Vanadium redox flow battery soc prediction method based on artificial neural network

Abstract

The present application belongs to the field of liquid flow battery, and particularly relates to a state of charge (SOC) prediction method for all-vanadium redox flow battery based on artificial neural network, which is mainly applied to predict the SOC of electrolyte of all-vanadium redox flow battery. The method comprises the following steps: performing charge-discharge cycle test on electrolyte of all-vanadium redox flow battery at different SOC to obtain data of charge-discharge cycle; extracting charge-discharge characteristics in each charge-discharge cycle as a feature vector X; taking SOC as a target function y to establish a database; dividing data in the database into a training set and a test set; training data in the training set by using artificial neural network to establish an SOC prediction model for all-vanadium redox flow battery; and using data in the test set to evaluate the trained positive electrode or negative electrode SOC prediction model for all-vanadium redox flow battery. The method can effectively predict the SOC of all-vanadium redox flow battery, and realize monitoring of the SOC of electrolyte of all-vanadium redox flow battery by using current-voltage data in the charge-discharge process.

Description

Technical Field

[0001] This invention belongs to the field of flow batteries, specifically a method for online prediction of the state of charge (SOC) of an all-vanadium redox flow battery system based on artificial neural networks. Background Technology

[0002] Predicting the state of charge (SOC) is crucial for the operation of vanadium redox flow battery systems. Accurate online SOC prediction can significantly improve system stability and reduce operation and maintenance costs. Currently, the main SOC estimation methods in vanadium redox flow battery demonstration applications include the ampere-hour integration method, open-circuit voltage method, Kalman filtering method, and AC impedance method. However, during operation, electrolyte migration can cause electrolyte imbalance, and the accuracy of estimating the SOC after electrolyte imbalance using the above methods is not high. Furthermore, other reported SOC detection methods are limited by the complexity and high cost of instruments and equipment, and are mostly confined to the laboratory research stage, unable to be applied to large-scale demonstration and commercial operation of vanadium redox flow battery systems. Summary of the Invention

[0003] The purpose of this invention is to provide an online prediction method for the state of charge (SOC) of a vanadium redox flow battery system. This method predicts the SOC of the vanadium redox flow battery system using single-cell data without introducing additional data acquisition and testing instruments.

[0004] To address the limitations and complexity of existing SOC detection methods, this invention proposes a method for online SOC prediction using only charge-discharge cycle data from vanadium redox flow batteries, without introducing additional complex acquisition instruments and measurement equipment. Artificial neural networks can theoretically be used to fit arbitrarily complex functions; therefore, this invention proposes to offline configure a series of positive and negative electrolytes with different SOCs, utilize the charge-discharge characteristics of a single cell during the charge-discharge process, and use artificial neural network modeling to predict the SOC of a single vanadium redox flow battery. Applying this model to a vanadium redox flow battery system enables online monitoring of the battery's SOC. This method is low-cost and easy to implement; data can be obtained through single-cell charge-discharge testing. The established vanadium redox flow battery positive electrode SOC prediction model can be applied to large-scale demonstration and commercial systems. The vanadium redox flow battery SOC prediction model designed using this invention can significantly improve the operational stability of vanadium redox flow battery systems and reduce system and maintenance costs.

[0005] The technical solution adopted by this invention to achieve the above objectives is: a method for predicting the state of charge (SOC) of an all-vanadium redox flow battery based on artificial neural networks, comprising the following steps:

[0006] The charge-discharge characteristics in the charge-discharge cycle are obtained as feature vector X, and the positive electrode SOC is used as target y to establish a database. The second charge-discharge cycle data under each charge-discharge current density in the database are used as training dataset to train the model. The data of the third charge-discharge cycle are used as test dataset to verify the accuracy of the model.

[0007] The training set is input into the neural network model for training, and the trained vanadium redox flow battery cathode SOC prediction model is obtained.

[0008] The charge and discharge characteristics during the charge and discharge cycle are collected and used as feature vector X. This feature vector X is then input into the SOC prediction model of the vanadium redox flow battery cathode to obtain the SOC of the vanadium redox flow battery cathode.

[0009] The process of obtaining charge-discharge characteristics during a charge-discharge cycle includes the following steps:

[0010] By configuring vanadium electrolytes with different positive electrode SOCs and keeping the negative electrode SOC at 0, single cells were assembled for charge-discharge cycle testing.

