Method for analyzing performance of flow battery based on current density distribution measurement

Abstract

The application discloses a liquid flow battery performance analysis method based on current density distribution measurement and relates to the technical field of batteries.Compared with previous battery performance measurement technologies, the application solves the problems of narrow data collection dimension, lack of dynamic adaptability of sampling frequency, lack of consideration of different operation conditions and state of charge, lack of comprehensive performance evaluation and prediction under multi-parameter fusion, etc.Combining real-time environmental parameters and block data, the application enriches and comprehensively collects data dimensions, analyzes key performance indicators of the battery in real time, adjusts the sampling frequency according to the dynamic change of the indicators in the charging and discharging process, analyzes the optimal operation condition of the liquid flow battery, and improves the subsequent optimization ability of the overall performance of the battery.Further, the application integrates multiple technical means, comprehensively evaluates the performance of the battery from different aspects, and can also mine the optimal operation mode and potential problems of the battery under different operation conditions based on the analysis and prediction results.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of batteries, and particularly relates to a flow battery performance analysis method based on current density distribution measurement. BACKGROUND

[0002] As an advanced electrochemical energy storage technology, flow batteries have great potential applications in renewable energy storage (such as solar energy, wind energy) and grid peak shaving. Its unique structure and working principle enable it to achieve large-capacity and long-time energy storage. However, the internal processes of flow batteries involve complex electrochemical processes, including ion diffusion in electrolyte, charge transfer at electrode surface, and electrolyte flow in the battery system. These processes interact with each other and dynamically change during the charging and discharging process of the battery, making it a challenging task to accurately evaluate and track the performance of flow batteries.

[0003] Traditional battery performance test methods often focus on the measurement of macroscopic performance indicators, such as simply recording charging and discharging time, voltage and current to calculate energy and power-related parameters. These methods are of limited help in understanding the microscopic electrochemical processes inside the flow battery. Moreover, in actual operation, the performance of flow batteries will be dynamically affected by various factors (such as temperature changes, electrolyte concentration fluctuations, current density changes, etc.), and traditional measurement indicators cannot capture the impact of these changes on battery performance in real time.

[0004] The existing Chinese patent with publication number CN116148681A discloses a method for predicting the performance of iron-chromium flow batteries, which includes: obtaining target iron-chromium flow battery energy storage stack charging and discharging experimental data, wherein the charging and discharging experimental data is obtained under the same system complete charging and discharging cycle state; arranging the charging and discharging experimental data and converting the charging and discharging experimental data into independent variables; according to the mechanism of flow batteries, the independent variables are used as characteristic input variables to construct a multiple linear regression model, and the battery performance indicators are predicted by the multiple linear regression model. The present application can predict the performance of iron-chromium flow batteries, especially the Coulomb efficiency, voltage efficiency and energy efficiency of iron-chromium flow batteries, by establishing a multiple linear regression model to predict battery performance, which can reduce the experimental cost of the research and development team, shorten the research and development time, and improve the economic benefit.

[0005] Although the above methods can meet most scenarios, research and practical application of the above methods and existing technologies have revealed that the above methods and existing technologies have at least the following defects:

[0006] 1. Only the macroscopic overall parameters of the battery during operation are measured, and the data dimension is relatively single, which makes it difficult to fully reflect the actual performance of the battery during the entire charge and discharge cycle and the comprehensive status under different working conditions.

[0007] 2. During the data collection process, the sampling frequency cannot be flexibly adjusted according to the actual battery conditions, resulting in the inability to accurately capture data under certain rapidly changing battery conditions, or excessive data is collected when it is not necessary, wasting resources.

[0008] 3. The diversity of battery performance under different operating conditions and states of charge is not fully considered, and the accuracy and applicability of the prediction method under different working conditions will be limited.

[0009] 4. It mainly relies on simulation and corresponding flow velocity distribution analysis. The technical means are relatively simple, making it difficult to predict potential fault hazards in battery operation and the degree of impact on the overall performance of the battery. There are deficiencies in the comprehensive control of battery performance.

[0010] In order to solve the above problems, the present invention proposes a flow battery performance analysis method based on current density distribution measurement. Summary of the Invention

[0011] The purpose of the present invention is to propose a flow battery performance analysis method based on current density distribution measurement to solve the problems raised in the background technology:

[0012] The data acquisition dimension is narrow; the sampling frequency lacks dynamic adaptability; different operating conditions and charge states are not taken into account; and there is a lack of comprehensive performance evaluation and prediction under multi-parameter fusion.

