A multi-dimensional information monitoring system inside a vanadium redox flow battery stack

Abstract

The application discloses a kind of full vanadium flow battery stack internal multidimensional information monitoring system, including sensor array, data acquisition and processing unit, working condition identification unit, multidimensional information correlation diagnosis unit and monitoring output unit.Sensor array synchronously acquires voltage distribution signal, temperature distribution signal, pressure difference signal and flow signal;Multidimensional information correlation diagnosis unit is built in the electrical and temperature, fluid and electrical and temperature and fluid correlation constraint model based on physical mechanism, according to operating phase, each correlation constraint model is calculated respectively Normalized coupling residual, and when the normalized coupling residual exceeds corresponding dynamic threshold, correlation abnormal signal is generated, and the fault type and fault location are matched and output through residual combination mode base.The application realizes cross-dimension coupling diagnosis, can effectively identify early anomaly and reduce false alarm.

Description

Technical Field

[0001] This invention relates to the field of flow battery energy storage technology, specifically to a multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack. Background Technology

[0002] Vanadium redox flow batteries (VRBs), as a long-term energy storage technology decoupling power and capacity, have significant application value in large-scale renewable energy grid integration and grid peak shaving and frequency regulation. The fuel cell stack is the core energy conversion component of a VRB, and its internal operating state directly determines the system's efficiency, lifespan, and safety. The stack operation involves strong coupling effects of multiple physical fields, including electrochemical reactions, mass and heat transfer, and fluid distribution; any abnormality in any of these processes can lead to performance degradation or even safety accidents.

[0003] Several fuel cell stack monitoring schemes have been proposed, but they generally suffer from the following shortcomings: First, the monitoring dimensions are limited. Existing solutions mostly collect and analyze only one type of parameter, such as temperature, voltage, or flow rate, without incorporating multi-dimensional information such as electrical, thermal, and fluid parameters into a unified monitoring framework, making it difficult to comprehensively reflect the internal operating status of the fuel cell stack.

[0004] Second, anomaly detection relies on single-parameter thresholds. Existing solutions typically set independent alarm thresholds for each parameter, triggering an alarm when any parameter exceeds its limit. However, inherent coupling relationships exist between various physical quantities during fuel cell stack operation. A single parameter exceeding its limit could originate from a genuine fault, or be caused by changes in operating conditions, sensor noise, or transient disturbances, leading to a high false alarm rate. More importantly, some early faults may not yet show significant deviations in a single parameter, but the normal coupling relationships between multiple parameters have already changed, and the single-threshold method cannot identify such coupling anomalies. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack, which addresses the above-mentioned technical deficiencies and includes: A sensor array is used to synchronously acquire voltage distribution signals, temperature distribution signals, differential pressure signals, and flow signals inside the fuel cell stack; The data acquisition and processing unit is connected to the sensor array and is used to perform time synchronization processing and feature extraction on the signals of each sensor, and output a multi-dimensional time-series feature vector. The operating condition identification unit is used to obtain information on the current operating stage of the fuel cell stack. The operating stage includes at least constant current charging, constant voltage charging, constant current discharging, constant voltage discharging, and shutdown. The multidimensional information correlation diagnosis unit has built-in electrical and temperature correlation constraint models, fluid and electrical correlation constraint models, and temperature and fluid correlation constraint models pre-established based on the physical mechanism of the fuel cell stack. The multidimensional information correlation diagnosis unit is used to calculate the normalized coupling residual for each correlation constraint model according to the information of the operation stage, and generate a correlation anomaly signal when the normalized coupling residual exceeds the corresponding dynamic threshold. The monitoring output unit is used to output the associated abnormal signal and, when the fault mode matching condition is met, output the fault type and fault location.

[0006] To further optimize this technical solution, the sensor array includes a distributed voltage sensor group arranged along the electrolyte flow direction of the fuel cell stack, a temperature sensor group arranged along the direction from the fuel cell stack inlet to the outlet, a pressure sensor group arranged at the fuel cell stack inlet and outlet, and a flow sensor arranged in the main inlet pipeline.

[0007] To further optimize this technical solution, the data acquisition and processing unit includes a time synchronization module. The time synchronization module uses an internal clock as a reference to stamp a unified timestamp on each sensor signal and performs forward compensation on the timestamp based on the response delay parameters of each sensor.

[0008] To further optimize this technical solution, the data acquisition and processing unit uses a sliding time window to segment the time-synchronized signal, and extracts the voltage distribution range, temperature distribution non-uniformity, flow fluctuation coefficient and pressure difference change slope from each sliding time window to form a multi-dimensional time series feature vector.

