A method and system for coordinated management of liquid flow battery electrolyte inventory

Abstract

The application discloses a kind of liquid flow battery electrolyte inventory collaborative management method and system, method includes: real-time monitoring the instantaneous flow data of electrolyte in liquid flow battery system and the state of charge data of battery, generates the real-time data sequence of electrolyte operating state;Based on real-time data sequence, combine the historical cycle attenuation characteristics of electrolyte with current operating parameter, determine the actual available inventory of electrolyte and predict short-term inventory demand trend by dynamic coupling analysis model;According to actual available inventory and short-term inventory demand trend, generate electrolyte cycle dynamic adjustment strategy using multi-objective optimization algorithm;Execute electrolyte cycle dynamic adjustment strategy, dynamically adjust the flow and pressure distribution in electrolyte circulation loop, realize the collaborative management of liquid flow battery electrolyte inventory.Utilize the embodiment of the application, can realize accurate perception and collaborative control to electrolyte inventory, effectively improve the operating efficiency, cycle life and energy utilization rate of liquid flow battery.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of liquid flow batteries, and particularly relates to a liquid flow battery electrolyte inventory collaborative management method and system. BACKGROUND

[0002] In recent years, liquid flow batteries have been widely applied in large-scale energy storage fields due to their large scale, long service life, high safety and other advantages. As the core active substance of liquid flow batteries, the effective management of the inventory of electrolyte is directly related to the system operation efficiency and service life. At present, the traditional electrolyte management is mainly based on fixed threshold or simple feedback for pump speed adjustment, which is difficult to adapt to the dynamic changes of the electrolyte state in the charging and discharging process in real time, and is easy to cause uneven distribution of electrolyte in the system, decrease of cycle efficiency, and even cause problems such as accelerated local side reaction and capacity decay. SUMMARY

[0003] The application aims to provide a liquid flow battery electrolyte inventory collaborative management method and system to solve the problems in the prior art and realize accurate perception and collaborative regulation of the electrolyte inventory, and effectively improve the operation efficiency, cycle life and energy utilization rate of the liquid flow battery system.

[0004] One embodiment of the application provides a liquid flow battery electrolyte inventory collaborative management method, which comprises the following steps:

[0005] Real-time monitoring of the instantaneous flow data of the electrolyte in the liquid flow battery system and the state of charge data of the battery to generate a real-time data sequence of the electrolyte operation state;

[0006] Based on the real-time data sequence, the historical cycle attenuation characteristics of the electrolyte and the current working condition parameters are combined to determine the actual available inventory of the electrolyte and predict the short-term inventory demand change trend through a dynamic coupling analysis model;

[0007] According to the actual available inventory and the short-term inventory demand change trend, a multi-objective optimization algorithm is used to generate an electrolyte circulation dynamic adjustment strategy, wherein the strategy at least includes frequency adjustment of the electrolyte circulation pump and collaborative control of the valve opening degree;

[0008] The electrolyte circulation dynamic adjustment strategy is executed to dynamically adjust the flow and pressure distribution in the electrolyte circulation loop, so as to realize collaborative management of the liquid flow battery electrolyte inventory.

[0009] Another embodiment of the application provides a liquid flow battery electrolyte inventory collaborative management system, which comprises the following steps:

[0010] The monitoring module is used for monitoring the instantaneous flow data of the electrolyte and the state of charge data of the battery in the flow battery system in real time, and generating a real-time data sequence of the running state of the electrolyte.

[0011] The determining module is used for determining the actual available stock of the electrolyte and predicting a short-term stock demand change trend by means of a dynamic coupling analysis model based on the real-time data sequence, in combination with the historical cycle attenuation characteristics of the electrolyte and the current working condition parameters.

[0012] The generating module is used for generating an electrolyte circulation dynamic adjustment strategy by means of a multi-objective optimization algorithm according to the actual available stock and the short-term stock demand change trend, wherein the strategy at least includes frequency adjustment of the electrolyte circulation pump and valve opening degree cooperative control.

[0013] The adjusting module is used for executing the electrolyte circulation dynamic adjustment strategy, dynamically adjusting the flow and pressure distribution in the electrolyte circulation loop, and realizing cooperative management of the electrolyte stock of the flow battery.

[0014] Another embodiment of the present application provides a storage medium having a computer program stored therein, wherein the computer program is configured to execute the method described in any one of the above embodiments when running.

[0015] Another embodiment of the present application provides an electronic device comprising a memory and a processor, wherein the memory has a computer program stored therein, and the processor is configured to execute the computer program to execute the method described in any one of the above embodiments.

[0016] Compared with the prior art, the flow battery electrolyte stock cooperative management method provided by the present application can realize accurate perception and cooperative regulation of the electrolyte stock, and effectively improve the running efficiency, cycle life and energy utilization rate of the flow battery system. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 FIG. 1 is a hardware structure block diagram of a computer terminal of a flow battery electrolyte stock cooperative management method according to an embodiment of the present application;

[0018] Figure 2 FIG. 2 is a flowchart of a flow battery electrolyte stock cooperative management method according to an embodiment of the present application;

[0019] Figure 3 FIG. 3 is a structure diagram of a flow battery electrolyte stock cooperative management system according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] The embodiments described below with reference to the drawings are exemplary and are only used to explain the present application, and cannot be explained as a limitation of the present application.

[0021] Figure 1 This is a hardware structure block diagram of a computer terminal for a collaborative management method of electrolyte inventory in a flow battery, provided as an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0022] See Figure 2 The present invention provides a method for collaborative management of electrolyte inventory in a flow battery, which may include the following steps:

[0023] S201, real-time monitoring of the instantaneous flow rate data of the electrolyte and the state of charge data of the battery in the flow battery system, generating a real-time data sequence of electrolyte operating status;

[0024] Specifically, flow sensors can be deployed at key nodes in the electrolyte circulation loop, pressure sensors can be installed at the positive and negative electrode inlets and outlets of the battery stack, and the state of charge data of the battery management system can be collected to obtain multi-source raw monitoring data.

[0025] I. Key Nodes and Flow Sensor Deployment in the Electrolyte Circulation Loop

[0026] Criteria for selecting key nodes: The electrolyte circulation loop of a flow battery includes a closed-loop path of "storage tank → circulation pump → filter → battery stack → heat exchanger → storage tank". Key nodes need to cover "flow change sensitive areas" and "high-failure areas". Specifically, four core nodes were selected:

[0027] Node 1: Electrolyte tank outlet (monitors the total flow rate from the tank to the loop, reflecting the tank's discharge rate); Node 2: Circulation pump outlet (monitors the actual output flow rate of the pump, determining the pump's efficiency and whether cavitation occurs); Node 3: Battery stack inlet (monitors the effective flow rate entering the reaction zone, directly affecting the battery reaction efficiency); Node 4: Battery stack outlet (monitors the return flow rate of the electrolyte after the reaction, determining whether there is blockage within the stack).

[0028] Flow sensor selection and parameters: An electromagnetic flow sensor (suitable for conductive liquids, such as vanadium electrolyte) is selected, model ABBFEP300, with the following core parameters:

[0029] Range: 0~50 L / min (cover the rated flow 15 L / min of this 10 kW system, reserve 3 times redundancy to avoid overload); Accuracy: ±0.5% FS (full scale accuracy, absolute error ≤0.075 L / min at 15 L / min, meet the flow control requirements); Response time: ≤100 ms (quickly capture flow fluctuations, such as flow mutations caused by pump frequency changes); Output signal: 4~20 mA analog signal (for subsequent AD acquisition) + RS485 digital signal (transmit raw data);

[0030] Installation requirements: The lengths of the straight pipe sections in front of and behind the sensor are 5 times the pipe diameter (front) and 3 times the pipe diameter (back), respectively. The pipe diameter at node 1 is DN25 (25 mm), so the front straight pipe section is 125 mm, and the back straight pipe section is 75 mm, to avoid measurement errors caused by flow field disturbances.

[0031] Deployment example: The flow sensor at node 3 (battery stack inlet) is installed on the DN25 pipe between the filter and the battery stack, with the sensor axis coinciding with the pipe axis. The seal uses a fluororubber gasket (resistant to vanadium electrolyte corrosion), and the wiring end is connected to the data acquisition module through a shielded wire. The sampling frequency is set to 10 Hz (10 data points collected per second, balancing real-time performance and data volume).

[0032] II. Deployment of battery stack inlet and outlet pressure sensors

[0033] Pressure monitoring significance: The pressure difference at the inlet and outlet of the positive and negative electrodes of the battery stack directly reflects the patency of the flow channel in the stack (a large pressure difference may indicate flow channel blockage, and a small pressure difference may indicate sealing leakage). Therefore, one pressure sensor is required at the inlet and outlet of each positive and negative electrode, for a total of four sensors (positive electrode inlet, positive electrode outlet, negative electrode inlet, and negative electrode outlet).

[0034] Pressure sensor selection and parameters: A diffused silicon pressure transmitter, model Rosemount 3051CG, is selected, with the following core parameters:

[0035] Range: 0~1 MPa (the system normally operates at 0.1~0.3 MPa, with a safety reserve); Accuracy: ±0.075% FS (absolute error ≤0.225 kPa at 0.3 MPa, ensuring accurate pressure difference measurement); Response time: ≤50 ms (quickly capture pressure fluctuations, such as sudden pressure changes caused by valve opening changes); Output signal: 4~20 mA analog signal (consistent with the flow sensor, facilitating unified acquisition); Corrosion resistance: The liquid contact material is 316L stainless steel (resistant to acidic substances in the vanadium electrolyte, with a service life of ≥5 years).

[0036] Deployment example: The positive inlet pressure sensor is installed at the positive inlet flange of the battery stack, using a threaded connection (M20x1.5). The sensor measurement end directly contacts the electrolyte, and the cable passes through the explosion-proof gland to access the control cabinet. The sampling frequency is set to 5Hz (pressure changes slower than flow, 5Hz can meet the monitoring requirements).

[0037] Three, the collection mechanism of state of charge (SOC) data

[0038] Definition and collection principle of SOC: SOC represents the percentage of remaining capacity to rated capacity (0%~100%). The rated capacity of the 10kW flow battery is 20kWh, and SOC=(current remaining capacity / 20kWh) x 100%. BMS collects SOC through "coulomb counting method + open circuit voltage correction":

[0039] Coulomb counting method: The shunt resistor (0.1Ω, accuracy ±0.1%) is connected in series in the loop to collect the charging and discharging current, and the capacity change is calculated by integration. The formula is SOC(t)=SOC(t0)-(1 / Cn) x ∫I(τ)dτ (t0 is the initial time, Cn=20kWh / Vn, Vn is the rated voltage 1.5V, so Cn≈13333Ah);

[0040] Open circuit voltage correction: stop for 10s every 30 minutes, measure the open circuit voltage (OCV) of the battery stack, and correct the cumulative error of the coulomb counting method through the OCV-SOC curve (such as OCV=1.45V corresponds to SOC=50%, OCV=1.55V corresponds to SOC=100%).

