value of the lithium battery;
the second calculation module is used for calling the trained BP neural network model stored in the database, taking the measurement variance matrix and the variance matrix of the state measurement value calculated by the first calculation module as the input layer of the BP neural network model, and calculating through the BP neural network model to obtain the correction value of the SOC value;
and the third calculation module is used for adding the SOC value of the lithium battery calculated by the first calculation module and the corrected value of the SOC calculated by the second calculation module to obtain an SOC estimated value.
An electronic device comprising a memory and a processor, the memory having stored therein a computer program for running on the processor, the computer program, when running on the processor, performing the method of the invention.
A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method of the invention.
The above-mentioned embodiments only express the specific embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for those skilled in the art, without departing from the technical idea of the present application, several changes and modifications can be made, which are all within the protection scope of the present application.
Claims (10)
1. A lithium battery SOC estimation method based on a neural network is characterized by comprising the following steps:
step 1: acquiring sampling data at the moment of k +1, and inputting the acquired sampling data into a UKF algorithm to obtain a measurement variance matrix at the moment of k +1 and a variance matrix of a state and a measured value; obtaining the SOC value of the lithium battery based on a UKF algorithm;
and 2, step: updating a system state variable value and an error variance matrix based on the obtained measurement variance matrix and the variance matrix of the state and the measured value;
and step 3: taking the system state variable and the error variance matrix as the input of a trained BP neural network, and obtaining a corrected value of the SOC value based on the BP neural network;
and 4, step 4: and adding the corrected value of the SOC value and the SOC value of the lithium battery obtained by the UKF algorithm to obtain an SOC estimated value.
2. The method for estimating the SOC of the lithium battery based on the neural network as claimed in claim 1, wherein the specific steps of obtaining the measurement variance matrix and the variance matrix of the state and the measured value are as follows:
A. predicting a system state variable and an error variance matrix at the k +1 moment based on a UKF algorithm, and simultaneously obtaining a prediction mean value of the system state variable, wherein the specific calculation mode is as follows:
in the formula: i =1,2,3.. 2n +1;
is the predicted value of the system state variable at the moment k +1,
is a sampling point one-step prediction value of the UKF algorithm,
is that
Corresponding weight, P xx,k+1|k Is the state of charge estimation error at time k +1The variance matrix is used to determine the variance of the received signal,
weight of the predicted value joint covariance of x, Q k+1 Is gaussian noise;
B. updating the average weight of the measured value at the moment k +1 and the measured variance matrix to obtain the state at the moment k +1 and the variance matrix of the measured value:
in the formula:
is a predicted value for each of the sampling points,
is that
And is an input variable, P yy,k+1 Is the average weight at time k +1 and the variance matrix of the measurements; p xy,k+1 Is the variance matrix of the state and measurements at time k +1; r k+1 Is the noise at time k + 1.
3. The method for estimating the SOC of the lithium battery based on the neural network as claimed in claim 1, wherein the step 2 comprises the steps of:
step 2.1: obtaining a Kalman gain based on the measurement variance matrix at the moment of k +1 and the variance matrix of the state and the measurement value:
step 2.2: updating a system state variable value and an error variance matrix based on Kalman gain:
in the formula:
is a value of a system state variable at time K +1 based on Kalman gain, K k+1 Is Kalman gain, y k+1 Is the measured value at time k +1, P xx,k+1|k+1 Is based on the system state variable (state of charge estimation) error variance matrix at time k +1 of the kalman gain.
4. The method for estimating the SOC of the lithium battery based on the neural network as claimed in claim 1, wherein the training step of the BP neural network is as follows:
A. acquiring historical sampling data, and constructing a training data set based on the historical sampling data;
B. calculating a training data set based on a UKF algorithm to obtain a Kalman filter gain matrix and a state of charge estimation matrix error;
C. and training the BP neural network based on the obtained Kalman filter gain matrix and the charge state estimation matrix error as an input layer of the BP neural network, and correcting the weight and the threshold of the BP neural network model based on the training result to finally obtain the trained BP neural network model.
5. The lithium battery SOC estimation method based on the neural network as claimed in claim 4, wherein the threshold and activation function of the current node of the BP neural network are as follows:
in the formula: omega ej Represents the weight between e and node j; b is a mixture of j A threshold representing node j; f (S) j ) Is an activation function; a is a j Representing the output value of node j.
6. The lithium battery SOC estimation method based on the neural network as claimed in claim 4, wherein the error function of the BP neural network is established based on a gradient descent method, specifically as follows:
in the formula: e (ω, b) refers to the error function of the BP neural network with respect to the weight and the threshold;
is to make a partial derivation of it, d j Representing the results of the output layer; y is j The compensation correction value of the UKF algorithm is represented; f (S) j ) Is an activation function;
the weights between the hidden layer and the output layer are derived from the error function:
in the formula: omega ej And (3) representing the weight between e and the node j, namely, obtaining the weight update between the hidden layer and the output layer, and l represents the learning rate.
