Training method for pulverized coal flame form prediction model and pulverized coal flame form prediction method

Abstract

To accurately and promptly predict a pulverized coal flame combustion form.SOLUTION: In regard to a technical field of combustion analysis, a training method for a pulverized coal flame prediction model and a pulverized coal flame form prediction method are disclosed. The training and prediction methods include: acquiring a sample pulverized coal flame image, a sample combustion parameter and a sample flame form type; pretreating the sample pulverized coal flame image and constructing a training set and a validation set; inputting the flame image to an initial pulverized coal flame form prediction model constructed beforehand and extracting main features of the flame image; constructing a feature vector by combining the main features with a training combustion parameter, performing dispersion analysis of a training flame form different from the feature vector, and sorting an influence parameter; and training the initial pulverized coal flame form prediction model by using the influence parameter, the training flame form and the validation set and obtaining a target pulverized coal flame form prediction model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the technical field of combustion analysis, and in particular to a method for training a pulverized coal flame morphology prediction model and a method for predicting pulverized coal flame morphology. [Background technology]

[0002] The combustion conditions of a coal-fired boiler have a significant impact on complete fuel combustion and pollutant emissions. Good combustion conditions mean that pulverized coal can achieve complete combustion under appropriate temperature, oxygen concentration, and mixing conditions, thereby promoting complete combustion and reducing the generation of harmful substances such as carbon monoxide, particulate matter, and volatile organic compounds. Flame combustion morphology is one of the direct indicators of a boiler's combustion conditions and reflects the effectiveness and stability of fuel combustion. A rational and stable flame combustion morphology generally indicates good combustion conditions, effectively burning fuel and converting it into thermal energy, and sufficiently transferring this thermal energy to the boiler walls or other heat-dissipating surfaces, thereby uniformly distributing the heat load and improving the boiler's thermal efficiency. In contrast, an unstable flame morphology can lead to flame flashover and instability, shortening the equipment's service life and increasing the risk of accidents. Therefore, accurately predicting pulverized coal flame combustion morphology is essential for evaluating combustion efficiency, controlling pollutant emissions, and monitoring safety and stability.

[0003] In related technology, flame monitoring equipment such as cameras and infrared cameras installed in thermal power plants can monitor and record flame combustion conditions in real time. These equipment are usually installed near the combustion area to capture local combustion images of pulverized coal flames. However, these local images do not clearly reflect the overall flame combustion morphology in the furnace and require judgment based on a combination of relevant theoretical knowledge and operator experience. Pulverized coal flame morphology predictions are usually performed by specialized personnel who observe local flame morphology images and then make judgments based on relevant theoretical knowledge and experience. This prediction is prone to subjective errors, has difficulty capturing subtle changes, and is unable to handle large-scale data and complex situations. Furthermore, it requires high training costs for human resources. Summary of the Invention

[0004] In view of this, the present invention provides a method for training a pulverized coal flame morphology prediction model and a method for predicting pulverized coal flame morphology, in order to solve the problem of high dependency on operator experience and insufficient accuracy of flame monitoring.

[0005] In a first aspect, the present invention provides a method for training a pulverized coal flame morphology prediction model, the method comprising: obtaining a sample pulverized coal flame image, sample combustion parameters corresponding to the sample pulverized coal flame image, and a sample flame morphology type; preprocessing the sample pulverized coal flame image to obtain a preprocessed flame image; Constructing a training set and a validation set using the preprocessed flame images, corresponding sample combustion parameters, and sample flame morphology types as samples, wherein the training set includes training flame images, training combustion parameters, and training flame morphologies corresponding to the training flame images, and the validation set includes validation flame images, validation combustion parameters, and validation flame morphologies corresponding to the validation flame images; Inputting the training flame image into a pre-constructed initial pulverized coal flame morphology prediction model and extracting key features of the training flame image; Combining the main features with the training combustion parameters to construct a feature vector, and performing an analysis of variance on the feature vector and different training flame forms to select influential parameters, the feature vector is used to characterize each sample feature parameter, and the influential parameters are used to characterize multiple training combustion parameters that have the greatest impact on distinguishing the training flame forms; training an initial pulverized coal flame morphology prediction model using the influence parameters, the training flame morphology and the verification flame image, and the verification combustion parameters and the verification flame morphology corresponding to the verification flame image, to obtain a target pulverized coal flame morphology prediction model, which is used to perform pulverized coal flame morphology prediction.

[0006] In this invention, the quality and uniformity of the collected pulverized coal flame images are improved by preprocessing the pulverized coal flame images. Then, a training set and a sample set are constructed to train a pulverized coal flame morphology prediction model. The model's ability to extract flame image features is trained to enable the model to better extract the main features of the flame images. Next, by combining combustion parameters such as the characteristics of pulverized coal, combustion conditions, and the structure of combustion equipment that cause differences in image morphology, the influence of the images and each parameter is comprehensively considered, making the classification more accurate. Next, a variance analysis is used to distinguish between different flame morphology types, which has a greater impact. By screening out the influencing factors that affect the flame morphology, further reducing the feature dimension and simplifying the complexity of the model, and then performing cross-validation using the validation set, a support vector machine classifier is introduced to classify the predicted flame morphology, which more directly represents the flame combustion situation in the furnace. This gives the trained target pulverized coal flame morphology prediction model the ability to predict flame morphology more comprehensively and accurately, which can more accurately and quickly predict pulverized coal flame combustion morphology. The model has strong feature extraction and generalization capabilities, provides a more effective means for predicting flame combustion morphology, and has wide application prospects in practical applications.