[0041] SOC data collection parameters:

[0042] Collection frequency: 1Hz (SOC changes slowly, 1Hz can meet the requirements, avoid data redundancy); accuracy: ±2% (coulomb counting method error ±3%, after OCV correction, reduced to ±2%); data transmission: through CAN bus (baud rate 250kbps) from BMS to data acquisition system, each frame of data contains "timestamp + SOC value + correction state" (such as "20250923-153000-000, 65.2%, CALIBRATED" indicates that the corrected SOC is 65.2%).

[0043] Four, multiple source raw monitoring data example

[0044] The multiple source raw data at a certain time (20250923-153000-000) is as follows:

[0045] Flow data (unit: L / min): node 1=14.8, node 2=15.0, node 3=14.7, node 4=14.6 (the difference between node 2 and node 3 is caused by the pressure loss of the filter).

[0046] Pressure data (unit: kPa): positive electrode in = 280, positive electrode out = 250, negative electrode in = 275, negative electrode out = 245 (positive and negative electrode inlet and outlet pressure difference is 30 kPa, the internal flow channel is unobstructed);

[0047] SOC data: 65.2%, correction status = corrected, current = 10A (discharge state);

[0048] The data storage format is TXT file, each data contains timestamp (millisecond level), sensor type, node / position, value, unit, and ensures that the original data is traceable.

[0049] The multi-source original monitoring data is processed by time series alignment and dimensionless, and the sliding window filtering algorithm is used to eliminate measurement noise to form the pretreated standardized monitoring data;

[0050] I. Time series alignment processing (time stamp unification)

[0051] Time series difference analysis: the sampling frequencies of multi-source data are different, resulting in that the data timestamps of the same physical moment do not coincide (for example, the flow sensor 10Hz sampling, each data corresponds to the timestamp t=0ms, 100ms, 200ms…; the pressure sensor 5Hz sampling, the timestamp is t=0ms, 200ms, 400ms…; the SOC 1Hz sampling, the timestamp is t=0ms, 1000ms, 2000ms…), which needs to be aligned to the timestamp of the highest sampling frequency (10Hz, interval 100ms).

[0052] Alignment method: linear interpolation method: for pressure data (5Hz): between two adjacent original data points (such as t=0ms 280kPa, t=200ms 282kPa), the pressure value of t=100ms is interpolated according to the time interval 100ms, t=100ms pressure value=280kPa+(282kPa-280kPa)×(100ms / 200ms)=281kPa;

[0053] For SOC data (1Hz): between t=0ms 65.2% and t=1000ms 65.0%, the SOC value of t=100ms~900ms is calculated by linear interpolation, t=500ms SOC value=65.2%-(65.2%-65.0%)×(500ms / 1000ms)=65.1%;

[0054] Alignment timestamp example: the alignment data at t=100ms is: flow node 3=14.7L / min, positive electrode inlet pressure=281kPa, SOC=65.18% (more accurate interpolation result), ensuring that each 100ms timestamp corresponds to a complete set of flow, pressure, SOC data.

[0055] II. Dimensionless processing (normalization)

[0056] Dimension difference problem: the numerical range of flow (0~50L / min), pressure (0~1000kPa), and SOC (0~100%) is different from the physical meaning, and direct calculation will lead to the dominance of large numerical parameters (such as pressure) in subsequent analysis, which needs to be normalized to map all data to the [0,1] interval to eliminate the influence of dimension.

[0057] Normalization model: linear normalization (Min-Max standardization): formula is x_norm=(x-x_min) / (x_max-x_min), where x is the original data, x_min is the minimum theoretical value of the parameter, and x_max is the maximum theoretical value;

[0058] Normalization data verification: after normalization, all data are in the [0,1] interval, and the relative size relationship of the original data is preserved (such as flow 14.7L / min<15.0L / min, normalized 0.294<0.300), ensuring that the physical meaning of the data is not lost.

[0059] III. Sliding window filtering algorithm (noise elimination)

[0060] Noise source and influence: there are two types of noise in the original data: "spike noise" caused by electromagnetic interference (such as flow suddenly jumping to 20L / min, lasting for 1 sampling point), and "random noise" of the sensor itself (such as pressure fluctuating at 281kPa±0.5kPa), which needs to be smoothed by filtering to avoid the influence of noise on subsequent feature calculation.

[0061] Filtering parameter design:

[0062] Window size N: set to 5 sampling points (corresponding to 500ms, taking into account the smoothing effect and real-time performance, too large window size leads to lag, too small filtering is not sufficient);

[0063] Weight distribution: adopt "equal weight moving average" (calculate the arithmetic mean of 5 data in the window), formula is x_filtered=(x1+x2+x3+x4+x5) / N, which is suitable for the scene where noise is randomly distributed;

[0064] Boundary processing: When the window of the first or last data point is less than 5 points, use "mirror extension" to avoid distortion of boundary data.

[0065] Four, example of standardized monitoring data after preprocessing

[0066] The normalized and filtered standardized data from t=100ms to t=500ms are as follows:

[0067] t=100ms: Flow rate node 3=0.294 (14.7L / min), positive pressure=0.281 (281kPa), SOC=0.652 (65.2%);

[0068] t=200ms: Flow rate node 3=0.310 (15.5L / min), positive pressure=0.282 (282kPa), SOC=0.651 (65.1%);

[0069] t=300ms: Flow rate node 3=0.309 (15.44L / min), positive pressure=0.283 (283kPa), SOC=0.651 (65.1%);

[0070] t=400ms: Flow rate node 3=0.308 (15.4L / min), positive pressure=0.282 (282kPa), SOC=0.650 (65.0%);

[0071] t=500ms: Flow rate node 3=0.306 (15.3L / min), positive pressure=0.281 (281kPa), SOC=0.650 (65.0%);

[0072] The data is stored in CSV format, with the path " / data / standardized / 20250923 / 1530.csv", and each data contains timestamp, parameter type, normalized value, and original value, which is convenient for subsequent feature calculation and backtracking.

[0073] Based on the standardized monitoring data, the instantaneous change rate of electrolyte flow and the change trend of state of charge are calculated, and a dynamic feature index data set is established;

[0074] One, calculation of electrolyte flow instantaneous change rate

[0075] Definition and physical meaning of instantaneous change rate: The instantaneous change rate of flow rate represents the change amplitude of flow rate per unit time, with the unit of L / (min?s), positive number indicates flow rate increase, negative number indicates flow rate decrease, reflecting the dynamic adjustment speed of flow rate (such as the frequency of circulating pump increases from 50 Hz to 55 Hz, the flow rate increases from 15 L / min to 16 L / min, the change rate is positive).

[0076] Calculation method: two-point difference method: based on the standardized flow rate data of adjacent two time stamps, the formula is r_Q=(Q_norm(k)-Q_norm(k-1)) / Δt, where: Q_norm(k) is the standardized flow rate at time k, Q_norm(k-1) is the standardized flow rate at time k-1; Δt is the time interval (100 ms=0.1 s, because the sampling frequency is 10 Hz); after calculation, it needs to be converted to the original flow rate change rate (multiplied by the flow rate range 50 L / min), finally r_Q_raw=r_Q×50 L / min / Δt.

[0077] Calculation example (take node 3 flow rate as an example):

[0078] Standardized flow rate data: Q_norm=0.310 at t=200 ms (k=2), Q_norm=0.294 at t=100 ms (k=1);

[0079] Standardized change rate r_Q=(0.310-0.294) / 0.1 s=0.16 / s;

[0080] Original flow rate change rate r_Q_raw=0.16 / s×50 L / min=8 L / (min?s) (indicating that the flow rate increases by 8 L / min per second, which is consistent with the dynamic characteristics of pump frequency increase);

[0081] Abnormal value processing: if r_Q_raw>20 L / (min?s) (exceeding the maximum adjustment capacity of the pump), it is determined as noise, and the change rate of the previous time is replaced (such as r_Q_raw=25 L / (min?s), replaced by 8 L / (min?s) of the previous time).

[0082] II. Analysis of state of charge (SOC) change trend

[0083] Definition and analysis method of change trend: SOC change trend reflects the change direction and rate of SOC with time, linear regression fitting is used to analyze the SOC data of the last N sampling points, the slope of the fitted straight line is used to judge the trend: slope>0 is "upward trend" (charging), slope=0 is "stable trend", slope<0 is "downward trend" (discharging), N is set to 10 points (corresponding to 1000 ms, considering the trend stability and real-time performance).

[0084] Linear regression fitting calculation: Let N SOC data be (x1, y1), (x2, y2), …, (xN, yN), where xi is the timestamp (unit s, such as t = 100 ms = 0.1 s), yi is the normalized SOC value, and the straight line equation is y = ax + b. The calculation formula of the slope a is: a = [NΣ (xiyi) - ΣxiΣyi] / [NΣ (xi²) - (Σxi)²]. Trend analysis example:

[0085] The last 10 SOC data (t = 0 ms ~ 900 ms, normalized value): 0.652, 0.6518, 0.651, 0.651, 0.650, 0.650, 0.649, 0.649, 0.648, 0.648; Calculate Σxi = 0 + 0.1 + 0.2 + … + 0.9 = 4.5s, Σyi = 0.652 + 0.6518 + … + 0.648 ≈ 6.5008; Σxiyi = 0 × 0.652 + 0.1 × 0.6518 + … + 0.9 × 0.648 ≈ 2.924; Σxi² = 0² + 0.1² + … + 0.9² = 2.85s²; Slope a = [10 × 2.924 - 4.5 × 6.5008] / [10 × 2.85 - 4.5²] = [29.24 - 29.2536] / [28.5 - 20.25] =

[0086] (-0.0136) / 8.25 ≈ -0.00165 / s; Trend determination: slope a ≈ -0.00165 / s < 0, "downward trend", original SOC change rate = |a| × 100% / s = 0.165% / s (indicating that the SOC decreases by 0.165% per second, which is consistent with the rate of 10A discharge current).

[0087] III. Construction of dynamic characteristic index data set

[0088] Index composition: The dynamic characteristic index data set contains "static basic indicators" and "dynamic derived indicators", a total of 12 indicators, covering the static values and dynamic characteristics of flow, pressure and SOC:

[0089] Static basic indicators (6): 4 node normalized flow, positive electrode inlet and outlet normalized pressure difference (positive electrode inlet - positive electrode outlet), normalized SOC;

[0090] Dynamic derived indicators (6): Instantaneous change rate (original value) of 4 node flow, SOC change trend slope (normalized), SOC change rate (original value, % / s);

[0091] Dynamic derived indicators: Node 3 flow rate change rate = 7.5 L / (min·s), Node 4 flow rate change rate = 7.0 L / (min·s), SOC trend slope = -0.00165 / s, SOC change rate = 0.165% / s;

[0092] Dataset format: The dataset is stored in JSON format, and each record contains a timestamp, the names of 12 indicators, numerical values, and units.

[0093] Dataset update frequency: The dataset is updated at the same frequency as the standardized data (10 Hz), providing complete feature input for subsequent dynamic coupling analysis.

[0094] Integrate the dynamic feature indicator dataset with timestamp information to construct a real-time data sequence of electrolyte operating state containing flow rate, pressure, state of charge, and their change rates.

