Real-time SOC prediction method for all-vanadium redox flow battery system

Abstract

The invention discloses a real-time SOC prediction method for an all-vanadium redox flow battery system. The method comprises the following steps: sampling positive and negative electrolyte in the charge-discharge cycle of the all-vanadium redox flow battery system to determine the concentration of positive tetravalent and pentavalent vanadium ions and the concentration of negative divalent and trivalent vanadium ions; recording the charging and discharging parameters of the positive and negative electrolyte sampling moments in the charging and discharging cycle; taking the charging and discharging parameters as feature vectors X; taking the concentration of positive tetravalent and pentavalent vanadium ions and the concentration of negative divalent and trivalent vanadium ions as a target function y, and establishing a database; dividing data in the database into a training set and a test set; and training the data in the training set by adopting an artificial intelligence algorithm, and establishing a prediction model for the concentration of the positive electrode tetravalent and pentavalent vanadium ions and the concentration of the negative electrode divalent and trivalent vanadium ions of the all-vanadium redox flow battery. According to the method, the concentration of the active substances in the positive and negative electrolyte of the all-vanadium redox flow battery system in the charge-discharge cycle process can be predicted in real time, and the real-time monitoring of the SOC of the all-vanadium redox flow battery system is realized.

Description

Technical Field

[0001] The present application relates to a method for real-time SOC prediction of a vanadium redox flow battery system, belonging to the technical field of vanadium redox flow batteries. Background Art

[0002] Monitoring the state of charge (SOC) of a vanadium redox flow battery system is very important for the safety and stability of the system. A reliable real-time prediction system for the SOC of a vanadium redox flow battery can significantly improve the stability of the system and reduce the operation and maintenance costs. At present, the main methods for estimating the SOC of a vanadium redox flow battery system include: discharge experiment method, ampere-hour integration method, open-circuit voltage method, Kalman filtering method, etc. The difficulty in real-time SOC prediction of a vanadium redox flow battery system lies in that during the operation of the vanadium redox flow battery system, the migration of the electrolyte will cause the imbalance of the positive and negative electrolytes, making the real-time estimated SOC inaccurate. The discharge experiment method has high accuracy but cannot be measured online; the ampere-hour integration method and the open-circuit voltage method cannot consider the accuracy degradation caused by electrolyte migration; the Kalman filtering method depends greatly on the accuracy of the battery model and cannot consider the influence caused by electrolyte migration during the operation of the vanadium redox flow battery. Therefore, these methods cannot achieve real-time prediction of the system SOC in a large-scale demonstration and commercial operation vanadium redox flow battery system. Summary of the Invention

[0003] According to one aspect of the present application, a method for real-time SOC prediction of a vanadium redox flow battery system is provided. Aiming at the limitations and complexity of the existing SOC detection methods, the present application only uses the charge and discharge parameter data during the charge and discharge process of the vanadium redox flow battery system to realize real-time online prediction of SOC without introducing additional complex acquisition instruments and measurement devices, and solves the problem that it is difficult to balance high accuracy and real-time performance of the existing SOC detection methods.

[0004] The present application adopts the following technical solutions:

[0005] A method for real-time SOC prediction of a vanadium redox flow battery system includes the following steps:

[0006] S1. Perform charge and discharge cycle tests on the vanadium redox flow battery system, take samples of the electrolyte in the electrolyte storage tank several times during the charge and discharge cycle process, obtain the vanadium ion concentration in the electrolyte, and record the charge and discharge parameters at the sampling moment;

[0007] S2. Establish a database with the charge and discharge parameters recorded in step S1 as the feature vector X and the vanadium ion concentration in the electrolyte obtained in step S1 as the objective function y. Divide the data in the database into a training set and a test set, and use an artificial intelligence algorithm to train the data in the training set and perform data modeling. After data verification and model parameter tuning, a prediction model for predicting the vanadium ion concentration in the electrolyte is obtained;

[0008] During the operation of the all-vanadium redox flow battery system, charge and discharge parameters are collected in real time, and the vanadium ion concentration in the electrolyte in real time is predicted through the prediction model in step S2, and then the real-time SOC of the all-vanadium redox flow battery system is calculated according to formula (1):

[0009]

[0010] In the formula, [V 2+ , [V 3+ , [VO 2+ and respectively represent the divalent, trivalent, tetravalent and pentavalent vanadium ion concentrations.