[0007] In an alternative embodiment, the initial pulverized coal flame morphology prediction model includes a convolutional layer, a pooling layer, and a fully connected layer; Inputting the training flame image into a pre-built initial pulverized coal flame morphology prediction model and extracting the main features of the training flame image is Input the training flame image into a convolutional layer and perform a nonlinear transformation using an activation function to obtain a first feature map; inputting the first feature map into a pooling layer, downsampling the first feature map, and obtaining a second feature map; and inputting the second feature map into a fully connected layer for learning to extract key features of the training flame images.

[0008] In this embodiment, a convolutional neural network model is used to learn the spatial features of pulverized coal flame images and capture important information such as flame texture, shape, and edges. The model also has powerful image processing capabilities, allowing it to effectively extract flame image features and improve prediction accuracy. The introduction of activation functions between convolutional layers promotes information flow, strengthens the model's depth and expressive power, and fully represents the complexity and diversity of pulverized coal flame combustion morphology.

[0009] In an alternative embodiment, the steps of combining the main features with the training combustion parameters to construct a feature vector, performing a variance analysis on the feature vector and different training flame configurations, and selecting the influential parameters include: The method includes calculating the variance between the feature vector and different training flame configurations, and using the training combustion parameters with the largest variance as the influence parameters.

[0010] In this embodiment, by introducing variance analysis, relatively important characteristic parameters are selected and it is determined which parameters have a greater impact on distinguishing different flame morphology types, and pulverized coal flame combustion morphologies are classified using fewer influencing factors, thereby reducing the cost of calculation and storage, simplifying the complexity of the model, and improving the training speed and inference speed of the model.

[0011] In an alternative embodiment, training an initial pulverized coal flame morphology prediction model using the influence parameters, the training flame morphology and the validation flame image, and the validation combustion parameters and the validation flame morphology corresponding to the validation flame image to obtain a target pulverized coal flame morphology prediction model includes: Using the influence parameters and the training flame morphology, a vector machine classifier is trained, and a verification flame image, a verification combustion parameter corresponding to the verification flame image, and a verification flame morphology are combined to perform cross-validation on the vector machine classifier, and training is performed to obtain a target vector machine classifier; and adjusting parameters of the initial pulverized coal flame morphology prediction model using a target vector machine classifier to obtain a target pulverized coal flame morphology prediction model.

[0012] In this embodiment, by introducing a support vector machine classifier to deal with nonlinear problems, problems that were nonlinearly separable in a low-dimensional feature space become linearly separable in the new high-dimensional feature space, making it possible to effectively deal with complex image classification problems and more directly represent the combustion status of flames in a furnace.

[0013] In a second aspect, the present invention provides a method for predicting pulverized coal flame morphology, the method comprising: Acquiring a pulverized coal flame image to be measured and combustion parameters to be measured corresponding to the pulverized coal flame image to be measured; inputting a pulverized coal flame image to be measured and combustion parameters to be measured corresponding to the pulverized coal flame image to be measured into a pulverized coal flame morphology prediction model to obtain a flame morphology type corresponding to the pulverized coal flame image to be measured, wherein the pulverized coal flame morphology prediction model is obtained by training using the training method for a pulverized coal flame morphology prediction model according to any one of the first aspect.

[0014] In the present invention, by inputting a pulverized coal flame image and combustion parameters corresponding to the pulverized coal flame image into a trained pulverized coal flame morphology prediction model, the flame morphology corresponding to the pulverized coal flame image can be accurately predicted, factors affecting combustion stability can be timely discovered and adjusted, and the effectiveness of the adjustment results can be evaluated, providing important support for the optimization of the combustion process of coal-fired boilers and environmental protection.

[0015] In a third aspect, the present invention provides an apparatus for training a pulverized coal flame morphology prediction model, the apparatus comprising: a first data acquisition module used to acquire a sample pulverized coal flame image, sample combustion parameters corresponding to the sample pulverized coal flame image, and a sample flame morphology type; a data preprocessing module used to preprocess the sample pulverized coal flame image to obtain a preprocessed flame image; a sample set construction module used to construct a training set and a validation set using the preprocessed flame images, corresponding sample combustion parameters, and sample flame morphology types as samples, wherein the training set includes training flame images, training combustion parameters corresponding to the training flame images, and training flame morphologies, and the validation set includes validation flame images, validation combustion parameters corresponding to the validation flame images, and validation flame morphologies; a feature extraction module used to input the training flame image into a pre-built initial pulverized coal flame morphology prediction model and extract key features of the training flame image; an influence parameter selection module that combines the main features with the training combustion parameters to construct a feature vector, and performs an analysis of variance on the feature vector and different training flame forms to select influence parameters, wherein the feature vector is used to characterize the sample feature parameters, and the influence parameters are used to characterize multiple training combustion parameters that have the greatest impact on distinguishing the training flame forms; a model training module used to train an initial pulverized coal flame morphology prediction model using the influence parameters, the training flame morphology and the verification flame image, and the verification combustion parameters and the verification flame morphology corresponding to the verification flame image, to obtain a target pulverized coal flame morphology prediction model, wherein the target pulverized coal flame morphology prediction model is used by the model training module to perform pulverized coal flame morphology prediction.

