A SOC evaluation method for zinc-bromine flow batteries based on linear regression prediction

Abstract

The present invention relates to a method for assessing the state of charge (SOC) of zinc-bromine flow batteries based on linear regression prediction, aiming to improve the accuracy and efficiency of state estimation during the battery's charge and discharge processes. The method constructs a database (DB1) to record data such as charge quantity, current, open-circuit voltage, temperature, and rest time during the charge and discharge process. The battery management system regularly performs linear regression predictions, extracts historical data using a data matching algorithm, calculates coulombic efficiency (CE) and energy efficiency (EE), and stores the results and the current cycle count in a database (DB2). The linear regression prediction algorithm is then executed using DB2 to obtain linear equations representing the coulombic efficiency, energy efficiency, and cycle count, thereby predicting subsequent coulombic efficiency and energy efficiency, and improving the accuracy of SOC assessment. This invention has the advantages of improving the accuracy of battery state prediction, efficiently calculating historical data, enhancing adaptability, optimizing battery management strategies, and extending battery life. It also possesses significant technical value in the field of battery applications.

Description

Technical Field

[0001] The invention relates to a method for evaluating the SOC of a zinc-bromine flow battery, and the linear regression prediction technology used involves the field of machine learning. Background Art

[0002] The current estimation of the SOC of zinc-bromine flow batteries is mainly based on the ampere-hour integration method, but the static loss of zinc-bromine flow batteries is relatively serious, and relying solely on ampere-hour integration for SOC estimation will result in large errors. The usual practice is to introduce other measurement data to correct the SOC. The commonly used method is to introduce the battery OCV to correct the SOC. Due to the uneven bromine concentration in the electrolyte of the zinc-bromine flow battery, the OCV curve at both ends of the stack is more discrete, making it difficult to correct the SOC. For example, in a zinc-bromine flow battery real-time SOC detection method disclosed in publication number CN201811249665 on the announcement date of 2020.05.05, a zinc-bromine flow battery SOC estimation method is proposed. The principle is to estimate the SOC value by measuring the conductivity of the battery electrolyte, and then calculate the concentration of the electrolyte by measuring the electrolyte conductivity. Finally, the battery SOC is calculated based on the concentration of the electrolyte. This method requires a sensor to measure the electrolyte concentration. However, due to the highly corrosive nature of the electrolyte in zinc-bromine flow batteries, the sensor cannot be placed in the electrolyte for long periods of time for real-time measurement. It can only be measured offline by manually opening the cover, which has no engineering application value. For example, a method for online calibration of the SOC of zinc-bromine flow batteries, disclosed in publication number CN201710418502 and announced on August 4, 2017, proposes an online calibration method for the SOC of zinc-bromine flow batteries. Based on the ampere-hour integration method, the current of the battery stack is integrated during the charge and discharge process to obtain the total charge Ec during charging and the total discharge Ed during discharging. The total charge of the battery is corrected based on Ec and Ed. This method requires deep charging and discharging of the battery, which is rarely used in actual engineering applications, making it difficult to achieve the conditions and objectives of calibration. Summary of the Invention

[0003] To address the current problem of large errors and difficulty in correcting the SOC (State of Charge) assessment of zinc-bromine flow batteries, the present invention proposes a method based on machine learning and linear regression prediction to improve the accuracy of zinc-bromine flow battery SOC assessment. The implementation scheme of the present invention is as follows: A zinc-bromine flow battery state of charge (SOC) assessment method based on linear regression prediction, the steps of which include:

[0004] a) Collect the current and voltage at both ends of the battery stack and perform power integration to obtain the power during each charge and discharge process;

[0005] b) Calculate the internal resistance of the battery stack;

[0006] c) Calculate the open circuit voltage of the stack based on the internal resistance, current, and measured voltage;

[0007] d) constructing a battery information database DB1 using the charge, current, open circuit voltage, temperature, and rest time during the charge and discharge process;

[0008] e) The battery management system regularly performs linear regression predictions and extracts historical data from the battery information database DB1 based on the real-time open circuit voltage to perform data matching and obtain the interval data of successful matching;

[0009] f) Calculate the coulombic efficiency and energy efficiency using the matched interval data and the open circuit voltage, and store the calculated results and the current cycle number in the database DB2;

[0010] g) Based on the observation data in the database DB2, a linear regression prediction algorithm is executed to obtain a linear equation of coulombic efficiency, energy efficiency and cycle number;

[0011] h) Predict the subsequent coulombic efficiency and energy efficiency based on the fitted linear equation.

[0012] The method steps for data matching in step e are as follows:

[0013] a) extracting the last DB1 record that was successfully matched previously from the battery information database DB1, and using this record as the first DB1 record for this matching;

[0014] b) traversing the remaining DB1 records in the battery information database DB1 in order, extracting each record one by one and comparing them with the first record of DB1;

[0015] c) When two records are matched successfully, the data set between the two DB1 records is returned and the matching is completed.