[0011] The configured electrolyte was subjected to single-cell charge-discharge cycle tests. The discharge mode in the charge-discharge cycle test method was constant power discharge or constant current discharge.

[0012] The charging and discharging characteristics are obtained by testing the current and voltage characteristics of the charging and discharging process.

[0013] The feature vector X includes the current density I during constant current charging and discharging, and the starting charging voltage V during constant current charging. C_s The current density I at the start of constant voltage charging C_s Charging capacity C c The initial voltage of constant current discharge V d_s and the average voltage V of the resting period relax3 .

[0014] The feature vector X further includes at least one of the following: the average voltage V during rest. relax1 Average voltage V during rest relax2 .

[0015] The neural network model has 1 to 5 hidden layers, and the number of neurons in each hidden layer is an integer. The activation function is any one of Sigmoid, Tanh, ReLU, Leaky ReLU, ELU, PReLU, Softmax, Swish, Maxout, and Softplus. The optimizer of the neural network is any one of SGD, adagrad, adadelta, and adam.

[0016] A vanadium redox flow battery SOC prediction system based on artificial neural networks includes:

[0017] The dataset construction module is used to obtain the charge and discharge characteristics in the charge and discharge cycle as the feature vector X, and the positive electrode SOC as the target y to build a database; the second charge and discharge cycle data under each charge and discharge current density in the database is used as the training dataset to train the model; the data of the third charge and discharge cycle is used as the test dataset to test the model accuracy.

[0018] The neural network model training module is used to input the training set into the neural network model for training, and obtain the trained vanadium redox flow battery cathode SOC prediction model.

[0019] The positive electrode SOC prediction module is used to collect charge and discharge characteristics during charge and discharge cycles, which are used as feature vector X and input into the positive electrode SOC prediction model of the vanadium redox flow battery to obtain the positive electrode SOC of the vanadium redox flow battery.

[0020] A vanadium redox flow battery SOC prediction device based on artificial neural networks includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the vanadium redox flow battery SOC prediction method based on artificial neural networks when the computer program is executed.

[0021] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the SOC prediction method for vanadium redox flow batteries based on artificial neural networks as described in any one of claims 1-5.

[0022] The present invention has the following beneficial effects and advantages:

[0023] 1. This invention addresses the problem that existing SOC prediction methods cannot accurately predict the SOC after electrolyte imbalance. The design does not introduce any additional data acquisition devices or testing instruments. Utilizing test data from a single cell, it incorporates an artificial neural network to predict the SOC of a vanadium redox flow battery under different cathode SOC conditions. Furthermore, the SOC prediction model can be applied to vanadium redox flow battery systems. This invention has the advantages of simple experimental methods, low cost, and easy operation and maintenance.

[0024] 2. The online SOC prediction method for the all-vanadium redox flow battery system described in this invention can use real-time data from the charging and discharging process to predict the SOC state of the electrolyte after the positive and negative electrodes become unbalanced. The prediction is sensitive, accurate, and robust. Attached Figure Description

[0025] Figure 1 The flowchart of the implementation steps of the vanadium redox flow battery SOC prediction method based on artificial neural networks of this invention;

[0026] Figure 2aFigure 1. Results of the SOC prediction model for the cathode of a vanadium redox flow battery established using artificial neural network modeling with 32 neurons in each hidden layer.

[0027] Figure 2b Figure 1. Prediction model of SOC of vanadium redox flow battery cathode when the number of neurons in each hidden layer is 64, modeled using artificial neural network.

[0028] Figure 2c The result of the SOC prediction model for the cathode of a vanadium redox flow battery, which is established using artificial neural network modeling with 128 neurons in each hidden layer. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0030] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The technical solutions in the embodiments of the present invention are clearly and completely described. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] This invention is primarily applied to predicting the State of Charge (SOC) of vanadium redox flow batteries. It includes the following steps: performing charge-discharge cycle tests on the electrolyte of the vanadium redox flow battery under different positive or negative electrode SOCs to obtain charge-discharge cycle data; extracting the charge-discharge characteristics from each charge-discharge cycle as a feature vector X; establishing a database with the positive or negative electrode SOC as the objective function y; dividing the data in the database into a training set and a test set; using an artificial neural network to train the data in the training set to establish a prediction model for the positive or negative electrode SOC of the vanadium redox flow battery; and evaluating the trained prediction model using data from the test set. This method can effectively predict the SOC of vanadium redox flow batteries, enabling the monitoring of the electrolyte SOC using current and voltage data from the charge-discharge process.