[0016] In a fourth aspect, the present invention provides an apparatus for predicting pulverized coal flame morphology, the apparatus comprising: a second data acquisition module used to acquire to-be-measured pulverized coal flame images and to-be-measured combustion parameters corresponding to the to-be-measured pulverized coal flame images; a flame morphology prediction module used to input a pulverized coal flame image to be measured and combustion parameters to be measured corresponding to the pulverized coal flame image to be measured into a pulverized coal flame morphology prediction model to obtain a flame morphology type corresponding to the pulverized coal flame image to be measured, wherein the pulverized coal flame morphology prediction model includes the flame morphology prediction module obtained by training using a training device for a pulverized coal flame morphology prediction model of the third aspect.

[0017] In a fifth aspect, the present invention provides a computer device including a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method for training a pulverized coal flame morphology prediction model according to the first aspect or any corresponding embodiment, or the method for predicting pulverized coal flame morphology according to the second aspect.

[0018] In a sixth aspect, the present invention provides a computer-readable storage medium having stored thereon computer instructions, the computer instructions being used to cause a computer to execute the method for training a pulverized coal flame morphology prediction model according to the first aspect or any corresponding embodiment thereof, or the method for predicting pulverized coal flame morphology according to the second aspect.

[0019] In a seventh aspect, the present invention provides a computer program product, comprising computer instructions for use in causing a computer to perform the method for training a pulverized coal flame morphology prediction model according to the first aspect or any corresponding embodiment thereof, or the method for predicting pulverized coal flame morphology according to the second aspect. [Brief explanation of the drawings]

[0020] In order to more clearly describe the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings used in the description of the specific embodiments or the prior art. It should be understood that the drawings in the following description are only some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative efforts. [Figure 1] 1 is a flowchart of a method for training a pulverized coal flame morphology prediction model according to an embodiment of the present invention. [Figure 2] 1 is a flowchart of a pulverized coal flame combustion mode prediction algorithm according to an embodiment of the present invention. [Figure 3] 1 is a flowchart of training a pulverized coal flame combustion morphology prediction model according to an embodiment of the present invention. [Figure 4] 1 is a flowchart of another method for training a pulverized coal flame morphology prediction model according to an embodiment of the present invention. [Figure 5] FIG. 1 is a schematic diagram illustrating the structure of a convolutional neural network model according to an embodiment of the present invention. [Figure 6] 1 is a flowchart of a pulverized coal flame morphology prediction method according to an embodiment of the present invention. [Figure 7] FIG. 1 is a block diagram showing the structure of a training device for a pulverized coal flame morphology prediction model according to an embodiment of the present invention. [Figure 8] 1 is a block diagram showing the structure of a pulverized coal flame morphology prediction device according to an embodiment of the present invention; [Figure 9] 1 is a schematic diagram illustrating a hardware structure of a computer device according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0021] In order to more clearly describe the objectives, technical solutions and advantages of the embodiments of the present invention, the following clearly and completely describes the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments that can be obtained by those skilled in the art without any creative efforts shall fall within the protection scope of the present invention.

[0022] In related technology, flame monitoring equipment such as cameras and infrared cameras installed in thermal power plants can monitor and record flame combustion conditions in real time. These equipment are usually installed near the combustion area to capture local combustion images of pulverized coal flames. However, these local images do not clearly reflect the overall flame combustion morphology in the furnace and require judgment based on a combination of relevant theoretical knowledge and operator experience. Pulverized coal flame morphology predictions are usually performed by specialized personnel who observe local flame morphology images and then make judgments based on relevant theoretical knowledge and experience. This prediction is prone to subjective errors, has difficulty capturing subtle changes, and is unable to handle large-scale data and complex situations. Furthermore, it requires high training costs for human resources.

[0023] To solve the above problems, embodiments of the present invention provide a pulverized coal flame morphology prediction model training method and a pulverized coal flame morphology prediction method applied to a computer device, and as described, the execution entity may be a pulverized coal flame morphology prediction model training device and a pulverized coal flame morphology prediction device, and the device may be realized as part or all of a computer device by software, hardware, or a combination of hardware and software, where the computer device may be a terminal, client, or server, and the server may be a single server or a server cluster consisting of multiple servers, and the terminal in the embodiments of the present application may be a smartphone, personal computer, tablet computer, or other intelligent hardware device. In the following method embodiments, all of the execution entities will be described using a computer device.

[0024] The computer device in this embodiment is suitable for use in situations where local combustion images of pulverized coal flames are collected and the overall combustion state of the in-furnace flame is determined based on the local combustion images. In the method for training a pulverized coal flame morphology prediction model according to the present invention, preprocessing is performed on the pulverized coal flame images to improve the quality and uniformity of the collected pulverized coal flame images, a training set and a sample set are then constructed to train the pulverized coal flame morphology prediction model, and the model's ability to extract flame image features is trained to enable the model to better extract the main features of the flame images. Next, combustion parameters such as the pulverized coal characteristics, combustion conditions, and combustion equipment structure that cause differences in image morphology are combined to comprehensively consider the effects of the images and each parameter, resulting in more accurate classification. Next, analysis of variance is performed to distinguish different flame morphology types. By doing so, the influencing factors with greater influence are selected, the feature dimension is further reduced, and the complexity of the model is simplified. Furthermore, by performing cross-validation using the validation set, a support vector machine classifier is introduced to classify the predicted flame morphology, which more directly represents the flame combustion situation in the furnace. This gives the trained target pulverized coal flame morphology prediction model the ability to predict flame morphology more comprehensively and accurately, which can make it possible to predict pulverized coal flame combustion morphology more accurately and quickly. It has strong feature extraction and generalization capabilities, provides a more effective means for predicting flame combustion morphology, and has wide application prospects in practical applications.

[0025] According to an embodiment of the present invention, there is provided an embodiment of a method for training a pulverized coal flame morphology prediction model, wherein the steps illustrated in the flowcharts of the drawings may be performed, for example, on a computer system including a series of computer-executable instructions, and although a logical order is shown in the flowcharts, the steps may, in some cases, be performed in a different order than that shown in the specification.

[0026] In this embodiment, a method for training a pulverized coal flame morphology prediction model that can be used on a computer device is provided. FIG. 1 is a flowchart of the method for training a pulverized coal flame morphology prediction model according to an embodiment of the present invention. As shown in FIG. 1, the process includes the following steps S101 to S106.

[0027] In step S101, a sample pulverized coal flame image, sample combustion parameters corresponding to the sample pulverized coal flame image, and a sample flame configuration type are obtained.

[0028] In one example, pulverized coal flame images are collected by flame monitoring equipment, such as cameras or infrared cameras, installed in thermal power plants. The flame monitoring equipment is typically installed near the combustion zone, and the sample pulverized coal flame image has a size of 36*36. The sample combustion parameters may include related parameters such as pulverized coal characteristics (coal type, quality, sulfur content), combustion conditions (air velocity and distribution, combustion temperature, furnace pressure), combustion equipment structure, and combustor operating conditions. Flame morphologies are classified into laminar diffusion combustion, laminar group combustion, turbulent diffusion combustion, turbulent group combustion, and turbulent explosion combustion, and are numbered sequentially from 0 to 4, with each number corresponding to one operating condition type.

[0029] Specifically, current flame combustion morphology classifications typically only consider appearance characteristics such as flame spatial distribution, size, and color, while ignoring the dynamics of gas-solid two-phase reactions during pulverized coal combustion. The flame morphology classification comprehensively considers the gas dynamics and morphology of the flame. In the laminar diffusion combustion region, the injection velocity and powder mass concentration are so low that sufficient volatile components cannot accumulate to generate a bright, wide gas-phase flame, resulting in localized, single-particle homogeneous ignition. In the laminar group combustion region, the injection velocity is low but the powder mass concentration is high, resulting in sufficient volatile components to generate needle-shaped or linear group combustion flames in the furnace. In the turbulent diffusion combustion region, the powder mass concentration is low but the injection velocity is high, resulting in intense turbulence suction, diluting the volatile components and completely breaking down the particle clusters. In the turbulent group combustion region, the group combustion flame exhibits obvious fluctuations and various morphologies, indicating vigorous turbulence caused by the high injection velocity. In the turbulent explosive combustion region, high-volatility coal with a high mass concentration of powder feed creates a violent flame pattern formed at high injection speeds, resulting in unstable combustion patterns, posing a huge safety risk. Therefore, by classifying flame patterns into the above five types, it is possible to distinguish the combustion conditions inside a boiler.

[0030] In step S102, the sample pulverized coal flame image is preprocessed to obtain a preprocessed flame image.

[0031] In one example, the collected sample pulverized coal flame images are preprocessed, including scaling, cropping, and grayscaling, to convert the pixel range of the sample pulverized coal flame images from the [0,255] interval to the [0,1] interval, thereby ensuring the quality and uniformity of the data.

[0032] In step S103, a training set and a validation set are constructed using the preprocessed flame images, the corresponding sample combustion parameters, and the sample flame morphology types as samples.

[0033] In an embodiment of the present invention, the training set includes training flame images, training combustion parameters corresponding to the training flame images, and training flame morphology, and the validation set includes validation flame images, validation combustion parameters corresponding to the validation flame images, and validation flame morphology.

[0034] In one example, a dataset of preprocessed flame images is divided into a training set, a validation set, and a test set in a 6:2:2 ratio. A CNN model is trained using the training set, and cross-validation is performed in combination with the validation set to optimize the model's parameters, ensuring good generalization ability and accuracy. The trained model is evaluated and validated using the test set, and the difference between predicted and actual results is calculated to obtain the optimal training model.

[0035] In step S104, the training flame image is input into a pre-constructed initial pulverized coal flame morphology prediction model, and the main features of the training flame image are extracted.

[0036] In one example, a pre-constructed initial pulverized coal flame morphology prediction model includes a convolutional neural network (CNN). The CNN model is used as the core of the pulverized coal flame morphology prediction algorithm to learn spatial features of pulverized coal flame images. The pulverized coal flame morphology prediction model structure includes two 2D convolutional layers with a 5x5 convolution kernel size, two max-pooling layers with a 2x2 pooling window, four activation layers, and three fully connected layers. A ReLU activation function is then used to introduce nonlinearity after each convolutional layer. First, training flame images are input to the input layer. Residual connections are then introduced to solve the vanishing gradient problem and enhance information transfer. Next, an attention mechanism is applied after the output of the convolutional layer. An attention weighting network is used to calculate attention weights for each position. The attention weights are then multiplied by the output of the convolutional layer. A weighted sum is used to focus on important regions of the training flame image, resulting in a feature map. The feature map is then downsampled using a pooling layer to further reduce its size and retain key features. Batch normalization is performed after each convolutional or fully connected layer to speed up convergence and improve generalization. Finally, high-level features are learned through the fully connected layer, and L2 regularization is introduced after the fully connected layer to control the model's parameter size and reduce the risk of overfitting. This design and process allows the convolutional neural network to automatically learn and extract feature information by focusing on important regions of the flame image, while capturing related information such as the flame's texture, shape, and edges. As a result, the model's ability to represent and extract flame image features is improved.