[0095] I. Precision and format specification of timestamp

[0096] Timestamp precision requirement: The flow rate and pressure change response time of the flow battery electrolyte is within 100 ms, and the SOC change response time is within 1 s, therefore the timestamp needs to achieve "millisecond-level precision" (format "YYYYMMDD-HHMMSS-SSS", where SSS is milliseconds, 000-999), ensuring that the time position of each data point is unique and accurate, avoiding time deviation in time series analysis.

[0097] Time synchronization mechanism: Use "Network Time Protocol (NTP)" to synchronize the time of all sensors and data acquisition systems, with a synchronization accuracy of ≤10 ms (satisfying the time alignment requirement of 10 Hz sampling frequency). For example, the flow sensor, pressure sensor, and BMS all obtain time from the data acquisition system through NTP, ensuring that all data timestamps at the same physical time are consistent (e.g., t=20250923-153000-500 represents 15:30:00 500 milliseconds).

[0098] II. Structure design of real-time data sequence

[0099] Overall structure of the sequence: The real-time data sequence is a "timestamp-driven array structure", with each array element corresponding to a timestamp, containing four sub-modules: "basic information", "flow-related indicators", "pressure-related indicators", and "SOC-related indicators". Each sub-module contains static values and dynamic characteristics of the corresponding parameters, ensuring clear classification of indicators and facilitating model analysis.

[0100] Detailed composition of sub-modules:

[0101] Basic information: timestamp, data collection status (normal / abnormal), device running status (pump running / stopped, valve open / closed); flow-related indicators: raw flow of 4 nodes (L / min), normalized flow, instantaneous change rate (L / (min-s)); pressure-related indicators: raw pressure at inlet and outlet of positive and negative electrodes (kPa), normalized pressure, pressure difference between positive and negative electrodes (kPa); SOC-related indicators: raw SOC (%), normalized SOC, change trend slope (1 / s), change rate (% / s); data type and precision: all numerical data are stored as double-precision floating-point numbers (64 bits) to ensure calculation accuracy; status data are stored as strings (e.g. "normal" "pump running") for easy manual identification.

[0102] III. Generation and update mechanism of real-time data sequence

[0103] Sequence generation process: Step 1: read all indicators at the current time from the dynamic feature indicator dataset every 100ms (10Hz); Step 2: read device status signals (e.g. running frequency of circulating pump 50Hz, valve opening 80%) and supplement to the basic information module; Step 3: organize data in the order of "basic information -> flow -> pressure -> SOC" to generate sequence elements for the current timestamp; Step 4: append the current element to the end of the real-time data sequence, with a maximum length of 36000 elements (corresponding to 1 hour of data, with the oldest elements automatically deleted to avoid memory overflow).

[0104] Storage method: real-time data sequence is maintained as a dynamic array in memory, and is also stored as a JSON file by minute to ensure real-time data calling and historical backtracking.

[0105] IV. Integrity and validity verification of data sequence

[0106] Integrity verification: every 100 sequence elements are generated (corresponding to 10 seconds), check whether each element contains all 12 dynamic feature indicators, if a certain indicator is missing (e.g. node 3 flow change rate), mark it as "data missing" and use linear interpolation to complete (e.g. use the average of node 2 and node 4 change rates instead);

[0107] Validity verification: check the numerical range of key indicators (e.g. flow 0-50 L / min, SOC 0%-100%), if it exceeds the range (e.g. SOC=105%), mark it as "data abnormal" and replace it with the valid data at the previous time;

[0108] Verification result output: generate a "data quality report" every 1 minute, including completeness rate (≥99% is qualified), abnormal rate (≤1% is qualified), example report: "20250923-1530, completeness rate 99.8%, abnormal rate 0.2%, data quality qualified", to ensure the reliability of real-time data sequence.

[0109] S202, based on the real-time data sequence, combining the historical cycle attenuation characteristics of the electrolyte and the current working condition parameters, determining the actual available inventory of the electrolyte and predicting the short-term inventory demand trend through dynamic coupling analysis model;

[0110] Specifically, the historical cycle attenuation characteristic data of the electrolyte can be extracted from the battery management database and sorted into a historical attenuation characteristic vector;

[0111] I. Selection criteria and data types of historical attenuation characteristics

[0112] The core principle of selection is to select characteristics that are "directly related to the actual available inventory of the electrolyte, quantifiable monitoring, and showing regular changes with cycle number", and exclude irrelevant or highly fluctuating parameters, finally determine 5 core characteristics:

[0113] Cycle count (Cycle Count): the total number of charge and discharge cycles experienced by the electrolyte, which directly affects the degree of attenuation (the more the cycle number, the more serious the attenuation);

[0114] Capacity fade rate (Capacity Fade Rate): the percentage difference between the actual capacity of the current electrolyte and the initial capacity, reflecting the degree of loss of active substances in the electrolyte;

[0115] Viscosity change rate (Viscosity Change Rate): the percentage difference between the current viscosity of the electrolyte and the initial viscosity, an increase in viscosity will increase the flow resistance and indirectly affect the utilization efficiency of the inventory;

[0116] Vanadium ion concentration deviation (Vanadium Concentration Deviation): the difference between the vanadium ion concentration in the positive and negative electrolyte and the initial concentration (1.5 mol / L), a decrease in concentration will directly reduce the available active substances and reduce the actual inventory;

[0117] Coulombic efficiency fade rate (Coulombic Efficiency Fade Rate): the percentage difference between the current charge and discharge coulombic efficiency and the initial efficiency (95%), efficiency decay reflects the intensification of electrolyte side reactions, which is indirectly related to inventory loss.

[0118] Data source: BMS database automatically records the above characteristic data once every 10 cycles, and the storage format is a relational database (MySQL) table structure, with fields including "record ID, cycle number, capacity fade rate, viscosity change rate, vanadium concentration deviation, coulomb efficiency fade rate, and record time".

[0119] II. Extraction and cleaning process of historical data

[0120] Extraction condition setting: For the current electrolyte to be analyzed (device number FVB-2024-001), extract all historical records from initial use to the present, and set the SQL query condition; data cleaning process: missing value filling; outlier removal; data smoothing.

[0121] III. Construction and standardization of historical decay feature vector

[0122] Vector construction: Select the cleaned data of the "last complete record" (cycle 1800, record time September 20, 2025) to construct a 5-dimensional historical decay feature vector V_hist = [C_cycle, C_fade, Visc_change, V_conc_dev, Eff_fade], where the definition and example values of each element are as follows:

[0123] C_cycle (cycle number): 1800 (dimensionless, directly take the integer);

[0124] C_fade (capacity fade rate): 4.5% (percentage, reflecting the actual capacity of 95.5% of the initial capacity, the initial capacity corresponds to 500L electrolyte, and the decayed capacity corresponds to 477.5L);

[0125] Visc_change (viscosity change rate): 8.3% (initial viscosity 1.2cP, current viscosity 1.2996cP, (1.2996-1.2) / 1.2x100%=8.3%);

[0126] V_conc_dev (vanadium concentration deviation): -0.12mol / L (current concentration 1.38mol / L, 1.38-1.5=-0.12mol / L, negative value indicates concentration decrease);

[0127] Eff_fade (coulomb efficiency fade rate): 3.2% (initial coulomb efficiency 95%, current 91.92%, (91.92-95) / 95x100%=-3.2%, take absolute value as 3.2%);

[0128] Standardization: To unify the standardization format of subsequent real-time data sequences, the "Min-Max normalization" is used to map the vector elements to the interval [0, 1], and the minimum value (x_min) and maximum value (x_max) of each feature are set based on the electrolyte full life cycle data:

[0129] C_cycle: x_min = 0 times, x_max = 2000 times, normalized = 1800 / 2000 = 0.9;

[0130] C_fade: x_min = 0%, x_max = 20% (fade rate at end of life), normalized = 4.5 / 20 = 0.225;

[0131] Visc_change: x_min = 0%, x_max = 30%, normalized = 8.3 / 30 ≈ 0.277;

[0132] V_conc_dev: x_min = -0.5 mol / L (severe fade), x_max = 0 mol / L, normalized = (-0.12 - (-0.5)) / (0 - (-0.5)) = 0.38 / 0.5 = 0.76;

[0133] Eff_fade: x_min = 0%, x_max = 10%, normalized = 3.2 / 10 = 0.32;

[0134] The final normalized historical fade feature vector V_hist_norm = [0.9, 0.225, 0.277, 0.76, 0.32] is stored in JSON format.

[0135] The electrolyte running state real-time data sequence is fused with the historical fade feature vector, combined with the working condition parameters including the current temperature and load, to form a multi-dimensional feature input set;

[0136] I. Logic and specific method of feature fusion

[0137] Fusion core logic: Real-time data sequences reflect the "current 1-hour dynamic running state" of the electrolyte (such as flow changes, SOC trends), and historical fade feature vectors reflect the "long-term cycle-based fade basis". Both need to be fused by "time dimension alignment + feature dimension splicing" - taking the "latest timestamp" (such as 20250923-153000-500) of the real-time data sequence as the basis, the real-time feature corresponding to this timestamp is directly spliced with the historical fade feature vector to form a "real-time + historical" combined feature, and the current working condition parameters are added to form a complete input set.

[0138] Real-time feature extraction: From the real-time data sequence of the electrolyte running state, 12 dynamic feature indicators with the latest timestamp (20250923-153000-500) are extracted, and the standardized values are as follows:

[0139] Flow-related (4): node1_flow_norm=0.296, node2_flow_norm=0.300, node3_flow_norm=0.306, node4_flow_norm=0.304;

[0140] Pressure-related (2): pos_diff_norm=0.001, neg_diff_norm=0.0012;

[0141] SOC-related (2): soc_norm=0.650, soc_rate_norm=0.165 (SOC change rate 0.165% / s, normalized x_min=0% / s, x_max=1% / s);

[0142] Rate-related (4): node1_rate_norm=0.375, node2_rate_norm=0.4, node3_rate_norm=0.375, node4_rate_norm=0.35 (flow rate change 7.5 L / (min·s), normalized x_min=0, x_max=20);

[0143] The above 12 real-time features form a vector V_real=[0.296, 0.300, 0.306, 0.304, 0.001, 0.0012, 0.650, 0.165, 0.375, 0.4, 0.375, 0.35].

[0144] Feature concatenation and fusion: The real-time feature vector V_real (12 dimensions) and the normalized historical decay feature vector V_hist_norm (5 dimensions) are directly concatenated in the order of "real-time first, history last", forming a 17-dimensional "real-time + history" combined feature vector V_comb=[V_real[0], V_real[1],..., V_real

[11] , V_hist_norm[0],..., V_hist_norm[4]], the first 5 elements are [0.296, 0.300, 0.306, 0.304, 0.001], and the last 5 elements are [0.9, 0.225, 0.277, 0.76, 0.32].

