Method and device for identifying combustion state of solid oxide fuel cell

Abstract

The invention provides a solid oxide fuel cell combustion state identification method and device, and belongs to the field of fuel cell safety, and the method comprises the steps: inputting an after-combustion chamber flame image of a to-be-detected solid oxide fuel cell into a combustion state identification network which is completely trained, performing coarse-grained feature extraction on the post-combustion chamber flame image to obtain coarse-grained features, performing fine-grained feature extraction on the coarse-grained features to obtain fine-grained features, and performing adaptive feature alignment and feature fusion on the coarse-grained features and the fine-grained features to obtain fusion features; and performing classification prediction output on the fusion features to obtain a combustion state recognition result. According to the method, overall features including shape information are captured through coarse-grained feature extraction, detail features including texture information are captured through fine-grained feature extraction, and important features of each level are integrated through adaptive feature alignment and feature fusion, so that the discrimination capability of the network is enhanced, and the robustness of the network is improved. And accurate identification of the flame combustion state of the afterburner is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cell safety, and particularly to a method and device for identifying the combustion state of a solid oxide fuel cell. Background Art

[0002] A solid oxide fuel cell is a highly efficient energy conversion system that can convert chemical energy into electrical energy. At the same time, it can also be combined with various cycle systems and gas turbines or steam turbines to achieve an energy conversion efficiency of up to 70%. In the design of a solid oxide fuel cell, the afterburner is one of the key components. It is not only used for waste heat recovery to enhance the overall energy efficiency of the system but also plays a core role in regulating the thermal environment of the solid oxide fuel cell. However, the fluctuations of the high-temperature flue gas inside the afterburner and the possible flashback or blowout phenomena during the combustion process pose challenges to the safety and thermoelectric stability of the system. Therefore, during the operation of a solid oxide fuel cell, accurately detecting and judging the combustion state of the afterburner in real time is the basis for ensuring safe operation and thermoelectric stability.

[0003] In the related research on judging the flame combustion state, the traditional temperature sensing method relies relatively on single temperature data and actually cannot comprehensively and accurately reflect the flame combustion condition inside the afterburner. In particular, only setting a temperature threshold cannot effectively predict phenomena such as flameout and flame backflow that may occur during the combustion process. Given the limitations of the traditional method, non-invasive image processing methods have been widely used. For example, in existing research, there are studies on the combustion characteristics of a cylindrical premixed porous burner through digital images captured by a CCD (Charge Coupled Device) camera and color processing technology. By observing the change in flame color, the optimal temperature combustion conditions are determined, and certain effects have been achieved. However, in the judgment of the combustion state of the afterburner of a solid oxide fuel cell, compared with the color change characteristics, the change in the combustion state is often more intuitively reflected in the shape and texture of the flame. For example, when the combustion state of the flame is changing, at this time, the overall shape and detailed texture of the flame will first show subtle changes, and then the flame color will change with the change in the proportion of combustion gas and temperature. However, in the existing combustion state recognition, it often only focuses on the color characteristics of the flame and ignores the shape characteristics and texture information of the combustion image, resulting in a low recognition accuracy.

[0004] Therefore, the prior art has the technical problem of ignoring the shape information and texture information in the afterburner flame combustion image, resulting in a low recognition accuracy, and needs to be improved. Summary of the Invention

[0005] In view of this, it is necessary to provide a method and device for identifying the combustion state of a solid oxide fuel cell to solve the technical problem in the prior art that the shape information and texture information in the combustion image of the afterburner are ignored, resulting in low recognition accuracy.

[0006] In a first aspect, the present invention provides a method for identifying the combustion state of a solid oxide fuel cell, including: Input the afterburner flame image of the solid oxide fuel cell to be measured into a trained combustion state recognition network, extract coarse-grained features from the afterburner flame image to obtain coarse-grained features, extract fine-grained features from the coarse-grained features to obtain fine-grained features, and perform adaptive feature alignment and feature fusion on the coarse-grained features and fine-grained features to obtain fused features; Perform classification prediction on the fused features to output a combustion state recognition result.

