SOFC state monitoring and health diagnosis method and system based on deep learning

Abstract

The invention provides an SOFC state monitoring and health diagnosis method and system based on deep learning, and relates to the technical field of solid oxide battery state monitoring, and the method comprises the steps: collecting the operation data of an SOFC; preprocessing the operation data, and performing feature extraction on the preprocessed operation data; inputting the operation data into a deep learning diagnosis model, the deep learning diagnosis model adopting a hybrid architecture of a convolutional neural network and a long-short term memory network, and combining an attention mechanism to output an adaptive health status index HSI; and performing real-time monitoring according to the HSI and the change rate thereof, and if the HSI is lower than a preset threshold value or the change rate of the HSI exceeds the preset threshold value, giving out an early warning. According to the technical scheme, the key problems of data complexity, insufficient precision, poor real-time performance and the like in SOFC state monitoring and health diagnosis can be effectively solved, and an innovative technical scheme is provided for efficient operation and long-life management of the SOFC.

Description

Technical Field

[0001] The present invention relates to the technical field of solid oxide battery state monitoring technology, and in particular to a SOFC state monitoring and health diagnosis method and system based on deep learning. Background Art

[0002] Solid oxide fuel cells (SOFCs) are a highly efficient, low-emission energy conversion technology. Due to their high energy efficiency and fuel adaptability, SOFCs have significant potential in distributed power generation, aerospace, and industrial energy systems.

[0003] However, in existing technologies, SOFC condition monitoring and health diagnosis still have the following problems: First, data complexity and feature extraction difficulties. The data generated by SOFC operation is high-dimensional, time-varying, and strongly coupled. Traditional physical models and shallow machine learning methods have difficulty effectively extracting its spatiotemporal features.

[0004] Second, the diagnostic accuracy and robustness are insufficient. Existing technologies are prone to false positives or missed positives when multiple faults coexist or when operating conditions are complex.

[0005] Third, the real-time performance is insufficient. Faults in SOFC operation often develop rapidly, and traditional monitoring systems have a long response time, making it difficult to achieve early warning.

[0006] Fourth, there is a lack of full lifecycle management. Existing technologies mostly focus on a single step (such as fault detection) and lack a closed-loop system from data collection to diagnosis and optimization feedback. This prevents intelligent and continuous SOFC health management. Summary of the Invention

[0007] In order to at least partially solve the technical problems in the related art, the present invention provides a SOFC state monitoring and health diagnosis method and system based on deep learning.

[0008] In order to achieve the above object, the technical solution adopted by the present invention includes: According to a first aspect of the present invention, a method for SOFC state monitoring and health diagnosis based on deep learning is provided, comprising the following steps: Step S1: collecting SOFC operating data, wherein the operating data includes temperature, total stack voltage, operating current, hydrogen flow rate, oxygen flow rate and SOFC internal pressure; Step S2: pre-processing the collected operating data and extracting features from the pre-processed operating data, wherein the features include voltage decay rate, voltage mean, voltage standard deviation, and voltage peak value; Step S3: Input the pre-processed operating data into a deep learning diagnostic model, which uses a hybrid architecture of convolutional neural networks and long short-term memory networks, and combines the attention mechanism to output an adaptive health status index. ; in, The calculation process specifically includes: Step S3-1: Calculate the basic adaptive health status index : Step S3-2: Calculate the rate of change: Step S3-3: Adjust weights: Step S3-4: Calculation : Where, First, set the weight. Set the weight for the second, is the voltage decay rate, is the SOFC operating temperature, is the optimal operating temperature of SOFC, is the maximum allowable operating temperature of SOFC, It is the adaptive health status index of the previous moment, which is used to calculate the rate of change and reflect the temporal correlation of the status; For the first adjustment weight, Adjust weights for the second; Step S4: According to and its rate of change in real time. Below a preset threshold or If the rate of change exceeds the preset threshold, an early warning is issued.

[0009] Optionally, the SOFC state monitoring and health diagnosis method based on deep learning further includes: Step S5: In response to the early warning result, the health status of the SOFC is displayed through a visual interface and optimization suggestions are provided.

[0010] Optionally, step S5 specifically includes: In response to the early warning results, the health status of the SOFC is displayed through a visual interface; Output the corresponding optimization suggestions according to the following formula: Where, For optimization suggestions, is the actual operating temperature of SOFC.

[0011] Optionally, step S2 specifically includes: Step S2-1: performing denoising processing on the collected operating data based on wavelet transform; Step S2-2: normalizing the denoised running data; Step S2-3: Perform feature extraction on the normalized operating data.

[0012] Optionally, the deep learning diagnostic network model includes a CNN part and an LSTM part; The CNN part includes two convolutional layers, each of which is followed by a ReLU activation function, and the convolution operation is followed by a maximum pooling layer for dimensionality reduction. The LSTM part includes two LSTM layers, and each LSTM layer is followed by a Dropout layer to prevent overfitting.

[0013] Optionally, the attention mechanism is implemented in the following way: Step S3-2-1: Calculate the feature vector score output by the long short-term memory network; Step S3-2-2: normalize the feature vector score; Step S3-2-3: Weight the key features to highlight their Contribution of evaluation.

[0014] Optionally, step S4 specifically includes: Calculate according to the following formula Descent rate: Where, is the adaptive health status index at time t - k, is the time interval; when When the value of is greater than a first preset threshold, a first warning signal is output; when When it is less than a second preset threshold, a second warning signal is output.

[0015] Optionally, the The value is 5, and the first preset threshold is set to 0.05; and / or, The second preset threshold is set to 0.7.

[0016] According to a second aspect of the present invention, a deep learning-based SOFC state monitoring and health diagnosis system is provided, which is used to execute the deep learning-based SOFC state monitoring and health diagnosis method described in any technical solution of the first aspect of the present invention. The deep learning-based SOFC state monitoring and health diagnosis system includes: Data acquisition module, used to collect SOFC operation data; A preprocessing module is used to preprocess the collected operation data and perform feature extraction on the preprocessed operation data; A deep learning diagnostic model, using a hybrid architecture of convolutional neural networks and long short-term memory networks, combined with an attention mechanism, is used to analyze data and output an adaptive health status index. ; Condition monitoring and early warning module, used to Real-time monitoring and early warning of the data and its change rate.

[0017] Optionally, the deep learning-based SOFC state monitoring and health diagnosis system further includes: The visualization and feedback module is used to respond to the early warning results, display the health status of the SOFC through a visual interface, and provide optimization suggestions.

[0018] Beneficial effects: 1. Through the above technical solution, firstly, this invention combines convolutional neural networks (CNN) and long short-term memory networks (LSTM) for the first time, designing a new deep learning model for multi-dimensional time series data of SOFCs. CNN is used to extract spatial features, and LSTM is used to capture temporal dependencies. Compared to the existing HY-CNN-NARX paradigm and intelligent diagnostic methods, this hybrid model excels in spatiotemporal feature analysis, enabling more accurate health status assessment.

[0019] Second, the adaptive health status index proposed by the present invention The algorithm dynamically adjusts parameter weights to quantify the health of SOFCs in real time. Compared to existing multi-label diagnostic methods, this algorithm can self-optimize based on operating conditions, effectively improving diagnostic flexibility and accuracy.

[0020] Third, the method of the present invention integrates data acquisition, preprocessing, diagnosis, and early warning steps, effectively and accurately establishing a closed-loop management system for the entire SOFC lifecycle. Compared with conventional diagnostic methods and cloud platform solutions in existing technologies, the closed-loop design of the present invention effectively reduces manual intervention and significantly improves the system's intelligence and operational continuity.

[0021] Fourth, the method of this invention effectively improves the robustness of the model in complex environments through data augmentation technology and attention mechanisms. Furthermore, by combining algorithm optimization with hardware acceleration, this method can achieve rapid response (for example, within 0.5 seconds) and a fault prediction window of several hours, far exceeding the real-time performance of existing related technologies.

[0022] In summary, the present invention introduces deep learning diagnostic model, adaptive health status index The full life cycle closed-loop system can effectively solve key problems such as data complexity, insufficient accuracy and poor real-time performance in SOFC status monitoring and health diagnosis, and provide an innovative technical solution for the efficient operation and long life management of SOFC.

[0023] 2. Other beneficial effects or advantages of the present invention will be described in detail in the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for describing the embodiments. 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 paying any creative labor.

[0025] in: Figure 1 1 is a schematic diagram of the workflow of a deep learning-based SOFC state monitoring and health diagnosis system provided by an exemplary embodiment of the present invention; Figure 2 1 is a schematic diagram of the CNN-LSTM model structure provided by an exemplary embodiment of the present invention; Figure 3 is a schematic diagram of a training accuracy curve provided by an exemplary embodiment of the present invention; Figure 4 1 is a time-dependent comparison curve of HIS provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0027] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0028] In order to facilitate relevant technical personnel to have a clearer and more accurate understanding of the technical solutions of the present invention, the technical problems existing in the prior art are described in more detail below with reference to examples.

[0029] Because SOFC operates at high temperatures (e.g., 600°C to 1000°C) and in a complex chemical environment, it is susceptible to factors such as material degradation, thermal stress, and uneven fuel distribution, leading to performance degradation and even system failure.

[0030] Existing SOFC condition monitoring technologies primarily rely on physical models (e.g., thermodynamic and electrochemical analysis models) and simple threshold determination, for example, by monitoring voltage drops or temperature anomalies for early warning. While these condition monitoring methods were effective in early SOFC applications, their limitations have become increasingly apparent as SOFC systems become more complex and their application scenarios expand.

[0031] In recent years, with the development of artificial intelligence and data processing technology, deep learning methods have begun to be introduced into the field of SOFC condition monitoring and fault diagnosis. For example, Mohamadali et al. (2024) proposed a deep learning method based on the HY-CNN-NARX paradigm for transient modeling of SOFC, demonstrating the potential of neural networks in dynamic performance analysis [1]. Zhang et al. (2021) developed an intelligent diagnosis method based on deep learning to achieve simultaneous fault identification of SOFC systems, promoting the intelligent process of SOFC health management [2]. In terms of patent technology, the Chinese patent document with publication number CN112373352B proposed a fault diagnosis and fault-tolerant control method for fuel cell systems [5], and the Chinese patent document with publication number CN113594510B designed a SOFC fault diagnosis solution based on a cloud platform and introduced cloud computing technology [6]. The review by Ren et al. (2022) comprehensively reviewed the development of SOFC control technology and provided a background reference for monitoring and diagnosis [3].

[0032] Despite the progress made in the above research and technologies, SOFC condition monitoring and health diagnosis still face many problems and challenges, mainly including the following aspects: First, data complexity and feature extraction are difficult.

[0033] The data generated by SOFC operation is high-dimensional, time-varying, and strongly coupled. Traditional physical models and shallow machine learning methods have difficulty effectively extracting its spatiotemporal features. For example, traditional threshold-based methods cannot capture weak abnormal signals in dynamic changes, and the HY-CNN-NARX paradigm in the literature [1], while good at modeling, does not focus on diagnosis.

[0034] Second, the diagnostic accuracy and robustness are insufficient.

[0035] Existing methods are prone to false positives or missed positives when multiple faults coexist or when operating conditions are complex. Although the deep learning methods in references [2] and [4] have improved diagnostic capabilities, their robustness in dealing with noise interference and high-dimensional data still needs to be improved. The traditional diagnostic method in Chinese patent document CN112373352B[5] lacks adaptability to complex fault modes.

[0036] Third, it lacks real-time performance.

[0037] Faults in SOFC operation often develop rapidly, and traditional monitoring systems have a long response time, making it difficult to achieve early warning. For example, the cloud platform solution in Chinese patent document CN113594510B is affected by network latency and has limited real-time performance. The control technology mentioned in document [3] also does not solve the problem of real-time diagnosis.

[0038] Fourth, there is a lack of full life cycle management.

[0039] Current technologies are mostly focused on a single link (such as fault detection), and lack a closed-loop system from data collection to diagnosis to optimization feedback, resulting in the inability to achieve intelligent and continuous health management of SOFC.

[0040] In view of this, how to design a new SOFC state monitoring and health diagnosis solution to at least partially solve the above problems has become an urgent issue to be solved.

[0041] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0042] According to the first aspect of the present invention, this embodiment provides a method for SOFC state monitoring and health diagnosis based on deep learning, comprising the following steps: Step S1: collecting SOFC operating data, including temperature, total stack voltage, operating current, hydrogen flow rate, oxygen flow rate and SOFC internal pressure; Step S2: pre-processing the collected operating data and extracting features from the pre-processed operating data, the features including voltage decay rate, voltage mean, voltage standard deviation and voltage peak value; Step S3: Input the pre-processed operating data into the deep learning diagnostic model, which uses a hybrid architecture of convolutional neural network and long short-term memory network, and combines the attention mechanism to output the adaptive health status index. ; in, The calculation process specifically includes: Step S3-1: Calculate the basic adaptive health status index : Step S3-2: Calculate the rate of change: Step S3-3: Adjust weights: Step S3-4: Calculation : Where, First, set the weight. Set the weight for the second, is the voltage decay rate, is the SOFC operating temperature, is the optimal operating temperature of SOFC, is the maximum allowable operating temperature of SOFC, It is the adaptive health status index of the previous moment, which is used to calculate the rate of change and reflect the temporal correlation of the status; For the first adjustment weight, Adjust weights for the second; Step S4: According to and its rate of change in real time. Below a preset threshold or If the rate of change exceeds the preset threshold, an early warning is issued.

[0043] Through the above technical solution, firstly, this invention combines convolutional neural networks (CNN) and long short-term memory networks (LSTM) for the first time, designing a new deep learning model for multi-dimensional time series data of SOFCs. CNN is used to extract spatial features, while LSTM is used to capture temporal dependencies. Compared to existing HY-CNN-NARX paradigms and intelligent diagnostic methods, this hybrid model excels in spatiotemporal feature analysis, enabling more accurate health status assessment.

[0044] Second, the adaptive health status index proposed by the present invention The algorithm dynamically adjusts parameter weights to quantify the health of SOFCs in real time. Compared to existing multi-label diagnostic methods, this algorithm can self-optimize based on operating conditions, effectively improving diagnostic flexibility and accuracy.

[0045] Third, the method of the present invention integrates data acquisition, preprocessing, diagnosis, and early warning steps, effectively and accurately establishing a closed-loop management system for the entire SOFC lifecycle. Compared with conventional diagnostic methods and cloud platform solutions in existing technologies, the closed-loop design of the present invention effectively reduces manual intervention and significantly improves the system's intelligence and operational continuity.

[0046] Fourth, the method of this invention effectively improves the robustness of the model in complex environments through data augmentation technology and attention mechanisms. Furthermore, by combining algorithm optimization with hardware acceleration, this method can achieve rapid response (for example, within 0.5 seconds) and a fault prediction window of several hours, far exceeding the real-time performance of existing related technologies.

[0047] In summary, the present invention introduces deep learning diagnostic model, adaptive health status index The full life cycle closed-loop system can effectively solve key problems such as data complexity, insufficient accuracy and poor real-time performance in SOFC status monitoring and health diagnosis, and provide an innovative technical solution for the efficient operation and long life management of SOFC.

[0048] The method of the present invention is described below with reference to an exemplary embodiment.

[0049] 1. Experimental stack: The implementation of the present invention relies on precise experimental equipment and controlled environmental conditions to ensure the accuracy of data collection and the stability of the system.

[0050] The following are the detailed equipment and environment configurations: SOFC stack (using a 1kW flat-plate solid oxide fuel cell (SOFC) stack, which is composed of 20 single cells connected in series). The specifications of a single battery are: Active area: 100cm²; Anode material: nickel-zirconia (Ni-YSZ), thickness about 500μm; Cathode material: lanthanum strontium manganese oxide (LSM), thickness about 50μm; Electrolyte: Zirconium oxide (YSZ), thickness about 150μm; Collected operating data: Temperature: 800°C, maintained by a built-in ceramic heater, with a fluctuation range of ±5°C; Fuel: pure hydrogen (purity 99.99%), flow range 0-10L / min; Oxidant: air (containing 21% oxygen), flow range 0-5L / min; Initial voltage: 1.2V (single battery about 0.06V); Current range: 0-50A; Fuel supply: Hydrogen is provided through a high-pressure steel cylinder and regulated to 0.1 MPa through a pressure reducing valve; Air is supplied by a compressor and enters the fuel cell stack after being filtered.

[0051] 2. Implementation Steps The implementation of the present invention can be divided into four steps, from data collection to operation optimization. The specific operations, technical parameters and implementation details of each step are described below.

[0052] Step 1: Data Collection The operation time is: 30 consecutive days (i.e., 720 hours); the collected data includes temperature (anode inlet, cathode outlet, and stack center, unit: °C), voltage (total stack voltage, unit: V), current (operating current, unit: A), flow rate (hydrogen flow rate and air flow rate, unit: L / min), pressure (system internal pressure, unit: kPa), and sampling frequency (1 Hz, i.e., once per second, for a total of approximately 25.92 million data points). The operating conditions include: normal operation (voltage 1.2V, current 20A, temperature 800°C), load switching (current increased from 20A to 30A, lasting 12 hours), and aging stage (voltage gradually decreased to 1.15V, 25 days). The purpose of this step is to obtain multi-dimensional data of the entire life cycle of SOFC operation to provide a basis for subsequent analysis.

[0053] Step 2: Data Preprocessing Denoising (the method used is wavelet transform, and the number of decomposition layers is 5); the calculation formula is: Where, is the noise standard deviation, is the standard deviation coefficient of the noise, , is the signal length, (data points per minute); to remove sensor noise (for example, voltage jitter ±0.01V) and preserve running trends.

[0054] Normalization (the method used is linear normalization); the calculation formula is: Where, is the normalized data, is the original data, is the minimum value in the original data, is the maximum value in the original data; the parameter ranges are: temperature (600-1000℃), voltage (0V-1.2V), current (0A-50A), hydrogen flow rate (0L / min-10L / min), air flow rate (0L / min-5L / min), and pressure (0kPa-200kPa); the data are standardized to 0-1 to eliminate dimensional differences.

[0055] Feature extraction (extract voltage attenuation rate, voltage mean, voltage standard deviation and voltage peak value respectively): Among them, for the voltage attenuation rate, the calculation formula is: Where, is the voltage decay rate, is the initial voltage, is the current voltage, used to quantify the performance degradation; illustratively, ; ; .

[0056] For the voltage mean, the calculation formula is: Where, is the number of samples, is the voltage of each cell in the stack, used to indicate the overall level; for example, , the voltage data within the minute is [1.18,1.19,…,1.17], and the calculated voltage mean is 1.18V.

[0057] For the voltage standard deviation, the calculation formula is: , which is used to measure the voltage fluctuation amplitude; for example, the calculated standard deviation value is 0.008V.

[0058] For the voltage peak, the calculation formula is: , to capture abnormal voltage values; for example, the peak value within one minute is 1.19V.

[0059] Step 3: Model training (for accuracy changes during training, see Figure 3 ),exist Figure 3 The graph shows the evolution of training and validation accuracy. This curve illustrates the performance of the CNN-LSTM model during training, with the horizontal axis representing training epochs (1 to 100) and the vertical axis representing accuracy (0 to 1). The solid blue line represents training accuracy, and the dashed orange line represents validation accuracy. Both curves gradually stabilize with increasing epochs, indicating strong model learning capabilities and no significant overfitting. The grid lines and legend enhance the readability of the curves, demonstrating the convergence of model training and high performance (e.g., accuracy close to 1), providing intuitive verification of the system's prediction accuracy.

[0060] 1) For the model structure, CNN (Convolutional Neural Network): First layer: 32 3×3 convolution kernels, stride 1, padding "same", activation function ReLU; Second layer: 64 3×3 convolution kernels, stride 1, padding "same", activation function ReLU; Pooling layer: 2×2 maximum pooling, reducing the dimension of features to 30×64; Input: 60×6 (60 seconds × 6 parameters); Output: time series feature vector; LSTM (Long Short-Term Memory Network): First layer: 128 units, return sequence, Dropout rate 0.2; Second layer: 64 units, returns a single vector, Dropout rate 0.2; Input: CNN output (30×64); Output: 64-dimensional time series features; Attention Mechanism: The calculation formula is: , used to measure the importance of features; where, To calculate the score, is the LSTM output (64 dimensions), is the trainable weight vector; Output: weighted feature vector; 2) The training parameters are: input dimension is 60×6 (60 seconds × 6 parameters), batch size is 32, learning rate is 0.001, number of training rounds is 100 rounds, and loss function is mean square error (MSE), where, , where is the actual experimental value, is the predicted value, and the optimizer is ),in, is the exponential decay rate of the first-order moment estimate, is the exponential decay rate of the second-order moment estimate, is a very small number used for numerical stability to prevent division by zero in calculations; the data is divided into 80% training set (576 hours) and 20% validation set (144 hours).

[0061] 3) The training results are: Validation set performance: HSI prediction error <5%, correlation coefficient 0.95 (HSI and actual decay rate), MSE < 0.01.

[0062] Step 4: Run and optimize Among them, the monitoring mechanism: calculate: The calculation formula is: ; Among them, the parameters: 、 ; Initial weight: , ; Dynamic Adjustment: Calculate the rate of change: ; like , ; Warning conditions: Rate of change: ;Threshold: rate>0.05; Output: "Advanced Warning"; Threshold judgment: ; Output: "sound and light alarm"; Optimization strategy: When HSI < 0.9, reduce the current; When the temperature is >900℃, increase the cooling air by 10%; Optimization effect: After the temperature anomaly, the HSI rebounded to 0.88; Avoid further damage after the voltage decays.

[0063] 3. Experimental Results and Invention Effects Experimental results Case 1: Abnormal temperature; Time: Day 15, Hour 360; Scenario: Temperature increases from 800°C to 920°C, HSI decreases from 0.95 to 0.85; System response: Detected 2 hours in advance (358th hour) ; Output "Advanced Warning" and recommend increasing the cooling air flow by 10%; Result: The temperature dropped to 810℃ and HSI rebounded to 0.88, avoiding overheating damage.

[0064] Case 2: Voltage decay; Time: Day 25, 600 hours; Scenario: The voltage drops from 1.2V to 1.14V (ΔV=8%\Delta V = 8\%ΔV=8%), and HSI drops to 0.72; System response: Trigger "sound and light alarm"; It is recommended to reduce the current to 80%; Result: During the shutdown and maintenance, it was found that the electrodes were aging, which prevented further damage to the battery stack.

[0065] Based on the above exemplary embodiments, the effects of the present invention include: First, it enables high-precision monitoring. Specifically, the HSI prediction error is less than 5%. Using a CNN-LSTM model and an attention mechanism, the method of the present invention effectively and accurately predicts the health status of SOFCs, with a correlation coefficient of 0.95 with the actual decay rate. Compared to traditional threshold methods (error of 15%-20%) and single CNNs (error of 8%-10%), the method of the present invention achieves superior prediction accuracy, effectively ensuring early detection of anomalies.

[0066] Second, it can meet real-time requirements (see Figure 4 , the system responded in advance at a 50-hour temperature anomaly. Specifically, the method of the present invention has an average response time of 0.42 seconds (<0.5 seconds), enabling rapid response through hardware acceleration and algorithm optimization. Furthermore, it can provide early warning 2-4 hours in advance (for example, a 920°C anomaly can be detected 2 hours in advance). Compared to traditional methods, which can lag for several minutes, the method of the present invention offers greater real-time performance, providing ample time for maintenance.

[0067] Among them, Figure 4 The HSI changes of the present invention and conventional methods are compared. This curve shows the evolution of the SOFC health status index (HSI) over 100 hours of operation, with operating time (hours) on the horizontal axis and HSI (0 to 1) on the vertical axis. The solid line represents the prediction results of the present invention (CNN-LSTM + adaptive HSI), the dashed line represents the conventional threshold method, and the dash-dotted line represents the single CNN model. The curve simulates a scenario where the voltage decreases linearly from 1.2V (0.001V per hour) and the temperature increases from 800°C to 850°C over 50 hours. The HSI of the present invention decreases smoothly, responding to the temperature anomaly (850°C) 50 hours in advance. The conventional threshold method drops abruptly to 0.7 after 60 hours when the voltage is <1.0V or the temperature is >840°C, showing significant hysteresis. The single CNN model exhibits large fluctuations due to noise (approximately 10% error). The grid lines and legend enhance readability, demonstrating the advantages of the present invention in accuracy (smooth prediction) and real-time performance (early response), providing intuitive evidence of the technical advantages of the present invention.

[0068] Third, the method of the present invention exhibits excellent robustness. Specifically, its noise immunity is significantly improved (for example, by 10%). Furthermore, data augmentation and denoising techniques ensure that HSI prediction errors are less than 5% in complex environments (e.g., 80% humidity). This allows the method to adapt to the noise and fluctuating operating conditions of industrial scenarios, reducing false positives.

[0069] Fourth, the method of the present invention can effectively improve the lifespan and efficiency of SOFCs. Specifically, through real-time optimization (e.g., reducing current, increasing cooling), the aging rate can be slowed, effectively extending the SOFC lifespan by, for example, 15% (from 2400 hours to 2760 hours). Furthermore, the method of the present invention can optimize operating parameters to improve energy conversion efficiency, effectively increasing SOFC efficiency by, for example, 5%-10% (from 85% to 90%). This effectively extends the lifespan of the equipment and reduces maintenance costs.

[0070] Fifth, the method of the present invention enables closed-loop management. Specifically, it enables full lifecycle management, from data collection to optimization recommendations. Compared to traditional segmented monitoring, it offers a higher degree of automation and reduces manual intervention.

[0071] Sixth, the method of this invention can achieve high application value in specific fields. For example, in industrial power generation, it can effectively improve SOFC reliability and reduce downtime losses. In the aerospace field, it can provide real-time early warning to ensure the safety of critical missions. Furthermore, in many application areas, it can extend lifespan, improve efficiency, and reduce total cost of ownership.

[0072] In one embodiment of the present invention, the SOFC state monitoring and health diagnosis method based on deep learning of the present invention may further include: Step S5: In response to the early warning result, the health status of the SOFC is displayed through a visual interface and optimization suggestions are provided.

[0073] In this way, through this step S5, the health status of the SOFC can be intuitively displayed on the visual interface according to the early warning result, and targeted optimization suggestions can be provided.

[0074] In one embodiment of the present invention, step S5 of the present invention specifically includes: displaying the health status of the SOFC through a visual interface in response to the early warning result; Output the corresponding optimization suggestions according to the following formula: Where, For optimization suggestions, is the actual operating temperature of SOFC.

[0075] This dynamic feedback mechanism enables a closed loop of "monitoring-diagnosis-optimization." Specifically, when the HSI is less than 0.9, current regulation (load reduction protection) is automatically triggered, and when the temperature exceeds 900°C, the cooling system is automatically adjusted (thermal management), forming a real-time parameter optimization closed loop.

[0076] In one embodiment of the present invention, step S2 of the present invention may specifically include: Step S2-1: performing denoising processing on the collected operating data based on wavelet transform; Step S2-2: normalizing the denoised running data; Step S2-3: Perform feature extraction on the normalized operating data.

[0077] In this implementation, wavelet transforms are used for multi-scale decomposition, effectively distinguishing high-frequency noise from actual operating fluctuation signals, thereby improving the signal-to-noise ratio. Furthermore, normalization creates a unified dimension space, eliminating the interference of magnitude differences on model training.

[0078] In one embodiment of the present invention, the deep learning diagnostic network model of the present invention includes a CNN part and an LSTM part; the CNN part includes two convolutional layers, each convolutional layer is followed by a ReLU activation function, and the convolution operation is followed by a maximum pooling layer for dimensionality reduction; the LSTM part includes two LSTM layers, each LSTM layer is followed by a Dropout layer to prevent overfitting.

[0079] The structure of the deep learning diagnostic network model can be found in Figure 2 As shown, in this embodiment, for the deep learning diagnostic network model, the formula for its convolution operation is: Where, is the output value at coordinate (i, j) in the convolutional layer output feature map, They are the horizontal and vertical indices of the convolution kernel, which are used to traverse the elements of the convolution kernel. The value range here is 0 to 2, indicating that the convolution kernel size is 3×3. is the feature value at the coordinate (i+m,j+n) in the input feature map, is the weight value of the position with coordinate (mn) in the convolution kernel, is used as bias to extract spatial features.