[0037] In step S105, the main features are combined with the training combustion parameters to construct a feature vector, and a variance analysis is performed on the feature vector and different training flame configurations to select the influential parameters.

[0038] In an embodiment of the present invention, the feature vector is used to characterize each sample feature parameter, and the influence parameter is used to characterize a plurality of training combustion parameters that have the greatest influence on distinguishing training flame morphologies.

[0039] For example, simply classifying the overall flame combustion morphology using local flame images can still result in significant errors. Therefore, comprehensive consideration of combustion parameters, such as pulverized coal characteristics, combustion conditions, and combustion equipment structure, can further improve prediction accuracy. When the above-mentioned combustion parameters, such as pulverized coal characteristics, combustion conditions, and combustion equipment structure, are comprehensively considered, some of the input combustion parameters may be irrelevant or redundant, resulting in poor model classification, overfitting, and low simulation efficiency. Therefore, the key features extracted by the convolutional neural network in the previous step are used to combine combustion parameters, such as pulverized coal characteristics, combustion conditions, and combustion equipment structure.

[0040] In step S106, the initial pulverized coal flame morphology prediction model is trained using the influencing parameters, the training flame morphology and the verification flame image, and the verification combustion parameters and verification flame morphology corresponding to the verification flame image, to obtain a target pulverized coal flame morphology prediction model.

[0041] In an embodiment of the present invention, a target pulverized coal flame morphology prediction model is used to predict pulverized coal flame morphology.

[0042] In one example, the pre-constructed initial pulverized coal flame morphology prediction model further includes a support vector machine classifier. Using the influence parameters selected in the previous step as input, along with the flame morphology type labels corresponding to each sample, a support vector machine (SVM) classifier is trained and tested. Five-fold cross-validation is used to find the optimal parameter settings for the SVM classifier model. By finding the optimal hyperplane, the SVM can clearly separate each type of flame image, even when multiple influence parameters (such as pulverized coal characteristics, combustion conditions, and combustion equipment structure) are considered, achieving the classification goal. The trained SVM classifier is then evaluated and optimized, and its performance is evaluated using evaluation metrics (such as accuracy, precision, and recall). Depending on the evaluation results, the parameters of the pulverized coal flame morphology prediction model are adjusted or other techniques are used to improve classification performance, resulting in the desired pulverized coal flame morphology prediction model.

[0043] In one implementation scenario, Figure 2 is a flowchart of a pulverized coal flame combustion morphology prediction algorithm according to an embodiment of the present invention. As shown in Figure 2, preprocessed flame images, corresponding sample combustion parameters, and sample flame morphology types are taken as samples, and the sample data are divided into a training data set, a test data set, and a validation set. In the model training stage, a standardization process is performed using the training data set, feature extraction, analysis of variance, and support vector machine classification algorithms are performed on the training data set using a convolutional neural network, feature extraction is performed on the validation set using a convolutional neural network, and flame morphology is classified through cross-validation, thereby training a pulverized coal flame morphology prediction model. Specifically, FIG. 3 is a flowchart of the training of a pulverized coal flame combustion morphology prediction model according to an embodiment of the present invention. As shown in FIG. 3 , the training of the pulverized coal flame combustion morphology prediction model specifically includes: preprocessing a sample pulverized coal flame image to generate a single-channel image; combining the single-channel image with corresponding combustion parameters and flame morphology types to jointly construct a training set, a validation set, and a test set; constructing and training a flame image classification model based on a convolutional neural network; performing feature extraction using the convolutional neural network model; combining the extracted image features with operating condition parameters to perform an analysis of variance to extract key influential parameters; constructing a new training set, a validation set, and a test set using the key influential parameters, the training flame image, and the training flame morphology; and constructing and training a flame image classification model based on an SVM algorithm. The SVM-based flame image classification model and the convolutional neural network model jointly constitute the pulverized coal flame combustion morphology prediction model. After the pulverized coal flame morphology prediction model is obtained through training, in the model testing stage, the mean and standard deviation of each feature in the training data set are used to standardize the test data through the formula X = (X-mean) / std, thereby ensuring consistency in the data distribution of the training data set and the test data set, where Mean and Std respectively represent the mean and standard deviation of each feature in the training data.The model is tested using a test dataset to obtain the optimal target pulverized coal flame combustion morphology prediction model.

[0044] In the pulverized coal flame morphology prediction model method according to this embodiment, preprocessing is performed on the pulverized coal flame images to improve the quality and uniformity of the collected pulverized coal flame images. Then, a training set and a sample set are constructed to train the pulverized coal flame morphology prediction model, and the model's ability to extract flame image features is trained so that the model can better extract the main features of the flame images. Next, by combining combustion parameters such as the characteristics of pulverized coal, combustion conditions, and the structure of combustion equipment that cause differences in image morphology, the effects of the images and each parameter are comprehensively considered, making the classification more accurate. Then, different flame morphology types are distinguished by variance analysis. By selecting influential factors that have a greater impact, further reducing the feature dimension and simplifying the complexity of the model, and then performing cross-validation using the validation set, a support vector machine classifier is introduced to classify the predicted flame morphology, which more directly represents the flame combustion situation in the furnace. This gives the trained target pulverized coal flame morphology prediction model the ability to predict flame morphology more comprehensively and accurately, which can more accurately and quickly predict pulverized coal flame combustion morphology. It has strong feature extraction and generalization capabilities, provides a more effective means for predicting flame combustion morphology, and has wide application prospects in practical applications.