[0011] Optionally, in step S1, the obtaining of the vanadium ion concentration in the electrolyte includes:

[0012] Obtaining the concentrations of tetravalent and pentavalent vanadium ions in the positive electrolyte storage tank, and / or obtaining the concentrations of divalent and trivalent vanadium ions in the negative electrolyte storage tank.

[0013] Optionally, in step S1, the charge and discharge parameters are selected from at least one of current, voltage, average open-circuit voltage during standby, charge capacity during charging process, and discharge capacity during discharge process.

[0014] Optionally, in step S1, the charge and discharge cycles include n groups of charge and discharge cycles, where n>1;

[0015] The sampling process includes:

[0016] During the charge cycle, at least 4 samplings are performed, with the first sampling within 1 minute after the start of charging and the last sampling within 1 minute before the end of charging;

[0017] During the discharge cycle, at least 4 samplings are performed, with the first sampling within 1 minute after the start of charging and the last sampling within 1 minute before the end of charging.

[0018] Optionally, in step S1, the process of obtaining the vanadium ion concentration in the electrolyte after sampling is: detecting the vanadium ion concentration in the electrolyte by titration method. In step S1, the process of obtaining the vanadium ion concentration in the electrolyte after sampling is: detecting the vanadium ion concentration in the electrolyte by titration method. For the vanadium ion concentration corresponding to the moment when no sampling is performed, it is calculated by interpolation according to the vanadium ion concentrations titrated in two adjacent samplings and the capacity difference of charge and discharge according to formula (2):

[0019]

[0020] In the formula, c(t 1 ), c(t 2 ), c(t i) represent the vanadium ion concentrations at the first sampling point, the second sampling point, and at time t i respectively, C(t 1 ), C(t 2 ), C(t i ) represent the charge or discharge capacities at the first sampling point, the second sampling point, and at time t i respectively.

[0021] Optionally, in step S2, using the continuous n sets of charge-discharge parameters recorded in step S1 as the feature vector X, and using the vanadium ion concentration in the electrolyte corresponding to the nth charge-discharge cycle as the objective function y to establish a database.

[0022] Optionally, using any m 1 sets of cycle data in the database as the training set for the charge cycle;

[0023] using any z 1 sets of cycle data in the database as the test set for the charge cycle;

[0024] using any m 2 sets of cycle data in the database as the training set for the discharge cycle;

[0025] using any z 2 sets of cycle data in the database as the test set for the discharge cycle;

[0026] wherein, m 1 , z 1 , m 2 , z 2 are independently less than n.

[0027] Optionally, in step S2, the artificial intelligence algorithm is selected from one of the autoregressive integrated moving average algorithm (ARIMA), Prophet algorithm, gradient boosting machine algorithm (GBM), Seq2Seq class algorithm, TCN algorithm, WaveNet algorithm, Informer algorithm, Transformer algorithm.

[0028] Optionally, the Seq2Seq class algorithm is selected from one of the recurrent neural network RNN, long short-term memory neural network LSTM, autoregressive recurrent neural network DeepAR.

[0029] Optionally, when the artificial intelligence algorithm is the long short-term memory neural network (LSTM, Long short-term Memory) algorithm, the activation function is "relu", the optimizer of the neural network is "Adam", the number of hidden layers is at least 1 layer, the number of neurons in each layer is at least 8, and a part of the neurons are randomly discarded during the training process to prevent overfitting.

[0030] Optionally, it is characterized in that in step S2, the data modeling process further includes using a regularization method to prevent overfitting;

[0031] The regularization method is selected from one or at least any two composite regularization methods of L1 regularization, L2, and Dropout.