[0016] The extracted DB1 records are assigned different weights according to their data types. The data in the two DB1 records are compared. If there is identical data, the corresponding weight is increased. When the weight exceeds a set threshold, the two DB1 records are considered to be matched successfully.

[0017] The steps of the linear regression prediction algorithm are as follows:

[0018] a) retrieving all DB2 records from a database DB2, wherein the DB2 records contain data on coulombic efficiency, energy efficiency, and cycle number;

[0019] b) Remove the DB2 records with large deviations and sort the remaining DB2 records according to the number of cycles. The DB2 records with larger cycle numbers have higher weights.

[0020] c) listing the remaining DB2 records into a system of equations and solving a fitting function;

[0021] d) Predict the efficiency of the next few cycles based on the fitted function.

[0022] The linear regression prediction algorithm needs to eliminate records with large deviations from the data in the database DB2, and the elimination criteria are dynamically adjusted according to historical data and the accuracy of the prediction.

[0023] The open circuit voltage of the battery is calculated according to the formula: open circuit voltage = measured voltage - current * stack internal resistance.

[0024] The calculation formula of the coulombic efficiency is:

[0025] Coulombic efficiency = discharge capacity / charge capacity.

[0026] The energy efficiency is calculated as follows:

[0027] Energy efficiency = discharge capacity / charge capacity.

[0028] The linear regression prediction algorithm uses a univariate linear equation or a multivariate linear equation to fit the data in the database DB2.

[0029] The data of power, current, open circuit voltage, temperature and rest time in the database DB1 are collected and recorded in real time by the battery management system.

[0030] Beneficial effects: A linear regression algorithm is used to evaluate and correct the SOC of zinc-bromine flow batteries. The battery management system automatically records data during operation and regularly executes the linear regression prediction algorithm without the need for user intervention or additional calibration operations. Correction through the regression prediction algorithm greatly improves its accuracy. Compared with the method of measuring SOC by measuring electrolyte concentration, this method does not require additional hardware and reduces costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is the overall flow chart of the SOC evaluation method of zinc-bromine flow battery based on linear regression prediction.

[0032] Figure 2 This is the matching algorithm flow chart of the SOC evaluation method of zinc-bromine flow battery based on linear regression prediction.

[0033] Figure 3 This is the regression algorithm flow chart of the SOC evaluation method of zinc-bromine flow battery based on linear regression prediction.

[0034] Figure 4 This is a data graph of the regression prediction algorithm for coulombic efficiency and battery cycle number in zinc-bromine flow batteries.

[0035] Figure 5 Data graph of the regression prediction algorithm for energy efficiency and battery cycle number in zinc-bromine flow batteries.

[0036] Figure 6 It is a regression curve diagram of energy efficiency and coulombic efficiency in the fuel cell stack in the embodiment. DETAILED DESCRIPTION

[0037] The following will describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of this application.

[0038] Accuracy test of the method

[0039] 1) After using the sensing modules at both ends of the battery stack to collect basic battery information such as current, voltage, and internal resistance, the amount of electricity in each charge and discharge process is calculated using the ampere-hour integration method, and the open-circuit voltage of the battery stack is calculated.

[0040] 2) Use the power, current, open circuit voltage, temperature, and rest time during the charging and discharging process to build a battery information database DB1.

[0041] 3) The battery management system regularly performs linear regression predictions and extracts historical data from the battery information database DB1 based on the real-time open circuit voltage to perform data matching and obtain the interval data of successful matching;

[0042] 4) Calculate the coulombic efficiency and energy efficiency using the matched interval data and open circuit voltage, and store the calculated results and the current cycle number in the database DB2;

[0043] 5) Based on the observation data in database DB2, a linear regression prediction algorithm is executed to obtain a linear equation about coulombic efficiency, energy efficiency and cycle number;

[0044] 6) Predict the subsequent coulombic efficiency and energy efficiency based on the fitted linear equation.

[0045] 7) The coulombic efficiency and energy efficiency of the battery at 100, 200, 300 and 400 cycles were statistically analyzed, and the measured data were used for regression analysis and prediction.

[0046] 8) If Figure 4 and Figure 5As shown, the regression prediction algorithm corrects the coulombic efficiency and energy efficiency. Although the initial efficiency deviation is large, as the number of cycles increases, the program accumulates, analyzes and predicts the data, and can effectively converge the predicted values ​​to the actual values. The regression predictions of coulombic efficiency and energy efficiency gradually approach the actual values. In addition, because each fuel cell stack has its own database, the data input to the algorithm is most realistic, and the accuracy is significantly superior to traditional offline regression prediction algorithms.

[0047] Example

[0048] This method was applied to a zinc-bromine liquid flow battery module system to estimate the SOC of each battery stack. After the module was assembled, the coulombic efficiency and energy efficiency of the battery stack were set to 90% and 75%, respectively, based on historical experience.

[0049] During the first five cycles, the battery management system estimates the stack SOC based on the programmed performance parameters. During battery operation, the battery management system processes the measured data using the aforementioned method, records the data, and saves it in a time-series database DB1.