[0013] In order to achieve the above object, the present invention adopts the following technical solutions:

[0014] The flow battery performance analysis method based on current density distribution measurement includes the following steps:

[0015] S1: Import real-time environmental parameter data related to battery operation, preset the initial sampling frequency and perform data collection, and set the sampling frequency adjustment rules;

[0016] S2: Run the preset charge and discharge program, measure and store data based on the printed circuit board with block grid design, adjust the operating conditions, and record the battery data and each block data under different operating conditions;

[0017] S3: Calculate the charge and discharge capacity and energy under different operating conditions, calculate the pump work based on the electrolyte flow rate, calculate the coulombic efficiency, voltage efficiency, energy efficiency and system efficiency under the different operating conditions, and draw the charge and discharge curves and efficiency curves to analyze the optimal operating conditions of the battery;

[0018] S4: Filtering the block data recorded under the above conditions, and drawing the current density and voltage curve of each block as the battery state of charge changes based on the area of ​​each block; generating a current density distribution map based on the position of each block and the measurement principle;

[0019] S5: Calculate the current density distribution uniformity coefficient under different operating conditions and different battery states of charge, draw a curve of the current density distribution uniformity coefficient as it changes with the battery state of charge, and calculate its average value;

[0020] S6: Select seconds as the time characteristic period and extract the time characteristics of the current density, voltage curve and current density distribution uniformity coefficient; use the minimum-maximum normalization method to normalize the current density and voltage data to obtain a data set, and divide the data set into a training set, a validation set and a test set;

[0021] S7: Build a first machine learning model based on the long short-term memory network, use mean square error, mean absolute error, and root mean square error to evaluate the prediction performance of the first machine learning model, and predict the current density and voltage change trends of each block and the change trend of the overall current density distribution uniformity coefficient;

[0022] S8: Build an anomaly monitoring model using an unsupervised learning algorithm to monitor the prediction results of the first machine learning model for anomalies. If no anomaly is detected during monitoring, continue to monitor the prediction results of the first machine learning model for anomalies. If an anomaly is detected during monitoring, classify the detected anomaly data based on the anomaly classification model built based on the decision tree to obtain the anomaly type.

[0023] S9: Screening the current density distribution cloud map under different operating conditions; extracting current density distribution data based on the screened current density distribution cloud map, performing data cleaning and normalization on the current density distribution data, marking the flow dead zone in the current density distribution cloud map, and dividing the training set, validation set, and test set based on this;

[0024] S10: Build a second machine learning model based on a convolutional neural network, select cross entropy loss and squared absolute error loss as loss functions, use Adam as the optimizer, train the second machine learning model using the training set, adjust the hyperparameters of the second machine learning model using the validation set, and verify the accuracy of the second machine learning model using the test set;

[0025] S11: Use the second machine learning model obtained through training optimization to process the current density distribution data under different operating conditions, automatically mark the flow dead zone location within the electrode frame, and summarize it;

[0026] S12: All processed and analyzed data are classified and saved according to time sequence and data type, and the stored data is visualized based on a visualization interface.

[0027] Preferably, the environmental parameter data in S1 includes temperature, humidity and pressure;

[0028] The sampling frequency adjustment rules described in S1 are as follows:

[0029] When the key performance indicator is within the preset threshold range, the sampling frequency is gradually reduced according to the preset time interval until it is reduced to the preset minimum sampling frequency;

[0030] When the key performance indicator exceeds the preset threshold range, the sampling frequency is increased until the performance stabilizes again or reaches the preset high-frequency sampling frequency;

[0031] In the sampling frequency adjustment rule, the minimum sampling frequency and the high frequency sampling frequency are optimized and adjusted based on a machine learning algorithm.

[0032] Preferably, in S2, during the data acquisition process, the key performance indicators of the battery are analyzed in real time, and corresponding sampling frequency adjustment rules are executed according to the key performance indicators; the key performance indicators include current density change rate, voltage fluctuation amplitude and power change rate.

[0033] Preferably, the printed plate in S2 is designed with several block grids on its surface, the interval between each block grid is not less than 0.5 mm and not more than 2 mm, the groove of each block grid is filled with epoxy resin, and after it solidifies, the back of the bipolar plate is cut until the solidified resin is exposed to obtain a printed plate based on the block grid design.

[0034] Preferably, the voltage of each block in S5 is the voltage collected by the data acquisition card, the current of each block is the voltage of each block divided by the resistance of the resistor connected to each block, and the current density is the current of each block divided by the area corresponding to each block.

[0035] Preferably, the current density distribution uniformity in S6 is calculated as follows:

[0036]

[0037] Where U is the current density distribution uniformity coefficient, i a is the average current density of each block, and i is the current density of each block.

[0038] Preferably, the data set in S8 is also augmented by a generative adversarial network before partitioning.

[0039] Preferably, the unsupervised learning algorithm in S10 is implemented by an autoencoder, and the state data and the environmental parameter data are input into the autoencoder to calculate a reconstruction error; when the reconstruction error exceeds a preset abnormal threshold, the data is determined to be abnormal.

[0040] Preferably, the training method of the first machine learning model comprises:

[0041] K sets of experimental training data are collected in advance, and the experimental training data include current density and voltage data, current density and voltage variation trends of each sub-block, and variation trends of overall current density distribution uniformity coefficients;

[0042] Each set of experimental training data is taken as an input of the first machine learning model, the first machine learning model takes the current density and voltage variation trends of each sub-block corresponding to each set of current density and voltage data and the variation trends of overall current density distribution uniformity coefficients as outputs, takes actual current density and voltage variation trends of each sub-block corresponding to each set of current density and voltage data and the variation trends of overall current density distribution uniformity coefficients as prediction targets, and takes minimization of a sum of prediction errors of the current density and voltage variation trends of all sub-blocks and the variation trends of overall current density distribution uniformity coefficients as a training target; the first machine learning model is trained until the sum of prediction errors reaches convergence.

