A SOC estimation method based on deep fusion neural network and unscented Kalman filter

Abstract

The application discloses a SOC estimation method based on a deep fusion neural network and a UKF (Unscented Kalman Filter), relates to the technical field of all-vanadium redox flow batteries, and aims to realize high-precision estimation of the state of charge of a battery. The method comprises the following steps: a second-order equivalent circuit model of the battery is established, and parameter identification is performed on the second-order equivalent circuit model; a group of sigma points are obtained through unscented transformation, the propagation state estimation and the propagation state prediction of each sigma point are calculated, the first-order statistical moments of the prior state estimation and the prior state prediction are obtained through unscented transformation, the cross covariance is obtained, and the observation difference and the state update difference are obtained; the observation difference and the state update difference are input into a neural network system to obtain Kalman gain, the prior state estimation is updated through the Kalman gain, and the state of charge of the battery is predicted.

Description

Technical Field

[0001] This invention relates to the field of vanadium redox flow battery technology, and in particular to a SOC estimation method using a deep fusion neural network unscented Kalman filter. Background Technology

[0002] In recent years, with the rapid development of battery technology, research on the estimation of battery state of charge (SOC) has received increasing attention. SOC is one of the core parameters of a battery management system and an important indicator for measuring the remaining capacity or charge / discharge state of a battery.

[0003] Accurate estimation of the battery's state of charge (SOC) helps the battery management system (BMS) develop more efficient scheduling strategies, thereby preventing overcharging or over-discharging and contributing to extending battery life. However, the accuracy of current SOC estimations is generally low. Summary of the Invention

[0004] This invention provides a SOC estimation method based on deep fusion neural network unscented Kalman filtering, aiming to achieve high-precision estimation of battery state of charge.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] This invention provides a deep fusion neural network unscented Kalman filter-based SOC estimation method, comprising: S1: establishing a second-order equivalent circuit model of the battery and identifying parameters of the second-order equivalent circuit model using recursive least squares method; S2: obtaining a set of fixed-weight sigma points through unscented transformation, calculating the propagation state estimate and propagation state prediction for each sigma point, and obtaining the first-order statistical moments of the prior state estimate and the first-order statistical moments of the prior state prediction, as well as the cross-covariance, through unscented transformation, to obtain the observation difference and state update difference; S3: inputting the observation difference and state update difference into the neural network system to obtain the Kalman gain, and updating the prior state estimate using the Kalman gain to obtain the prediction of the battery's state of charge; the neural network system includes long short-term memory network units; S4: when new input variables and observation variables are obtained, updating the data of the neural network system, updating the first-order statistical moments and cross-covariance of the prior state prediction, and iteratively executing the process of obtaining the observation difference and state update difference, obtaining the Kalman gain, and obtaining the prediction of the battery's state of charge.

[0007] In some embodiments, step S1 includes S11: with a fixed battery flow rate, measuring the open-circuit voltage of the battery using a low-current charge-discharge method to obtain the nonlinear relationship between the battery's open-circuit voltage and its state of charge. S12: establishing a second-order equivalent circuit model based on the nonlinear relationship between the battery's open-circuit voltage and its state of charge.

[0008] In some embodiments, measuring the open-circuit voltage of a battery using the low-current charge-discharge method means: charging the battery with a small current when the battery is fully discharged, and recording the voltage and state-of-charge values ​​when the battery is working; and discharging the battery with a small current when the battery is fully charged, and recording the voltage and state-of-charge values ​​when the battery is working.

[0009] In some embodiments, each sigma point includes an initial state.

[0010] Calculating the propagation state estimate and propagation state prediction for each sigma point refers to:

[0011] Using transfer function Propagate the initial state of each sigma point to By measuring function Calculate its predicted value :

[0012]

[0013] The first-order statistical moments of the prior state estimate are obtained through unscented transformation. First-order statistical moments of prior state prediction and cross covariance , and obtain the observation difference and state update difference ,include:

[0014]

[0015]

[0016]

[0017]

[0018] in, For the observed values, This represents the initial state value.

[0019] In some embodiments, the neural network system includes a first fully connected layer; in step S3, the observation difference is... and state update difference Input to the first fully connected layer, according to The output of the first fully connected layer is obtained; where, For the activation function ReLU, These are the parameters of the first fully connected layer. This is the output of the first fully connected layer; the parameters of the first fully connected layer include the dimension, the number of state variables, and the number of observation variables.