[0050] After the module cycles five times, the battery management system obtains enough data and starts executing the matching algorithm. It traverses DB1 to find the DB1 record pair with the highest matching degree. After a successful search, it extracts the data of the two DB1 record intervals, statistics the charging capacity, charging energy, discharge capacity, and discharge energy in the DB1 records, calculates the coulombic efficiency and energy efficiency of the interval, and saves the calculation results in the database DB2.

[0051] When the data in database DB2 exceeds 3, linear regression prediction is performed on database DB2 to predict the coulombic efficiency and energy efficiency of subsequent cycles. The following table is the DB2 data table of one of the fuel cell stacks. Since the first record has a large deviation, it is marked as an invalid record. Only the last 3 data are linearly fitted. The fitting results are as follows: Figure 6 As stated.

[0052] The linear equation for coulombic efficiency and cycle number is Ec = -0.0002x + 0.8811, and the fitted function for energy efficiency is Ee = -0.0003x + 0.7306. These two equations can be used to calculate the efficiency of subsequent charge and discharge cycles, thereby accurately estimating the stack's SOC.

[0053] Table 1 Kunlun efficiency and energy efficiency recorded in the DB2 database as the number of cycles increases

[0054]

Claims

1. A method for evaluating the SOC of a zinc-bromine flow battery based on linear regression prediction, characterized in that: The steps include: a) Collect the current and voltage at both ends of the battery stack and perform power integration to obtain the power during each charge and discharge process; b) Calculate the internal resistance of the battery stack; c) Calculate the open circuit voltage of the stack based on the internal resistance, current, and measured voltage; d) constructing a battery information database DB1 using the charge, current, open circuit voltage, temperature, and rest time during the charge and discharge process; e) The battery management system regularly performs linear regression predictions and extracts historical data from the battery information database DB1 based on the real-time open circuit voltage to perform data matching and obtain the interval data of successful matching; f) Calculate the coulombic efficiency and energy efficiency using the matched interval data and open circuit voltage, and store the calculated results and the current cycle number into the database DB2; g) Based on the observation data in the database DB2, a linear regression prediction algorithm is executed to obtain a linear equation about coulombic efficiency, energy efficiency and cycle number; h) Predict the subsequent coulombic efficiency and energy efficiency based on the fitted linear equation; The data matching method in step e is as follows: a) extracting the last DB1 record that was successfully matched in the previous time from the battery information database DB1, and using this record as the first DB1 record matched this time; b) traversing the remaining DB1 records in the battery information database DB1 in order, taking out each record and comparing them one by one with the first record of DB1; c) When two records are matched successfully, the data set between the two DB1 records is returned and the matching is completed; The data matching method in step e is as follows: the extracted DB1 records are assigned different weights according to their data types, and the data in the two DB1 records are compared. If there is identical data, the corresponding weight is increased. When the weight exceeds a set threshold, the two DB1 records are considered to be matched successfully.

2. The method for evaluating the SOC of a zinc-bromine flow battery based on linear regression prediction according to claim 1, characterized in that: The steps of the linear regression prediction algorithm are as follows: a) retrieving all DB2 records from a database DB2, wherein the DB2 records contain data on coulombic efficiency, energy efficiency, and cycle number; b) Remove the DB2 records with large deviations and sort the remaining DB2 records according to the number of cycles. The DB2 records with larger cycle numbers have higher weights. c) listing the remaining DB2 records into a system of equations and solving a fitting function; d) Predict the efficiency of the next cycle based on the fitted function.

3. The method for evaluating the SOC of a zinc-bromine flow battery based on linear regression prediction according to claim 1 or 2, characterized in that: The linear regression prediction algorithm needs to eliminate records with large deviations from the data in the database DB2, and the elimination criteria are dynamically adjusted according to historical data and the accuracy of the prediction.

4. The method for evaluating the SOC of a zinc-bromine flow battery based on linear regression prediction according to claim 1, wherein: The open circuit voltage of the battery is calculated according to the formula: open circuit voltage = measured voltage - current * stack internal resistance.

5. The method for evaluating the SOC of a zinc-bromine flow battery based on linear regression prediction according to claim 1, characterized in that: The calculation formula of the coulombic efficiency is: coulombic efficiency=discharge capacity / charge capacity.

6. The method for evaluating the SOC of a zinc-bromine flow battery based on linear regression prediction according to claim 1, characterized in that: The energy efficiency is calculated as follows: energy efficiency = discharge power / charge power.

7. The method for evaluating the SOC of a zinc-bromine flow battery based on linear regression prediction according to claim 1, characterized in that: The linear regression prediction algorithm uses a univariate linear equation or a multivariate linear equation to fit the data in the database DB2.

8. The method for evaluating the SOC of a zinc-bromine flow battery based on linear regression prediction according to claim 1, characterized in that: The data of power, current, open circuit voltage, temperature and rest time in the database DB1 are collected and recorded in real time by the battery management system.

Images