[0145] II. Collection and standardization of current operating parameters

[0146] Working condition parameter selection and collection: Working condition parameters reflect the "current environment and load on the immediate impact of electrolyte inventory demand", select 2 core parameters:

[0147] Electrolyte current temperature (T): affects the viscosity and ion diffusion rate of electrolyte, too low temperature will increase the flow resistance, indirectly reduce the actual available inventory, through the platinum resistance temperature sensor (PT100, precision ±0.1℃) installed in the electrolyte tank, sampling frequency 1Hz, the current value is 25.3℃;

[0148] Battery current load (P): reflects the real-time consumption rate of electrolyte, the higher the load, the faster the active material reaction, the more urgent the inventory demand, through the BMS to collect the inverter output power, the current value is 10.2kW (the system rated load is 10kW, slightly overrated);

[0149] Standardization: adopt Min-Max normalization consistent with other features, set reasonable range of parameters:

[0150] Temperature T: x_min=15℃ (minimum operating temperature), x_max=40℃ (maximum operating temperature), after normalization=(25.3-15) / (40-15)=10.3 / 25=0.412;

[0151] Load P: x_min=0kW (no load), x_max=12kW (overload protection threshold), after normalization=10.2 / 12=0.85;

[0152] The normalized working condition parameter vector V_cond=[0.412,0.85].

[0153] Three, construction and verification of multi-dimensional feature input set

[0154] Input set structure: combine the 17-dimensional combined feature vector V_comb and the 2-dimensional working condition parameter vector V_cond again to form a 19-dimensional multi-dimensional feature input set X=[V_comb[0],...,V_comb

[16] ,V_cond[0],V_cond[1]], the meaning and example value of each dimension in the input set are as follows (in order):

[0155] node1_flow_norm: 0.296 (node 1 normalized flow);

[0156] node2_flow_norm: 0.300 (node 2 normalized flow);

[0157] node3_flow_norm: 0.306 (node 3 normalized flow);

[0158] node4_flow_norm: 0.304 (Node 4 normalized flow);

[0159] pos_diff_norm: 0.001 (Positive electrode normalized pressure difference);

[0160] neg_diff_norm: 0.0012 (Negative electrode normalized pressure difference);

[0161] soc_norm: 0.650 (Normalized SOC);

[0162] soc_rate_norm: 0.165 (Normalized SOC change rate);

[0163] node1_rate_norm: 0.375 (Node 1 normalized flow change rate);

[0164] node2_rate_norm: 0.4 (Node 2 normalized flow change rate);

[0165] node3_rate_norm: 0.375 (Node 3 normalized flow change rate);

[0166] node4_rate_norm: 0.35 (Node 4 normalized flow change rate);

[0167] C_cycle_norm: 0.9 (Normalized cycle count);

[0168] C_fade_norm: 0.225 (Normalized capacity fade rate);

[0169] Visc_change_norm: 0.277 (Normalized viscosity change rate);

[0170] V_conc_dev_norm: 0.76 (Normalized vanadium concentration deviation);

[0171] Eff_fade_norm: 0.32 (Normalized coulomb efficiency fade rate);

[0172] T_norm: 0.412 (Normalized temperature);

[0173] P_norm: 0.85 (Normalized load);

[0174] Dimension verification and storage: Verify the input set through the "feature dimension checklist" for missing dimensions (19 dimensions in total, consistent with the design), no outliers (all elements are in the [0,1] interval), and store it as a Numpy array format (float32 type). At the same time, generate input set documentation, label the meaning and source of each dimension, and provide traceability for model input.

[0175] Input the multi-dimensional feature input set into the dynamic coupling analysis model, which uses a long short-term memory network architecture to analyze the nonlinear relationship between electrolyte inventory and operating parameters and calculate the actual available inventory of electrolyte;

[0176] LSTM architecture selection basis: The relationship between electrolyte inventory and operating parameters has "time sequence correlation" (such as the current inventory being affected by the flow change in the past 1 hour) and "nonlinearity" (such as the impact of concentration decay on inventory showing nonlinear growth with cycle number), traditional linear models (such as linear regression) cannot capture these features, while LSTM can effectively remember long-term time series information through the gating mechanism of "forget gate, input gate, and output gate", making it suitable for this nonlinear time series modeling scenario.

[0177] Specific architecture parameters: The model uses a four-layer structure of "input layer → LSTM hidden layer → fully connected layer → output layer", with the following parameters for each layer:

[0178] Input layer: Number of neurons = number of dimensions of multi-dimensional feature input set = 19, receives a 19-dimensional X vector, uses "ReLU" (linear rectifier function to avoid gradient disappearance) as the activation function, and the input data format is (batch_size, time_steps, input_dim), where batch_size = 32 (number of samples for each training), time_steps = 60 (takes the feature sequence of the past 60 timestamps, corresponding to 6000ms = 6 seconds, to capture short-term time sequence dependence), and input_dim = 19;

[0179] LSTM hidden layer: Set up a 2-layer stacked structure (to enhance the model's expression ability), with 32 neurons in each layer, "Sigmoid" activation function for forget gate, input gate, and output gate (output in the [0,1] interval to control information transmission), "tanh" activation function for cell state update (output in the [-1,1] interval to adjust state value), and dropout_rate = 0.2 (to prevent model overfitting, randomly discard 20% of neuron connections);

[0180] Fully connected layer: Set up 1 layer with 16 neurons, using "ReLU" as the activation function, to map the output of the LSTM layer (32 dimensions) to 16 dimensions and further extract high-order features;

[0181] Output layer: number of neurons = 1 (output actual available inventory, unit L), no activation function (regression task, directly output continuous value), output range set to [0, 500L] (not exceeding the nominal volume of the tank).

[0182] Model inference process: the multi-dimensional feature input set X (19 dimensions) is expanded to an input tensor of "32 (batch_size) x 60 x 19" by time_steps = 60 (the insufficient part is filled with 0, and batch_size = 1 because it is currently a single sample inference, the actual input is 1 x 60 x 19), and the trained LSTM model is input. The model calculates through the following steps:

[0183] First step: the input layer maps the 19-dimensional features to a 19-dimensional ReLU activated output;

[0184] Second step: the LSTM hidden layer processes the 60-time-step feature sequence through the gating mechanism and outputs a 32-dimensional hidden state;

[0185] Third step: the fully connected layer maps the 32-dimensional hidden state to a 16-dimensional ReLU output;

[0186] Fourth step: the output layer outputs one continuous value, which is the prediction of the actual available inventory of the current electrolyte.

[0187] Example of calculation result: the actual available inventory prediction value output by the model inference is 482.3L, which means that the current electrolyte tank has a nominal volume of 500L, and after deducting the loss due to circulation attenuation (22.5L loss corresponding to a capacity attenuation rate of 4.5%), the flow loss due to increased viscosity (about 3.2L), and the active material loss due to concentration deviation (about 2.0L), the effective volume available for reaction is 482.3L, which is basically consistent with the theoretical estimated value (500-22.5-3.2-2.0=472.3L, there is a small difference, because the model considers the dynamic influence of real-time working conditions, such as the current temperature of 25.3°C which is slightly higher than the standard temperature of 25°C, which reduces the viscosity loss).

[0188] Based on the output of the dynamic coupling analysis model, a time series prediction algorithm is used to predict the inventory demand trend in the future preset period, and an inventory demand prediction curve is output.

[0189] I. Selection and adaptation logic of time series prediction algorithm

[0190] Algorithm selection basis: The electrolyte inventory demand changes are affected by "periodic factors" (such as regular changes in daily load peaks) and "trend factors" (such as long-term inventory decline with cycle decay), and may be affected by "special events such as holidays". The Prophet algorithm of Facebook can flexibly capture these factors through a three-component model of "trend, seasonality, and holiday items", and is robust to missing data and outliers, suitable for this scenario; compared with traditional ARIMA algorithm (high requirement for data stationarity, difficult to handle periodicity and trend superposition), Prophet is more suitable for short-term prediction (within 1 hour) in this scenario.

[0191] Three-component model of Prophet algorithm: the predicted value y(t) is composed of trend item g(t), seasonality item s(t), holiday item h(t), and error item ε(t), the formula is y(t)=g(t)+s(t)+h(t)+ε(t), the meanings of each component are as follows:

[0192] Trend item g(t): reflects the long-term change trend of inventory (such as slow decline due to decay), uses "piecewise linear trend" model, and captures trend changes (such as sudden increase in load leading to steep trend slope) through changepoints;

[0193] Seasonality item s(t): reflects the periodic change of inventory (such as periodic consumption of inventory due to hourly load fluctuations), uses "Fourier series" modeling, the formula is s(t)=Σ(a_kcos(2πkt / P)+b_ksin(2πkt / P)), where P is the period (here P=3600 seconds=1 hour, corresponding to short-term load period);

[0194] Holiday item h(t): reflects special events (such as sudden drop in inventory demand due to system maintenance downtime), here the prediction period is short (1 hour), there are no special events, h(t)=0;

[0195] Error term ε(t): follows normal distribution N(0,σ²), represents random fluctuations.

[0196] II. Prediction data preparation and parameter setting

[0197] Historical time series data preparation: extract the actual available inventory time series data of electrolyte in the past 24 hours, sampling frequency=1 time / minute (total 1440 data points), data source is the historical output value of dynamic coupling analysis model, example data segment (part of time points):

[0198] 2024-09-23 14:30:00: 485.2 L; 2024-09-23 14:45:00: 484.0 L; 2024-09-23 15:00:00: 483.1 L; 2024-09-23 15:15:00: 482.7 L; 2024-09-23 15:30:00: 482.3 L (current time, model output value); The data format is arranged as "ds (timestamp), y (inventory value)" two columns required by Prophet, and stored as a CSV file.

[0199] Prediction parameter settings:

[0200] Default prediction period: 1 hour (3600 seconds), prediction step = 5 minutes (total of 12 prediction data points, balancing prediction accuracy and data volume); Trend item parameters: set changepoint_prior_scale = 0.05 (control change point sensitivity, the smaller the value, the smoother the trend, avoiding overfitting short-term fluctuations); Seasonal item parameters: set yearly_seasonality = False (1-hour prediction does not require annual cycles), weekly_seasonality = False (no weekly cycles), daily_seasonality = True (need to capture daily load cycles), seasonality_prior_scale = 10 (control seasonal item strength, the larger the value, the more obvious the seasonal fluctuations);

[0201] Confidence interval: set interval_width = 0.95 (output 95% confidence interval, reflecting the prediction uncertainty, i.e. 95% of the true values will fall within this interval).

[0202] Three, generation and interpretation of inventory demand prediction curve

[0203] Model training and prediction execution: input the past 24 hours of time series data into the Prophet model, after training, call the predict() function to generate the prediction results for the next 1 hour, including "ds (prediction timestamp), yhat (predicted value), yhat_lower (95% confidence interval lower limit), yhat_upper (95% confidence interval upper limit)" four columns.

[0204] Prediction curve generation: take "time (ds)" as the horizontal axis (from 15:30 to 16:30, a total of 60 minutes), and "inventory value (L)" as the vertical axis (range 474 L ~ 485 L), draw three curves:

[0205] Predicted value curve (yhat): blue solid line, smoothly decreasing from 482.3 L (15:30) to 475.7 L (16:30) with a slope of about -0.11 L / min (6.6 L / hour decrease), reflecting the inventory consumption rate under the current 10.2 kW load;

[0206] Confidence interval curve (yhat_lower / yhat_upper): gray dashed lines, the area between the two curves is the 95% confidence interval, with a width of about 1.3 L (e.g., at 15:35, the interval width = 483.1-480.5 = 2.6 L, and the interval slightly widens as the prediction time increases, due to increased uncertainty);

[0207] Trend interpretation: the predicted curve shows a "smooth downward trend", as the current load is stable at 10.2 kW (slightly exceeding the rated load), the electrolyte active material reaction rate is stable, and the inventory consumption rate is uniform; if the load increases to 12 kW (overload), the predicted curve slope will become steeper (e.g., -0.15 L / min), and the inventory will decrease faster; if the load decreases to 5 kW, the slope will become slower (e.g., -0.06 L / min), and the inventory will decrease slower. This trend provides a clear demand orientation for subsequent adjustment of pump frequency and valve opening (e.g., to slow down the inventory decrease, the flow rate needs to be appropriately reduced to reduce active material consumption).