[0043] The training method of the second machine learning model comprises:

[0044] G sets of model training data are collected in advance, and the model training data include current density distribution data and flow dead zones corresponding to the current density distribution data;

[0045] Each set of model training data is taken as an input of the second machine learning model, the second machine learning model takes the flow dead zones corresponding to each set of current density distribution data as outputs, takes actual flow dead zones corresponding to each set of current density distribution data as prediction targets, and takes minimization of a sum of prediction errors of all flow dead zones as a training target; the second machine learning model is trained until the sum of prediction errors reaches convergence.

[0046] Compared with the prior art, the liquid flow battery performance analysis method based on current density distribution measurement provided by the application has the following beneficial effects:

[0047] The present invention combines real-time environmental parameters related to battery operation and collects block data under different operating conditions. The data collection dimensions are rich and comprehensive, and can characterize the overall operating status of the battery from multiple angles. It also analyzes the key performance indicators of the battery in real time during the data collection process, and further adjusts the sampling frequency and analyzes the optimal operating conditions of the battery based on the dynamic changes of these indicators during the charging and discharging process, thereby improving the subsequent optimization capabilities of the overall battery performance. Through dynamic adaptive sampling frequency adjustment, when battery performance changes rapidly, sufficient data is obtained in a timely manner to analyze the details of the performance changes. Various parameters under different operating conditions and different battery charge states are calculated in detail, such as the current density distribution uniformity coefficient, and their impact on battery performance is analyzed, which can more accurately grasp the characteristics of the battery under various operating conditions. Data is measured and collected on a printed board based on a block grid design. Combined with the generated current density distribution map, problems such as uneven current distribution and flow dead zones within the battery can be intuitively discovered. It integrates multiple technical means to comprehensively evaluate battery performance from different levels, and can also explore the optimal operating mode and potential problems of the battery under different operating conditions based on the analysis and prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the method mentioned in Example 1 of the present invention;

[0049] Figure 2 Schematic diagram of the coulombic efficiency, voltage efficiency and energy efficiency of the battery mentioned in Example 1 of the present invention;

[0050] Figure 3 Schematic diagram of the charge and discharge curve mentioned in Example 1 of the present invention;

[0051] Figure 4 This is a schematic diagram of the current density curve mentioned in Example 1 of the present invention;

[0052] Figure 5 This is a schematic diagram of the voltage curve mentioned in Example 1 of the present invention;

[0053] Figure 6 Schematic diagram of the current density distribution uniformity coefficient curve mentioned in Example 1 of the present invention. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0055] The present invention combines real-time environmental parameters related to battery operation and collects block data under different operating conditions. The data collection is rich and comprehensive, capable of depicting the overall operating status of the battery from multiple perspectives. It also analyzes the battery's key performance indicators in real time during data collection, and further adjusts the sampling frequency and analyzes the battery's optimal operating conditions based on the dynamic changes of these indicators during the charge and discharge process, thereby improving the ability to subsequently optimize the battery's overall performance. Through dynamic adaptive sampling frequency adjustment, sufficient data is obtained in a timely manner to analyze the details of performance changes when battery performance changes rapidly. Various parameters, such as the current density distribution uniformity coefficient, are calculated in detail under different operating conditions and different battery states of charge, and their impact on battery performance is analyzed, enabling a more accurate understanding of the battery's characteristics under various operating conditions. Data is measured and collected using a printed circuit board designed with a block grid. Combined with the generated current density distribution map, problems such as uneven current distribution and flow dead zones within the battery can be intuitively identified. The invention integrates multiple technical means to comprehensively evaluate battery performance from different levels, and can also identify the battery's optimal operating mode and potential problems under different operating conditions based on the analysis and prediction results. Specifically, it includes the following contents.

[0056] Example 1:

[0057] See also Figure 1 The present invention provides a flow battery performance analysis method based on current density distribution measurement, comprising the following steps:

[0058] S1: Import real-time environmental parameter data related to battery operation, preset the initial sampling frequency and perform data collection, and set the sampling frequency adjustment rules; the details are as follows:

[0059] Changing the software's read path to the Excel file containing the data allows for single or batch data import. Specifically, real-time environmental parameter data related to battery operation (such as temperature, humidity, and pressure) facilitates subsequent comprehensive analysis. Filtering algorithms (such as Kalman filtering) can be used to reduce noise in the imported data, eliminating any interference from the measurement process and ensuring its accuracy.

[0060] Based on the battery type and estimated performance changes, an initial sampling frequency is set to begin data collection. During data collection, changes in key battery performance indicators, including current density change rate, voltage fluctuation amplitude, and power change rate, are analyzed in real time.