[0080] In LSTM update, the calculation formula of the forget gate is , which controls the history retention. The calculation formula of the input gate is , which controls the addition of new information. The calculation formula for candidate states is , which is used to generate updated content. The calculation formula of the output gate is , which controls the current output. The calculation formula for updating the unit state is , which is used to record long-term memory. The calculation formula for generating the hidden state is , which is used to output the analysis results.

[0081] Where, is the output of the forget gate. It is a value between 0 and 1. It is the Sigmoid function, which is used to compress the function value to the range of 0 to 1. is the weight matrix, Indicates that the previous moment is hidden and current moment input Splice it together, is the bias term, is the cell state at time t, is the candidate state at time t, Is the output of the output gate (OutputGate).

[0082] This paper combines a convolutional neural network (CNN) with a long short-term memory (LSTM) network for the first time. The resulting design combines a two-layer CNN (32 3×3 convolution kernels in the first layer, 64 in the second) with a two-layer LSTM (128 cells in the first layer, 64 in the second). This approach specifically optimizes SOFC operating data for its multidimensional characteristics (including temperature, voltage, current, flow rate, etc.) and time-varying characteristics (performance degradation over time). The CNN extracts spatial features through convolution operations, while the LSTM analyzes time series using a gating mechanism to output a health status index (HSI).

[0083] Compared to a single CNN or traditional machine learning (SVM) approach, the hybrid model significantly improves its ability to capture the spatiotemporal correlations in SOFC data. In experiments, HSI prediction error was reduced to below 5%, with a correlation coefficient of 0.95. This compares to a single CNN error of approximately 8%-10%, and SVM, which can only process static data and has an accuracy of less than 70%. For example, when voltage decreases slowly while temperature fluctuates, the hybrid model can comprehensively analyze the long-term impact of both, whereas traditional methods may overlook temporal effects.

[0084] This innovation improves the accuracy and depth of SOFC condition monitoring, addressing the inability of traditional methods to recognize complex operating modes. For example, in distributed power generation, the system can accurately identify early signs of performance degradation and prevent large-scale failures. This technological breakthrough is unprecedented in SOFC monitoring and fills a gap in the application of deep learning in fuel cell dynamic analysis.

[0085] In one embodiment of the present invention, the attention mechanism of the present invention is implemented in the following way: Step S3-2-1: Calculate the feature vector score output by the long short-term memory network; Step S3-2-2: normalize the feature vector score; Step S3-2-3: Weight the key features to highlight their Contribution of evaluation.

[0086] In this implementation, a dynamic feature weighting system based on an attention mechanism is constructed, significantly improving the ability to capture key fault features in SOFC condition monitoring. Specifically, first, a learnable feature score calculation module is used to automatically generate attention weights using the time-series feature vectors output by the LSTM. Compared to traditional fixed weight assignment methods, this method significantly improves the accuracy of sensitive feature recognition. Second, by normalizing the feature vector scores, irrelevant noise interference can be effectively suppressed. Third, key features can be weighted, enabling dynamic weight adjustment.

[0087] Among them, key characteristics may be battery voltage, flow, temperature and pressure.

[0088] In one embodiment of the present invention, step S4 of the present invention may specifically include: Calculate according to the following formula Descent rate: Where, is the adaptive health status index at time tk, is the time interval (i.e., step size); when When the value of is greater than a first preset threshold, a first warning signal is output; when When it is less than a second preset threshold, a second warning signal is output.

[0089] Thus, in this embodiment, by The formula dynamically tracks the performance degradation trend and can improve the success rate of early monitoring of SOFC performance degradation compared to the fixed threshold monitoring method.

[0090] In one embodiment of the present invention, the present invention The value is 5, that is, , the first preset threshold is set to 0.05; and / or, the second preset threshold is set to 0.7.

[0091] According to the second aspect of the present invention, Figure 1As shown, a SOFC state monitoring and health diagnosis system based on deep learning is also provided, which is used to execute the SOFC state monitoring and health diagnosis method based on deep learning of any technical solution in the first aspect of the present invention. The SOFC state monitoring and health diagnosis system based on deep learning includes a data acquisition module, a preprocessing module, a deep learning diagnosis model and a state monitoring and early warning module. Among them, the data acquisition module is used to collect the operating data of SOFC. The preprocessing module is used to preprocess the collected operating data and perform feature extraction on the preprocessed operating data. The deep learning diagnosis model adopts a hybrid architecture of convolutional neural network and long short-term memory network, and combines the attention mechanism. The deep learning diagnosis model is used to analyze data and output an adaptive health status index The status monitoring and early warning module is used to Real-time monitoring and early warning of the data and its change rate.