[0009] To further optimize this technical solution, the voltage distribution range is the difference between the maximum and minimum values ​​collected by each voltage sensor within the sliding time window; the temperature distribution non-uniformity is the standard deviation of the values ​​collected by each temperature sensor within the sliding time window; the flow fluctuation coefficient is the ratio of the variance of the flow signal to the square of the mean of the flow signal within the sliding time window; and the pressure difference change slope is the slope of the straight line obtained by linearly fitting the pressure difference sequence between the inlet and outlet within the sliding time window.

[0010] To further optimize this technical solution, the electrical and temperature correlation constraint model is constructed based on the negative correlation mechanism that the equivalent internal resistance of the fuel cell stack decreases with increasing temperature. Its actual feature vector consists of the voltage distribution range and the average temperature. Its correlation constraint feature vector characterizes the correspondence between the voltage distribution range and the average temperature under healthy conditions. The average temperature is the average value of the temperature sensor data collected within the sliding time window. The fluid-electrical correlation constraint model is constructed based on the mechanism that the decrease in electrolyte flow leads to the increase in voltage fluctuation. Its actual feature vector consists of the flow fluctuation coefficient and the voltage distribution range. Its correlation constraint feature vector characterizes the correspondence between the flow fluctuation coefficient and the voltage distribution range under healthy conditions. The temperature and fluid correlation constraint model is constructed based on the mechanism that the increase in internal flow channel resistance of the fuel cell stack causes an increase in pressure difference and temperature distribution non-uniformity. Its actual feature vector is composed of the slope of pressure difference change and temperature distribution non-uniformity. Its correlation constraint feature vector characterizes the correspondence between the slope of pressure difference change and temperature distribution non-uniformity under healthy conditions. The correlation constraint feature vectors of each model have the same dimension as the corresponding actual feature vectors, and are obtained through experimental calibration and / or fitting of historical running data.

[0011] To further optimize this technical solution, the multidimensional information association diagnostic unit also outputs a corresponding binary residual abnormality flag when generating the associated abnormal signal; the flag is 1 if it exceeds the dynamic threshold and 0 if it does not. The multidimensional information association diagnostic unit also includes a threshold adaptive update module, which records the normalized coupling residual values ​​of each model for multiple charge-discharge cycles when the stack is in a healthy state, calculates the residual mean μ0 and standard deviation σ0, and uses μ0+kσ0 as the initial dynamic threshold Th0, where k is a preset multiple; during operation, it statistically analyzes the recent residual sequence at a preset period, calculates the current residual mean μ, and the residual drift Δ=μ-μ0; when |Δ| exceeds the preset threshold Δth, it adjusts the dynamic threshold according to the degradation correction coefficient α, and the adjusted dynamic threshold Th=Th0×(1+α·|Δ| / Δth), where the dimension of Δth is consistent with Δ and is obtained by experimental calibration, and the degradation correction coefficient α is determined according to a preset grading table or piecewise function; independent initial dynamic thresholds and degradation correction coefficients are maintained for different operating stages.

[0012] To further optimize this technical solution, the multidimensional information association diagnosis unit also includes a residual combination pattern library, which stores binary coded fault pattern vectors corresponding to various typical faults constructed by physical mechanisms and fault sample statistics. After generating the associated abnormal signal, the multidimensional information association diagnosis unit combines the residual abnormal flags of each model into a current residual pattern vector. The current residual pattern vector and the fault pattern vector are encoded in the same dimension. The matching degree is calculated with each fault pattern vector in the residual combination pattern library, and the fault type and fault location are output according to the matching degree.

[0013] To further optimize this technical solution, the matching degree is calculated using weighted Hamming distance; when there is a fault mode vector with a matching degree exceeding a preset threshold, the fault type and fault location with the highest matching degree are output; when all matching degrees are below the preset threshold, only the associated abnormal signal is output.