[0061] When the battery performance is in a relatively stable state and the change rate of key performance indicators is within the preset stability threshold range, the sampling frequency is gradually reduced according to the preset time interval (such as every several sampling cycles), but the minimum sampling frequency limit is maintained to ensure that important information is not missed.

[0062] Once it is detected that the rate of change of key performance indicators exceeds the stability threshold, such as a sudden increase in the rate of change of current density or a large fluctuation in voltage, the sampling frequency is immediately increased to quickly collect more data points until the performance stabilizes again or reaches the preset high-frequency sampling frequency.

[0063] The minimum and high-frequency sampling frequencies are optimized and adjusted based on machine learning algorithms. The sampling frequency adjustment history can also be regularly reviewed to optimize the sampling frequency threshold and adjustment step size based on the long-term performance trends of the battery to adapt to the characteristics of different batteries at different stages of use.

[0064] S2: Run the charge and discharge program, measure and store data based on the printed circuit board with a block grid design, adjust the operating conditions such as load current density and electrolyte flow rate, and record the battery data and each block data under different operating conditions; the details are as follows:

[0065] A pre-set charge and discharge program is initiated, which charges and discharges the battery according to a specific current control strategy to obtain detailed electrochemical information during the charge and discharge process. A printed circuit board with a segmented grid design is then used, with each segment capable of measuring the electrical parameters of its corresponding area. A data acquisition card connected to the printed circuit board collects voltage data from each segment, thereby providing detailed information on different localized regions of the battery. The manufacturing process for this segmented grid printed circuit board includes: grooves are machined between the grids on the upper surface of the bipolar plate; the grooves do not extend completely through the plate. The grid grooves are filled with epoxy resin. After the epoxy resin solidifies, the back of the bipolar plate is cut until the solidified resin is exposed, resulting in a fully segmented bipolar plate and ensuring electrical insulation between the grids. The same segmentation process is performed on the side of the printed circuit board that contacts the bipolar plate. The PCB segments can be slightly smaller than the bipolar plate segments. To reduce contact resistance, the PCB segments are gold-plated. Each block on the printed circuit board (PCB) is connected to a resistor of precise value. A pair of pin headers connects to each resistor, drawing out the potential across it. These pin headers are arranged on both sides of the PCB. A large current collector is designed on the back of the PCB to ensure that all current flowing into and out of the blocks converges there before flowing into the collector plate. To measure the current density of each grid, the PCB is designed to ensure that the internal circuitry is: block on the front of the PCB – resistor connected to each block – current collector on the back of the PCB. The pin headers on the PCB are connected to a data acquisition card via wires, which in turn is connected to a computer via a data cable. During battery operation, the data acquisition card records and stores the voltage signals across each resistor.

[0066] Data is measured and collected using a printed board based on the aforementioned grid-based design. During the data collection process, key battery performance indicators (KPIs) are analyzed in real time, and corresponding sampling frequency adjustment rules are implemented based on these KPIs. During the charge and discharge process, the load current density can be changed manually or through an automated control system to observe the battery's response at different current densities. The electrolyte flow rate is also adjusted, as it affects processes such as material transport within the battery. For each different combination of load current density and electrolyte flow rate (i.e., different operating conditions), detailed data from each block measured on the grid-based printed board is recorded. This data will subsequently be used for various analyses.

[0067] S3: Calculate the charge and discharge capacity and energy under different operating conditions, calculate the pump work based on the electrolyte flow rate, calculate the coulombic efficiency, voltage efficiency, energy efficiency and system efficiency under the different operating conditions, and draw the charge and discharge curves and efficiency curves to analyze the optimal operating conditions of the battery; the details are as follows:

[0068] By integrating the current and time data collected during the charge and discharge process, the battery's charge and discharge capacity (usually expressed in ampere-hours, Ah) can be calculated. Combined with the voltage data, the energy calculation formula (energy = voltage × capacity) is then integrated to calculate the energy value during the charge and discharge process. These calculations are performed for different operating conditions (such as load current density and electrolyte flow rate) to understand their impact on battery capacity and energy.

[0069] Based on the electrolyte flow rate and related physical parameters (such as electrolyte density, pipeline resistance coefficient, etc.), using the power calculation formula in fluid mechanics (pump work = flow rate × pressure difference and other related forms), combined with the flow of electrolyte in the battery system, the pump work required to maintain a specific electrolyte flow rate is calculated.

[0070] Calculate the coulombic efficiency, voltage efficiency and energy efficiency of the battery and draw the corresponding change diagram, which can be referred to Figure 2 , the coulombic efficiency, voltage efficiency and energy efficiency of the battery can be obtained intuitively:

[0071] The coulombic efficiency is calculated as follows:

[0072]

[0073] Among them, η C is the Coulomb efficiency; Q fd is the amount of charge passing through the battery during discharge; Q cd is the amount of charge passing through the battery during the charging process; T fd is the discharge time; T cd is the charging time; I fd is the average current during the discharge process; Icd is the average current during the charging process;

[0074] Coulombic efficiency, a key indicator of charge utilization during a battery's charge and discharge processes, reflects the reversibility of the electrochemical reactions within the battery and its health by comparing the discharge output charge with the charge input charge. As the battery cycles and ages, the coulombic efficiency may gradually decrease. This may be due to factors such as reduced electrode material activity, electrolyte decomposition, or changes in the battery's internal structure. For example, when a battery's coulombic efficiency drops from an initial 98% to below 90%, it indicates that the battery may be experiencing severe aging issues, such as the shedding of active material from the electrode surface or the generation of a large number of irreversible reaction products in the electrolyte. This provides important evidence for battery fault diagnosis and life prediction.