[0007] In some possible implementation manners, extracting coarse-grained features from the afterburner flame image to obtain coarse-grained features includes: Performing continuous residual feature convolutions on the afterburner flame image to obtain coarse-grained features.

[0008] In some possible implementation manners, extracting fine-grained features from the coarse-grained features to obtain fine-grained features includes: Performing continuous residual feature convolutions and feature downsampling on the coarse-grained features to obtain fine-grained features.

[0009] In some possible implementation manners, performing adaptive feature alignment and feature fusion on the coarse-grained features and fine-grained features to obtain fused features includes: Performing adaptive average pooling and feature convolution on the coarse-grained features to obtain coarse-grained aligned features; Performing feature convolution and bilinear interpolation upsampling on the fine-grained features to obtain fine-grained aligned features; Performing convolution fusion on the coarse-grained aligned features and fine-grained aligned features to obtain fused features.

[0010] In some possible implementation manners, performing classification prediction on the fused features to output a combustion state recognition result includes: Performing adaptive average pooling, flattening operation, and fully connected layer output on the fused features in sequence to obtain a combustion state recognition result.

[0011] In some possible implementation manners, the trained combustion state recognition network is obtained by training an initial combustion state recognition network. Training the initial combustion state recognition network includes: Obtaining combustion state training data; Input the combustion state recognition training data into the initial combustion state recognition network to obtain the combustion state prediction output. Determine the cross-entropy loss based on the combustion state prediction output, and iteratively optimize the initial combustion state recognition network according to the cross-entropy loss until the model performance meets the preset requirements.

[0012] In some possible implementation manners, obtaining the combustion state recognition training data includes: Collect combustion images of the combustion chamber in a preset experimental environment under different working conditions to obtain initial combustion images; Perform data augmentation processing on the initial combustion images to obtain the combustion state recognition training data; Wherein, the preset experimental environment combustion chamber is the afterburner simulation environment of the solid oxide fuel cell to be measured, and the data augmentation processing includes image cropping, image transformation, and adding noise.

[0013] In a second aspect, the present invention provides a solid oxide fuel cell combustion state recognition device, including: A feature extraction unit, configured to input the afterburner flame image of the solid oxide fuel cell to be measured into the trained complete combustion state recognition network, perform coarse-grained feature extraction on the afterburner flame image to obtain coarse-grained features, perform fine-grained feature extraction on the coarse-grained features to obtain fine-grained features, and perform adaptive feature alignment and feature fusion on the coarse-grained features and the fine-grained features to obtain fused features; An identification output unit, configured to perform classification prediction output on the fused features to obtain the combustion state recognition result.

[0014] In a third aspect, the present invention provides a fuel cell detection device, including a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above-mentioned solid oxide fuel cell combustion state recognition method is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned solid oxide fuel cell combustion state recognition method is implemented.

[0016] The beneficial effects of adopting the above embodiments are as follows: In the solid oxide fuel cell combustion state recognition method provided by the present invention, the overall features containing flame shape information are captured through coarse-grained feature extraction, the detailed features containing flame texture information are captured through fine-grained feature extraction, and the coarse-grained features and the fine-grained features are subjected to adaptive feature alignment and feature fusion to retain and integrate the important features of each level, thereby enhancing the discrimination ability of the network and enabling accurate recognition of the combustion state of the afterburner flame. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those skilled in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0018] Figure 1 It is a schematic flowchart of an embodiment of the method for identifying the combustion state of a solid oxide fuel cell proposed in the embodiment of the present invention; Figure 2 It is a schematic flowchart of the adaptive feature alignment and feature fusion in the embodiment of the present invention; Figure 3 It is a schematic flowchart of training the initial combustion state recognition network in the embodiment of the present invention; Figure 4 It is a schematic flowchart of step S301 in the embodiment of the present invention; Figure 5 It is a schematic structural diagram of an embodiment of the device for identifying the combustion state of a solid oxide fuel cell proposed in the embodiment of the present invention; Figure 6 It is a schematic structural diagram of an embodiment of the fuel cell detection device proposed in the embodiment of the present invention. Detailed implementation manners

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0020] It should be understood that the schematic drawings are not drawn to scale. The flowcharts used in the present invention illustrate the operations implemented according to some embodiments of the present invention. It should be understood that the operations in the flowchart may not be implemented in sequence, and steps without logical context relationships may be reversed or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of the present invention. Some of the block diagrams shown in the drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor systems and / or microcontroller systems.