[0014] To further optimize this technical solution, the sensor array also includes a reference electrode, which is arranged on the electrode frames on the positive and negative sides of the fuel cell stack through pre-set mounting holes, for collecting the potential of the positive half-cell and the potential of the negative half-cell. The multi-dimensional information association diagnostic unit also calculates the correlation coefficient between the half-cell potential change trend and the voltage distribution signal change trend, and compares the relationship between the half-cell potential offset and the voltage distribution range. When the half-cell potential on a certain side shows a continuous offset and is related to the voltage distribution change in the corresponding area, the abnormal half-cell side is determined and a half-cell-level early warning signal is output.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. Multidimensional coupling diagnosis effectively identifies early anomalies. The multidimensional information association diagnosis unit incorporates three physical mechanism-based correlation constraint models, using normalized coupling residuals for cross-dimensional diagnosis. This effectively identifies early anomalies where a single parameter has not yet exceeded its limit, but the multidimensional coupling relationship has mismatched, compensating for the blind spots of traditional single threshold methods. 2. Adaptive operating condition and dynamic threshold adjustment significantly reduce false alarms. The operating condition identification unit calls the corresponding correlation constraint feature vector and dynamic threshold for the operating stage, and the threshold adaptive update module corrects the dynamic threshold based on long-term health drift, significantly reducing the false alarm rate caused by operating condition switching or normal equipment aging. 3. Residual pattern matching enables joint and accurate determination of fault type and location. The residual combination pattern library matches the current residual pattern vector of each model's residual anomaly markers. Combined with physical mechanisms and fault sample statistics, it outputs clear fault types and locations, improving the targeting and accuracy of diagnosis. 4. Non-intrusive main body design is suitable for online upgrades of already operational fuel cell stacks. Voltage, temperature, pressure, and flow sensors can be installed through existing fuel cell stack interfaces or non-invasive methods without modifying the fuel cell stack's sealing structure; the reference electrode is installed through pre-set mounting holes, making the overall solution easy to deploy on already operational fuel cell stacks; 5. It has half-cell-level monitoring capabilities. By introducing a reference electrode into the sensor array and combining it with the correlation analysis of half-cell potential and voltage distribution signals, it can distinguish early anomalies on the positive and negative electrode sides, achieving finer-grained fault location. Attached Figure Description

[0016] Figure 1 This is a structural block diagram of the multi-dimensional information monitoring system inside the vanadium redox flow battery stack provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of the workflow of the multidimensional information association diagnostic unit provided in an embodiment of the present invention.

[0018] Figure 3 This is a schematic diagram of the fault determination process based on the residual combination pattern library provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] Combination Figures 1 to 3 As shown, this embodiment provides a multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack. It should be noted that, in this invention, "multi-dimensional information within the stack" refers to multi-dimensional information about the stack's internal operating status. Voltage, temperature, pressure, and flow sensors can be installed via the stack's pre-installed interface or externally, without needing to penetrate the stack's sealed flow channels, thus enabling online monitoring of the stack's internal state.

[0021] The sensor array includes a distributed voltage sensor group, a temperature sensor group, a pressure sensor group, and a flow sensor. The distributed voltage sensor group includes multiple voltage sensors arranged at different locations on the fuel cell stack along the electrolyte flow direction. The temperature sensor group includes multiple temperature sensors arranged on the electrode frames at the inlet, middle, and outlet sides of the fuel cell stack. The pressure sensor group includes an inlet pressure sensor and an outlet pressure sensor, arranged at the fuel cell stack inlet and outlet, respectively. The flow sensor is located in the main inlet pipeline. The sensor array may also include a reference electrode, preferably a dynamic hydrogen electrode, arranged on the electrode frames on the positive and negative sides of the fuel cell stack through pre-set mounting holes. The mounting holes are equipped with sealing gaskets and insulating sleeves to ensure isolation from the electrolyte environment and to avoid disrupting the flow field structure. This reference electrode is used to collect the positive and negative half-cell potentials.

[0022] The data acquisition and processing unit amplifies, levels, and digitally filters the raw signals from each sensor. This unit includes a time synchronization module that timestamps all sensor signals using an internal clock and performs forward compensation based on the response delay parameters of each sensor. The unit uses a sliding time window approach to process the time-synchronized signals in segments; the window length and sliding step size can be set according to actual operational requirements. For each time window, the following characteristic parameters are extracted: voltage distribution range (the difference between the maximum and minimum values ​​collected by each voltage sensor within the window); temperature distribution non-uniformity (the standard deviation of the values ​​collected by each temperature sensor within the window); flow rate fluctuation coefficient (the ratio of the variance of the flow rate signal to the square of the mean flow rate signal within the window); and pressure difference slope (the slope of the straight line obtained by linearly fitting the pressure difference sequence between the inlet and outlet within the window). These characteristic parameters constitute a multi-dimensional time-series feature vector, which is then sent to the multi-dimensional information association and diagnostic unit.