[0032] Optionally, the parameter of the regularization method is (0, 1].

[0033] Optionally, the range of neurons randomly discarded is (0, 1).

[0034] Optionally, the prediction model for predicting the concentration of vanadium ions in the electrolyte is a concentration prediction model for tetravalent and pentavalent vanadium ions, or a concentration prediction model for divalent and trivalent vanadium ions.

[0035] The beneficial effects that can be produced by this application include:

[0036] The real-time SOC prediction method for a vanadium redox flow battery system takes parameters such as current, voltage, and charge-discharge capacity during the charge-discharge process of the vanadium redox flow battery system as input features, and establishes a concentration prediction model for vanadium ions of each valence state at the positive and negative electrodes through an artificial intelligence algorithm. It can utilize the real-time data during the charge-discharge process to perform online prediction on the electrolyte SOC state of the vanadium redox flow battery after the positive and negative electrodes are unbalanced. The prediction is sensitive and the result is accurate. This method has low cost, does not introduce any additional data acquisition devices and detection instruments, is easy to implement and maintain. The SOC prediction model for the positive electrode of the vanadium redox flow battery established through the parameter data of the charge-discharge process of a single cell can be applied to large-scale demonstration and commercial systems. Description of the Drawings

[0037] Figure 1 It is a schematic flow chart of the implementation steps of the real-time SOC prediction method for the vanadium redox flow battery system of this application.

[0038] Figure 2 It is a comparison between the model prediction results and the true values of the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode during the charging process in Example 1 of this application.

[0039] Figure 3 It is a comparison between the model prediction results and the true values of the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode during the discharging process in Example 1 of this application. Detailed Embodiments

[0040] The following describes this application in detail with reference to the embodiments, but this application is not limited to these embodiments.

[0041] Unless otherwise specified, the raw materials in the embodiments of this application are all purchased through commercial channels.

[0042] Unless otherwise specified, the test methods are all conventional methods, and the instrument settings are all the settings recommended by the manufacturer.

[0043] Example 1

[0044] As Figure 1 shown in the flow of the implementation steps, for the prediction model of the tetravalent and pentavalent vanadium ion concentrations at the positive electrode of the all-vanadium redox flow battery system, the method for realizing the real-time SOC prediction of the all-vanadium redox flow battery system is as follows:

[0045] Step 1: Perform charge and discharge tests on the all-vanadium redox flow battery system, and take samples of the positive and negative electrode electrolytes during the charge and discharge tests for titration to determine the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode and the concentrations of divalent and trivalent vanadium ions at the negative electrode. In this example, the SOC of the all-vanadium redox flow battery system is determined based on the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode; the rated power of the system is 5 kW. The charge and discharge test process is as follows: First, set aside for 30 seconds, then perform constant power charging at 5 kW, and when the charging reaches the cut-off voltage of 1.55 V, switch to constant voltage charging at 1.53 V, charge until 100% SOC, and set aside for 30 seconds; finally, perform constant power discharge until the cut-off voltage is 1 V to complete 1 charge and discharge cycle; the sampling time for titration is as follows: For one charging cycle, at least 4 samples are taken, 1 sample is taken within 1 minute after the start of charging, 1 sample is taken within 1 minute before the end of charging, and at least 2 more samples are taken during the charging process; for one discharge cycle, at least 4 samples are taken, 1 sample is taken within 1 minute after the start of discharge, 1 sample is taken within the last 1 minute before the end of discharge, and at least 2 more samples are taken during the discharge process.

[0046] Step 2: Based on the titrated concentrations of tetravalent and pentavalent vanadium ions at the positive electrode, calculate the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode in this cycle by interpolation according to the charge and discharge capacity; for the vanadium ion concentrations corresponding to the moments without sampling, they are calculated by interpolation according to the vanadium ion concentrations of the two adjacent sampling titrations and the charge and discharge capacity difference according to formula (2):

[0047]

[0048] In the formula, c(t 1 ), c(t 2 ), c(t i ) respectively represent the vanadium ion concentrations at the first sampling point, the second sampling point, and the t i moment, and C(t 1 ), C(t 2 ), C(t i ) respectively represent the charge or discharge capacities at the first sampling point, the second sampling point, and the t i moment.