[0075] The voltage efficiency is calculated as follows:

[0076]

[0077] Among them, η U is the voltage efficiency; U fd is the average voltage during the discharge process; U cd is the average voltage during the charging process;

[0078] Voltage efficiency reflects the voltage loss during the battery's charge and discharge processes. It is calculated by comparing the average discharge voltage with the average charge voltage. Voltage loss is primarily caused by polarization within the battery (including ohmic polarization, concentration polarization, and activation polarization). For example, during charging, polarization causes the battery's charge voltage to exceed its equilibrium voltage; while during discharge, the discharge voltage falls below the equilibrium voltage. By calculating voltage efficiency, the extent of this voltage loss can be quantified, thereby assessing the magnitude of energy loss within the battery.

[0079] The energy efficiency is calculated as follows:

[0080]

[0081] Among them, η E is energy efficiency; E cd E is the charging energy; fd is the discharge energy.

[0082] Energy efficiency is the product of coulombic efficiency and voltage efficiency. It comprehensively considers the charge utilization and voltage loss of the battery during the charging and discharging process, and can more comprehensively evaluate the performance of the battery.

[0083] Reference Figure 3, extract the charge and discharge capacity and charge and discharge voltage, and draw the charge and discharge curve. The charge and discharge curve (voltage-capacity curve) can intuitively show the changes in voltage and capacity of the battery during the charge and discharge process. The shape of the curve can show some basic performance characteristics of the battery, such as the battery's open circuit voltage, platform voltage, cut-off voltage, and capacity. Various efficiency curves can be drawn, such as coulombic efficiency, energy efficiency, etc., which change with operating conditions. By observing and analyzing these curves, find the combination of operating conditions that can make the battery perform best in terms of energy conversion, performance stability, etc., such as determining the most appropriate parameters such as load current density and electrolyte flow rate.

[0084] S4: Filter the block data recorded under the above conditions, and plot the current density and voltage curves of each block as the battery state of charge changes based on the area of ​​each block; generate a current density distribution map based on the position of each block and the measurement principle; the details are as follows:

[0085] The block data previously recorded under different operating conditions are processed using a filtering algorithm to remove noise interference, making the data smoother and facilitating subsequent accurate analysis.

[0086] The voltage of each block is the voltage collected by the data acquisition card, the current of each block is the voltage divided by the resistance of the resistor connected to the block, and the current density is the current of each voltage divided by the area corresponding to each block. According to the area information of each block and the collected current and voltage data, the current density (current divided by area) corresponding to each block is calculated. Using a certain state parameter of the electrolyte (such as electrolyte concentration, temperature, etc.) as the horizontal coordinate, and the current density and voltage as the vertical coordinate, a curve of each block changing with the battery charge state is drawn. You can refer to Figure 4 、 Figure 5 , visually presenting the changes in the electrical characteristics of each block at different battery states of charge. Based on the position information of each block in the battery, combined with the measured value of current density and related measurement principles (such as the distribution pattern of current in different areas), the current density values ​​of each block are graphically displayed using drawing software to generate a current density distribution map, which can clearly show the spatial distribution of current density inside the battery.

[0087] S5: Calculate the current density distribution uniformity coefficient under different operating conditions and different battery states of charge, draw a curve of its change with the battery state of charge, and calculate its average value; the details are as follows:

[0088] The current density distribution uniformity is calculated as follows:

[0089]

[0090] Where U is the current density distribution uniformity coefficient, ia is the average current density of each block, and i is the current density of each block.

[0091] Use the battery state of charge parameter as the horizontal axis and the calculated current density distribution uniformity coefficient as the vertical axis to draw a curve of the change with the battery state of charge using the drawing tool. For details, please refer to Figure 6 This curve shows the impact of battery state-of-charge changes on current density distribution uniformity. The average value of all current density distribution uniformity coefficients obtained under different operating conditions and battery states of charge is calculated. This average value can be used as an overall reference indicator to compare the approximate level of current density distribution uniformity under different operating conditions or different experimental batches.

[0092] S6: Select seconds as the time characteristic period and extract the time characteristics of the current density, voltage curve and current density distribution uniformity coefficient; use the minimum-maximum normalization method to normalize the current density and voltage data to obtain a data set, and divide the data set into a training set, a validation set and a test set; the details are as follows:

[0093] From the previously drawn current density curve, voltage curve and calculated current density distribution uniformity coefficient data, we analyze their temporal variation patterns and extract time-related characteristic information such as periodic variation characteristics (if any), variation trends within a specific time period (increase, decrease, etc.). We explicitly select seconds as the basic time period characteristic to facilitate the unification of the time scale for subsequent data processing and model analysis.