[0021] In the embodiments of the present invention, the descriptions such as "first" and "second" are only for descriptive purposes, and cannot be construed as indicating or implying their relative importance or implicitly specifying the quantity of the indicated technical features. Therefore, the technical features defined with "first" and "second" may explicitly or implicitly include at least one such feature.

[0022] Reference to "embodiment" in this text means that the specific features, structures, or characteristics described in connection with the embodiment may be included in at least one embodiment of the present invention. The phrase appearing in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein may be combined with other embodiments.

[0023] The present invention provides a method and device for identifying the combustion state of a solid oxide fuel cell, which will be described separately below.

[0024] Figure 1 It is a schematic flowchart of an embodiment of the method for identifying the combustion state of a solid oxide fuel cell proposed in the embodiments of the present invention. As Figure 1 shown, the method for identifying the combustion state of a solid oxide fuel cell includes: S101. Input the post-combustion chamber flame image of the solid oxide fuel cell to be measured into a trained combustion state recognition network, perform coarse-grained feature extraction on the post-combustion chamber flame image to obtain coarse-grained features, perform fine-grained feature extraction on the coarse-grained features to obtain fine-grained features, and perform adaptive feature alignment and feature fusion on the coarse-grained features and the fine-grained features to obtain fusion features; It should be noted that the combustion state recognition network is a recognition network based on a neural network that can identify the combustion state of the post-combustion chamber flame image. Being trained completely means that after the combustion state recognition network is trained on the training data set, the recognition effect on the test set reaches the preset accuracy requirement.

[0025] Among them, in order to extract the shape information and texture information that can effectively distinguish the flame combustion state, the embodiment sets coarse-grained feature extraction and fine-grained feature extraction to respectively extract the coarse-grained shape information and fine-grained texture information in the image. Among them, the main structures of the coarse-grained feature extraction and the fine-grained feature extraction are both composed of several residual blocks. The difference lies in the number of input and output channels of the residual blocks and the convolution kernel parameters, so as to adapt to different levels of feature extraction; the adaptive feature alignment performs different feature alignment processing on the features of different levels of the coarse-grained and fine-grained according to the characteristics of the features to be retained, so as to enhance the corresponding feature representation, and finally the aligned features are fused and used as the input of the classifier.

[0026] S102. Classify and predict the fused features to output the combustion state recognition result.

[0027] It should be noted that adaptive feature alignment refers to performing corresponding feature processing steps on the coarse-grained features and fine-grained features respectively for the feature levels to be retained as needed, so as to align the sizes of the feature images, retain important features, and be used for subsequent fusion operations.

[0028] Among them, the fused features are feature images that retain the flame combustion shape information and texture information. The classification prediction is based on a designed classifier, which includes operations such as pooling processing and feature flattening, aiming to convert the fused features into a format suitable for classification; the combustion state recognition result includes three categories: "flashback limit", "stable combustion", and "flame lift-off" in the embodiment. The embodiment determines the final judged combustion state recognition result by outputting the classification probabilities of different categories.

[0029] Compared with the prior art, in the solid oxide fuel cell combustion state recognition method provided by the embodiment of the present invention, the overall features containing flame shape information are captured through coarse-grained feature extraction, the detailed features containing flame texture information are captured through fine-grained feature extraction, and the coarse-grained features and fine-grained features are subjected to adaptive feature alignment and feature fusion to retain and integrate the important features of each level, thereby enhancing the discriminative ability of the network and enabling accurate recognition of the combustion state of the afterburner flame.

[0030] In some embodiments of the present invention, the extraction of coarse-grained features from the afterburner flame image includes: Performing continuous residual feature convolutions on the afterburner flame image to obtain coarse-grained features.

[0031] Among them, the coarse-grained feature extraction stage includes two BasicBlock residual blocks. The first residual block maintains the size and depth of the image, and the second residual block doubles the number of channels and halves the size of the feature map to 45×45×128 to extract the rough features of the flame image, including but not limited to the shape contour of the flame, the overall color of the flame, and the combustion area of the flame, etc. Each BasicBlock residual block sequentially includes batch normalization, activation function, residual convolution, batch normalization, and residual convolution processing steps. The sizes of the two residual convolution kernels in the first residual block are both 3×3, the size of the first residual convolution kernel in the second residual block is 3×3, and the size of the second residual convolution kernel is 1×1.

[0032] Preferably, before the coarse-grained feature extraction stage, a preprocessing step for image features is also set. In the preprocessing stage, the input image is first processed by a convolutional layer equipped with 64 filters, followed by batch normalization, ReLU activation function, and max pooling layer to extract a high-level abstract representation of the image, laying the foundation for coarse-grained feature extraction. The preprocessing stage starts from an input image of 360×360×3. First, the convolutional layer reduces the feature map to 180×180×64, then batch normalization and ReLU activation are performed to keep the size unchanged, and finally, max pooling further reduces it to 90×90×64.

[0033] In some embodiments of the present invention, fine-grained features are obtained by performing fine-grained feature extraction on the coarse-grained features, including: Performing continuous several times of residual feature convolution and feature downsampling on the coarse-grained features to obtain fine-grained features.

[0034] Among them, in the fine-grained feature extraction stage, it includes two BasicBlock residual block structures similar to those in the coarse-grained feature extraction stage. However, different from the coarse-grained feature extraction stage, to ensure consistent feature dimensions, a downsampling module is added to each BasicBlock residual block in the fine-grained feature extraction stage. In fine-grained extraction, the first BasicBlock residual block reduces the feature map to 23×23×256, and the second BasicBlock residual block reduces the feature map to 12×12×512, thereby extracting the detailed features of the flame image, including but not limited to the texture of the flame and the distribution details of the flame color.

[0035] To further strengthen and integrate important features at different levels, in some embodiments of the present invention, Figure 2 is a schematic flow diagram of the adaptive feature alignment and feature fusion for the embodiments of the present invention. As Figure 2 shown, performing adaptive feature alignment and feature fusion on the coarse-grained features and fine-grained features to obtain fused features, including: S201. Performing adaptive average pooling and feature convolution on the coarse-grained features to obtain coarse-grained aligned features; S202. Performing feature convolution and bilinear interpolation upsampling on the fine-grained features to obtain fine-grained aligned features; S203. Performing convolutional fusion on the coarse-grained aligned features and fine-grained aligned features to obtain fused features.

[0036] Among them, the feature fusion step aims to integrate the coarse-grained features and fine-grained features extracted in the early stage of the network. Before feature fusion, it is necessary to perform an alignment operation on the features.

[0037] For the coarse-grained features, in the embodiment, an adaptive average pooling layer is adopted to compress the feature map from the rough feature layer to a size of 1×1. Subsequently, a 1×1 convolutional layer is used to increase the number of channels from 128 to 256, completing the preliminary feature transformation. In this process, the adaptive average pooling can adaptively pool each position in the input and suppress some useless features. Through the adaptive average pooling and the convolutional layer, the embodiment can more effectively retain the contour shape features of the flame in the image during the feature alignment process.

[0038] For the fine-grained features, in the embodiment, first, a 1×1 convolutional layer is used to reduce the dimension from 512 channels to 256 channels, and then upsampling is performed by means of bilinear interpolation to expand the size to match the size of other feature maps. In this process, the bilinear interpolation can effectively remove the jaggedness in the feature map, enhance the smoothness of the features, and retain and highlight the texture information of the flame in the fine-grained features.

[0039] Finally, in the fusion stage, the aligned feature maps are fused after passing through another 1×1 convolutional layer. This convolutional layer maintains 512 channels, ensuring the integrity and efficiency of the information. The fusion process integrates the important features of different levels, enhances the representation ability of the network, and enables it to identify complex patterns with high accuracy. The formula for the fusion process is expressed as:

[0040] where, represents the fusion operation. For example, in this embodiment, the method of concatenating and then performing convolution is adopted to merge and refine the feature maps into a unified representation, thereby enhancing the discriminative ability of the network.