[0023] The operating condition identification unit obtains real-time charging and discharging status signals and operating mode signals from the battery management system, or automatically identifies the operating stage based on the collected current and voltage signals, and generates an operating stage identifier, at least distinguishing between constant current charging, constant voltage charging, constant current discharging, constant voltage discharging and shutdown states.

[0024] The multidimensional information correlation diagnosis unit has built-in electrical and temperature correlation constraint models, fluid and electrical correlation constraint models, temperature and fluid correlation constraint models, threshold adaptive update module, and residual combination mode library.

[0025] The workflow of the multidimensional information association diagnostic unit is as follows: it receives the operating stage identifier sent by the operating condition identification unit, and retrieves the association constraint feature vectors and their corresponding initial dynamic thresholds for each model under the corresponding operating stage from the parameter configuration table. After receiving the multidimensional time-series feature vector, the voltage distribution range and average temperature are used to form the first actual feature vector, which is then input into the electrical-temperature correlation constraint model. Using the correlation constraint feature vector (i.e., the first correlation constraint feature vector) corresponding to this model retrieved from the parameter configuration table, the normalized coupling residual between the first actual feature vector and the first correlation constraint feature vector is calculated as the electrical-temperature coupling residual. The flow fluctuation coefficient and voltage distribution range are used to form the second actual feature vector, which is then input into the fluid-electrical correlation constraint model. Using the correlation constraint feature vector (i.e., the second correlation constraint feature vector) corresponding to this model retrieved from the parameter configuration table, the normalized coupling residual is calculated as the fluid-electrical coupling residual. The pressure difference slope and temperature distribution non-uniformity are used to form the third actual feature vector, which is then input into the temperature-fluid correlation constraint model. Using the correlation constraint feature vector (i.e., the third correlation constraint feature vector) corresponding to this model retrieved from the parameter configuration table, the normalized coupling residual is calculated as the temperature-fluid coupling residual. The normalized coupling residuals of the three correlation constraint models are calculated using the following formula: R = ||W·(X_actual - X_constraint)|| / ||X_constraint||, where W is the weight matrix, a diagonal matrix to balance the differences in the dimensions of different features, and the weight coefficients are obtained through experimental calibration, with the norm set to L2. To avoid division by zero, when ||X_constraint|| is less than a preset minimum value, R = ||W·(X_actual - X_constraint)|| is used as a substitute. X_actual and X_constraint of each model are substituted into their respective actual feature vectors and correlation constraint feature vectors. When any normalized coupling residual exceeds the dynamic threshold, a binary residual anomaly flag is output (1 for exceeding, 0 for not exceeding). The residual anomaly flags of each model are summarized; if at least one model outputs an anomaly flag, a correlation anomaly signal is generated.

[0026] The threshold adaptive update module operates as follows: The stack health state refers to the operating state where all performance indicators are within the normal range during the initial commissioning phase or after maintenance confirmation. This health state corresponds to the operating interval within a few initial operating cycles during which no anomaly judgment is triggered. It can be manually calibrated or automatically determined by the system based on initial operating parameters. During the stack health state, the normalized coupled residual values ​​of each model for multiple charge-discharge cycles are recorded, and the residual mean μ0 and standard deviation σ0 are calculated. μ0 + kσ0 is used as the initial dynamic threshold Th0, where k is a preset multiple. During operation, the recent residual sequence is statistically analyzed at preset intervals, and the current residual mean μ is calculated. The residual drift Δ = μ - μ0.

[0027] When |Δ| exceeds the preset threshold Δth, the dynamic threshold is adjusted according to the degradation correction coefficient α. The adjusted dynamic threshold Th = Th0 × (1 + α·|Δ| / Δth), where the dimension of Δth is consistent with Δ and is obtained by experimental calibration. The degradation correction coefficient α is determined according to a preset grading table or piecewise function. Independent initial dynamic thresholds and degradation correction coefficients are maintained for different operating stages.

[0028] When associated anomaly signals exist, the binary residual anomaly flags of each model are combined into a current residual pattern vector composed of the residual flags of the three models, and input into the residual combined pattern library for matching. The residual combined pattern library stores various binary coded fault pattern vectors corresponding to typical faults constructed from physical mechanisms and fault sample statistics. Each coded bit of the fault pattern vector corresponds to the residual anomaly state of the three associated constraint models, and has the same dimension as the current residual pattern vector. The matching degree is calculated using weighted Hamming distance. When a fault pattern vector has a matching degree exceeding a preset confidence threshold, the fault type and fault location with the highest matching degree are output; if all matching degrees are below the confidence threshold, only the associated anomaly signal is output. The fault type and fault location are determined by the residual pattern matching results.