[0032] This invention provides a method for predicting the state of charge (SOC) of an all-vanadium redox flow battery based on an artificial neural network. The method is illustrated below using the prediction of the SOC of the cathode in an all-vanadium redox flow battery as an example, and specifically includes the following steps:

[0033] Step 1: Prepare vanadium electrolytes with different positive electrode SOCs and assemble single cells to test charge-discharge curves. In this embodiment, the SOC of the negative electrode in the prepared electrolyte is 0, corresponding to a trivalent vanadium ion concentration of 1.6 M; the SOCs of the positive electrodes are 0, 0.2, 0.4, and 0.6 respectively (if the positive electrode SOC is x, then the tetravalent vanadium ion concentration is 1.6*(1-x) M, and the pentavalent vanadium ion concentration is 1.6*x M).

[0034] Step 2: Perform single-cell charge-discharge cycle tests on the prepared electrolyte. The discharge mode in the charge-discharge cycle test method can be constant power discharge or constant current discharge. This embodiment uses constant current discharge as an example for specific explanation. A complete charge-discharge cycle method is as follows: First, use constant current charging to charge to the cutoff voltage of 1.55V, rest for 30 seconds, and record the initial charging voltage V at the start of constant current charging. C_s and the average voltage V during rest relax1 Then, constant voltage charging is used until the open circuit voltage reaches 1.5V. After resting for 30 seconds, the current density I at the beginning of constant voltage charging is recorded. C_s The current density I at the end of constant voltage charging C_e Average voltage V during rest relax2 Charging capacity C c Finally, constant current discharge was used until the cutoff voltage was 1V, then the circuit was left to stand for 30 seconds to complete one complete charge-discharge cycle. The initial voltage V of the constant current discharge was recorded. d_s and the average voltage V of the resting period relax3 Three charge-discharge cycles were performed at each current density, and the data from the first charge-discharge cycle were discarded. The current density I during constant current charging and discharging was 80 mA cm⁻¹. -2 100mA cm -2 120mA cm -2 140mA cm -2 160mA cm -2 .

[0035] Step 3: Extract the charge and discharge features recorded in each charge and discharge cycle in Step 2 as feature vector X, and establish a database with the positive electrode SOC as the target y (the range of y is 0-1). Feature vector X may also include: the current density I of constant current charge and discharge, and the starting charging voltage V of constant current charging. C_s Average voltage V during rest relax1、 Current density I at the start of constant voltage charging C_s The current density I at the end of constant voltage charging C_e Average voltage V during rest relax2 Charging capacity C c The initial voltage V of constant current discharge d_s and the average voltage V of the resting periodrelax3 Several or all of them. The eigenvector X must include the current density I during constant current charging and discharging, and the starting charging voltage V during constant current charging. C_s The current density I at the start of constant voltage charging C_s The initial voltage of constant current discharge V d_s and the average voltage V of the resting period relax3 .

[0036] Step 4: Use the data from the second charge-discharge cycle at each charge-discharge current density in the database as the training dataset to train the model; use the data from the third charge-discharge cycle as the test dataset to verify the model's accuracy.

[0037] Step 5: Train the data in the training set using a neural network algorithm to establish a prediction model for the SOC of the positive electrode of the vanadium redox flow battery. The parameters that can be configured for the neural network algorithm include: the number of hidden layers, the number of neurons in each hidden layer, the activation function, the optimizer used for fitting the neural network algorithm, and the number of training iterations. In this example, the artificial neural network has 3 hidden layers, with 32, 64, and 128 neurons in each hidden layer, respectively; the activation function is "ReLU", and the optimizer is "SGD". The result of the neural network modeling is as follows: Figures 2a-2c As shown in Table 1, "Train" represents the training set, and the model training accuracy is shown in Table 1. The definition of model accuracy achievement is as follows: the coefficient of determination (R²) for the training set... 2 The mean square error (MSE) is greater than 0.95, the mean square error (MSE) is less than 0.0025, and the mean absolute deviation (MAE) is less than 0.05.