[0020] In some embodiments, the neural network system further includes a second fully connected layer; after obtaining the output of the first fully connected layer, step S3 further includes: according to Obtain the input to the Long Short-Term Memory (LSTM) network units; where, This is the storage vector for the previous time step. As input to the Long Short-Term Memory (LSTM) network units; according to Update the state vector of the long short-term memory unit; where This is the state vector of the previous time step. Represents the updated state vector of a long-term memory unit; calculates the output vector of the memory unit. The output vector of the input gate The output vector of the forget gate and the output vector of the output gate ;according to Obtain the output of the Long Short-Term Memory network unit ;in, For the Tanh activation function; according to The output of the Long Short-Term Memory network unit Transformed into Kalman gain through the second fully connected layer. ,in, For the activation function ReLU, These are the parameters for the second fully connected layer.

[0021] In some embodiments, updating the prior state estimate using Kalman gain to obtain a prediction of the battery's state of charge includes: updating the state based on the Kalman gain and the observation difference. Update the first-order statistical moments of the current prior state estimate ;according to Obtain the final state estimate .

[0022] In some embodiments, updating the data of the neural network system, updating the first-order statistical moments and cross-covariance of the prior state predictions, and iteratively executing the steps of obtaining the observation difference and state update difference, obtaining the Kalman gain, and obtaining the prediction of the battery state of charge, includes: based on Obtain the covariance matrix of the battery at the next time step. ;in The cross covariance is the prediction of the propagation state.

[0023] In some embodiments, step S4 is followed by step S5: establishing the overall loss function of the second-order equivalent circuit model and training the neural network system.

[0024] In some embodiments, the overall loss function of the established second-order equivalent circuit model and the overall loss function when training the neural network system are: .

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] The SOC estimation method based on deep fusion neural network unscented Kalman filtering provided in this application uses a second-order equivalent circuit model, which, compared to a first-order model, more accurately reflects the dynamic characteristics of the battery (such as voltage delay, multiple time constants, etc.), providing a more reliable physical basis for subsequent state estimation. Parameter identification is performed using recursive least squares, which can adapt online or offline to changes in battery parameters due to aging, temperature, etc., maintaining model accuracy. Unscented transformation is used to handle nonlinear problems, which, compared to linearized extended Kalman filtering (EKF), can more accurately capture the state distribution, especially in higher-order models, reducing errors caused by linearization. Long short-term memory network units can learn the complex, time-varying characteristics of the Kalman gain, which may be affected by various factors such as battery aging, temperature changes, and charge / discharge rates, making it difficult for traditional unscented Kalman filtering (UKF) to precisely set. Through learning, the gain can be dynamically adjusted, resulting in more accurate state updates. Therefore, a system capable of online adaptive, high-precision, and highly robust estimation of battery SOC is formed. It not only takes into account the physical characteristics of the battery, but also uses data intelligence to optimize key control parameters, thereby achieving a more accurate measurement of the battery's state of charge. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart of a deep fusion neural network unscented Kalman filter SOC estimation method provided in this application embodiment;

[0029] Figure 2 A schematic diagram of an all-vanadium redox flow battery structure provided in this application embodiment;

[0030] Figure 3 A schematic diagram of a second-order equivalent circuit provided in an embodiment of this application.

[0031] Figure 4 This is a schematic diagram illustrating an estimation of the state of charge of a battery, as provided in an embodiment of this application.

[0032] Wherein: 1 is the power supply, 2 is the load, 3 is the converter, 4 is the pipeline, 5 is the electrolyte storage tank, 6 is the ion exchange membrane, 7 is the electrode, and 8 is the circulation pump. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0034] In the description of the invention, it should be understood that the terms "upper," "lower," "left," "right," "front," "rear," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or relative positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Unless otherwise specified, the above-mentioned orientational descriptions can be flexibly set in practical applications, provided that the relative positional relationships shown in the accompanying drawings are satisfied.

[0035] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0036] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "communication" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection. They can refer to a direct connection or an indirect connection through an intermediate medium, or a communication between the internal components of two elements. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0037] In embodiments of the invention, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, article, or apparatus that includes that element.