7. The lithium battery SOC estimation method based on the neural network as claimed in claim 4, wherein the weight optimization method of the BP neural network is as follows:
in the formula: omega ej (k + 1) is a weight optimization expression of the BP neural network at the next time, ω ej (k) For the weight optimization expression at this time, mu a Mu as the learning rate factor rise coefficient b For learning rate factor reduction factor, E (k + 1) and E (k) are custom comparison values.
8. A lithium battery SOC estimation system based on a neural network is characterized by comprising:
the data acquisition module is used for acquiring parameters of the equivalent circuit;
the first calculation module calls a UKF algorithm stored in a database, and substitutes the obtained equivalent circuit parameters into the UKF algorithm for calculation to obtain a measurement variance matrix, a variance matrix of a state measurement value and an SOC value of the lithium battery;
the second calculation module is used for calling the trained BP neural network model stored in the database, taking the measurement variance matrix and the variance matrix of the state measurement value calculated by the first calculation module as the input layer of the BP neural network model, and calculating through the BP neural network model to obtain the correction value of the SOC value;
and the third calculation module is used for adding the SOC value of the lithium battery calculated by the first calculation module and the SOC correction value calculated by the second calculation module to obtain an SOC estimation value.
9. An electronic device comprising a memory and a processor, the memory having stored thereon a computer program for execution on the processor, wherein the computer program, when executed on the processor, performs the method of any one of claims 1 to 7.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211372630.2A CN115792625A (en) | 2022-11-03 | 2022-11-03 | Lithium battery SOC estimation method, system, equipment and medium based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211372630.2A CN115792625A (en) | 2022-11-03 | 2022-11-03 | Lithium battery SOC estimation method, system, equipment and medium based on neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115792625A true CN115792625A (en) | 2023-03-14 |
Family
ID=85435383
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211372630.2A Pending CN115792625A (en) | 2022-11-03 | 2022-11-03 | Lithium battery SOC estimation method, system, equipment and medium based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115792625A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117148161A (en) * | 2023-08-29 | 2023-12-01 | 深圳市今朝时代股份有限公司 | Battery SOC estimation method and device based on cloud neural network |
-
2022
- 2022-11-03 CN CN202211372630.2A patent/CN115792625A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117148161A (en) * | 2023-08-29 | 2023-12-01 | 深圳市今朝时代股份有限公司 | Battery SOC estimation method and device based on cloud neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112034356B (en) | GP-UKF-based online SOH estimation method for electric vehicle power battery | |
| CN105116343B (en) | The electrokinetic cell state of charge method of estimation and system of least square method supporting vector machine | |
| CN111428433B (en) | Lithium ion battery state calculation method based on hybrid filtering | |
| CN109031147B (en) | SOC estimation method of lithium iron phosphate battery pack | |
| CN112630659A (en) | Lithium battery SOC estimation method based on improved BP-EKF algorithm | |
| CN113128672B (en) | Lithium ion battery pack SOH estimation method based on transfer learning algorithm | |
| CN114740357A (en) | Joint estimation method for branch current, charge state and power state of parallel battery pack | |
| Cui et al. | Equivalent Circuit Model of Lead-acid Battery in Energy Storage Power Station and Its State-of-Charge Estimation Based on Extended Kalman Filtering Method. | |
| CN113805075A (en) | BCRLS-UKF-based lithium battery state of charge estimation method | |
| CN114217234B (en) | IDE-ASRCKF-based lithium ion battery parameter identification and SOC estimation method | |
| CN115327415A (en) | Lithium battery SOC estimation method based on limited memory recursive least square algorithm | |
| CN115656848A (en) | Lithium battery SOC estimation method based on capacity correction | |
| CN115792625A (en) | Lithium battery SOC estimation method, system, equipment and medium based on neural network | |
| CN110441691A (en) | It is a kind of based on the SOC estimation method for simplifying particle Unscented transform | |
| CN112946480B (en) | Lithium battery circuit model simplification method for improving SOC estimation real-time performance | |
| CN109085509A (en) | The parameter identification method and system of lithium ion battery open-circuit voltage and SOC relationship | |
| CN115656839A (en) | Battery state parameter collaborative estimation method based on BP-DEKF algorithm | |
| CN115469228B (en) | Reconfigurable network type energy storage system battery state of charge estimation method | |
| CN112364471A (en) | Research on lithium battery SOC estimation method based on Thevenin model and unscented Kalman filter | |
| CN112649734A (en) | SOC estimation method based on improved PNGV model and extended Kalman algorithm | |
| CN114295987B (en) | Battery SOC state estimation method based on nonlinear Kalman filtering | |
| CN113109725B (en) | Parallel battery state-of-charge estimation method based on state noise matrix self-adjustment | |
| CN112255545B (en) | Lithium battery SOC estimation model based on square root extended Kalman filtering | |
| CN115128461A (en) | SOC estimation method based on evanescent factors EKF and FFRLS | |
| CN114779105A (en) | Lithium battery pack inconsistency estimation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |