An AI-based method and system for SOFC condition monitoring and health diagnosis

Abstract

This invention provides an artificial intelligence-based method and system for SOFC state monitoring and health diagnosis, relating to the field of solid oxide battery state monitoring technology. The method includes: real-time acquisition of SOFC operating parameters; weighted processing of the acquired operating parameters using an algorithm based on extended Kalman filtering and information entropy optimization; real-time monitoring of battery state, prediction of expected lifespan, and identification of anomalies based on an improved hybrid neural network model and reinforcement learning algorithm, and dynamic adjustment of model parameters through sliding window online learning and self-correction mechanisms; and generation of feedback control signals by combining monitoring results, health predictions, and anomaly warning signals to dynamically adjust the SOFC operating parameters. This invention features high precision, strong robustness, and adaptability, making it suitable for intelligent management of SOFCs in energy storage and high-efficiency power generation.

Description

Technical Field

[0001] This invention relates to the field of solid oxide battery state monitoring technology, specifically to an artificial intelligence-based method and system for SOFC state monitoring and health diagnosis. Background Technology

[0002] Solid oxide fuel cells (SOFCs) have significant application value in energy storage and clean power generation due to their high energy conversion efficiency, environmental friendliness, and long lifespan. Although current technologies have made considerable progress in SOFC monitoring, the following problems still exist in the field of SOFC condition monitoring and health diagnosis:

[0003] First, the limitations of data acquisition. Existing technologies are mostly limited to single or a few parameters (such as voltage and current), lacking comprehensive acquisition and integration of key SOFC operating parameters (such as oxygen partial pressure and ambient humidity), resulting in incomplete information.

[0004] Second, the limitations of feature extraction. Existing technologies mostly use static features (such as mean and variance), which are difficult to capture the time-frequency dynamic characteristics of SOFC aging process, thus limiting the prediction accuracy.

[0005] Third, the models lack adaptability. Most AI models used in existing technologies are statically trained and lack online self-correction capabilities, making them unable to adapt to the nonlinear degradation throughout the entire lifecycle of SOFC.

[0006] Fourth, the lack of closed-loop optimization. Existing technologies mostly remain at the monitoring and diagnosis level, failing to form a closed-loop system from prediction to operational optimization. Summary of the Invention

[0007] To at least partially address the technical problems in related technologies, this invention provides an artificial intelligence-based method and system for SOFC state monitoring and health diagnosis. This invention aims to solve the technical challenges of low state monitoring accuracy and insufficient health prediction capabilities in solid oxide battery systems under complex operating conditions, providing an efficient artificial intelligence-based method and system to achieve accurate monitoring of battery operating status and durability prediction. This invention comprehensively captures the nonlinear degradation process of SOFCs through multi-source data fusion and dynamic feature extraction, achieving high-precision monitoring and durability prediction throughout the entire lifecycle. Compared to traditional methods, this invention fully utilizes the powerful data processing capabilities of artificial intelligence, combined with the operating characteristics of solid oxide batteries, significantly improving the real-time nature of monitoring and the accuracy of prediction, providing technical support for intelligent management in the fields of energy storage and high-efficiency power generation.

[0008] To achieve the above objectives, the technical solution adopted by the present invention includes:

[0009] According to a first aspect of the present invention, an artificial intelligence-based method for SOFC status monitoring and health diagnosis is provided, comprising the following steps:

[0010] Step S1: Real-time acquisition of SOFC operating parameters, including temperature, voltage, current, internal resistance, oxygen partial pressure, and ambient humidity;

[0011] Step S2: The collected operating parameters are weighted using an algorithm based on extended Kalman filtering and information entropy to optimize data consistency and noise resistance.

[0012] Step S3: Based on the improved hybrid neural network model and reinforcement learning algorithm, monitor the battery status in real time, predict the expected lifespan and identify anomalies, and dynamically adjust the model parameters through online learning via a sliding window and a self-correction mechanism;

[0013] Step S4: Combine monitoring results, health predictions, and abnormal warning signals to generate feedback control signals and dynamically adjust the operating parameters of the SOFC to extend battery life.

[0014] Optionally, step S4 further includes: outputting the monitoring results, health predictions, and abnormal warning signals in a visual form.

[0015] Optionally, in step S1, operating parameters are collected in real time through a sensor network.

[0016] Optionally, step S2 specifically includes:

[0017] Step S2-1: Process the collected operating parameters according to the following formula:

[0018]

[0019]

[0020]

[0021]

[0022]

[0023] In the formula, Let F be the predicted state, and F be the state transition matrix. This represents the updated state at time k-1. To predict covariance, Let the updated covariance be at time k-1. Let F be the transpose of F, and Q be the process noise covariance. H is the Kalman gain, and H is the observation matrix. Let H be the transpose of H, and R be the measurement noise covariance. To update the status, For the observed values, To update the covariance, I is the identity matrix;

[0024] Step S2-2: Calculate the information entropy and weights of the collected operating parameters according to the following formula:

[0025]

[0026]

[0027] In the formula, E is the information entropy. Let be the normalized value of the i-th sensor data. The weight of the i-th sensor data is... Let i be the information entropy corresponding to the i-th sensor data. The information entropy corresponding to the data from the j-th sensor;

[0028] Step S2-3: Calculate the fusion value according to the following formula:

[0029]

[0030] In the formula, This is the fused value obtained by fusing data from multiple sensors. This represents the data from the i-th sensor.