[0208] Four, uncertainty analysis and correction of prediction results

[0209] Uncertainty sources: prediction errors mainly come from two aspects: one is the measurement error of historical data (MAE of dynamic coupling model ≤3.5 L), and the other is the uncertainty of future working conditions (e.g., sudden changes in load, temperature fluctuations), leading to the existence of 95% confidence interval;

[0210] Real-time correction mechanism: every 5 minutes (consistent with the prediction step) use the latest output value (actual available inventory) of the dynamic coupling model to update the historical time series data, retrain the Prophet model and update the predicted curve, e.g., at 15:35, the actual available inventory calculated by the model is 481.9 L (with an error of 0.1 L from the predicted value 481.8 L), which is added to the historical data to re-predict the trend from 15:35 to 16:35, and the corrected predicted curve has the same slope but the starting point is updated to 481.9 L, ensuring that the prediction is synchronized with the actual state;

[0211] Abnormal correction: if a sudden change in working conditions is detected (e.g., at 15:40, the load suddenly decreases to 5 kW), an "abnormal re-prediction" is triggered immediately, the seasonal term parameters are re-set, and the predicted curve slope is corrected from -0.11 L / min to -0.06 L / min, avoiding prediction failure due to sudden changes in working conditions and ensuring that the predicted trend is always consistent with the actual demand.

[0212] S203, generating an electrolyte circulation dynamic adjustment strategy by using a multi-objective optimization algorithm according to the actual available inventory and the short-term inventory demand change trend, wherein the strategy at least includes frequency adjustment of the electrolyte circulation pump and valve opening degree collaborative control;

[0213] Specifically, a multi-objective optimization function can be established with the minimization of energy consumption, the maximization of inventory utilization rate, and the optimization of system stability as the targets, wherein the frequency of the electrolyte circulation pump and the opening degree of the valve are defined as decision variables.

[0214] I. Definition and constraint range of decision variables

[0215] Electrolyte circulation pump frequency (f):

[0216] Definition: The operating frequency of the circulation pump (unit: Hz), which determines the output flow rate of the pump (the higher the frequency, the greater the flow rate), is the core variable for adjusting the electrolyte circulation rate.

[0217] Value range: Based on the physical characteristics of the pump, the minimum frequency f_min=20Hz is set (to avoid pump cavitation and ensure stable flow rate), and the maximum frequency f_max=60Hz is set (to avoid pump cavitation and ensure stable flow rate), so f∈[20, 60]Hz.

[0218] Practical significance: Taking the 10kW system as an example, the pump frequency of 20Hz corresponds to an output flow rate of 8L / min, and the pump frequency of 60Hz corresponds to an output flow rate of 24L / min. The flow rate and frequency are linearly related (Q=0.4f-0, R²=0.99, fitting error ≤0.5L / min).

[0219] Valve opening degree (θ): Definition: The opening degree of the regulating valve in the electrolyte circulation loop (unit: %), which controls the resistance of the loop (the smaller the opening degree, the greater the resistance, and the smaller the flow rate), assists the pump frequency in achieving coordinated adjustment of flow rate and pressure; Value range: The minimum opening degree θ_min=30% is set (to avoid excessive opening degree leading to sudden pressure rise and damage to the pipeline), and the maximum opening degree θ_max=100% is set (full opening state, minimum resistance), so θ∈[30, 100]%;

[0220] Practical significance: At a pump frequency of 50Hz (flow rate of 20L / min), a valve opening degree of 30% corresponds to a loop pressure of 0.4MPa (exceeding the safety threshold of 0.3MPa), an opening degree of 80% corresponds to a pressure of 0.25MPa (within the safety range), and the opening degree and pressure are negatively correlated (P=-0.005θ+0.55, R²=0.98).

[0221] II. Mathematical modeling of three target functions

[0222] Energy consumption minimization target function (F1(f, θ)):

[0223] Objective: Minimize the operating energy consumption (unit: kW・h) of the electrolyte circulating pump, reduce system operating costs; Energy consumption characteristics: Pump energy consumption is approximately proportional to the cube of frequency (similar law of fluid machinery), while affected by valve opening (the smaller the opening, the greater the resistance, and the additional energy consumption), the mathematical expression is: F1(f,θ)=P0×(f / f n )³×(1 / (k×θ / 100))×t, where: P0=1.5kW (rated power of pump, power when f n =60Hz); f n =60Hz (rated frequency of pump); k=0.85 (opening influence coefficient, empirical value, reflecting the correction effect of opening on energy consumption); t=1h (calculation period, unified energy consumption calculation dimension).

[0224] Stock utilization rate maximization objective function (F2(f,θ)):

[0225] Objective: Maximize the utilization rate of the actual available stock of electrolyte (unit: %), avoid stock idling or excessive consumption;

[0226] Utilization rate definition: The ratio of the actual amount of electrolyte participating in the reaction to the total available stock, positively related to the circulating flow (the greater the flow, the more electrolyte participating in the reaction per unit time), the mathematical expression is: F2(f,θ)=[Q(f,θ)×t×ρ×C] / (V_total×ρ×C)×100%=(Q(f,θ)×t) / V_total×100%, where: Q(f,θ)=0.4f×(θ / 100) (fitting relationship between flow and frequency, opening, unit L / min, 0.4 is the frequency coefficient, and θ / 100 is the opening correction); t=60min (calculation period, unified time dimension); V_total=500L (total available stock of electrolyte, effective volume of storage tank); ρ=1.2g / cm³ (electrolyte density), C=1.5mol / L (vanadium ion concentration), the numerator and denominator cancel out, which does not affect the utilization rate calculation.

[0227] System stability optimization objective function (F3(f,θ)):

[0228] Objective: Minimize the pressure fluctuation and flow fluctuation of the electrolyte circulating loop to ensure stable system operation, measured by "fluctuation coefficient" (the smaller the fluctuation coefficient, the better the stability);

[0229] The fluctuation coefficient is defined as the weighted sum of the pressure fluctuation and the flow fluctuation, and the mathematical expression is: F3(f, θ) = a x σ_P(f, θ) + β x σ_Q(f, θ). Wherein: a = 0.6 (pressure fluctuation weight, pressure has greater impact on system safety, and higher weight); β = 0.4 (flow fluctuation weight); σ_P(f, θ) = |P(f, θ) - P_target| / P_target (pressure fluctuation coefficient, P_target = 0.25 MPa is the target pressure); σ_Q(f, θ) = |Q(f, θ) - Q_target| / Q_target (flow fluctuation coefficient, Q_target = 16 L / min is the target flow).

[0230] III. Standardization processing of objective function

[0231] Because the dimensions and value ranges of the three objective functions are different (F1: 1~3kW・h, F2: 50%~100%, F3: 0~0.2), direct solution will lead to the dominance of optimization results of the objective with large dimension, and all objectives need to be mapped to the [0,1] interval through "linear normalization". The processing formula is as follows:

[0232] Minimization objective (F1, F3): F'_i = (F_i_max - F_i) / (F_i_max - F_i_min), where F_i_max is the maximum value of the objective, and F_i_min is the minimum value of the objective;

[0233] Maximization objective (F2): F'_i = (F_i - F_i_min) / (F_i_max - F_i_min);

[0234] Example processing (take F1 as an example): F1_min = 1.5 x (20 / 60)³ x (1 / (0.85 x 1)) x 1 ≈ 0.08 kW・h, F1_max = 1.5 x (60 / 60)³ x (1 / (0.85 x 0.3)) x 1 ≈ 5.88 kW・h, when F1 = 1.26 kW・h, F'_1 = (5.88 - 1.26) / (5.88 - 0.08) ≈ 0.797 (normalized value, the closer to 1 indicates the better energy consumption).

[0235] Based on the actual available inventory and inventory demand prediction curve, set optimization constraint conditions, wherein the conditions include inventory safety threshold and equipment operation limit;

[0236] I. Inventory safety threshold constraint

[0237] Definition of actual available inventory: calculated by dynamic coupling analysis model, refers to the effective inventory in the electrolyte tank at the current time that can participate in the circulation (after deducting the amount of tank bottom sediment and pipeline residue), the actual available inventory S_actual of the 10kW system in the example is 420L (total volume 500L, deducting 80L unusable part).

[0238] Inventory safety threshold setting: the minimum safety inventory S_min=100L (to avoid the pump from sucking air due to low tank level, causing cavitation failure); the maximum safety inventory S_max=480L (to avoid electrolyte overflow due to high tank level, while reserving expansion space); constraint meaning: during optimization, it is necessary to ensure that the actual inventory S(t) ∈ [S_min, S_max] at any time, combined with the relationship between inventory change and flow (S(t)=S_actual-∫Q(f,θ)dt+∫Q_returndt, Q_return is the return flow, Q_return=Q(f,θ) in steady state, so S(t)≈S_actual), converted to flow constraint: Q(f,θ)≤(S_actual-S_min) / t (t=1h, to avoid the inventory falling below S_min within 1 hour);

[0239] Mathematical expression: Q(f,θ)≤(420-100) / 60≈5.33L / min (but combined with the minimum flow of the pump 8L / min, the actual Q≥8L / min, so this constraint is automatically satisfied when the pump frequency≥20Hz, no additional restriction is needed).

[0240] Short-term inventory demand trend constraint:

[0241] Short-term demand prediction curve: output by dynamic coupling analysis model, the inventory demand in the next 1 hour (preset period) shows a "first rising and then stable" trend, the demand rises from 18L / min to 20L / min in 0~30min, and stabilizes at 20L / min in 30~60min;

[0242] Constraint meaning: the optimized flow Q(f,θ) needs to follow the demand trend, i.e. Q(f,θ)≥Q_demand_min (Q_demand_min=18L / min, minimum demand), to avoid insufficient flow leading to unsatisfied inventory demand;

[0243] Mathematical expression: Q(f,θ)=0.4f×(θ / 100)≥18L / min, substituting f≥20Hz, θ≥30%, we get 0.4×20×0.3=2.4L / min<18L / min, which needs to be adjusted by decision variables to meet (such as f=50Hz, θ=90%, Q=0.4×50×0.9=18L / min, which just meets the requirement).