[0045] In this embodiment, a method for training a pulverized coal flame morphology prediction model that can be used on a computer device is provided. FIG. 4 is a flowchart of the method for training a pulverized coal flame morphology prediction model according to an embodiment of the present invention. As shown in FIG. 4, the process includes the following steps S401 to S406.

[0046] In step S401, a sample pulverized coal flame image, sample combustion parameters corresponding to the sample pulverized coal flame image, and sample flame morphology types are obtained. For details, refer to step S101 in the embodiment shown in Figure 1, and the details will not be repeated here.

[0047] In step S402, the sample pulverized coal flame image is pre-processed to obtain a pre-processed flame image, for details see step S102 in the embodiment shown in Figure 1, and will not be repeated here.

[0048] In step S403, the preprocessed flame images, corresponding sample combustion parameters, and sample flame morphology types are used as samples to construct a training set and a validation set. For details, see step S103 in the embodiment shown in Figure 1, and the details will not be repeated here.

[0049] In step S404, the training flame image is input into a pre-constructed initial pulverized coal flame morphology prediction model, and the main features of the training flame image are extracted.

[0050] Specifically, the initial pulverized coal flame morphology prediction model includes a convolution layer, a pooling layer and a fully connected layer, and the above step S404 includes: Step S4041: inputting the training flame image into a convolutional layer, performing nonlinear transformation using an activation function, and obtaining a first feature map; Step S4042 of inputting the first feature map into a pooling layer and downsampling the first feature map to obtain a second feature map; and step S4043 of inputting the second feature map into a fully connected layer for learning and extracting key features of the training flame image.

[0051] In one example, a convolutional neural network model is used as the core of a prediction algorithm to learn the spatial features of pulverized coal flame images. First, a flame image is input to the input layer. The model structure includes two 2D convolutional layers with a 5x5 convolution kernel, two max-pooling layers with a 2x2 pooling window, four activation layers, and three fully-connected layers. A ReLU activation function is then used to introduce nonlinearity after each convolutional layer. Residual connections are then introduced to solve the vanishing gradient problem and enhance information transfer. An attention mechanism is then applied after the output of the convolutional layers. An attention weighting network is used to calculate attention weights for each position, and these are multiplied by the output of the convolutional layers to focus on important regions of the flame image using a weighted sum. A pooling layer is then used to downsample the feature map, further reducing its size and preserving key features. Batch normalization is performed after each convolutional or fully-connected layer to speed up convergence and improve generalization ability. Finally, high-level features are learned through a fully connected layer, and L2 regularization is introduced after the fully connected layer to control the model's parameter size and reduce the risk of overfitting. This design and process allows the convolutional neural network to automatically learn and extract feature information by focusing on important regions of the flame image, while capturing related information such as the flame's texture, shape, and edge, thereby improving the model's ability to represent and extract flame image features.

[0052] 5 is a schematic diagram showing the structure of a convolutional neural network model according to an embodiment of the present invention, in which @ indicates a convolution operation. For example, 1@36×36 means that a convolution operation is performed by applying one convolution kernel to an input image of 36×36 size. As shown in FIG. 5, the layers are numbered (1) to (8) from the front to the back as follows:

[0053] (1) Input layer: A flame image of size 36x36 is accepted as input. Since the flame image is a gray image, the number of input channels is 1.

[0054] (2) The first convolution layer uses a 5x5 convolution kernel, with 16 output channels and a stride of 1. After the convolution, a nonlinear transformation is performed using the ReLU (Rectified Linear Unit) activation function, and the calculation formula is as follows: TIFF2025164653000002.tif7170In the formula, x is the position and channel that generate the feature map after the convolution operation of the convolution layer, (x) + indicates that the function is non-negative.

[0055] (3) First max pooling layer: A max pooling operation is performed with a 2x2 pooling window, stride is 2, and the feature maps are downsampled.

[0056] (4) The second convolutional layer uses a 5x5 convolution kernel, with 32 output channels and a stride of 1. After the convolution, a nonlinear transformation is performed using the ReLU activation function.

[0057] (5) Second max pooling layer: A max pooling operation is performed using a 2x2 pooling window with a stride of 2 to downsample the feature maps.

[0058] (6) Fully connected layer 1: A fully connected layer with 1152 inputs and 256 outputs. A nonlinear transformation is performed using the ReLU activation function.

[0059] (7) Fully connected layer 2: This fully connected layer has 256 inputs and 128 outputs, and performs nonlinear transformation using the ReLU activation function.

[0060] (8) Fully connected layer 3: This fully connected layer has 128 inputs and 5 outputs, and is used to output classification prediction results.

[0061] In this embodiment, a convolutional neural network model is used to learn the spatial features of pulverized coal flame images and capture important information such as flame texture, shape, and edges. The model also has powerful image processing capabilities, allowing it to effectively extract flame image features and improve prediction accuracy. The introduction of activation functions between convolutional layers promotes information flow, strengthens the model's depth and expressive power, and fully represents the complexity and diversity of pulverized coal flame combustion morphology.

[0062] In step S405, the main features are combined with the training combustion parameters to construct a feature vector, and then a variance analysis is performed on the feature vector and different training flame configurations to select the influential parameters.