[0031] Optionally, step S3 specifically includes:

[0032] Step S3-1: Perform time-frequency analysis on the voltage and internal resistance signals using discrete wavelet transform to extract transient features, and calculate the capacity decay rate using statistical methods; the calculation formula for discrete wavelet transform is as follows:

[0033]

[0034]

[0035] In the formula, Let be the wavelet coefficients at the j-th scale and the k-th time shift. For input signal, For wavelet basis functions, Let be the approximation coefficients for the j-th layer. Let be the detail coefficients of the j-th layer. For scaling function, It is a wavelet function;

[0036] Step S3-2: Based on a hybrid neural network model and reinforcement learning algorithm, monitor battery status in real time, predict expected lifespan, and identify anomalies. The hybrid neural network model is a 4-layer convolutional, 3×3 kernel CNN and a 3-layer, 128-unit LSTM hybrid neural network model. The CNN is used to extract spatial features from multi-dimensional data, and the LSTM is used to capture long-term dependencies in the time series. The model outputs the Performance Degradation Index (PDI) and the Predicted Expected Lifespan (EEC). The calculation formulas for the PDI and EEC are as follows:

[0037]

[0038]

[0039]

[0040]

[0041] In the formula, This is a capacity-based performance degradation index, where C is the current capacity. This is the initial capacity; This is a performance degradation index based on environmental factors. This is the temperature compensation factor, where T is the current temperature. This is a factor affecting oxygen partial pressure. This represents the current partial pressure of oxygen. These are the corresponding weighting coefficients; This is the performance failure threshold. β is the decay rate, β is the cyclic stress factor caused by temperature, and N is the number of cycles.

[0042] Optionally, the Set it to 20%.

[0043] Optionally, step S3-2 further includes:

[0044] The model weights are updated every certain number of iterations to dynamically adjust parameters using the latest data, and the prediction strategy is optimized in conjunction with the reward function to improve long-term stability.

[0045] Optionally, in step S4, the feedback control signal is used to adjust the charging rate of the SOFC and the battery control parameters.

[0046] According to a second aspect of the present invention, an AI-based SOFC condition monitoring and health diagnosis system is also provided, for executing the AI-based SOFC condition monitoring and health diagnosis method described in any of the technical solutions of the first aspect of the present invention, wherein the AI-based SOFC condition monitoring and health diagnosis system comprises:

[0047] The data acquisition module is used to collect SOFC's operating parameters in real time.

[0048] The data fusion unit is used to fuse the collected operating parameters.

[0049] The feature adaptive filtering unit is used to extract multi-scale features related to battery health from the fused data and perform dynamic filtering.

[0050] The self-calibrating AI module is used to diagnose battery status and predict durability in real time based on artificial intelligence. It achieves model self-calibration through online learning and identifies abnormal states.

[0051] The feedback control unit is used to generate feedback control signals based on diagnostic results to optimize battery operating parameters.

[0052] Optionally, the self-calibrating AI module uses an improved hybrid neural network and reinforcement learning algorithm to detect battery status, predict durability, and identify anomalies in real time. The feedback control unit generates feedback control signals based on the prediction results to dynamically adjust the battery charging rate and battery control parameters.

[0053] Beneficial effects:

[0054] 1. Through the above technical solution, the present invention can achieve the following technical effects:

[0055] First, high-precision condition monitoring. The method of this invention achieves data consistency of over 98% and reduces monitoring error to below 2% by fusing multi-source data (such as temperature, voltage, current, internal resistance, oxygen partial pressure, and ambient humidity) and combining extended Kalman filtering (EKF) and information entropy weighting algorithm. Compared with traditional single-parameter monitoring methods (such as voltage-based methods with an error of about 5-10%), this significantly improves the monitoring accuracy of SOFCs at high temperatures (600-1000℃) and during dynamic degradation processes.

[0056] Second, accurate health prediction. The method of this invention utilizes a CNN-LSTM hybrid neural network and reinforcement learning algorithms. This invention can predict the expected lifespan (EEC) of SOFC in real time with a prediction accuracy of up to 95% and an error controlled within ±10 cycles. Through a sliding window online learning mechanism, the model adaptively adjusts parameters to ensure accurate capture of the nonlinear aging process throughout its entire lifespan, improving long-term stability by 20%.

[0057] Third, it extends battery life. The method of this invention uses closed-loop feedback optimization to dynamically adjust operating parameters based on diagnostic results (such as reducing the charging rate to 70% and controlling the temperature at 600-700℃), effectively slowing down the capacity decay rate. Tests show that after 1000 cycles, the capacity decay rate of the feedback-optimized group is reduced to 5% (compared to 8% in the non-feedback group), battery life is extended by 15-20% (approximately 150-200 additional cycles), and operating efficiency is improved by 10%.

[0058] Fourth, strong robustness and real-time performance. The dynamic feature extraction technology of this invention extracts multi-scale features through time-frequency analysis and wavelet transform, improving feature representation capability by 30%, reducing computational complexity by 70%, and shortening computation time from about 500ms to less than 150ms. This ensures the system's real-time monitoring capability under high loop counts (>1000 times) and complex environments, and the feature stability is 20% better than traditional static methods.

[0059] Fifth, intelligent management support. This invention provides visualized result outputs (internal resistance growth rate graph, temperature influence factor graph, abnormal alarm) and feedback control signals, providing technical support for the intelligent management of SOFC in energy storage and high-efficiency power generation, significantly improving the system's economic benefits and practicality.

[0060] In summary, this invention achieves high precision, high efficiency, and intelligence in SOFC condition monitoring and health diagnosis through technological integration. It not only solves the limitations of traditional methods under complex conditions but also provides an innovative solution for the long-term stable operation and lifespan optimization of SOFCs, demonstrating significant application value.

[0061] 2. Other beneficial effects or advantages of the present invention will be described in detail in the specific embodiments. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] in:

[0064] Figure 1 This is a schematic diagram of the workflow of an AI-based SOFC status monitoring and health diagnosis system provided in an exemplary embodiment of the present invention.

[0065] Figure 2 This is a schematic diagram illustrating the change in internal resistance growth rate with the number of cycles, provided by an exemplary embodiment of the present invention. Figure 2In the diagram, the horizontal axis represents the "number of cycles" and the vertical axis represents the "growth rate (%)", reflecting the dynamic changes in internal resistance during battery aging.

[0066] Figure 3 This is a schematic diagram illustrating the change of temperature influence factor over time according to an exemplary embodiment of the present invention. Figure 3 In the graph, the horizontal axis represents "time (hours)" and the vertical axis represents "influence factor," reflecting the real-time impact of temperature on performance. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0068] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0069] To facilitate a clearer and more accurate understanding of the technical solutions of this invention by those skilled in the art, the following will first provide a more detailed explanation of the technical problems existing in the prior art in conjunction with examples.

[0070] Solid oxide fuel cells (SOFCs) have significant application value in energy storage and clean power generation due to their high energy conversion efficiency, environmental friendliness, and long lifespan. The development of SOFC technology has evolved from traditional electrochemical monitoring to data-driven intelligent monitoring. Early research primarily relied on equivalent circuit models and threshold detection methods to analyze battery state, but these methods showed limitations in capturing the complex characteristics of SOFCs under high temperatures (600-1000℃) and dynamic degradation processes. With the rise of data-driven technologies and artificial intelligence (AI), the field of battery health monitoring has ushered in new development opportunities.

[0071] While existing technologies have driven the development of battery monitoring technology, the following problems and challenges still exist in the field of SOFC state monitoring and health diagnosis:

[0072] First, there are limitations in data acquisition. Existing technologies are mostly limited to single or a few parameters (such as voltage and current), lacking comprehensive acquisition and integration of key SOFC operating parameters (such as oxygen partial pressure and ambient humidity), resulting in incomplete information. For example, the Chinese patent document with publication number CN112373352B only focuses on basic electrical parameters and does not involve the multidimensional effects of SOFC under high-temperature environments.

[0073] Second, there are shortcomings in feature extraction. Existing technologies mostly use static features (such as mean and variance), which are difficult to capture the time-frequency dynamic characteristics during the SOFC aging process, thus limiting the prediction accuracy. Although multi-scale analysis has been proposed in existing related technologies, it has not been optimized for the material and operational characteristics of SOFCs.

[0074] Third, the models lack adaptability. Most existing AI models are statically trained and lack online self-correction capabilities, making them unable to adapt to the nonlinear degradation throughout the entire lifecycle of SOFC. For example, although the Chinese patent document with announcement number CN113594510B incorporates a cloud platform, it does not achieve real-time adaptive adjustment.

[0075] Fourth, the lack of closed-loop optimization. Existing technologies mostly remain at the monitoring and diagnosis level, failing to form a closed-loop system from prediction to operational optimization. For example, although the Chinese patent document with publication number CN112373352B improves predictive capabilities, it does not feed the results back to system control, resulting in limited lifespan extension.

[0076] To address the aforementioned challenges, this invention proposes an AI-based method and system for SOFC condition monitoring and health diagnosis, possessing significant technical importance and application value. This invention achieves high-precision real-time monitoring and durability prediction of SOFCs under complex conditions through multi-source data acquisition and fusion, dynamic feature extraction, self-correcting AI diagnosis, and closed-loop feedback optimization. Compared to existing related technologies, this invention fully utilizes the powerful data processing capabilities of AI, combined with the high-temperature operating characteristics of SOFCs, significantly improving monitoring robustness and prediction accuracy. Its technical advantages include:

[0077] 1. Comprehensiveness. It integrates multiple parameters such as temperature and oxygen partial pressure to compensate for the shortcomings of monitoring a single parameter.

[0078] 2. Dynamic nature. Through time-frequency analysis and adaptive feature selection, the degradation process is accurately captured.

[0079] 3. Adaptability. The AI ​​model learns online and adapts to changes throughout its entire lifecycle.

[0080] 4. Practicality. Closed-loop feedback optimizes the operation strategy and extends the lifespan of SOFC.

[0081] Therefore, this invention not only fills the technological gap in the field of SOFC health diagnosis, but also provides an innovative solution for intelligent management of energy storage and efficient power generation, which has significant scientific value and industrial application prospects.

[0082] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0083] According to a first aspect of the present invention, an artificial intelligence-based method for SOFC status monitoring and health diagnosis is provided, comprising the following steps:

[0084] Step S1: Real-time acquisition of SOFC operating parameters, including temperature, voltage, current, internal resistance, oxygen partial pressure, and ambient humidity;

[0085] Step S2: The collected operating parameters are weighted using an algorithm based on extended Kalman filtering and information entropy to optimize data consistency and noise resistance.

[0086] Step S3: Based on the improved hybrid neural network model and reinforcement learning algorithm, monitor the battery status in real time, predict the expected lifespan and identify anomalies, and dynamically adjust the model parameters through online learning via a sliding window and a self-correction mechanism;

[0087] Step S4: Combine monitoring results, health predictions, and abnormal warning signals to generate feedback control signals and dynamically adjust the operating parameters of the SOFC to extend battery life.

[0088] Through the above technical solution, this invention significantly improves the state monitoring accuracy, health prediction capability, and service life of solid oxide fuel cells (SOFCs) under complex operating conditions by integrating multi-source data acquisition and fusion, dynamic feature extraction, adaptive AI diagnosis, and closed-loop feedback optimization technologies. This invention achieves the following technical effects:

[0089] First, high-precision condition monitoring. The method of this invention achieves data consistency of over 98% and reduces monitoring error to below 2% by fusing multi-source data (such as temperature, voltage, current, internal resistance, oxygen partial pressure, and ambient humidity) and combining extended Kalman filtering (EKF) and information entropy weighting algorithm. Compared with traditional single-parameter monitoring methods (such as voltage-based methods with an error of about 5-10%), this significantly improves the monitoring accuracy of SOFCs at high temperatures (600-1000℃) and during dynamic degradation processes.

[0090] Second, accurate health prediction. The method of this invention utilizes a CNN-LSTM hybrid neural network and reinforcement learning algorithms. This invention can predict the expected lifespan (EEC) of SOFC in real time with a prediction accuracy of up to 95% and an error controlled within ±10 cycles. Through a sliding window online learning mechanism, the model adaptively adjusts parameters to ensure accurate capture of the nonlinear aging process throughout its entire lifespan, improving long-term stability by 20%.

[0091] Third, it extends battery life. The method of this invention uses closed-loop feedback optimization to dynamically adjust operating parameters based on diagnostic results (such as reducing the charging rate to 70% and controlling the temperature at 600-700℃), effectively slowing down the capacity decay rate. Tests show that after 1000 cycles, the capacity decay rate of the feedback-optimized group is reduced to 5% (compared to 8% in the non-feedback group), battery life is extended by 15-20% (approximately 150-200 additional cycles), and operating efficiency is improved by 10%.

[0092] Fourth, strong robustness and real-time performance. The dynamic feature extraction technology of this invention extracts multi-scale features (e.g., voltage harmonic components, internal resistance transient response) through time-frequency analysis and wavelet transform, improving feature representation capability by 30%, reducing computational complexity by 70%, and shortening computation time from approximately 500ms to less than 150ms. This ensures the system's real-time monitoring capability under high cycle counts (>1000 times) and complex environments, and the feature stability is 20% better than traditional static methods.

[0093] Fifth, intelligent management support. This invention provides visualized result outputs (internal resistance growth rate graph, temperature influence factor graph, abnormal alarm) and feedback control signals, providing technical support for the intelligent management of SOFC in energy storage and high-efficiency power generation, significantly improving the system's economic benefits and practicality.

[0094] In summary, this invention achieves high precision, high efficiency, and intelligence in SOFC condition monitoring and health diagnosis through technological integration. It not only solves the limitations of traditional methods under complex conditions but also provides an innovative solution for the long-term stable operation and lifespan optimization of SOFCs, demonstrating significant application value.

[0095] The present invention will now be described with reference to an exemplary embodiment.

[0096] I. Hardware Configuration.

[0097] 1) Temperature sensor: K-type thermocouple (0-1300℃, ±0.2℃), sampling rate 5Hz.

[0098] 2) Voltage sensor: differential amplifier (0-6V, ±0.005V) with anti-interference filter.