[0094] Using the minimum-maximum normalization method, for the current density and voltage data, first find their respective minimum and maximum values, and then use the formula (normalized value = (original value - minimum value) / (maximum value - minimum value)) to map the data to the [0,1] interval (it can also be further adjusted to [-1,1] intervals according to subsequent model requirements). This can eliminate the impact of different data magnitudes on subsequent machine learning model training and other operations, making the data more comparable and universal.

[0095] The normalized data is divided into training, validation, and test sets according to a specific ratio (commonly seen ratios include 7:2:1 or 8:1:1). The training set is used to train the machine learning model, allowing it to learn patterns in the data; the validation set is used to adjust the model's hyperparameters (such as the number of layers and nodes in the neural network) during training to prevent overfitting; and the test set is used to ultimately evaluate the model's performance on unseen data.

[0096] S7: Construct a first machine learning model based on the long short-term memory network, and use the mean square error, mean absolute error, and root mean square error to evaluate the prediction performance of the first machine learning model to predict the current density and voltage change trends of each block and the change trend of the overall current density distribution uniformity coefficient; the details are as follows:

[0097] Determine the structure of the LSTM network, including the number of input layer nodes (determined by the characteristic dimensions of the input data, such as the number of features like current density and voltage), the number of hidden layer layers and nodes (adjustable and optimized through experimentation and experience), and the number of output layer nodes (corresponding to the number of output targets to be predicted, such as current density, voltage trends, and uniformity coefficient trends). Use a suitable deep learning framework to build the LSTM network model architecture and perform initialization settings.

[0098] The constructed LSTM model is trained using the partitioned training set. The model parameters are adjusted through multiple iterations to enable the model to learn the time series characteristics and inherent patterns in the data. The mean squared error (MSE, calculated as the average of the squares of the differences between the predicted and true values), the mean absolute error (MAE, calculated as the average of the absolute values ​​of the differences between the predicted and true values), and the root mean squared error (RMSE, which is the square root of the MSE) are selected as indicators for evaluating the model's predictive performance. These indicators can reflect the degree of deviation between the model's predictions and the actual data from different perspectives.

[0099] The trained model is used to predict data over a short period of time, outputting the future trends of current density and voltage for each block, as well as the trend of the overall current density distribution uniformity coefficient. When analyzing the prediction results, focus on time points and corresponding regions where there are significant discrepancies between the predicted and true values. By further studying the causes of these discrepancies (such as data anomalies and special patterns not learned by the model), the model can be improved or a deeper understanding of battery operation can be obtained.

[0100] S8: Build an anomaly monitoring model using an unsupervised learning algorithm to monitor the prediction results of the first machine learning model for anomalies. If no anomalies are detected during monitoring, continue to monitor the prediction results of the first machine learning model for anomalies. If anomalies are detected during monitoring, classify the detected anomaly data using an anomaly classification model built based on a decision tree to obtain an anomaly type. The details are as follows:

[0101] The battery data features under normal operating conditions are learned through unsupervised learning algorithms (such as autoencoders) to build a data anomaly detection model. The real-time collected data is input into the autoencoder to calculate the reconstruction error. When the reconstruction error exceeds the preset threshold, the data is judged to be abnormal, that is, the battery may have a fault. For example, if local corrosion occurs on the electrode surface, the difference between the data features and the normal state will increase, thereby triggering an anomaly detection alarm. At the same time, machine learning algorithms (such as random forest classifiers) are used to classify abnormal data and determine the possible fault type (such as electrode failure, electrolyte failure, etc.) so that timely measures can be taken to repair or adjust it to ensure the safe and efficient operation of the battery system.

[0102] S9: In the current density distribution diagram under different operating conditions, the current density distribution cloud map is screened; the current density distribution data is extracted based on the screened current density distribution cloud map, the current density distribution data is cleaned and normalized, the flow dead zone in the current density distribution cloud map is marked, and the training set, validation set and test set are divided according to this; the details are as follows:

[0103] From the numerous current density distribution maps generated previously under different operating conditions, representative cloud maps that can clearly show the current density distribution characteristics are selected based on research objectives, analysis needs and other factors. These cloud maps present the spatial distribution of current density inside the battery in intuitive colors, shapes, etc., which facilitates subsequent observation and processing.

[0104] The selected current density distribution data is cleaned to remove any erroneous data (such as outliers caused by measurement errors) and duplicate data. An appropriate normalization method is then used (such as the same minimum-maximum normalization method used previously, which adjusts parameters to map the data to the [-1, 1] interval) to ensure that each block of data falls within the specified interval, facilitating subsequent model training and data scaling.

[0105] Based on the characteristics of the current density distribution cloud map, professionals identify and manually mark dead zones (i.e., areas where electrolyte flow is slow or almost non-existent, potentially affecting battery performance). The processed data is then divided into training, validation, and test sets according to a specific ratio (similar to the principles used in the previous dataset division), preparing for subsequent convolutional neural network model training and other operations.