[0092] In one embodiment of the present invention, the deep learning-based SOFC state monitoring and health diagnosis system may further include a visualization and feedback module. The visualization and feedback module is configured to respond to early warning results by displaying the health status of the SOFC through a visual interface and providing optimization suggestions.

[0093] The above are only specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A SOFC state monitoring and health diagnosis method based on deep learning, characterized in that: The steps include: Step S1: collecting SOFC operating data, wherein the operating data includes temperature, total stack voltage, operating current, hydrogen flow rate, oxygen flow rate and SOFC internal pressure; Step S2: pre-processing the collected operating data and extracting features from the pre-processed operating data, wherein the features include voltage decay rate, voltage mean, voltage standard deviation, and voltage peak value; Step S3: Input the pre-processed operating data into a deep learning diagnostic model, which uses a hybrid architecture of convolutional neural networks and long short-term memory networks, and combines the attention mechanism to output an adaptive health status index. ; in, The calculation process specifically includes: Step S3-1: Calculate the basic adaptive health status index : Step S3-2: Calculate the rate of change: Step S3-3: Adjust weights: Step S3-4: Calculation : Where, First, set the weight. Set the weight for the second one, is the voltage decay rate, is the SOFC operating temperature, is the optimal operating temperature of SOFC, is the maximum allowable operating temperature of SOFC, It is the adaptive health status index of the previous moment, which is used to calculate the rate of change and reflect the temporal correlation of the status; For the first adjustment weight, Adjust weights for the second; Step S4: According to and its rate of change in real time. Below a preset threshold or If the rate of change exceeds the preset threshold, an early warning is issued.

2. The SOFC state monitoring and health diagnosis method based on deep learning according to claim 1 is characterized in that: The SOFC state monitoring and health diagnosis method based on deep learning also includes: Step S5: In response to the early warning result, the health status of the SOFC is displayed through a visual interface and optimization suggestions are provided.

3. The SOFC state monitoring and health diagnosis method based on deep learning according to claim 2 is characterized in that: The step S5 specifically includes: In response to the early warning results, the health status of the SOFC is displayed through a visual interface; Output the corresponding optimization suggestions: Where, For optimization suggestions, is the actual operating temperature of SOFC.

4. The SOFC state monitoring and health diagnosis method based on deep learning according to claim 1 is characterized in that: The step S2 specifically includes: Step S2-1: performing denoising processing on the collected operating data based on wavelet transform; Step S2-2: normalizing the denoised running data; Step S2-3: Perform feature extraction on the normalized operating data.

5. The SOFC state monitoring and health diagnosis method based on deep learning according to claim 1 is characterized in that: The deep learning diagnostic network model includes a CNN part and an LSTM part; The CNN part includes two convolutional layers, each of which is followed by a ReLU activation function, and the convolution operation is followed by a maximum pooling layer for dimensionality reduction. The LSTM part includes two LSTM layers, and each LSTM layer is followed by a Dropout layer to prevent overfitting.

6. The SOFC state monitoring and health diagnosis method based on deep learning according to claim 1, characterized in that: The attention mechanism is implemented in the following way: Step S3-2-1: Calculate the feature vector score output by the long short-term memory network; Step S3-2-2: normalize the feature vector score; Step S3-2-3: Weight the key features to highlight their Contribution to evaluation.

7. The SOFC state monitoring and health diagnosis method based on deep learning according to claim 1, characterized in that: The step S4 specifically includes: Calculate according to the following formula Descent rate: Where, for Adaptive health status index at all times, is the time interval; when When the value of is greater than a first preset threshold, a first warning signal is output; when When it is less than a second preset threshold, a second warning signal is output.

8. The SOFC state monitoring and health diagnosis method based on deep learning according to claim 7 is characterized in that: described The value is 5, and the first preset threshold is set to 0.05; and / or, The second preset threshold is set to 0.

7.

9. A SOFC state monitoring and health diagnosis system based on deep learning, characterized in that: For executing the SOFC state monitoring and health diagnosis method based on deep learning according to any one of claims 1 to 8, the SOFC state monitoring and health diagnosis system based on deep learning comprises: Data acquisition module, used to collect SOFC operation data; A preprocessing module is used to preprocess the collected operation data and perform feature extraction on the preprocessed operation data; A deep learning diagnostic model, using a hybrid architecture of convolutional neural networks and long short-term memory networks, combined with an attention mechanism, is used to analyze data and output an adaptive health status index. ; Condition monitoring and early warning module, used to Real-time monitoring and early warning of the data and its change rate.

10. The deep learning-based SOFC state monitoring and health diagnosis system according to claim 9, characterized in that: The deep learning-based SOFC state monitoring and health diagnosis system also includes: The visualization and feedback module is used to respond to the early warning results, display the health status of the SOFC through a visual interface, and provide optimization suggestions.

Images