[0041] In some embodiments of the present invention, the fused features are classified and predicted to output the combustion state recognition result, including: The fused features are sequentially subjected to adaptive average pooling, flattening operation, and fully connected layer output to obtain the combustion state recognition result.

[0042] where, in the prediction output stage, the embodiment is implemented through a classifier. The classifier is a simple but powerful sequential module, which includes an adaptive average pooling layer, a flattening operation, and a fully connected layer in this step, aiming to convert the fused features into a format suitable for classification. The classifier outputs the probability that the input image belongs to a certain category, and the formula is expressed as:

[0043] where, is the flattened feature vector, represents the weights of the fully connected layer, are the parameters of the classifier, It is the output result, and then the softmax function converts these output results into the probabilities of each category to obtain the final classification output result.

[0044] In some embodiments of the present invention, Figure 3 It is a schematic flowchart for training the initial combustion state recognition network of the embodiments of the present invention. As Figure 3 shown, training the initial combustion state recognition network includes: S301. Obtain combustion state training data; S302. Input the combustion state recognition training data into the initial combustion state recognition network to obtain the combustion state prediction output, determine the cross-entropy loss according to the combustion state prediction output, and iteratively optimize the initial combustion state recognition network according to the cross-entropy loss until the model performance meets the preset requirements.

[0045] Among them, during the network training process, the embodiment first collects and obtains the combustion images of the afterburner of the solid oxide fuel cell and the corresponding combustion states to form the combustion state training data required for training the network. By inputting the combustion images into the initial combustion state recognition network for recognition and prediction, the combustion state prediction output is obtained, and the cross-entropy loss is calculated between the combustion state prediction output and the true class label. The network parameters are trained according to the cross-entropy loss through backpropagation. The cross-entropy loss formula is expressed as:

[0046] Among them, represents the one-hot encoding of the true label (if the sample belongs to the th class, then , otherwise it is 0), is the probability that the model predicts the sample belongs to the th class. This function calculates the loss by summing the logarithms of the probabilities of the labels, aiming to minimize the negative logarithm probability of the actual class, so that the prediction of the model is as close as possible to the true label.

[0047] When the model converges, the embodiment calculates the corresponding evaluation indicators in the validation set to save the best model parameters and prevent overfitting of the model at the same time.

[0048] To more objectively evaluate the performance of the network model, the embodiments adopt four evaluation metrics: Accuracy, Precision, Recall, and F1 score. Among them, Accuracy describes the proportion of samples correctly predicted by the model, which is the most intuitive performance metric but may not be comprehensive enough in the case of data imbalance; Precision represents the proportion of samples actually being positive among all samples predicted as positive by the model. A high Precision means fewer false positives; Recall is the proportion of samples actually being positive among all samples correctly predicted as positive by the model. A high Recall means fewer False Negatives (FN). The F1 score is the harmonic mean of Precision and Recall, which can take into account the balance between Precision and Recall. The higher the values of the four metrics, the better the accuracy of the method. The definitions of the four metrics are as follows:

[0049]

[0050]

[0051]

[0052] Among them, represents the correct positive example, represents the correct negative example, represents the wrong positive example, represents the wrong negative example, represents the number of image types. The embodiments test the model on the validation set with the above four metrics to save the optimal model parameters.

[0053] In some embodiments of the present invention, as Figure 4 shown, step S301 includes: S401. Collect combustion images of the combustion chamber in a preset experimental environment under different working conditions to obtain initial combustion images; S402. Perform data augmentation processing on the initial combustion images to obtain combustion state recognition training data; Among them, the preset experimental environment combustion chamber is the simulation environment of the afterburner of the solid oxide fuel cell to be measured, and the data augmentation processing includes image cropping, image transformation, and adding noise.

[0054] Among them, as a preferred embodiment of step S301, in the process of obtaining training data, this solution collects and obtains more high-quality training data through simulation experiments.