[0063] Specifically, step S405 is as follows: The method includes a step S4051 of calculating the variance between the feature vector and different training flame configurations, and using the training combustion parameters with the largest variance as influence parameters.

[0064] In one example, a feature vector describing the parameters of each sample is constructed by combining the main features of the flame image extracted using a convolutional neural network, the characteristics of the pulverized coal corresponding to the flame image, the combustion conditions, the structure of the combustion equipment, and other combustion parameters. The average value of each parameter is calculated, and the squared deviations for all sample parameters are accumulated and divided by the number of samples. The variance between the feature vector and different flame morphologies is calculated; a larger variance indicates a greater influence of the combustion parameter in distinguishing between different flame morphologies. Based on the results of the analysis of variance, the top 10 combustion parameters with the largest variance are selected as influential parameters.

[0065] In this embodiment, by introducing variance analysis, relatively important characteristic parameters are selected and it is determined which parameters have a greater impact on distinguishing different flame morphology types, and pulverized coal flame combustion morphologies are classified using fewer influencing factors, thereby reducing the cost of calculation and storage, simplifying the complexity of the model, and improving the training speed and inference speed of the model.

[0066] In step S406, the initial pulverized coal flame morphology prediction model is trained using the influencing parameters, the training flame morphology and the verification flame image, and the verification combustion parameters and verification flame morphology corresponding to the verification flame image, to obtain a target pulverized coal flame morphology prediction model.

[0067] Specifically, the above step S406 includes step S4061 and step S4062.

[0068] In step S4061, the influence parameters and the training flame morphology are used to train a vector machine classifier, and the verification flame image, the verification combustion parameters corresponding to the verification flame image, and the verification flame morphology are combined to perform cross-validation on the vector machine classifier, and a target vector machine classifier is obtained by training.

[0069] In step S4062, the target vector machine classifier is used to adjust the parameters of the initial pulverized coal flame morphology prediction model to obtain a target pulverized coal flame morphology prediction model.

[0070] In one example, a support vector machine (SVM) classifier is trained and tested using the influence parameters selected in the previous step, along with the flame morphology corresponding to each sample, and 5-fold cross-validation is used to find the optimal parameter settings for the SVM classification model. By finding the optimal hyperplane, the SVM can clearly separate each type of flame image, even when multiple influence parameters (such as pulverized coal characteristics, combustion conditions, and combustion equipment structure) are considered, achieving the classification goal. The trained SVM classifier is then evaluated and optimized, and its performance is assessed using evaluation metrics (such as accuracy, precision, and recall). Depending on the evaluation results, the model parameters are adjusted or other techniques are used to improve classification performance.

[0071] In this embodiment, by introducing a support vector machine classifier to deal with nonlinear problems, problems that were nonlinearly separable in a low-dimensional feature space become linearly separable in the new high-dimensional feature space, making it possible to effectively deal with complex image classification problems and more directly represent the combustion status of flames in a furnace.

[0072] In this embodiment, the training method for the pulverized coal flame morphology prediction model uses a convolutional neural network model to learn the spatial features of pulverized coal flame images and capture important information such as flame texture, shape, and edges. The model also has powerful image processing capabilities, effectively extracting flame image features and improving prediction accuracy. The introduction of activation functions between convolutional layers promotes information flow, enhancing the depth and expressiveness of the model and fully representing the complexity and diversity of pulverized coal flame combustion morphologies. Analysis of variance is used to select relatively important feature parameters and determine which parameters have a greater impact on distinguishing different flame morphology types. This allows for classification of pulverized coal flame combustion morphologies using fewer influential factors, reducing computational and storage costs, simplifying model complexity, and improving model training and inference speeds. The introduction of a support vector machine classifier to address nonlinear problems makes problems that are nonlinearly separable in low-dimensional feature spaces linearly separable in the new high-dimensional feature space, effectively addressing complex image classification problems and more directly representing the combustion conditions of furnace flames.

[0073] In this embodiment, a pulverized coal flame morphology prediction method that can be used on the above-mentioned mobile terminals, such as mobile phones and tablet computers, is provided. FIG. 6 is a flowchart of the pulverized coal flame morphology prediction method according to the embodiment of the present invention. As shown in FIG. 6, the process includes: Step S601 of acquiring a pulverized coal flame image to be measured and combustion parameters to be measured corresponding to the pulverized coal flame image to be measured; and step S602 of inputting the pulverized coal flame image to be measured and the combustion parameters to be measured corresponding to the pulverized coal flame image to be measured into a pulverized coal flame shape prediction model to obtain a flame shape type corresponding to the pulverized coal flame image to be measured.

[0074] In the embodiment of the present invention, the pulverized coal flame morphology prediction model is obtained by training according to the above-mentioned method for training a pulverized coal flame morphology prediction model.

[0075] In one example, the model obtained by the above training can accurately predict the resulting flame type by inputting the local feature map of the flame and related combustion parameters.

[0076] In the pulverized coal flame morphology prediction method according to this embodiment, a pulverized coal flame image and combustion parameters corresponding to the pulverized coal flame image are input into a trained pulverized coal flame morphology prediction model, thereby accurately predicting the flame morphology corresponding to the pulverized coal flame image, timely discovering and adjusting factors affecting combustion stability, and evaluating the effectiveness of the adjustment results, thereby providing important support for optimizing the combustion process of coal-fired boilers and protecting the environment.

[0077] This example further provides a training device for a pulverized coal flame morphology prediction model, which is used to realize the above examples and preferred embodiments, and will not be described again here. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. The device described in the following examples is preferably implemented in software, but may also be implemented in hardware or a combination of software and hardware.