[0244] II. Equipment operating limit constraints

[0245] Operating limit of circulating pump:

[0246] Power constraint: actual operating power of pump P(f)≤P_rated=1.5kW (rated power), combined with the relationship between power and frequency P(f)=P0×(f / f n )³, P(f)=1.5×(f / 60)³≤1.5, f≤60Hz (consistent with the upper limit of decision variable, already included in variable range);

[0247] Temperature rise constraint: pump operating temperature T(f)≤T_max=85℃ (motor insulation temperature tolerance), empirical formula T(f)=30+0.01f² (environmental temperature 30℃, the higher the frequency, the greater the temperature rise), when f=60Hz, T=30+0.01×3600=66℃<85℃, meet the constraint; when f=80Hz (exceed the upper limit), T=30+0.01×6400=94℃>85℃, so the constraint of f≤60Hz can ensure the safety of temperature rise.

[0248] Operating limit of valve and pipeline:

[0249] Pressure constraint: maximum pressure of loop P_max=0.3MPa (rated pressure of pipeline, exceeding will lead to pipeline leakage), combined with the relationship between pressure and opening, frequency P(f,θ)=0.001f-0.005θ+0.5 (empirical fitting formula), in the example f=60Hz, θ=30% when, P=0.001×60-0.005×30+0.5=0.41MPa>0.3MPa, need to supplement the pressure constraint: P(f,θ)≤0.3MPa, substitute the formula 0.001f-0.005θ+0.5≤0.3→0.001f-0.005θ≤-0.2→f-5θ≤-200 (mathematical expression), when f=60Hz, θ≥(60+200) / 5=52% (i.e. θ≥52%, need to adjust the valve opening to more than 52% to avoid pressure exceeding);

[0250] Flow impact constraint: flow rate change rate ≤5L / (min・s) (to avoid pipeline vibration caused by sudden flow change), combined with the relationship between flow and frequency Q=0.4f, flow rate change rate=0.4×Δf / Δt, set Δt≥1s (frequency adjustment interval), so Δf≤(5×1) / 0.4=12.5Hz (i.e. each frequency adjustment does not exceed 12.5Hz, such as adjusting from 50Hz to 60Hz, need to adjust twice: 50→56.25→60Hz, interval 1s).

[0251] III. Mathematical summary of constraint conditions

[0252] The mathematical expression of the final optimization constraint condition is as follows: decision variable constraint: 20≤f≤60Hz, 30≤θ≤100%; inventory safety constraint: S(t)≥100L (automatically satisfied in steady state), Q(f,θ)≥18L / min; equipment pressure constraint: f-5θ≤-200; flow impact constraint: |f(k)-f(k-1)|≤12.5Hz (k is the adjustment time, and k-1 is the previous time); power constraint: P(f)≤1.5kW (automatically satisfied).

[0253] A multi-objective particle swarm optimization algorithm is used to solve a multi-objective optimization function, and the optimal pump frequency and valve opening combination is found under the condition of meeting the optimization constraint condition. The energy efficiency, inventory utilization rate and system stability index of each candidate strategy are calculated, and a candidate optimization strategy set with performance evaluation index is generated;

[0254] I. Parameter setting of multi-objective particle swarm optimization (MOPSO) algorithm

[0255] Definition and value of algorithm core parameters: number of particles N=50 (each particle represents a group of decision variable combination (f,θ), too many will increase the calculation amount, too few will lead to insufficient search, and 50 is the optimal value); iteration number T=100 (the maximum number of iterations of the algorithm, 100 times can ensure search convergence and avoid falling into local optimum); inertia weight ω=0.7 (control the influence of particle historical speed on current speed, 0.7 is the classical value, balance global search and local search); cognitive coefficient c1=2.0 (particle learning ability to its own optimal position, 2.0 ensures that the particle can fully utilize its own experience); social coefficient c2=2.0 (particle learning ability to the optimal position of the group, 2.0 ensures that the particle can move towards the optimal direction of the group;

[0256] External archive capacity (elite solution set size)=30 (store non-dominated solutions generated during iteration, too large capacity will increase the calculation burden, 30 can cover the main optimal solution);

[0257] Speed boundary: pump frequency speed v_f∈[-5,5]Hz / iteration (frequency change does not exceed 5Hz per iteration, avoid large search jump), valve opening speed v_θ∈[-10,10]% / iteration (opening change does not exceed 10% per iteration, ensure stable search).

[0258] Parameter setting basis:

[0259] Number of particles and iteration number: combined with the decision variable dimension (2D) of the 10kW system, 50 particles can cover the main area of the decision space, and the calculation time of 100 iterations on an ordinary industrial controller (such as STM32H7) is about 500ms, which meets the real-time optimization requirement (optimization period 1s);

[0260] Inertia weight and coefficient: ω = 0.7 can maintain a larger search range at the beginning of iteration (global exploration), and gradually reduce it at the later stage (local refinement), c1 = c2 = 2.0 conforms to the classic setting of "cognitive-social" balance, avoiding premature convergence or divergence of particles.

[0261] II. Solution process of MOPSO algorithm (combined with constraint conditions)

[0262] Initialize the particle swarm: randomly generate 50 particles, and the decision variables (f, θ) of each particle need to meet the constraint conditions (20 ≤ f ≤ 60 Hz, 30 ≤ θ ≤ 100%, f - 5θ ≤ -200), for example, particle 1: f = 50 Hz, θ = 80% (satisfies 50 - 5 × 80 = -350 ≤ -200); particle 2: f = 55 Hz, θ = 52% (55 - 5 × 52 = -205 ≤ -200);

[0263] Initialize the speed (v_f, v_θ) of each particle, randomly take value within the speed boundary (such as particle 1's v_f = 2 Hz / iteration, v_θ = -5% / iteration);

[0264] Calculate the objective function value (F1, F2, F3) and the standardized value (F'1, F'2, F'3) of each particle, determine the "individual optimal position" (Pbest, initially as itself) of each particle, and add non-dominated particles to the external archive (elite solution set).

[0265] Iterative update process (take the first iteration as an example):

[0266] Step 1: update particle speed, formula is v_k+1 = ω × v_k + c1 × r1 × (Pbest_k - x_k) + c2 × r2 × (Gbest_k - x_k), where r1, r2 are [0, 1] random numbers, Gbest is the global optimal position in the external archive (selected by crowding distance, the greater the crowding distance, the better the diversity of the solution);

[0267] Step 2: update particle position, formula is x_k+1 = x_k + v_k+1, after updating, it needs to be verified whether it meets the constraints, if not, perform "boundary correction" (such as particle position exceeding f_max = 60 Hz, forced to be set to 60 Hz);

[0268] Step 3: calculate the objective function value of the new position, update the Pbest of the particle (if the new position is better, replace Pbest), and compare the new position with the solutions in the external archive, if it is a non-dominated solution, add it to the archive, and delete the solutions in the archive that are dominated by the new solution (to ensure that only non-dominated solutions are retained in the archive).

[0269] Iteration convergence criterion: When the average crowding distance change rate of solutions in the external archive of the last 10 iterations is less than or equal to 1%, the algorithm is considered to have converged, and the iteration is stopped.

[0270] III. Performance index calculation of candidate strategies and strategy set generation

[0271] Performance index definition: Energy efficiency η_energy = 1 - F1 / F1_max (F1_max = 5.88kW・h, the closer η_energy is to 1, the higher the energy efficiency); Stock utilization rate η_storage = F2 / 100% (F2 max is 100%, η_storage = 1 indicates the optimal utilization rate); System stability η_stability = 1 - F3 / F3_max (F3_max = 0.2, the closer η_stability is to 1, the better the stability).

[0272] Candidate optimization strategy set format: The strategy set is stored in CSV format, containing "strategy ID, pump frequency (Hz), valve opening degree (%), energy efficiency (%), stock utilization rate (%), system stability (%), whether the constraints are met" fields, and the strategy set covers the optimal trade-off scheme for different targets.

[0273] Based on the performance evaluation index of each candidate optimization strategy, the comprehensive score is calculated by using the weighted scoring method, and the strategy with the highest comprehensive score is selected as the electrolyte circulation dynamic adjustment strategy containing pump frequency adjustment value and valve opening degree control parameters.

[0274] I. Basis for weight allocation of performance indicators

[0275] Weight setting principle: The weight reflects the importance of each indicator, which needs to be determined in combination with the application scenario and operation demand of the flow battery. This 10kW system is used for energy storage power station, and the core demand is "system stable operation (avoiding fault shutdown) → high stock utilization rate (improving energy storage capacity) → low energy consumption (reducing operation cost)", so the weight allocation is as follows:

[0276] System stability weight w_stability = 0.4 (highest, stability is the basis of energy storage system, and the loss of fault shutdown is much higher than the cost of energy consumption);

[0277] Stock utilization rate weight w_storage = 0.3 (second highest, high utilization rate can improve the charge and discharge capacity of energy storage system and increase economic benefits);

[0278] Energy efficiency weight w_energy = 0.3 (lowest, energy consumption cost accounts for a relatively low proportion, which can be optimized on the premise of meeting the stability and utilization rate);

[0279] The weights satisfy w_stability+w_storage+w_energy=1 (0.4+0.3+0.3=1), ensuring the rationality of score calculation.

[0280] Weight verification: Verify the rationality of the weights by expert scoring method, invite 5 flow battery system engineers to score the importance of the indicators (1-10 points), the statistical results: stability average score 9.2 points, utilization rate 8.5 points, energy consumption 7.8 points, after normalization, the weight is 9.2 / (9.2+8.5+7.8)=0.37≈0.4, 8.5 / 25.5≈0.33≈0.3, 7.8 / 25.5≈0.31≈0.3, consistent with the preset weight, verify the rationality of the weight.

[0281] II. Calculation model and example of weighted score

[0282] Score formula: Comprehensive score S=w_stability×η_stability+w_storage×η_storage+w_energy×η_energy, where η_stability, η_storage, η_energy are substituted in decimal form (such as 78.6%=0.786), the value range of S is [0,1], the closer to 1, the better the comprehensive performance of the strategy.

[0283] Score ranking and tie-breaker rule:

[0284] Rank all 30 candidate strategies in descending order of comprehensive score, the scores of the top 5 strategies are 93.58, 92.85, 91.72, 90.97, and 90.84, respectively.

[0285] If there are strategies with the same score (such as strategy D and strategy A with a score of 93.58), use the "stability first" tie-breaker rule to select the strategy with higher η_stability; if the stability is the same, select the strategy with higher η_storage, to ensure a unique optimal strategy.

[0286] III. Determination and output of the final dynamic adjustment strategy

[0287] Strategy parameter extraction: Select the strategy A with the highest comprehensive score as the final strategy, and extract its decision variable parameters:

[0288] Pump frequency adjustment value: The current pump operating frequency is 48Hz, the target frequency is 50Hz, the adjustment value is +2Hz (need to adjust once, because Δf=2Hz≤12.5Hz, meets the flow impact constraint);

[0289] Valve opening control parameters: current valve opening is 75%, target opening is 80%, control parameter is +5% (1-time adjustment, meets opening rate constraint).