[0099] 3) Current sensor: Hall effect sensor (0-150A, ±0.05A), supports high frequency response.

[0100] 4) Oxygen partial pressure sensor: Zirconia sensor (0-1atm, ±0.01atm).

[0101] 5) Data acquisition system: 32-bit ARM processor, 16-bit ADC, storage capacity 1TB.

[0102] II. Data Fusion and Preprocessing.

[0103] 1) Employ a fusion algorithm based on extended Kalman filter (EKF) and information entropy weighting to reduce sensor noise and redundancy and improve data consistency.

[0104] 2) Mathematical principles:

[0105]

[0106]

[0107]

[0108]

[0109]

[0110] In the formula, Let F be the predicted state, and F be the state transition matrix. This represents the updated state at time k-1. To predict covariance, Let the updated covariance be at time k-1. Let F be the transpose of F, and Q be the process noise covariance. H is the Kalman gain, and H is the observation matrix. Let H be the transpose of H, and R be the measurement noise covariance. To update the status, For the observed values, To update the covariance, I is the identity matrix;

[0111]

[0112]

[0113] Where E is the information entropy, Let be the normalized value of the i-th sensor data. The weight of the i-th sensor data is... Let i be the information entropy corresponding to the i-th sensor data. The information entropy corresponding to the data from the j-th sensor;

[0114] 3) Calculation of fusion value:

[0115]

[0116] In the formula, This is the fused value obtained by fusing data from multiple sensors. This represents the data from the i-th sensor.

[0117] In this process, firstly, the state of multi-source data (e.g., temperature, oxygen partial pressure, etc.) is estimated by extended Kalman filtering to filter out high-frequency noise; then, the entropy value of each sensor data is calculated and weights are assigned; finally, a weighted average is taken to obtain the fusion result.

[0118] III. Feature Extraction and Time-Frequency Analysis.

[0119] 1) Use Discrete Wavelet Transform (DWT) to perform time-frequency analysis on voltage and internal resistance signals, extract transient features, and calculate the capacity decay rate using statistical methods.

[0120] 2) Mathematical principles:

[0121] Mathematical formula:

[0122] Discrete wavelet transform:

[0123]

[0124] In the formula, Let be the wavelet coefficients at the j-th scale and the k-th time shift. For input signal, For wavelet basis functions, "db6" can be selected.

[0125] Signal decomposition:

[0126]

[0127] In the formula, Let be the approximation coefficients for the j-th layer. Let be the detail coefficients of the j-th layer. For scaling function, It is a wavelet function;

[0128] For example, taking a 6-layer decomposition, the voltage signal is decomposed into low-frequency approximate components and high-frequency detail components. The amplitude of the high-frequency components is extracted as transient features to reflect the change in battery internal resistance; the low-frequency components are used to calculate the capacity decay rate.

[0129] IV. AI Model Training and Optimization.

[0130] 1) Model structure: CNN (4 convolutional layers, 3×3 kernels) + LSTM (3 layers, 128 units), input 5-dimensional features, output PDI and EEC.

[0131] 2) Training parameters: 800 sets of data (1500 iterations), batch size 32, learning rate 0.002.

[0132] 3) Online learning: The model is updated every 50 cycles, with a sliding window of 50 and a learning rate decay of 0.95.

[0133] V. Results Presentation and Feedback Optimization. (e.g.) Figure 2 (As shown)

[0134] 1) Output: Internal resistance growth rate graph, temperature influence factor graph, and abnormal alarm.

[0135] 2) Feedback: When PDI < 60%, the charging current drops to 70%, and the temperature is controlled at 600-700℃.

[0136] In this exemplary embodiment, the advantages of the present invention include:

[0137] First, high-precision multi-source data fusion. This invention achieves efficient fusion of multi-dimensional parameters such as temperature, voltage, current, internal resistance, and oxygen partial pressure through extended Kalman filtering and information entropy weighting algorithms, ensuring data quality stability under complex high-temperature environments. The fusion process dynamically adjusts the weights of each parameter, effectively filtering out noise interference. For example, under conditions of temperature fluctuations of ±50℃ and oxygen partial pressure changes of ±0.05 atm, data consistency remains above 98%. Exemplarily, the monitoring error of this invention can be reduced to below 2%, significantly improving data reliability compared to traditional single-parameter monitoring (e.g., voltage-based errors of approximately 5-10%), providing a solid foundation for subsequent feature extraction and AI diagnosis.

[0138] Second, the efficiency and real-time performance of dynamic feature extraction. This invention employs time-frequency analysis and wavelet transform techniques to extract multi-scale features, such as voltage harmonic components and internal resistance transient response. Key features are dynamically screened using random forests and mutual information methods to eliminate redundant information. The feature dimension is reduced from an initial 15 dimensions to 5 dimensions, and the computation time is shortened from approximately 500ms in traditional methods to less than 150ms, ensuring real-time monitoring capabilities. For example, the feature representation capability of this invention can be improved by 30%, the computational complexity can be reduced by approximately 70%, and the feature stability under high iteration counts (>1000 times) is about 20% better than static methods, significantly improving the system's ability to capture the aging process.