[0029] As an example, typical failure modes include: Mode 1, where both the fluid and electrical models and the temperature and fluid models are abnormal, and the temperature distribution increases along the flow direction, indicating flow channel blockage, with the fault location being the inlet side region; Mode 2, where only the electrical and temperature models are abnormal, and the voltage in a certain region remains consistently low, with the reference electrode showing a potential shift in the half-cell potential on that side, indicating single-cell performance degradation, with the fault location being that region; Mode 3, where only the electrical and temperature models are abnormal, and the temperature distribution is not significantly abnormal but the internal resistance continues to rise, indicating membrane fouling or abnormal contact resistance. The above typical failure modes are merely examples, and the residual combination mode library can be expanded to include other failure types.

[0030] The multi-dimensional information correlation diagnostic unit also receives the half-cell potential signal acquired by the reference electrode, calculates the correlation coefficient between the half-cell potential change trend and the voltage distribution signal change trend, and compares the relationship between the half-cell potential offset and the voltage distribution range. When a half-cell potential on a certain side shows a continuous offset and is related to the voltage distribution change in the corresponding area, it outputs a half-cell-level early warning signal to indicate the side of the abnormality. The system also has a periodic self-calibration process to correct the reference electrode drift by comparing it with historical half-cell potential benchmarks.

[0031] The monitoring output unit displays the trend curves of multi-dimensional time-series feature vectors and the residual status of each model in real time. Anomalies are categorized based on the magnitude of each normalized coupled residual exceeding the dynamic threshold and the number of models exhibiting anomalies: mild anomaly occurs when only one model is abnormal and the deviation is small; moderate anomaly occurs when two models are abnormal or a single model has a large deviation; and severe anomaly occurs when all three models are abnormal or have significant deviations. Different alerts and alarms are used for different levels. The monitoring output unit also sends monitoring data, associated anomaly signals, and fault determination results to the remote monitoring center.

[0032] Simulation results show that this solution can identify early faults such as blockage of the inlet-side flow channel, degradation of the middle single cell, and slight membrane contamination in advance, and accurately locate the fault location, effectively reducing false alarms during the switching of operating conditions.

[0033] Obviously, the above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art can make various modifications or substitutions without departing from the concept of the present invention, and all such modifications or substitutions should fall within the protection scope of the present invention.

Claims

1. A multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack, characterized in that, include: A sensor array is used to synchronously acquire voltage distribution signals, temperature distribution signals, differential pressure signals, and flow signals inside the fuel cell stack; The data acquisition and processing unit is connected to the sensor array and is used to perform time synchronization processing and feature extraction on the signals of each sensor, and output a multi-dimensional time-series feature vector. The operating condition identification unit is used to obtain information on the current operating stage of the fuel cell stack. The operating stage includes at least constant current charging, constant voltage charging, constant current discharging, constant voltage discharging, and shutdown. The multidimensional information correlation diagnosis unit has built-in electrical and temperature correlation constraint models, fluid and electrical correlation constraint models, and temperature and fluid correlation constraint models pre-established based on the physical mechanism of the fuel cell stack. The multidimensional information correlation diagnosis unit is used to calculate the normalized coupling residual for each correlation constraint model according to the information of the operation stage, and generate a correlation anomaly signal when the normalized coupling residual exceeds the corresponding dynamic threshold. The monitoring output unit is used to output the associated abnormal signal and, when the fault mode matching condition is met, output the fault type and fault location.

2. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 1, characterized in that, The sensor array includes a distributed voltage sensor group arranged along the electrolyte flow direction of the fuel cell stack, a temperature sensor group arranged along the direction from the fuel cell stack inlet to the outlet, a pressure sensor group arranged at the fuel cell stack inlet and outlet, and a flow sensor arranged in the main inlet pipeline.

3. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 1, characterized in that, The data acquisition and processing unit includes a time synchronization module, which uses an internal clock as a reference to stamp a unified timestamp on each sensor signal and performs forward compensation on the timestamp based on the response delay parameters of each sensor.

4. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 1, characterized in that, The data acquisition and processing unit uses a sliding time window to segment the time-synchronized signal, and extracts the voltage distribution range, temperature distribution non-uniformity, flow fluctuation coefficient and pressure difference change slope from each sliding time window to form a multi-dimensional time series feature vector.

5. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 4, characterized in that, The voltage distribution range is the difference between the maximum and minimum values ​​collected by each voltage sensor within the sliding time window; the temperature distribution non-uniformity is the standard deviation of the values ​​collected by each temperature sensor within the sliding time window; the flow fluctuation coefficient is the ratio of the variance of the flow signal to the square of the mean of the flow signal within the sliding time window; and the pressure difference change slope is the slope of the straight line obtained by linearly fitting the pressure difference sequence between the inlet and outlet within the sliding time window.

6. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 1, characterized in that: The electrical and temperature correlation constraint model is constructed based on the negative correlation mechanism that the equivalent internal resistance of the fuel cell stack decreases with increasing temperature. Its actual feature vector consists of the voltage distribution range and the average temperature. Its correlation constraint feature vector characterizes the correspondence between the voltage distribution range and the average temperature under healthy conditions. The average temperature is the average value of the temperature sensor data collected within the sliding time window. The fluid-electrical correlation constraint model is constructed based on the mechanism that the decrease in electrolyte flow leads to the increase in voltage fluctuation. Its actual feature vector consists of the flow fluctuation coefficient and the voltage distribution range. Its correlation constraint feature vector characterizes the correspondence between the flow fluctuation coefficient and the voltage distribution range under healthy conditions. The temperature and fluid correlation constraint model is constructed based on the mechanism that the increase in internal flow channel resistance of the fuel cell stack causes an increase in pressure difference and temperature distribution non-uniformity. Its actual feature vector is composed of the slope of pressure difference change and temperature distribution non-uniformity. Its correlation constraint feature vector characterizes the correspondence between the slope of pressure difference change and temperature distribution non-uniformity under healthy conditions. The associated constraint feature vectors of each model have the same dimension as the corresponding actual feature vectors, and are obtained through experimental calibration and / or fitting of historical running data.

7. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 1, characterized in that, When generating the associated abnormal signal, the multidimensional information association diagnostic unit also outputs the corresponding binary residual abnormal flag, which is 1 if it exceeds the dynamic threshold and 0 if it does not exceed it. The multidimensional information association diagnostic unit also includes a threshold adaptive update module, which records the normalized coupling residual values ​​of each model for multiple charge-discharge cycles when the stack is in a healthy state, calculates the residual mean μ0 and standard deviation σ0, and uses μ0+kσ0 as the initial dynamic threshold Th0, where k is a preset multiple; during operation, it statistically analyzes the recent residual sequence at a preset period, calculates the current residual mean μ, and the residual drift Δ=μ-μ0; when |Δ| exceeds the preset threshold Δth, it adjusts the dynamic threshold according to the degradation correction coefficient α, and the adjusted dynamic threshold Th=Th0×(1+α·|Δ| / Δth), where the dimension of Δth is consistent with Δ and is obtained by experimental calibration, and the degradation correction coefficient α is determined according to a preset grading table or piecewise function; independent initial dynamic thresholds and degradation correction coefficients are maintained for different operating stages.

8. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 1, characterized in that, The multidimensional information association diagnostic unit also includes a residual combination pattern library, which stores binary coded fault pattern vectors corresponding to various typical faults constructed by physical mechanisms and fault sample statistics. After generating the associated abnormal signal, the multidimensional information association diagnostic unit combines the residual abnormal flags of each model into a current residual pattern vector. The current residual pattern vector and the fault pattern vector are encoded in the same dimension. The matching degree is calculated with each fault pattern vector in the residual combination pattern library, and the fault type and fault location are output according to the matching degree.

9. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 8, characterized in that, The matching degree is calculated using weighted Hamming distance. When there is a fault mode vector with a matching degree exceeding a preset threshold, the fault type and fault location with the highest matching degree are output. When all matching degrees are below the preset threshold, only the associated abnormal signal is output.

10. The multi-dimensional information monitoring system for the internal structure of a vanadium redox flow battery stack according to claim 1, characterized in that, The sensor array also includes a reference electrode, which is arranged on the electrode frame on the positive and negative sides of the fuel cell stack through pre-set mounting holes, for collecting the positive half-cell potential and the negative half-cell potential; the multi-dimensional information association diagnostic unit also calculates the correlation coefficient between the half-cell potential change trend and the voltage distribution signal change trend, and compares the relationship between the half-cell potential offset and the voltage distribution range. When the half-cell potential on a certain side shows a continuous offset and is related to the voltage distribution change in the corresponding area, the abnormal half-cell side is determined and a half-cell-level early warning signal is output.

Images