[0049] Using the characteristics of consecutive n charge-discharge cycles as the input feature X, and the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode corresponding to the nth charge-discharge cycle as the objective function y, a database is established. The characteristics of the charge-discharge cycle include: current, voltage, average open-circuit voltage, charge capacity (charging process), discharge capacity (discharging process); n is an integer greater than or equal to 1.

[0050] Step 3: Divide the data in the database into a training set and a test set. In this example, for the charging cycle: use the data of the 9th, 15th, 55th, and 81st charging cycles as the training set, and use the data of the 88th charging cycle as the test set; for the discharging cycle: use the data of the 2nd, 21st, 29th, and 39th discharging cycles as the training set, and use the data of the 61st discharging cycle as the test set.

[0051] Step 4: Use an artificial intelligence algorithm to train the data in the training set to establish a prediction model for the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode during the charging and discharging processes of the all-vanadium redox flow battery. The artificial intelligence uses the long short-term memory neural network (LSTM, Long short-term Memory) algorithm, with the number of hidden layers being 1, the number of neurons being 32 respectively, the activation function being "relu", the optimizer of the neural network being "Adam", using the L2 regularization method to prevent overfitting, the parameter being 0.01, randomly discarding 20% of the neurons during the training process to prevent overfitting. The predicted results and actual results of the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode are as Figure 2 shown, and the training accuracy of the model is shown in Table 1.

[0052] Step 5: Use the data in the test set to evaluate the trained prediction model for the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode of the all-vanadium redox flow battery. The training results are as Figure 2 shown, and the accuracy of the test set is shown in Table 1.

[0053] Step 6: When the all-vanadium redox flow battery system is operating, collect the charge-discharge parameter data in real time, and predict the concentration of vanadium ions in the positive electrode electrolyte in real time through the prediction model of the concentrations of tetravalent and pentavalent vanadium ions at the positive electrode, and then calculate the real-time SOC of the all-vanadium redox flow battery system according to formula (1):

[0054]

[0055] In the formula, [V 2+ , [V 3+ , [VO 2+ and represent the concentrations of divalent, trivalent, tetravalent, and pentavalent vanadium ions respectively.

[0056] Table 1 Model accuracy of the prediction model for tetravalent and pentavalent vanadium at the positive electrode of the all-vanadium redox flow battery system

[0057]

[0058] As described above, these are only several embodiments of the present application and do not impose any form of limitation on the present application. Although the present application is disclosed above with preferred embodiments, it is not intended to limit the present application. Any person skilled in the relevant art can make some changes or modifications within the scope of the technical solution of the present application by using the disclosed technical content, and these are all equivalent to equivalent embodiments and fall within the scope of the technical solution.

Claims

1. A real-time SOC prediction method for a vanadium redox flow battery system, characterized in that, it includes the following steps: S1. Conduct charge-discharge cycle tests on the vanadium redox flow battery system, take several samples of the electrolyte in the electrolyte storage tank during the charge-discharge cycle process, obtain the vanadium ion concentration in the electrolyte, and record the charge-discharge parameters at the sampling time; S2. Use the charge-discharge parameters recorded in step S1 as the feature vector X, and use the vanadium ion concentration in the electrolyte obtained in step S1 as the objective function y to establish a database. Divide the data in the database into a training set and a test set, and use an artificial intelligence algorithm to train the data in the training set and perform data modeling. After data verification and model parameter tuning, obtain a prediction model for predicting the vanadium ion concentration in the electrolyte; S3. When the vanadium redox flow battery system is operating, collect the charge-discharge parameters in real time, and predict the vanadium ion concentration in the real-time electrolyte through the prediction model in step S2. Then calculate the real-time SOC of the vanadium redox flow battery system according to formula (1): wherein, [V 2+ , [V 3+ , [VO 2+ and respectively represent the divalent, trivalent, tetravalent and pentavalent vanadium ion concentrations.