[0055] To identify the afterburner flame combustion state of a solid oxide fuel cell under different operating conditions, the embodiment designs an independent afterburner experimental system, which is separated from the solid oxide fuel cell system for research. The independent afterburner experimental system includes four main parts: a gas supply area, a preheating and temperature-rising area, a combustion area, and an exhaust gas treatment area.

[0056] In the gas supply area, the mixed gas (CH4, H2, CO, N2, and CO2) is controlled in flow rate by a mass flowmeter and delivered to a gas preheating furnace, while air is supplied after passing through an air compressor and a dryer. Given that the experiment involves flammable gases, each gas pipeline is equipped with a one-way valve to ensure the safety of the system.

[0057] In the preheating and temperature-rising area, the gas and air are heated separately in two independent preheating furnaces to ensure experimental safety. The maximum set temperature of the air preheating furnace is 500 °C, while the gas heating furnace can reach 650 °C. Through an accurate temperature control system, the heating temperature error is guaranteed to be controlled within ±2% to ensure the accuracy of the experiment.

[0058] In the combustion area, the preheated mixed gas burns in the combustion chamber, and the ratio of air to gas and the water vapor content are adjusted to adapt to different operating conditions. In the embodiment, the combustion area uses a metal fiber burner with a porosity of 79.2%, a diameter of 50 mm, and a wall thickness of 0.6 mm.

[0059] In the exhaust gas treatment area, the combustion exhaust gas is cooled and then sent to a flue gas analyzer for analysis to evaluate the combustion efficiency. The embodiment uses a Testo-340 flue gas analyzer to measure various gas components such as CO, NOX, O2, and CO2. The measurement accuracy of CO is ±10 ppm, and the measurement accuracy of NOX is ±5 ppm.

[0060] The core of the afterburner experimental system is a combustion chamber furnace with a diameter of 500 mm and a height of 600 mm, equipped with a metal fiber burner. The furnace is wrapped with 150 mm thick thermal insulation material to maintain thermal efficiency and prevent the outer shell from overheating. The mixed gas enters the system after being fully mixed through a static mixer, which is adjacent to the cold air inlet to provide the necessary air flow to support the combustion process and regulate the combustion chamber temperature. An ignition device is installed below the combustion chamber to trigger combustion. Three temperature sensors are arranged along the vertical axis in the combustion chamber at intervals of 150 mm. The first temperature sensor is about 100 mm above the outlet to monitor the temperature of the experimental process at different levels. These temperature data are fed back to a PLC (Programmable Logic Controller) temperature control system and monitored and adjusted through a connected computer. A high-temperature industrial camera is installed at the top of the combustion chamber to observe and record the flame state in the porous medium burner in real time. Finally, the exhaust gas is discharged from the system through the exhaust port at the top. To ensure safety, the experiment is carried out under preheating adjustment. Before the experiment starts, the gas system is checked for airtightness, and the gas pressure is adjusted to below 0.2 MPa. After the furnace is preheated to the set temperature, the furnace and the combustion chamber are purged with nitrogen at a flow rate of 5 L / min for 5 minutes. Subsequently, methane is ignited and gradually replaced with syngas to the target operating conditions.

[0061] During the data acquisition process, the embodiments perform simulation experiments in the afterburner experimental system according to a number of preset different operating conditions and acquire the flame combustion images in the combustion chamber. The high-temperature industrial camera is used to record the flame combustion conditions under different operating conditions. After the furnace temperature is stable for 15 minutes and the burner burns continuously for 20 minutes, the system is regarded as reaching a stable state. At this time, the limit equivalence ratio and flame propagation speed of unstable combustion are recorded.

[0062] For the acquired images, the embodiments also perform relevant data enhancement processing steps to enrich the data volume. First, the resolution of the acquired original flame images is 1920×1920 (RGB). To eliminate the influence of different image sizes, we adjust the size of all images to 360×360. To enhance the fitting ability for different angles, scales, positions and noises, the image data is processed through data enhancement calculations, including but not limited to: Image rotation: Rotate the image at a random angle between ±8°; Image scaling: Rescale the image at a rate between 0.9 and 1.1; Shifting and flipping: Randomly crop the image to 240×240 and flip it with a probability of 50%; Adding noise: Add Gaussian noise with a mean of zero and a variance of 0.9 to each pixel value.