[0139] Third, the high accuracy and stability of adaptive AI prediction. This invention combines a CNN-LSTM hybrid neural network and reinforcement learning to analyze battery state in real time and predict expected endurance cycles (EEC). Through online learning and model self-correction every 50 cycles, it adapts to the nonlinear changes in the battery from initial operation to deep aging. For example, at a temperature of 650°C and 1500 cycles, the EEC prediction error is controlled within ±10 cycles. Exemplarily, the prediction accuracy reaches 95%, and the long-term stability is improved by 20%, making it better able to cope with the dynamic degradation of SOFC throughout its entire life cycle than traditional static AI models.

[0140] Fourth, the invention features intelligent closed-loop feedback optimization and extended battery life. Based on diagnostic results, it generates feedback signals to dynamically adjust the charging rate (e.g., reducing to 70% rated current when PDI < 60%) and temperature control (maintaining 600-700℃), forming a closed-loop optimization system. Tests show that after 1000 cycles, the battery capacity degradation rate of the feedback-optimized group is reduced to 5%, while that of the non-feedback group is 8%. For example, battery life can be extended by 15-20% (approximately 150-200 additional cycles), and operating efficiency can be improved by 10% (increased power output stability). Compared to non-feedback monitoring systems (such as traditional threshold control), this significantly improves economic efficiency and lifespan.

[0141] In the above embodiments, it should be noted that, in this invention, the extended Kalman filter estimates the state of the nonlinear system, filters out sensor noise, and ensures dynamic data consistency. The information entropy weighting calculates weights based on the information entropy of each parameter, dynamically adjusting the contribution of different data to enhance noise resistance. Compared to traditional single Kalman filtering or mean fusion, this method combines EKF and information entropy, maintaining a fusion accuracy of over 98% under complex conditions such as high temperature in SOFCs and changes in oxygen partial pressure.

[0142] In one embodiment of the present invention, step S4 may further include: outputting the monitoring results, health predictions and abnormal warning signals in a visual form.

[0143] In one embodiment of the present invention, in step S1, operating parameters are collected in real time through a sensor network.

[0144] In one embodiment of the present invention, step S2 may specifically include:

[0145] Step S2-1: Process the collected operating parameters according to the following formula:

[0146]

[0147]

[0148]

[0149]

[0150]

[0151] In the formula, Let F be the predicted state, and F be the state transition matrix. This represents the updated state at time k-1. To predict covariance, Let the updated covariance be at time k-1. Let F be the transpose of F, and Q be the process noise covariance. H is the Kalman gain, and H is the observation matrix. Let H be the transpose of H, and R be the measurement noise covariance. To update the status, For the observed values, To update the covariance, I is the identity matrix;

[0152] Step S2-2: Calculate the information entropy and weights of the collected operating parameters according to the following formula:

[0153]

[0154]

[0155] In the formula, E is the information entropy. Let be the normalized value of the i-th sensor data. The weight of the i-th sensor data is... Let i be the information entropy corresponding to the i-th sensor data. The information entropy corresponding to the data from the j-th sensor;

[0156] Step S2-3: Calculate the fusion value according to the following formula:

[0157]

[0158] In the formula, This is the fused value obtained by fusing data from multiple sensors. This represents the data from the i-th sensor.

[0159] In one embodiment of the present invention, step S3 may specifically include:

[0160] Step S3-1: Perform time-frequency analysis on the voltage and internal resistance signals using discrete wavelet transform to extract transient features, and calculate the capacity decay rate using statistical methods; the calculation formula for discrete wavelet transform is as follows:

[0161]

[0162]

[0163] In the formula, Let be the wavelet coefficients at the j-th scale and the k-th time shift. For input signal, For wavelet basis functions, Let be the approximation coefficients for the j-th layer. Let be the detail coefficients of the j-th layer. For scaling function, It is a wavelet function;

[0164] Step S3-2: Based on a hybrid neural network model and reinforcement learning algorithm, monitor battery status in real time, predict expected lifespan, and identify anomalies. The hybrid neural network model is a hybrid of a 4-layer convolutional, 3×3 kernel CNN and a 3-layer, 128-unit LSTM neural network model. The CNN is used to extract spatial features from multi-dimensional data, and the LSTM is used to capture long-term dependencies in the time series. The output is the Performance Degradation Index (PDI) and the Predicted Expected Lifespan (EEC). The calculation formulas for the PDI and EEC are as follows:

[0165]

[0166]

[0167]

[0168]

[0169] In the formula, This is a capacity-based performance degradation index, where C is the current capacity. This is the initial capacity; This is a performance degradation index based on environmental factors. This is the temperature compensation factor, where T is the current temperature. This is a factor affecting oxygen partial pressure. This represents the current partial pressure of oxygen. These are the corresponding weighting coefficients; This is the performance failure threshold. β is the decay rate, β is the cyclic stress factor caused by temperature, and N is the number of cycles.