[0038] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0039] In recent years, battery technology has developed rapidly, and electrochemical energy storage technology is an important part of it. Electrochemical energy storage technology is a technology that stores and releases electrical energy through electrochemical reactions. It converts electrical energy into chemical energy for storage and then converts the chemical energy back into electrical energy when needed. It is widely used in power system peak shaving and frequency regulation, renewable energy grid connection, emergency power supply, electric vehicles, and portable electronic devices.

[0040] Among various battery types, vanadium redox batteries (VRFB) are highly favored due to their unique characteristics such as independent energy capacity, power output, long lifespan, safety, resistance to deep discharge, and high efficiency. This application uses an all-vanadium redox flow battery as an example for illustration.

[0041] The State of Charge (SOC) of a battery is one of the core parameters of a battery management system. With the rapid development of battery technology, research on SOC estimation methods has received increasing attention.

[0042] The battery management system formulates scheduling strategies based on the estimated state of charge (SOC) of the battery. Therefore, a more accurate estimate of SOC helps the battery management system to formulate more scientific battery strategies, ensuring high-efficiency battery operation, preventing overcharging or over-discharging, and helping to extend battery life.

[0043] In view of this, embodiments of this application provide a SOC estimation method based on deep fusion neural network unscented Kalman filtering. For example, as shown... Figure 1 As shown, the SOC estimation method of deep fusion neural network unscented Kalman filter includes:

[0044] S1: Establish a second-order equivalent circuit model of the battery, and identify the parameters of the second-order equivalent circuit model by recursive least squares method.

[0045] Reference Figure 2 and Figure 3 , Figure 2 This application illustrates an all-vanadium redox flow battery structure according to an embodiment. Based on this all-vanadium redox flow battery structure, a battery system can be constructed as follows: Figure 3 The second-order equivalent circuit model is shown. Figure 2The vanadium redox flow battery structure includes an ion exchange membrane 6 and an electrode 7. The electrode 7 is the site of electrochemical reactions, while the ion exchange membrane 6 allows specific ions to pass through, such as H+. + This serves to isolate the positive and negative electrolytes. Two electrolyte storage tanks 5 are connected to the battery stack via pipes 4, allowing for the storage of electrolytes... / and / The electrolyte can be delivered to the battery stack. A circulation pump 8 promotes the circulation of the electrolyte between the electrolyte storage tank 5 and the battery stack, ensuring continuous reaction and allowing the electrolyte to continuously enter the battery stack to participate in the electrochemical reaction. The converter 3 of the vanadium redox flow battery connects the power source 1 and the load 2. During charging, it converts the electrical energy of the power source 1 into chemical energy for storage; during discharging, it converts the chemical energy into electrical energy to supply the load 2, achieving bidirectional energy conversion.

[0046] More specifically, in some embodiments, step S1 further includes steps S11 and S12.

[0047] S11: With a fixed battery flow rate, the open-circuit voltage of the battery is measured using a low-current charge-discharge method to obtain the nonlinear relationship between the open-circuit voltage and the state of charge of the battery.

[0048] As one possible implementation, the low-current charge-discharge method for measuring the open-circuit voltage of a battery includes: charging the battery with a small current when the battery is fully discharged and recording the voltage and state of charge values ​​during battery operation; and discharging the battery with a small current when the battery is fully charged and recording the voltage and state of charge values ​​during battery operation.

[0049] By recording the battery's voltage and state of charge (SOC) values ​​during operation when the battery is fully discharged and then charged with a small current, a curve of charging voltage versus SOC can be plotted. Similarly, by recording the battery's voltage and SOC values ​​during operation when the battery is fully charged and then discharged with a small current, a curve of discharging voltage versus SOC can be plotted. Averaging the charging voltage vs. SOC curve and the discharging voltage vs. SOC curve yields the non-linear relationship between the battery's open-circuit voltage and its SOC.