[0038] Step 6: Evaluate the trained vanadium redox flow battery cathode SOC prediction model using data from the test set. The training results are as follows: Figures 2a-2c As indicated by the Test annotations, the accuracy of the test set is shown in Table 1. The definition of model accuracy achievement is as follows: the coefficient of determination (R²) for the test set. 2 The mean square error is greater than 0.95, the root mean square error is less than 0.0025, and the mean absolute deviation is less than 0.05.

[0039] Table 1. Model accuracy of the vanadium redox flow battery cathode SOC prediction model established with three hidden layers, each containing 32, 64, and 128 neurons, respectively.

[0040]

[0041] Step 7: If the model's accuracy does not meet the set standard, it can be optimized by changing the number of hidden layers in the neural network, the number of neurons in each hidden layer, the activation function, the optimizer for fitting the neural network algorithm, the number of training iterations, etc., to achieve higher accuracy.

[0042] Step 8: Save the neural network model that meets the accuracy requirements for subsequent online prediction of the SOC of the cathode in vanadium redox flow batteries.

Claims

1. A method for predicting the state of charge (SOC) of an all-vanadium redox flow battery based on artificial neural networks, characterized in that, Includes the following steps: The charge-discharge characteristics in the charge-discharge cycle are obtained as feature vector X, and the positive electrode SOC is used as target y to establish a database. The second charge-discharge cycle data under each charge-discharge current density in the database are used as training dataset to train the model. The data of the third charge-discharge cycle are used as test dataset to verify the accuracy of the model. The training set is input into the neural network model for training, and the trained vanadium redox flow battery cathode SOC prediction model is obtained. The charge and discharge characteristics during the charge and discharge cycle are collected and used as feature vector X. This feature vector X is then input into the SOC prediction model of the vanadium redox flow battery cathode to obtain the SOC of the vanadium redox flow battery cathode. The process of obtaining charge-discharge characteristics during a charge-discharge cycle includes the following steps: By preparing all-vanadium electrolytes with different positive electrode SOCs and keeping the negative electrode SOC at 0, single cells were assembled for charge-discharge cycle testing. The prepared electrolyte was subjected to single-cell charge-discharge cycle tests. The discharge mode in the charge-discharge cycle test method was constant current charging, then constant voltage charging, and finally constant current discharging. The charge-discharge characteristics were obtained by measuring the current and voltage characteristics of the charge-discharge process. The feature vector X includes: the current density I during constant current charging and discharging, and the starting charging voltage V during constant current charging. C_s Average voltage V during rest relax1、 Current density I at the start of constant voltage charging C_s The current density I at the end of constant voltage charging C_e Average voltage V during rest relax2 Charging capacity C c The initial voltage V of constant current discharge d_s and the average voltage V of the resting period relax3。 2. The method for predicting the state of charge (SOC) of an all-vanadium redox flow battery based on an artificial neural network according to claim 1, characterized in that, The neural network model has 1 to 5 hidden layers, and each hidden layer contains an integer number of neurons; the activation function is any one of Sigmoid, Tanh, ReLU, Leaky ReLU, ELU, PReLU, Softmax, Swish, Maxout, and Softplus; and the neural network optimizer is any one of SGD, adagrad, adadelta, and adam.

3. A vanadium redox flow battery SOC prediction system based on artificial neural networks, the system being used to implement the vanadium redox flow battery SOC prediction method based on artificial neural networks as described in any one of claims 1-2, characterized in that, include: The dataset construction module is used to obtain the charge and discharge characteristics in the charge and discharge cycle as the feature vector X, and the positive electrode SOC as the target y to build a database; the second charge and discharge cycle data under each charge and discharge current density in the database is used as the training dataset to train the model; the data of the third charge and discharge cycle is used as the test dataset to test the model accuracy. The neural network model training module is used to input the training set into the neural network model for training, and obtain the trained vanadium redox flow battery cathode SOC prediction model. The positive electrode SOC prediction module is used to collect charge and discharge characteristics during charge and discharge cycles, which are used as feature vector X and input into the positive electrode SOC prediction model of the vanadium redox flow battery to obtain the positive electrode SOC of the vanadium redox flow battery.

4. A vanadium redox flow battery SOC prediction device based on artificial neural networks, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the vanadium redox flow battery SOC prediction method based on artificial neural networks as described in any one of claims 1-2 when the computer program is executed.

5. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the SOC prediction method for vanadium redox flow batteries based on artificial neural networks as described in any one of claims 1-2.

Images