[0290] Strategy feasibility verification:

[0291] Constraint verification: f=50Hz, θ=80% meets all constraints (20≤50≤60, 30≤80≤100, 50-5×80=-350≤-200, Q=16L / min≥18L / min, here Q=0.4×50×0.8=16L / min<18L / min, does not meet the inventory demand constraint, needs to fine-tune the strategy);

[0292] Strategy fine-tuning: find a strategy that meets Q≥18L / min and has the highest score in the external archive, find strategy D (f=55Hz, θ=82%), Q=0.4×55×0.82=18.04L / min≥18L / min, performance indicators: η_energy=0.76, η_storage=1, η_stability=0.98, comprehensive score S_D=0.4×0.98+0.3×1+0.3×0.76=0.92 (92 points, ranking 2nd), all constraints are met, determine strategy D as the final adjustment strategy;

[0293] Strategy output format: the final electrolyte circulation dynamic adjustment strategy is output in JSON format, including strategy ID, generation timestamp, pump frequency parameters (current value, target value, adjustment step), valve opening parameters (current value, target value, adjustment step), performance indicators, constraint satisfaction.

[0294] S204, execute the electrolyte circulation dynamic adjustment strategy to dynamically adjust the flow and pressure distribution in the electrolyte circulation loop, and realize the collaborative management of the electrolyte inventory of the flow battery.

[0295] Specifically, the electrolyte circulation dynamic adjustment strategy can be decomposed into specific control instructions, including the frequency set value of the circulating pump and the opening control value of each valve, forming a device control instruction set;

[0296] I. Logic rules and device mapping of strategy decomposition

[0297] Device mapping relationship: the electrolyte circulation loop includes "1 main circulating pump (responsible for electrolyte total delivery) + 3 regulating valves (V1: storage tank outlet valve, V2: battery stack inlet valve, V3: heat exchanger outlet valve)", "pump frequency" in the strategy corresponds to the main circulating pump, "valve opening" needs to be distributed to the three valves according to function, the distribution rules are as follows:

[0298] V1 (tank outlet valve): control the output flow of the tank to the loop, the opening degree is positively correlated with the pump frequency (the higher the pump frequency, the larger the V1 opening degree, to avoid the sudden drop of the tank pressure), the distribution ratio is 30% (when the strategy opening degree is 82%, the target opening degree of V1 = 82% x 30% ≈ 25%, because the initial opening degree of V1 is 20%, the adjustment range is +5%);

[0299] V2 (battery stack inlet valve): directly control the reaction flow entering the battery stack, which has the greatest influence on the battery performance, the distribution ratio is 50% (the target opening degree of V2 = 82% x 50% ≈ 41%, the initial opening degree is 35%, the adjustment range is +6%);

[0300] V3 (heat exchanger outlet valve): auxiliary regulation of loop pressure and flow distribution, balance the heat exchanger resistance, the distribution ratio is 20% (the target opening degree of V3 = 82% x 20% ≈ 16%, the initial opening degree is 15%, the adjustment range is +1%);

[0301] The sum of the distribution ratios is 100% (30% + 50% + 20%), which ensures the matching of total flow and pressure.

[0302] Decomposition logic: timing and safety priority:

[0303] Timing rule: adjust valve opening degree first, then adjust pump frequency (valve opens to target opening degree first to avoid sudden rise of loop pressure caused by pump frequency rise), time interval is set to 200ms (considering response speed and pressure stability);

[0304] Safety rule: the adjustment range of each instruction should be ≤ the maximum single adjustment amount of the equipment (the maximum single adjustment of pump frequency is 10Hz, the maximum single adjustment of valve opening degree is 10%), if the strategy adjustment range exceeds the limit, it is executed in multiple times (for example, if the pump frequency is adjusted from 48Hz to 55Hz, the range is 7Hz ≤ 10Hz, which can be executed once; if the range is 12Hz, it is executed in two times: 48→54→55Hz, interval 200ms).

[0305] II. Parameter composition and examples of control instructions

[0306] Circulating pump control instruction parameters: including "instruction ID, equipment type, equipment number, target frequency, initial frequency, adjustment range, execution timing, safety upper limit, safety lower limit, feedback requirement";

[0307] Adjusting valve control instruction parameters (take V2 as an example): including "instruction ID, equipment type, equipment number, target opening degree, initial opening degree, adjustment range, execution timing, action time, feedback requirement";

[0308] Feedback requirement: return the current opening degree every 100ms during execution, and return "action completed" identifier after execution is completed;

[0309] III. Format specification and storage of device control instruction set

[0310] Instruction set format: structured JSON format, including "instruction set ID, generation timestamp, device instruction list, execution priority, validity period", to ensure that the frequency converter and valve controller can be parsed by industrial software.

[0311] Storage and verification: instruction set is stored as a JSON file, and the parameter legality is verified by "instruction verification algorithm" (such as target frequency 55Hz within 20-60Hz range, adjustment amplitude 7Hz≤10Hz), if there are illegal parameters (such as target opening 110%), immediately marked as "invalid instruction" and suspended, to avoid device failure.

[0312] Through industrial bus protocol, control instruction set is delivered to circulating pump frequency converter and intelligent valve controller, to execute coordinated regulation of flow and pressure, and to obtain device execution state feedback;

[0313] I. Selection and parameter configuration of industrial bus protocol

[0314] Protocol selection basis: circulating pump frequency converter and intelligent valve controller both support Modbus-RTU protocol, which has the characteristics of "long transmission distance (≤1200m), strong anti-interference ability (differential signal), simple protocol easy to implement", suitable for the industrial environment of liquid flow battery system (existence of electromagnetic interference), so Modbus-RTU is selected as the communication protocol.

[0315] Protocol parameter configuration: baud rate: 9600bps (balance transmission speed and anti-interference ability, 9600bps within 100m transmission distance error rate ≤0.1%); data bits: 8 bits (standard Modbus data length); stop bit: 1 bit; check bit: even parity (reduce transmission errors); slave address: allocate unique slave address for each device (frequency converter slave address 1, V1 controller 2, V2 controller 3, V3 controller 4);

[0316] Register address: frequency converter: frequency setting register address 40001 (write target frequency), actual frequency feedback register address 40002 (read current frequency); valve controller: opening setting register address 40001 (write target opening), actual opening feedback register address 40002 (read current opening), status register address 40003 (read execution status: 0=standby, 1=executing, 2=completed, 3=fault).

[0317] II. Instruction delivery process and feedback mechanism

[0318] Instruction issuing sequence: issuing according to the sequence of "valve first, pump later", specific process:

[0319] Step 1: Issue valve opening degree instructions for V1, V2, and V3 (T time), write target opening degrees (25%, 41%, 16%) to register 40001 of slave addresses 2, 3, and 4;

[0320] Step 2: Wait for 200 ms (valve starts to act), issue circulating pump frequency instruction (T+200 ms), write target frequency 55 Hz to register 40001 of slave address 1;

[0321] Step 3: Poll the status register (40003) of all devices every 100 ms to obtain the execution status.

[0322] Feedback state analysis and processing:

[0323] Status "received" (register value 0): the device has received the instruction and has not started to act, continue to wait;

[0324] Status "in execution" (register value 1): the device is executing the action (such as valve motor rotation, pump frequency rise), continue to poll;

[0325] Status "completed" (register value 2): the device has completed the action, read the actual running parameters (such as pump actual frequency 55 Hz, V2 actual opening degree 41%), verify whether they are consistent with the target;

[0326] Status "fault" (register value 3): the device executes a fault (such as valve jam, pump overload), immediately issue a "stop instruction" (write the current value to the setting register, maintain the current situation), record the fault code (read from status register 40004, such as code 5 = motor overload) at the same time, trigger sound and light alarm.

[0327] Feedback example:

[0328] Issue V2 opening degree instruction (target 41%) at T time, status register 40003 of slave address 3 returns 1 (in execution);

[0329] Poll at T+100 ms, status register returns 1, actual opening degree feedback 40002 returns 37% (in action);

[0330] Poll at T+500 ms, status register returns 2 (completed), actual opening degree returns 41% (consistent with the target);

[0331] Issue pump frequency instruction (target 55 Hz) at T+200 ms, poll at T+300 ms, inverter actual frequency returns 50 Hz (rising), poll at T+700 ms, actual frequency returns 55 Hz (completed).

[0332] III. Timing of flow and pressure coordinated regulation

[0333] Coordinated regulation logic: valves are first adjusted to target opening to provide stable flow passage resistance for pump frequency increase, avoiding sudden pressure rise; after pump frequency increase, small opening correction (±1%) of valves is used to balance loop pressure, specific timing:

[0334] T+0ms: V1, V2, V3 start adjusting, target opening 25%, 41%, 16%;

[0335] T+500ms: all valves complete adjustment, loop flow passage resistance is stable;

[0336] T+500ms: pump starts to increase frequency, from 48Hz to 55Hz (rise rate 5Hz / s, 7Hz requires 1.4s);

[0337] T+1900ms: pump frequency reaches 55Hz, at this time the battery stack inlet pressure is monitored to rise to 0.27MPa (target 0.25MPa), V2 opening correction instruction is issued (41%→40%);

[0338] T+2400ms: V2 opening is adjusted to 40%, pressure drops to 0.25MPa, coordinated regulation is completed.

[0339] Pressure safety protection: during pump frequency increase, real-time monitoring of battery stack inlet pressure (collected by pressure sensor), if pressure exceeds 0.3MPa (safety threshold), immediately issue pump frequency reduction instruction (reduce 5Hz), at the same time issue valve opening increase instruction (V2+2%), until pressure drops to safety range, avoid pipe damage.

[0340] Real-time monitoring of adjusted electrolyte flow distribution and pressure distribution data, verify the consistency of actual regulation effect and expected target, generate regulation effect evaluation report, finally realize the coordinated management of liquid flow battery electrolyte inventory.

[0341] I. Selection and collection mechanism of monitoring parameters

[0342] Monitoring parameter selection: select "4 flow monitoring points (node 1: tank outlet, node 2: pump outlet, node 3: battery stack inlet, node 4: heat exchanger outlet) + 2 pressure monitoring points (battery stack inlet, outlet)", covering key links of electrolyte circulation, parameters as follows:

[0343] Flow parameters: Node 3 flow (reflecting the effective flow participating in the reaction, expected to be 18.04 L / min), Node 1-4 flow deviation (reflecting the loop leakage, expected to be ≤5%); Pressure parameters: stack inlet pressure (expected to be 0.25 MPa), inlet and outlet pressure difference (reflecting the flow passage resistance in the stack, expected to be 0.03 MPa).

[0344] Collection mechanism: Collection frequency: 10 Hz (consistent with the previous adjustment, ensuring real-time data); Data preprocessing: using sliding window filtering (window size 5 points) to eliminate noise, such as Node 3 flow raw data 18.1, 18.0, 17.9, 18.2, 18.0, filtered 18.04 L / min;

[0345] Data storage: stored in CSV format by minute, each data contains timestamp, parameter type, node, original value, filtered value.

[0346] II. Judgment criteria for adjustment effect verification

[0347] Consistency judgment index: flow deviation: | actual flow - expected flow | ≤2% (such as expected 18.04 L / min, actual 17.7~18.4 L / min are qualified); pressure deviation: | actual pressure - expected pressure | ≤5% (such as expected 0.25 MPa, actual 0.2375~0.2625 MPa are qualified); flow distribution uniformity: the maximum deviation of Node 1-4 flow ≤5% (such as Node 1=18.1 L / min, Node 4=17.9 L / min, deviation 1.1% ≤5%, qualified); stable time: after adjustment is completed, the time for parameters to stabilize in the target range ≤3s (to avoid long-term fluctuations).