[0106] S10: Build a second machine learning model based on a convolutional neural network, select cross entropy loss and squared absolute error loss as the loss function, use Adam as the optimizer, train the second machine learning model using the training set, adjust the hyperparameters of the second machine learning model using the validation set, and verify the accuracy of the second machine learning model using the test set; the details are as follows:

[0107] Determine the network architecture of CNN, including parameters such as the number of convolutional layers, convolution kernel size, step size, pooling layer settings (such as maximum pooling, average pooling, etc. and corresponding parameters), and the structure of the fully connected layer, etc., to construct a CNN model suitable for processing current density distribution cloud map data, so that it can automatically extract feature information in the image, such as the texture, shape and other features of the current density distribution and the correlation characteristics of the flow dead zone.

[0108] The cross-entropy loss function was selected for classification to distinguish between dead zones and normal flow areas. The squared absolute error loss function was selected for anchor boxes to mark the locations of dead zones. Model training was guided by calculating the difference between the predicted results and the true labels. The Adam optimizer was used, which adaptively adjusts parameters such as the learning rate to accelerate model convergence and improve training results. The constructed CNN model was trained multiple times using the partitioned training set, allowing the model to continuously learn the image features and classification patterns in the data.

[0109] During training, model hyperparameters (such as the number of channels in the convolutional layer and the number of nodes in the fully connected layer) are adjusted based on the model's performance on the validation set (e.g., changes in loss and accuracy) to optimize model performance and prevent overfitting. Finally, the test set is used to verify the model's ultimate accuracy. Metrics such as accuracy (the proportion of correctly predicted samples to the total number of samples) and recall (the proportion of samples that are actually positive and predicted as positive to the total number of positive samples) are used to comprehensively evaluate the model's performance in identifying flow dead zones.

[0110] S11: Use the second machine learning model obtained through training optimization to process the current density distribution data under different operating conditions, determine the area within the electrode frame in each current density distribution map, automatically mark the flow dead zone location within the electrode frame, and summarize it; the details are as follows:

[0111] Previously collected image data, such as large-scale current density distribution cloud maps, is fed into a trained machine learning model based on a convolutional neural network. Based on the learned characteristics and patterns, the model identifies the region within the electrode frame in each image and automatically marks the location of any dead zones. The marked dead zones across all images are then summarized and collated, such as the frequency and size of dead zones in different regions, to further analyze the electrolyte flow characteristics within the battery and their impact on battery performance.

[0112] S12: All processed and analyzed data are classified and stored according to time sequence and data type, and the stored data is dynamically visualized based on the visualization interface. The details are as follows:

[0113] All processed and analyzed data, including raw data, pre-processed data, performance calculation results at different time scales, current density distribution data, sampling frequency adjustment records, performance trend prediction results, and anomaly detection information, is stored in chronological order and classified by data type. Efficient data storage formats (such as HDF5) are used to facilitate subsequent query, analysis, and data mining.

[0114] Develop an intuitive visualization interface that can simultaneously display the evolution of battery performance at multiple time scales. For example, display the charge transfer rate changes at the microscopic time scale and the energy efficiency trends at the macroscopic time scale in the same chart, using different colors and line styles to distinguish different parameters and time scales.

[0115] The dynamic tracking of battery performance over time is displayed in the form of animations or dynamic charts, including adaptive changes in sampling frequency and performance indicator curves at different time scales. Anomaly detection results are intuitively displayed. When an anomaly is detected, the abnormal data point and corresponding fault type information are highlighted on the visual interface, making it easier for operators to promptly identify the anomaly and make appropriate decisions.