[0050] This application employs the Kalman filter method to estimate the state of charge (SOC) of the battery. The Kalman filter is an optimal recursive filtering algorithm based on linear dynamic systems. Its core assumptions include a state-space model, which comprises observation equations and state equations. In other words, establishing a second-order equivalent circuit model of the battery involves establishing state equations and observation equations. By establishing the state equations and observation equations, the system state can be optimally estimated through prediction and update steps. For example, the state equations and observation equations for the battery's state of charge are as follows:

[0051]

[0052]

[0053] Among them, the state variables are ; Sampling time, The input is the operating current, which is the terminal voltage of the second-order RC network. , It is the ohmic internal resistance, which characterizes the instantaneous ohmic polarization of the battery and is directly related to the electrolyte conductivity, electrode contact resistance, etc. This is an electrochemical polarization branch, characterizing the dynamic process of charge transfer reactions; This is the concentration polarization branch, which characterizes the dynamic process of ion diffusion in the electrolyte; Let be the open-circuit voltage, which is a nonlinear function of SOC. ; and Let be independent random variables, representing system noise and measurement noise, respectively; and Let be the noise covariance matrix.

[0054] S12: Establish a second-order equivalent circuit model based on the nonlinear relationship between the battery's open-circuit voltage and its state of charge.

[0055] After establishing the second-order equivalent circuit model, its parameters are identified using the recursive least squares method. Recursive least squares (RLS) is an adaptive algorithm for parameter identification, widely used in online parameter estimation of dynamic systems. By employing RLS, key parameters in the battery model can be determined, more accurately reflecting the battery's dynamic characteristics and thus improving the accuracy of SOC estimation.

[0056] As one possible implementation, extracting parameters from the second-order equivalent circuit model using the least squares method requires converting the aforementioned state equation expression into a regression form, i.e., obtaining the regression expression for the vanadium redox battery model. This embodiment utilizes Laplace transform and bilinear transform to obtain the regression expression for the vanadium redox battery model. The obtained regression expression is shown below:

[0057]

[0058] in, , For information vectors, For parameter vectors.

[0059] Updated using a recursive algorithm, via parameter vectors. The key parameters of the second-order equivalent circuit model, including resistance and capacitance, are obtained through inverse solving. Specifically, this includes the first resistor. Second resistor Third resistor First capacitor and the second capacitor .

[0060] S2: Obtain a set of fixed-weight sigma points through unscented transformation, calculate the propagation state estimate and propagation state prediction for each sigma point, and obtain the first-order statistical moments of the prior state estimate and the first-order statistical moments of the prior state prediction, as well as the cross covariance, through unscented transformation, to obtain the observation difference and state update difference.

[0061] In some embodiments, for time step t Sample a set of sigma points to estimate the initial state. Each sigma point represents a combination of state variables obtained from its prior distribution. Therefore:

[0062]

[0063] in, m =3 represents the number of state variables. and These represent the previous time step. t− The mean and covariance of 1. Parameters Defined as ,in =0.8 controls the distribution of sigma points around the mean during the sampling period. =0 is a minor scaling factor.

[0064] The sigma points are obtained through sampling, and the weights of each sigma point are obtained. and The calculation formula is:

[0065]

[0066] in, =2 is used to introduce high statistics about state variables.

[0067] For step S2, it is necessary to calculate the propagation state estimate and propagation state prediction for each sigma point based on the initial state of each sigma point. Specifically, this is done using the transfer function. Propagate the initial state of each sigma point to By measuring function Calculate its predicted value :

[0068]

[0069] As one possible implementation, the first-order statistical moments of the prior state estimate are obtained through unscented transformation. First-order statistical moments of prior state prediction and cross covariance , and obtain the observation difference and state update difference ,include:

[0070]

[0071]

[0072]

[0073]

[0074] in, The observed value refers specifically to the voltage value obtained by actual measurement of the battery using sensors or other means. These are the initial values ​​preset based on the second-order equivalent circuit model and the actual situation. It is the voltage value obtained by actual measurement of the battery using sensors, etc.

[0075] S3: Input the observation difference and state update difference into the neural network system to obtain the Kalman gain, and update the state of the prior state estimate through the Kalman gain to obtain the prediction of the battery state of charge; the neural network system includes long short-term memory network units.

[0076] A neural network is an algorithmic mathematical model that mimics the behavioral characteristics of animal neural networks to perform distributed parallel information processing. This type of network relies on the complexity of the system to adjust the relationships between a large number of interconnected nodes, thereby achieving the purpose of information processing.