2. The real-time SOC prediction method for a vanadium redox flow battery system according to claim 1, characterized in that, in step S1, the obtaining of the vanadium ion concentration in the electrolyte includes: obtaining the concentrations of tetravalent and pentavalent vanadium ions in the positive electrolyte storage tank, and / or obtaining the concentrations of divalent and trivalent vanadium ions in the negative electrolyte storage tank.

3. The real-time SOC prediction method for a vanadium redox flow battery system according to claim 1, characterized in that, in step S1, the charge-discharge parameters are selected from at least one of current, voltage, open-circuit voltage, charge capacity during the charging process, and discharge capacity during the discharging process.

4. The real-time SOC prediction method for a vanadium redox flow battery system according to claim 1, characterized in that, in step S1, the charge-discharge cycle includes n groups of charge-discharge cycles, where n > 1; the sampling process includes: during the charging cycle, at least 4 samplings are performed, with the first sampling within 1 minute after the start of charging and the last sampling within 1 minute before the end of charging; during the discharging cycle, at least 4 samplings are performed, with the first sampling within 1 minute after the start of charging and the last sampling within 1 minute before the end of charging.

5. The real-time SOC prediction method for a vanadium redox flow battery system according to claim 1, characterized in that, in step S1, the process of obtaining the vanadium ion concentration in the electrolyte after sampling is: detecting the vanadium ion concentration in the electrolyte by titration method. For the vanadium ion concentration corresponding to the time when no sampling is performed, it is calculated by interpolation according to formula (2) based on the vanadium ion concentrations titrated in two adjacent samplings and the capacity difference of charge and discharge; Wherein, c(t 1 )、c(t 2 )、c(t i ) respectively represent the vanadium ion concentrations at the first sampling point, the second sampling point and at time t i . C(t 1 )、C(t 2 )、C(t i ) respectively represent the charge or discharge capacities at the first sampling point, the second sampling point and at time t i .

6. The real-time SOC prediction method for a vanadium redox flow battery system according to claim 4, characterized in that, in step S2, use the continuous n groups of charge-discharge parameters recorded in step S1 as the feature vector X, and use the vanadium ion concentration in the electrolyte corresponding to the nth group of charge-discharge cycles as the objective function y to establish a database.

7. The real-time SOC prediction method for a vanadium redox flow battery system according to claim 6, characterized in that, Take any m 1 sets of cyclic data in the database as the training set for the charging cycle; Use any z 1 sets of cyclic data in the database as the test set for the charging cycle; Take any m 2 sets of cyclic data in the database as the training set for the discharge cycle; Any z in the database 2 The data of the cyclic group is used as the test set for the discharge cycle; where m 1 , z 1 , m 2 , z 2 are each independently less than n.

8. The real-time SOC prediction method for the all-vanadium redox flow battery system according to claim 1, in step S2, the artificial intelligence algorithm is selected from one of the autoregressive integrated moving average algorithm, Prophet algorithm, gradient boosting machine algorithm, Seq2Seq class algorithm, TCN algorithm, WaveNet algorithm, Informer algorithm, and Transformer algorithm; The Seq2Seq class algorithm is selected from one of the recurrent neural network RNN, long short-term memory neural network LSTM, and autoregressive recurrent neural network DeepAR.

9. The real-time SOC prediction method for the all-vanadium redox flow battery system according to claim 1, characterized in that, characterized in that, in step S2, the data modeling process further includes using a regularization method to prevent overfitting; The regularization method is selected from one of L1 regularization, L2, Dropout, or a composite regularization method of at least any two of them.

10. The real-time SOC prediction method for the all-vanadium redox flow battery system according to claim 2, characterized in that, the prediction model for predicting the concentration of vanadium ions in the electrolyte is a concentration prediction model for tetravalent and pentavalent vanadium ions, or a concentration prediction model for divalent and trivalent vanadium ions.

Images