[0116] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A flow battery performance analysis method based on current density distribution measurement, characterized in that: The steps include: S1: Import real-time environmental parameter data related to battery operation, preset the initial sampling frequency and perform data collection, and set the sampling frequency adjustment rules; S2: Run the preset charge and discharge program, measure and store data based on the block grid design printed board. Each block of the grid network design printed board measures the electrical parameters of the corresponding area respectively. Through the data acquisition card connected to the printed board, the voltage data on each block is collected to obtain detailed information on different local areas of the battery; adjust the load current density and electrolyte flow rate operating conditions, and record the battery data and each block data under different operating conditions; S3: Calculate the charge and discharge capacity and energy under different operating conditions, calculate the pump work based on the electrolyte flow rate, calculate the coulombic efficiency, voltage efficiency, energy efficiency and system efficiency under the different operating conditions, and draw the capacity-voltage charge and discharge curve and efficiency curve to analyze the optimal operating conditions of the battery; S4: Filtering the block data recorded under the above different operating conditions, and combining the area of ​​each block to draw the current density and voltage curve of each block as the battery state of charge changes; generating a current density distribution diagram based on the position of each block and the measurement principle; the current of each block is the voltage of each block divided by the resistance of the resistor connected to each block, and the current density is the current of each block divided by the area corresponding to each block; S5: Calculate the current density distribution uniformity coefficient under different operating conditions and different battery states of charge; the current density distribution uniformity is calculated as follows: in, U is the current density distribution uniformity coefficient, i a is the average value of all block current densities, i is the current density for each block; And draw a curve of its change with the battery state of charge, and calculate the average value of all current density distribution uniformity coefficients obtained under different working conditions and battery state of charge; S6: Use the minimum-maximum normalization method to normalize the current density and voltage data to obtain a data set, and divide the data set into a training set, a validation set, and a test set; S7: Build a first machine learning model based on the long short-term memory network, use mean square error, mean absolute error, and root mean square error to evaluate the prediction performance of the first machine learning model, and predict the current density and voltage change trends of each block and the change trend of the overall current density distribution uniformity coefficient; S8: Build an anomaly monitoring model using an unsupervised learning algorithm to monitor the prediction results of the first machine learning model for anomalies. If no anomaly is detected during monitoring, continue to monitor the prediction results of the first machine learning model for anomalies. If an anomaly is detected during monitoring, classify the detected anomaly data based on the anomaly classification model built based on the decision tree to obtain the anomaly type. S9: In the current density distribution diagrams under different working conditions, the current density distribution cloud map is screened; the current density distribution data is extracted based on the screened current density distribution cloud map, the current density distribution data is cleaned and normalized, the flow dead zone in the current density distribution cloud map is marked, and the training set, validation set, and test set are divided based on this; S10: Build a second machine learning model based on a convolutional neural network, select cross entropy loss and squared absolute error loss as loss functions, use Adam as the optimizer, train the second machine learning model using the training set, adjust the hyperparameters of the second machine learning model using the validation set, and verify the accuracy of the second machine learning model using the test set; S11: Use the second machine learning model obtained through training optimization to process the current density distribution data under different working conditions, automatically mark the flow dead zone location within the electrode frame, and summarize it; S12: All processed and analyzed data are classified and saved according to time sequence and data type, and the stored data is visualized based on a visualization interface.

2. The method for analyzing flow battery performance based on current density distribution measurement according to claim 1, characterized in that: The environmental parameter data in S1 include temperature, humidity and pressure; The sampling frequency adjustment rules described in S1 are as follows: When the key performance indicator is within the preset threshold range, the sampling frequency is gradually reduced according to the preset time interval until it is reduced to the preset minimum sampling frequency; When the key performance indicator exceeds the preset threshold range, the sampling frequency is increased until the performance stabilizes again or reaches the preset high-frequency sampling frequency; In the sampling frequency adjustment rule, the minimum sampling frequency and the high frequency sampling frequency are optimized and adjusted based on a machine learning algorithm.

3. The method for analyzing flow battery performance based on current density distribution measurement according to claim 2, characterized in that: In S2, during the data collection process, the key performance indicators of the battery are analyzed in real time, and corresponding sampling frequency adjustment rules are executed according to the key performance indicators; the key performance indicators include current density change rate, voltage fluctuation amplitude and power change rate.

4. The method for analyzing flow battery performance based on current density distribution measurement according to claim 1, characterized in that: The printed board described in S2 is designed with several block grids on the board surface, and the interval between each block grid is not less than 0.5mm and not more than 2mm. The groove of each block grid is filled with epoxy resin. After it solidifies, the back of the bipolar plate is cut until the solidified resin is exposed to obtain a printed board based on the block grid design.

5. The method for analyzing flow battery performance based on current density distribution measurement according to claim 1, characterized in that: The dataset in S8 is also expanded by generating adversarial networks before division.

6. The method for analyzing flow battery performance based on current density distribution measurement according to claim 1, characterized in that: The unsupervised learning algorithm in S8 is implemented using an autoencoder, and the state data and environmental parameter data are input into the autoencoder to calculate the reconstruction error; when the reconstruction error exceeds a preset abnormality threshold, the data is determined to be abnormal.

7. The method for analyzing flow battery performance based on current density distribution measurement according to claim 1, characterized in that: The training method of the first machine learning model includes: Collect K groups of experimental training data in advance. The experimental training data includes current density and voltage data, the current density and voltage change trends of each block, and the change trend of the overall current density distribution uniformity coefficient; Each set of experimental training data is used as the input of the first machine learning model. The first machine learning model uses the current density, voltage change trend and the change trend of the overall current density distribution uniformity coefficient of each block corresponding to each set of current density and voltage data as output, and uses the actual current density, voltage change trend and the change trend of the overall current density distribution uniformity coefficient of each block corresponding to each set of current density and voltage data as the prediction target; minimizing the sum of the prediction errors of the current density, voltage change trend and the change trend of the overall current density distribution uniformity coefficient of all blocks as the training target; training the first machine learning model until the sum of the prediction errors reaches convergence and the training is stopped; The training method of the second machine learning model includes: G groups of model training data are collected in advance, and the model training data includes current density distribution data and flow dead zones corresponding to the current density distribution data; Each set of model training data is used as the input of the second machine learning model. The second machine learning model uses the flow dead zone corresponding to each set of current density distribution data as output, and the actual flow dead zone corresponding to each set of current density distribution data as the prediction target; minimizing the sum of the prediction errors of all flow dead zones is used as the training goal; the second machine learning model is trained until the sum of the prediction errors reaches convergence and the training is stopped.

Images