[0077] Because this application performs Sigma sampling near the operating point, approximating the state probability density function as the Gaussian density function of the sampling point, it obtains the mean and variance of the second-order equivalent circuit model, making the Kalman filtering method applicable to linear assumptions also applicable to solving nonlinear system problems.

[0078] However, when noise statistics are unknown or change over time, there is a high risk of reduced filtering accuracy or even divergence in the implementation of unscented Kalman filtering. Given the inherent memory and nonlinear mapping capabilities of Long Short-Term Memory Neural Networks (also referred to as Long Short-Term Memory Network Units in this application), they are well-suited as internal memory elements for tracking Kalman gain. By embedding Long Short-Term Memory Units into the unscented Kalman filtering process, noise-related information can be captured and estimated, resulting in a more accurate final battery SOC estimation.

[0079] In some embodiments, the neural network system further includes a first fully connected layer and a second fully connected layer. Step S3 includes sub-steps S31, S32, S33, S34, and S35.

[0080] S31: The observation difference obtained in step S2 and state update difference Input to the first fully connected layer, according to The output of the first fully connected layer is obtained.

[0081] in, For the activation function ReLU, These are the parameters of the first fully connected layer. This is the output of the first fully connected layer; the parameters of the first fully connected layer include the dimension, the number of state variables, and the number of observation variables.

[0082] After obtaining the output of the first fully connected layer, proceed to step S32: based on... Obtain input from Long Short-Term Memory (LSTM) network units.

[0083] in, This is the storage vector for the previous time step. It serves as the input to the Long Short-Term Memory (LSTM) network unit. according to It is iterated over given arbitrary initial values ​​and contains high-dimensional noise-related information. This is the Tanh activation function.

[0084] S33: Calculate the output vector of the storage unit The output vector of the input gate The output vector of the forget gate and the output vector of the output gate and according to Update the state vector of the long short-term memory unit.

[0085] in This is the state vector of the previous time step. This represents the updated state vector of a long-term memory unit.

[0086] S34: According to Obtain the output of the Long Short-Term Memory network unit .in, This is the Tanh activation function.

[0087] S35: According to The output of the Long Short-Term Memory network unit Transformed into Kalman gain through the second fully connected layer. .

[0088] in, For the activation function ReLU, These are the parameters for the second fully connected layer.

[0089] The specific steps in step S3, which involve updating the prior state estimate using Kalman gain to obtain a prediction of the battery's state of charge, include:

[0090] Based on Kalman gain and observation difference Update the first-order statistical moments of the current prior state estimate ;

[0091] according to Obtain the final state estimate .

[0092] Final state estimate It can reflect the prediction of the battery's state of charge. It should be understood that in order to continuously iterate and predict the battery's state of charge at the next moment, the data involved in this application also needs to be updated:

[0093] S4: Update the data of the neural network system, update the first-order statistical moments and cross-covariance of the prior state prediction, and iteratively execute the steps of obtaining the observation difference and state update difference, obtaining the Kalman gain, and obtaining the prediction of the battery state of charge.

[0094] For example, according to Update the cross covariance.

[0095] The neural network system used in the battery state of charge estimation method provided in this application includes long short-term memory (LSM) units, which in turn include input gates, forget gates, output gates, and memory units. The input gates determine whether information is stored in the state. The forget gates determine whether information needs to be discarded. The output gates determine the output information, and the memory units store long-term information. When new input variables and observed variables are obtained, the system can predict and update the battery's propagation state estimate and covariance matrix for the next time step based on the information stored in the memory units.

[0096] Reference Figure 4 The method for estimating the state of charge of a battery provided in this application first samples a series of sigma points, and then uses a transfer function to obtain the final state of charge. Measurement function Obtain the first-order statistical moments of prior state estimates First-order statistical moments of prior state prediction and cross covariance The observation difference was calculated. and state update difference The Kalman gain is then estimated within the neural network system to ultimately obtain the predicted SOC value. Real-time prediction of the battery state is based on the continuous updating of the state equation and the observation equation, that is, based on the continuous updates of the input variables and state variables.

[0097]

[0098] Achieve and Continuous updates. Utilizing transfer functions. Measurement function and Realize the first-order statistical moments for prior state estimation First-order statistical moments of prior state prediction and cross covariance Update.

[0099] In some embodiments, step S4 is followed by step S5: establishing the overall loss function of the second-order equivalent circuit model and training the neural network system.