[0170] In this embodiment, firstly, for the hybrid neural network model, its model structure is a hybrid network of CNN and LSTM to form a spatiotemporal feature analysis capability. CNN is used to extract the spatial features of multidimensional data (e.g., voltage harmonic components, internal resistance transient response, etc.), and LSTM is used to capture long-term dependencies in the time series for analyzing SOFC aging trends.

[0171] Second, regarding the self-calibration mechanism, model self-calibration is achieved through online learning via a sliding window. The model weights are updated every certain number of iterations (e.g., 50 iterations), and the parameters are dynamically adjusted using the latest data. Simultaneously, a reward-based prediction strategy is combined to improve long-term stability.

[0172] Third, in this implementation, the output target of the hybrid neural network model is to monitor the Performance Degradation Index (PDI) and predict the Expected Endurance Cycle (EEC) in real time. Compared with fixed-parameter AI models (such as traditional CNNs), this method adapts to the changes throughout the SOFC lifecycle through CNN-LSTM and online learning, achieving a prediction accuracy of 95%.

[0173] In one embodiment of the present invention, the present invention It can be set to 20%.

[0174] In one embodiment of the present invention, step S3-2 may further include:

[0175] The model weights are updated every certain number of iterations to dynamically adjust parameters using the latest data, and the prediction strategy is optimized in conjunction with the reward function to improve long-term stability.

[0176] In one embodiment of the present invention, in step S4, the feedback control signal is used to adjust the charging rate of the SOFC and the battery control parameters.

[0177] According to a second aspect of the invention, please refer to Figure 1 Furthermore, an AI-based SOFC condition monitoring and health diagnosis system is provided, used to execute the AI-based SOFC condition monitoring and health diagnosis method of any of the technical solutions in the first aspect of the present invention. The AI-based SOFC condition monitoring and health diagnosis system includes:

[0178] The data acquisition module is used to collect SOFC's operating parameters in real time.

[0179] The data fusion unit is used to fuse the collected operating parameters.

[0180] The feature adaptive filtering unit is used to extract multi-scale features related to battery health from the fused data and perform dynamic filtering.

[0181] The self-calibrating AI module is used to diagnose battery status and predict durability in real time based on artificial intelligence. It achieves model self-calibration through online learning and identifies abnormal states.

[0182] The feedback control unit is used to generate feedback control signals based on diagnostic results to optimize battery operating parameters.

[0183] In this embodiment, an exemplary implementation of the present invention will be described, based on an artificial intelligence-based SOFC condition monitoring and health diagnosis system.

[0184] 1. Data Acquisition Module:

[0185] (1) Function: Collect multi-dimensional parameters such as temperature and oxygen partial pressure through a multi-sensor network (such as K-type thermocouples and zirconia sensors).

[0186] (2) Hardware: 32-bit ARM processor, 16-bit ADC, etc.

[0187] 2. Data Fusion Unit:

[0188] (1) Function: Weighted fusion of multi-source data to optimize consistency and noise resistance.

[0189] (2) Implementation: The EKF and information entropy algorithms in the above technical solutions can be used.

[0190] 3. Feature-adaptive filtering unit:

[0191] (1) Function: Combine time-frequency analysis (such as wavelet transform) and statistical methods to extract multi-scale features and dynamically filter them through importance scoring.

[0192] (2) Output: Feature set strongly correlated with SOFC health.

[0193] 4. Self-calibrating AI module:

[0194] (1) Function: Use AI technology (such as CNN-LSTM) to diagnose the state and predict EEC, and adapt by learning online.

[0195] (2) Implementation: Refer to the above technical solutions.

[0196] 5. Feedback control unit:

[0197] (1) Function: Convert diagnostic results into control signals (e.g., adjust charging rate, temperature) to form closed-loop optimization.

[0198] (2) Output: Visualized results and operating strategies.

[0199] In this exemplary embodiment, the system of the present invention integrates multi-source data processing, dynamic feature extraction, self-correcting AI, and closed-loop feedback, which is significantly superior to traditional single-module systems.

[0200] In one embodiment of the present invention, the self-calibrating AI module of the present invention uses an improved hybrid neural network and reinforcement learning algorithm to detect battery status, predict durability and identify anomalies in real time. The feedback control unit generates feedback control signals based on the prediction results to dynamically adjust the battery charging rate and battery control parameters.