[0063] After processing with the above data augmentation method, a dataset containing 4,400 non - consecutive frame images was obtained, and it was divided into a training set, a validation set, and a test set according to the ratio of 7:1.5:1.5. The labels of the dataset include three combustion states: "tempering limit", "stable combustion", and "flame lift - off".

[0064] In summary, in order to accurately identify the combustion state of the after - combustion chamber of a solid oxide fuel cell, the present invention captures the overall features containing flame shape information through coarse - grained feature extraction, captures the detailed features containing flame texture information through fine - grained feature extraction, and performs adaptive feature alignment and feature fusion on the coarse - grained features and fine - grained features to retain and integrate the important features at each level, thereby enhancing the discriminative ability of the network and enabling accurate identification of the combustion state of the after - combustion chamber flame.

[0065] To better implement the method for identifying the combustion state of a solid oxide fuel cell in the embodiments of the present invention, correspondingly, the embodiments of the present invention also provide a device for identifying the combustion state of a solid oxide fuel cell. As Figure 5 shown, the device 500 for identifying the combustion state of a solid oxide fuel cell includes: A feature extraction unit 501, configured to input the after - combustion chamber flame image of the solid oxide fuel cell to be measured into a well - trained combustion state recognition network, perform coarse - grained feature extraction on the after - combustion chamber flame image to obtain coarse - grained features, perform fine - grained feature extraction on the coarse - grained features to obtain fine - grained features, and perform adaptive feature alignment and feature fusion on the coarse - grained features and fine - grained features to obtain fused features; An identification and output unit 502, configured to perform classification prediction on the fused features and output the combustion state recognition result.

[0066] The device 500 for identifying the combustion state of a solid oxide fuel cell provided in the above - mentioned embodiment can implement the technical solutions described in the embodiment of the method for identifying the combustion state of a solid oxide fuel cell. For the specific implementation principles of the above - mentioned modules or units, reference can be made to the corresponding content in the embodiment of the method for identifying the combustion state of a solid oxide fuel cell, which will not be elaborated here.

[0067] As Figure 6 shown, the present invention also correspondingly provides a fuel cell detection device 600. The fuel cell detection device 600 includes a processor 601, a memory 602, and a display 603. Figure 6 Only some components of the fuel cell detection device 600 are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0068] The memory 602 can be an internal storage unit of the fuel cell detection device 600 in some embodiments, such as the hard disk or memory of the fuel cell detection device 600. The memory 602 can also be an external storage device of the fuel cell detection device 600 in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the fuel cell detection device 600.

[0069] The processor 601 can be a Central Processing Unit (CPU), a microprocessor or other data processing chips in some embodiments, and is used to run the program code stored in the memory 602 or process data, such as the method for identifying the combustion state of a solid oxide fuel cell in the present invention.

[0070] The display 603 can be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) toucher, etc. in some embodiments. The display 603 is used to display the information of the fuel cell detection device 600 and to display a visual user interface. The components 601 - 603 of the fuel cell detection device 600 communicate with each other through a system bus.

[0071] In some embodiments of the present invention, when the processor 601 executes the program for identifying the combustion state of a solid oxide fuel cell in the memory 602, the following steps can be implemented: Input the post-combustion chamber flame image of the solid oxide fuel cell to be measured into a well-trained combustion state recognition network, perform coarse-grained feature extraction on the post-combustion chamber flame image to obtain coarse-grained features, perform fine-grained feature extraction on the coarse-grained features to obtain fine-grained features, and perform adaptive feature alignment and feature fusion on the coarse-grained features and the fine-grained features to obtain fused features; Perform classification prediction on the fused features and output the combustion state recognition result.

[0072] It should be understood that when the processor 601 executes the program for identifying the combustion state of a solid oxide fuel cell in the memory 602, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the relevant method embodiments above.