[0348] III. Generation of adjustment effect evaluation report and closed-loop management

[0349] Report composition: the evaluation report contains "adjustment basic information, parameter comparison table, deviation analysis, effect judgment, optimization suggestion" five parts, for easy reference and archiving.

[0350] Closed-loop management mechanism: if the report determines "the effect is qualified", the adjustment data of this time (strategy parameters, execution process, monitoring results) will be stored in the battery management database for subsequent historical attenuation characteristic analysis and model optimization;

[0351] If the judgment is "the effect is unqualified" (such as the deviation of a certain parameter exceeds the threshold), the "secondary adjustment" process is triggered - the multi-objective optimization algorithm is called again, the strategy parameters are corrected based on the current monitoring data (such as increasing the valve opening when the pressure deviation is too large), a new control instruction set is generated and the steps are repeated until the adjustment effect is qualified.

[0352] Through the above adjustment and verification process, the flow and pressure of the electrolyte circulation loop reach the expected target, which is specifically manifested as: at the flow level, the flow at node 3 is stabilized at 18.02 L / min, which matches the short-term inventory demand (18 L / min), ensuring that the amount of electrolyte participating in the reaction per unit time meets the battery charging and discharging demand, and the inventory utilization rate is maintained at 100%; at the pressure level, the battery stack inlet pressure is stabilized at 0.248 MPa, and the inlet and outlet pressure difference is 0.032 MPa, which meets the safety operation requirement of the stack flow channel and has no risk of pressure surge or leakage; at the inventory level, the actual available inventory is 420 L, and under the current flow, the total amount of circulating electrolyte within 1 hour is 18.02 L / min x 60 min = 1081.2 L, which realizes full consumption of the inventory through circulation, and at the same time, the liquid level in the storage tank is maintained within the safe range (120 L > S_min = 100 L), avoiding pump cavitation;

[0353] Finally, through the complete process of "strategy decomposition - instruction issuance - monitoring and verification", the collaborative management of the liquid flow battery electrolyte inventory is realized, ensuring that the system maximizes the inventory utilization rate under the premise of low energy consumption (energy consumption efficiency 76%) and high stability (stability 98%), and meets the operation demand of the energy storage power station.

[0354] Another embodiment of the application provides a liquid flow battery electrolyte inventory collaborative management system, which is shown in Figure 3 The system can include:

[0355] The monitoring module 301 is configured to monitor the instantaneous flow data of the electrolyte and the state of charge data of the battery in the liquid flow battery system in real time, and generate a real-time data sequence of the electrolyte running state.

[0356] The determination module 302 is configured to determine the actual available inventory of the electrolyte and predict the short-term inventory demand change trend based on the real-time data sequence, in combination with the historical circulation attenuation characteristics of the electrolyte and the current working condition parameters, through a dynamic coupling analysis model.

[0357] The generation module 303 is configured to generate an electrolyte circulation dynamic adjustment strategy by using a multi-objective optimization algorithm according to the actual available inventory and the short-term inventory demand change trend, wherein the strategy at least includes frequency adjustment of the electrolyte circulation pump and valve opening degree collaborative control.

[0358] The adjustment module 304 is configured to execute the electrolyte circulation dynamic adjustment strategy, dynamically adjust the flow and pressure distribution in the electrolyte circulation loop, and realize the collaborative management of the liquid flow battery electrolyte inventory.

[0359] The above detailed description of the structure, features and effects of the present application is based on the embodiments shown in the drawings. The above description is only the preferred embodiments of the present application, but the present application is not limited to the embodiments shown in the drawings. Any changes or modifications made in accordance with the concept of the present application, or equivalent embodiments with equivalent changes, are still within the scope of the present application.

Claims

1. A method for coordinated management of liquid flow battery electrolyte inventory, the method comprising: The method comprises: ​ Real-time monitoring of the instantaneous flow data of the electrolyte in the flow battery system and the state of charge data of the battery generates a real-time data sequence of the electrolyte operating state; Based on the real-time data sequence, combined with the historical cycle attenuation characteristics of the electrolyte and the current operating condition parameters, the actual available inventory of the electrolyte is determined through a dynamic coupling analysis model and the short-term inventory demand trend is predicted; wherein the historical cycle attenuation characteristic data of the electrolyte is extracted from the battery management database and arranged into a historical attenuation characteristic vector; the real-time data sequence of the electrolyte operating state is feature-fused with the historical attenuation characteristic vector, combined with the operating condition parameters including the current temperature and load, to form a multi-dimensional feature input set; the multi-dimensional feature input set is input into the dynamic coupling analysis model, which adopts a long short-term memory network architecture to calculate the actual available inventory of the electrolyte by analyzing the nonlinear relationship between the electrolyte inventory and the operating parameters; based on the output of the dynamic coupling analysis model, a time series prediction algorithm is used to predict the inventory demand trend in the future preset period, and an inventory demand prediction curve is output; According to the actual available inventory and the short-term inventory demand trend, a multi-objective optimization algorithm is used to generate an electrolyte circulation dynamic adjustment strategy, wherein the strategy at least includes frequency adjustment of the electrolyte circulation pump and valve opening degree collaborative control; The electrolyte circulation dynamic adjustment strategy is executed to dynamically adjust the flow and pressure distribution in the electrolyte circulation loop, realizing collaborative management of the flow battery electrolyte inventory.

2. The method of claim 1, wherein, The real-time monitoring of the instantaneous flow data of the electrolyte in the flow battery system and the state of charge data of the battery generates a real-time data sequence of the electrolyte operating state, comprising: Deploying flow sensors at key nodes of the electrolyte circulation loop and installing pressure sensors at the inlet and outlet of the positive and negative electrodes of the battery stack, while collecting the state of charge data of the battery management system, to obtain multi-source original monitoring data; Performing time series alignment and dimension unification processing on the multi-source original monitoring data, using a sliding window filtering algorithm to eliminate measurement noise, to form standardized monitoring data after preprocessing; Based on the standardized monitoring data, the instantaneous change rate of the electrolyte flow and the change trend of the state of charge are calculated, and a dynamic characteristic index data set is established; Integrating the dynamic characteristic index data set with the timestamp information to construct a real-time data sequence of the electrolyte operating state including flow, pressure, state of charge and their change rates.

3. The method of claim 2, wherein, The multi-objective optimization algorithm is used to generate an electrolyte circulation dynamic adjustment strategy according to the actual available inventory and the short-term inventory demand trend, wherein the strategy at least includes frequency adjustment of the electrolyte circulation pump and valve opening degree collaborative control, comprising: Establishing a multi-objective optimization function with the minimum energy consumption, the maximum inventory utilization rate and the optimal system stability as the target, wherein the frequency of the electrolyte circulation pump and the valve opening degree are defined as decision variables; Based on the actual available inventory and the inventory demand prediction curve, setting optimization constraint conditions, wherein the conditions include inventory safety threshold and equipment operating limit; The multi-objective particle swarm optimization algorithm is used to solve a multi-objective optimization function, and an optimal pump frequency and valve opening combination is found under the condition of meeting the optimization constraint condition, energy consumption efficiency, inventory utilization rate and system stability indexes of each candidate strategy are calculated, and a candidate optimization strategy set with performance evaluation indexes is generated; Based on the performance evaluation indexes of each candidate optimization strategy, a weighted scoring method is used to calculate the comprehensive score, and the strategy with the highest comprehensive score is selected as the electrolyte circulation dynamic adjustment strategy containing the pump frequency adjustment value and the valve opening control parameter.

4. The method of claim 3, wherein, The electrolyte circulation dynamic adjustment strategy is executed to dynamically adjust the flow and pressure distribution in the electrolyte circulation loop, and the cooperative management of the electrolyte inventory of the flow battery is realized, including: The electrolyte circulation dynamic adjustment strategy is decomposed into specific control instructions, including the frequency set value of the circulating pump and the opening control value of each valve, and a device control instruction set is formed; The control instruction set is issued to the circulating pump frequency converter and the intelligent valve controller through the industrial bus protocol, the cooperative adjustment of the flow and the pressure is executed, and the device execution state feedback is obtained; The real-time monitoring of the adjusted electrolyte flow distribution and pressure distribution data verifies the consistency of the actual adjustment effect and the expected target, generates an adjustment effect evaluation report, and finally realizes the cooperative management of the electrolyte inventory of the flow battery.

5. A flow battery electrolyte inventory cooperative management system, characterized by, The system comprises: A monitoring module for real-time monitoring of the instantaneous flow data of the electrolyte and the state of charge data of the battery in the flow battery system, and generating a real-time data sequence of the electrolyte running state; A determination module for determining the actual available inventory of the electrolyte and predicting the short-term inventory demand trend based on the real-time data sequence, combining the historical circulation attenuation characteristics of the electrolyte and the current working condition parameters, through a dynamic coupling analysis model; wherein the historical circulation attenuation characteristic data of the electrolyte is extracted from the battery management database and arranged into a historical attenuation characteristic vector; the real-time data sequence of the electrolyte running state is feature fused with the historical attenuation characteristic vector, combined with the working condition parameters including the current temperature and load, to form a multi-dimensional feature input set; the multi-dimensional feature input set is input into the dynamic coupling analysis model, which adopts a long short-term memory network architecture, calculates the actual available inventory of the electrolyte by analyzing the nonlinear relationship between the electrolyte inventory and the operating parameters; based on the output of the dynamic coupling analysis model, a time series prediction algorithm is used to predict the inventory demand trend in a future preset period, and an inventory demand prediction curve is output; A generation module for generating an electrolyte circulation dynamic adjustment strategy using a multi-objective optimization algorithm according to the actual available inventory and the short-term inventory demand trend, wherein the strategy at least includes the frequency adjustment of the electrolyte circulation pump and the valve opening cooperative control; An adjustment module for executing the electrolyte circulation dynamic adjustment strategy to dynamically adjust the flow and pressure distribution in the electrolyte circulation loop, and realizing the cooperative management of the electrolyte inventory of the flow battery.

6. The system of claim 5, wherein, The monitoring module is specifically configured to: Deploy flow sensors at key nodes of the electrolyte circulation loop, install pressure sensors at the inlet and outlet of the positive and negative electrodes of the battery stack, and simultaneously collect the state of charge data of the battery management system to obtain multi-source original monitoring data; The multi-source original monitoring data are time-series aligned and dimensionally unified, measurement noise is eliminated by using a sliding window filtering algorithm, and standardized monitoring data after preprocessing are formed; Based on the standardized monitoring data, the instantaneous change rate of the electrolyte flow and the change trend of the state of charge are calculated, and a dynamic characteristic index data set is established; The dynamic characteristic index data set and the time stamp information are integrated to construct an electrolyte running state real-time data sequence containing flow, pressure, state of charge and change rate thereof.

7. A storage medium, characterized by The storage medium has a computer program stored therein, wherein the computer program is configured to execute the method of any one of claims 1-4 when running.

8. An electronic device comprising a memory and a processor, characterized in that, The memory has a computer program stored therein, and the processor is configured to execute the computer program to execute the method of any one of claims 1-4.

Images