[0201] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for SOFC condition monitoring and health diagnosis based on artificial intelligence, characterized in that, Includes the following steps: Step S1: Real-time acquisition of SOFC operating parameters, including temperature, voltage, current, internal resistance, oxygen partial pressure, and ambient humidity; Step S2: The collected operating parameters are weighted using an algorithm based on extended Kalman filtering and information entropy to optimize data consistency and noise resistance. Step S3: Based on the improved hybrid neural network model and reinforcement learning algorithm, monitor the battery status in real time, predict the expected lifespan and identify anomalies, and dynamically adjust the model parameters through online learning via a sliding window and a self-correction mechanism; Step S4: Combine monitoring results, health predictions, and abnormal warning signals to generate feedback control signals and dynamically adjust the operating parameters of the SOFC to extend battery life; Specifically, step S3 includes: Step S3-1: Perform time-frequency analysis on the voltage and internal resistance signals using discrete wavelet transform to extract transient features, and calculate the capacity decay rate using statistical methods; the calculation formula for discrete wavelet transform is as follows: In the formula, Let be the wavelet coefficients at the j-th scale and the k-th time shift. For input signal, For wavelet basis functions, Let be the approximation coefficients for the j-th layer. Let be the detail coefficients of the j-th layer. For scaling function, It is a wavelet function; Step S3-2: Based on a hybrid neural network model and reinforcement learning algorithm, monitor battery status in real time, predict expected lifespan, and identify anomalies. The hybrid neural network model is a hybrid neural network model consisting of a 4-layer convolutional, 3×3 kernel CNN and a 3-layer, 128-unit LSTM. The CNN is used to extract spatial features from multi-dimensional data, and the LSTM is used to capture long-term dependencies in the time series. The model outputs the Performance Degradation Index (PDI) and the Predicted Expected Lifespan (EEC). The calculation formulas for the PDI and EEC are as follows: In the formula, This is a capacity-based performance degradation index, where C is the current capacity. This is the initial capacity; This is a performance degradation index based on environmental factors. This is the temperature compensation factor, where T is the current temperature. This is a factor affecting oxygen partial pressure. This represents the current partial pressure of oxygen. These are the corresponding weighting coefficients; This is the performance failure threshold. denoted as the decay rate, β as the cyclic stress factor caused by temperature, and N as the number of cycles.

2. The method for SOFC status monitoring and health diagnosis based on artificial intelligence according to claim 1, characterized in that, Step S4 further includes: outputting the monitoring results, health predictions, and abnormal early warning signals in a visual form.

3. The method for SOFC status monitoring and health diagnosis based on artificial intelligence according to claim 1, characterized in that, In step S1, operating parameters are collected in real time through a sensor network.

4. The method for SOFC status monitoring and health diagnosis based on artificial intelligence according to claim 1, characterized in that, Step S2 specifically includes: Step S2-1: Process the collected operating parameters according to the following formula: In the formula, Let F be the predicted state, and F be the state transition matrix. This represents the updated state at time k-1. To predict covariance, Let the updated covariance be at time k-1. Let F be the transpose of F, and Q be the process noise covariance. H is the Kalman gain, and H is the observation matrix. Let H be the transpose of H, and R be the measurement noise covariance. To update the status, For the observed values, To update the covariance, I is the identity matrix; Step S2-2: Calculate the information entropy and weights of the collected operating parameters according to the following formula: In the formula, E is the information entropy. Let be the normalized value of the i-th sensor data. The weight of the i-th sensor data is... Let i be the information entropy corresponding to the i-th sensor data. The information entropy corresponding to the data from the j-th sensor; Step S2-3: Calculate the fusion value according to the following formula: In the formula, This is the fused value obtained by fusing data from multiple sensors. This represents the data from the i-th sensor.

5. The method for SOFC status monitoring and health diagnosis based on artificial intelligence according to claim 1, characterized in that, The Set it to 20%.

6. The method for SOFC status monitoring and health diagnosis based on artificial intelligence according to claim 1, characterized in that, Step S3-2 further includes: The model weights are updated every certain number of iterations to dynamically adjust parameters using the latest data, and the prediction strategy is optimized in conjunction with the reward function to improve long-term stability.

7. The method for SOFC status monitoring and health diagnosis based on artificial intelligence according to claim 1, characterized in that, In step S4, the feedback control signal is used to adjust the charging rate of the SOFC and the battery control parameters.

8. An AI-based SOFC condition monitoring and health diagnosis system, characterized in that, For performing the AI-based SOFC condition monitoring and health diagnosis method as described in any one of claims 1-7, the AI-based SOFC condition monitoring and health diagnosis system comprises: The data acquisition module is used to collect SOFC's operating parameters in real time. The data fusion unit is used to fuse the collected operating parameters. The feature adaptive filtering unit is used to extract multi-scale features related to battery health from the fused data and perform dynamic filtering. The self-calibrating AI module is used to diagnose battery status and predict durability in real time based on artificial intelligence. It achieves model self-calibration through online learning and identifies abnormal states. The feedback control unit is used to generate feedback control signals based on diagnostic results to optimize battery operating parameters.

9. The AI-based SOFC condition monitoring and health diagnosis system according to claim 8, characterized in that, The self-calibrating AI module uses an improved hybrid neural network and reinforcement learning algorithm to detect battery status, predict durability, and identify anomalies in real time. The feedback control unit generates feedback control signals based on the prediction results to dynamically adjust the battery charging rate and battery control parameters.

Images