[0073] On the other hand, the embodiments of the present invention also provide a computer-readable storage medium, which is used to store computer-readable programs or instructions. When the programs or instructions are executed by a processor, the steps or functions in the method for identifying the combustion state of a solid oxide fuel cell provided in the above method embodiments can be implemented.

[0074] Those skilled in the art can understand that all or part of the processes for implementing the methods of the above embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a disk, an optical disc, a read-only memory, or a random access memory, etc.

[0075] The above has introduced in detail the method and device for identifying the combustion state of a solid oxide fuel cell provided by the present invention. Specific examples are used herein to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for identifying the combustion state of a solid oxide fuel cell, characterized in that: include: Inputting an afterburner flame image of a solid oxide fuel cell to be tested into a well-trained combustion state recognition network, performing coarse-grained feature extraction on the afterburner flame image to obtain coarse-grained features, performing fine-grained feature extraction on the coarse-grained features to obtain fine-grained features, and performing adaptive feature alignment and feature fusion on the coarse-grained features and the fine-grained features to obtain fused features; The fusion features are classified, predicted and output to obtain a combustion state recognition result.

2. The solid oxide fuel cell combustion state identification method according to claim 1, characterized in that: The step of extracting coarse-grained features from the afterburner flame image to obtain coarse-grained features includes: The residual feature convolution is performed on the afterburner flame image several times continuously to obtain a coarse-grained feature.

3. The solid oxide fuel cell combustion state identification method according to claim 1, characterized in that: The step of extracting fine-grained features from the coarse-grained features to obtain fine-grained features includes: The coarse-grained features are subjected to continuous residual feature convolution and feature downsampling for several times to obtain fine-grained features.

4. The solid oxide fuel cell combustion state identification method according to claim 1, characterized in that: The step of adaptively aligning and fusing the coarse-grained features and the fine-grained features to obtain fused features includes: Performing adaptive average pooling and feature convolution on the coarse-grained features to obtain coarse-grained alignment features; Performing feature convolution and bilinear interpolation upsampling on the fine-grained features to obtain fine-grained alignment features; The coarse-grained alignment feature and the fine-grained alignment feature are convolutionally fused to obtain a fused feature.

5. The solid oxide fuel cell combustion state identification method according to claim 1, characterized in that: The classifying, predicting and outputting the fusion features to obtain the combustion state recognition result includes: The fusion features are subjected to adaptive average pooling, flattening operation and full connection layer output in sequence to obtain the combustion state recognition result.

6. The solid oxide fuel cell combustion state identification method according to claim 1, characterized in that: The fully trained combustion state recognition network is obtained by training an initial combustion state recognition network, and the trained initial combustion state recognition network includes: Obtain combustion state training data; The combustion state recognition training data is input into an initial combustion state recognition network to obtain a combustion state prediction output, a cross entropy loss is determined according to the combustion state prediction output, and the initial combustion state recognition network is iteratively optimized according to the cross entropy loss until the model performance meets the preset requirements.

7. The solid oxide fuel cell combustion state identification method according to claim 6, characterized in that: The step of acquiring combustion state recognition training data comprises: Collecting combustion images of a combustion chamber in a preset experimental environment under different working conditions to obtain an initial combustion image; Performing data enhancement processing on the initial combustion image to obtain combustion state recognition training data; The preset experimental environment combustion chamber is a simulated environment of an afterburner of a solid oxide fuel cell to be tested, and the data enhancement processing includes image cropping, image transformation and noise addition.

8. A solid oxide fuel cell combustion state identification device, characterized in that: include: A feature extraction unit is used to input the afterburner flame image of the solid oxide fuel cell to be tested into a well-trained combustion state recognition network, perform coarse-grained feature extraction on the afterburner flame image to obtain coarse-grained features, perform fine-grained feature extraction on the coarse-grained features to obtain fine-grained features, and perform adaptive feature alignment and feature fusion on the coarse-grained features and the fine-grained features to obtain fused features; The recognition output unit is used to classify, predict and output the fusion features to obtain a combustion state recognition result.

9. A fuel cell detection device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the solid oxide fuel cell combustion state identification method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the solid oxide fuel cell combustion state identification method according to any one of claims 1 to 7 is implemented.

Images