[0100] As one possible implementation, the overall loss function of the established second-order equivalent circuit model, and the overall loss function when training the neural network system, are: Further improve the accuracy of predictions through model training.

[0101] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for SOC estimation using a deep fusion neural network unscented Kalman filter, characterized in that, include: S1: Establish a second-order equivalent circuit model of the battery, and identify the parameters of the second-order equivalent circuit model by recursive least squares method. S2: Obtain a set of fixed-weight sigma points through unscented transformation, calculate the propagation state estimate and propagation state prediction for each sigma point, and obtain the first-order statistical moments of the prior state estimate and the first-order statistical moments of the prior state prediction, as well as the cross covariance, through unscented transformation, to obtain the observation difference and state update difference. S3: Input the observation difference and state update difference into the neural network system to obtain the Kalman gain, and update the prior state estimate using the Kalman gain to obtain a prediction of the battery state of charge; the neural network system includes a long short-term memory network unit; S4: When new input variables and observation variables are obtained, update the data of the neural network system, update the first-order statistical moments and cross-covariance of the prior state prediction, and iteratively execute to obtain the observation difference and state update difference, obtain the Kalman gain, and obtain the prediction of the battery state of charge.

2. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 1, characterized in that, Step S1 includes: S11: With a fixed battery flow rate, the open-circuit voltage of the battery is measured using a low-current charge-discharge method to obtain the nonlinear relationship between the open-circuit voltage of the battery and the state of charge of the battery. S12: Establish a second-order equivalent circuit model based on the nonlinear relationship between the open-circuit voltage of the battery and the state of charge of the battery.

3. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 2, characterized in that, The low-current charge-discharge method for measuring the open-circuit voltage of a battery refers to: With the battery fully discharged, charge the battery with a small current and record the voltage and state of charge values ​​during operation; with the battery fully charged, discharge the battery with a small current and record the voltage and state of charge values ​​during operation.

4. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 1, characterized in that, Each of the sigma points includes an initial state. The calculation of the propagation state estimate and propagation state prediction for each sigma point refers to: Using transfer function Propagate the initial state of each sigma point to By measuring function Calculate its predicted value : The first-order statistical moments of the prior state estimate are obtained through unscented transformation. First-order statistical moments of prior state prediction and cross covariance , and obtain the observation difference and state update difference ,include: in, For the observed values, This represents the initial state value.

5. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 4, characterized in that, The neural network system includes a first fully connected layer; In step S3 The observation difference and state update difference Input the first fully connected layer, according to The output of the first fully connected layer is obtained; where, For the activation function ReLU, These are the parameters of the first fully connected layer. The output of the first fully connected layer is denoted as ; the parameters of the first fully connected layer include the dimension, the number of state variables, and the number of observation variables.

6. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 5, characterized in that, The neural network system further includes a second fully connected layer; after obtaining the output of the first fully connected layer, step S3 further includes: according to The input to the Long Short-Term Memory (LSTM) network unit is obtained; among which, This is the storage vector for the previous time step. It serves as the input to the Long Short-Term Memory (LSTM) network units; according to Update the state vector of the long short-term memory unit; where This is the state vector of the previous time step. Represents the updated state vector of a long-term memory unit; Calculate the output vector of the storage unit The output vector of the input gate The output vector of the forget gate and the output vector of the output gate ; according to The output of the Long Short-Term Memory network unit is obtained. ;in, Use the Tanh activation function; according to The output of the Long Short-Term Memory network unit This is converted into Kalman gain through the second fully connected layer. ,in, For the activation function ReLU, These are the parameters of the second fully connected layer.

7. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 6, characterized in that, Step S3 further includes: Based on Kalman gain and observation difference Update the first-order statistical moments of the current prior state estimate ; according to Obtain the final state estimate .

8. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 7, characterized in that, Step S4 includes: according to Obtain the covariance matrix of the battery at the next time step. ;in The cross-covariance is the predicted propagation state.

9. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 1, characterized in that, After step S4, step S5 is also included: establishing the overall loss function of the second-order equivalent circuit model and training the neural network system.

10. The SOC estimation method for a deep fusion neural network unscented Kalman filter according to claim 9, characterized in that, The overall loss function is: in, The first-order statistical moments for predicting prior states. These are the observed values.

Images