[0078] This embodiment provides a training device for a pulverized coal flame morphology prediction model, and includes the following steps 701 to 704 as shown in FIG.

[0079] The first data acquisition module 701 is used to acquire a sample pulverized coal flame image, sample combustion parameters corresponding to the sample pulverized coal flame image, and sample flame morphology type. For details, refer to step S101 in the embodiment shown in Figure 1, and will not be repeated here.

[0080] The data pre-processing module 702 is used to pre-process the sample pulverized coal flame image to obtain a pre-processed flame image, for details see step S102 in the embodiment shown in Figure 1, which will not be repeated here.

[0081] The sample set construction module 703 is used to construct a training set and a validation set using the preprocessed flame images, corresponding sample combustion parameters, and sample flame morphology types as samples, where the training set includes training flame images, training combustion parameters corresponding to the training flame images, and training flame morphologies, and the validation set includes validation flame images, validation combustion parameters corresponding to the validation flame images, and validation flame morphologies. For details, see step S103 in the embodiment shown in Figure 1, and will not be repeated here.

[0082] The feature extraction module 704 is used to input the training flame image into a pre-built initial pulverized coal flame morphology prediction model and extract key features of the training flame image. For details, refer to step S104 in the embodiment shown in Figure 1, and will not be repeated here.

[0083] The influence parameter selection module 705 combines the main features with the training combustion parameters to construct a feature vector, and performs an analysis of variance on the feature vector and different training flame morphologies to select influence parameters, where the feature vector is used to characterize the sample feature parameters, and the influence parameters are used to characterize multiple training combustion parameters that have the greatest impact on distinguishing between training flame morphologies. For details, see step S105 in the embodiment shown in Figure 1, and will not be repeated here.

[0084] The model training module 706 is used to train an initial pulverized coal flame morphology prediction model using the influence parameters, the training flame morphology and the verification flame image, and the verification combustion parameters and verification flame morphology corresponding to the verification flame image, to obtain a target pulverized coal flame morphology prediction model, and the target pulverized coal flame morphology prediction model is used to perform pulverized coal flame morphology prediction. For details, see step S106 in the embodiment shown in Figure 1, and will not be repeated here.

[0085] In some alternative embodiments, the initial pulverized coal flame morphology prediction model includes a convolutional layer, a pooling layer, and a fully connected layer, and the feature extraction module 704: a first image input unit used to input training flame images into a convolutional layer and perform nonlinear transformation using an activation function to obtain a first feature map; a second image input unit used to input the first feature map into a pooling layer, downsample the first feature map, and obtain a second feature map; and a third image input unit that is used to input the second feature map into a fully connected layer for learning and extract key features of the training flame images.

[0086] In some alternative embodiments, the influence parameter selection module 705: The apparatus includes an influence parameter variance analysis unit that calculates the variance between the feature vector and different training flame configurations and uses a plurality of training combustion parameters with the largest variance as influence parameters.

[0087] In some alternative embodiments, the model training module 706: a basis vector classifier training unit that uses the influence parameters and the training flame morphology to train a vector machine classifier, and performs cross-validation on the vector machine classifier by combining the validation flame image, the validation combustion parameters corresponding to the validation flame image, and the validation flame morphology to obtain a target vector machine classifier by training; a parameter adjustment unit that is used to adjust parameters of the initial pulverized coal flame morphology prediction model using the target vector machine classifier to obtain a target pulverized coal flame morphology prediction model.

[0088] The further functional description of each module and unit described above is the same as the corresponding embodiment above, and will not be repeated here.

[0089] The pulverized coal flame morphology prediction model training device in this example is shown in the form of a functional unit, where the unit may be an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory executing one or more software or fixed programs, and / or other device capable of providing the above functionality.

[0090] This example further provides a pulverized coal flame morphology prediction device, which is used to realize the above examples and preferred embodiments, and will not be described again here. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. The device described in the following example is preferably implemented in software, but may also be implemented in hardware or a combination of software and hardware.

[0091] This embodiment provides a pulverized coal flame morphology prediction device, which includes the following 801 and 802 as shown in FIG.

[0092] The second data acquisition module 801 is used to acquire a pulverized coal flame image to be measured and a combustion parameter to be measured corresponding to the pulverized coal flame image to be measured. For details, refer to step S601 in the embodiment shown in Figure 6, and the details will not be repeated here.

[0093] The flame morphology prediction module 802 is used to input the to-be-measured pulverized coal flame image and the to-be-measured combustion parameters corresponding to the to-be-measured pulverized coal flame image into a pulverized coal flame morphology prediction model to obtain a flame morphology type corresponding to the to-be-measured pulverized coal flame image, where the pulverized coal flame morphology prediction model is obtained by training using the above-mentioned pulverized coal flame morphology prediction model training device. For details, see step S602 in the embodiment shown in Figure 6, and will not be repeated here.

[0094] The further functional description of each module and unit described above is the same as the corresponding embodiment above, and will not be repeated here.

[0095] The pulverized coal flame morphology prediction apparatus in this embodiment is shown in the form of a functional unit, where the unit may be an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory executing one or more software or fixed programs, and / or other device capable of providing the functionality described above.

[0096] An embodiment of the present invention further provides a computer device including the above-mentioned training device for pulverized coal flame morphology prediction model shown in FIG. 7 and the pulverized coal flame morphology prediction device shown in FIG.

[0097] Referring to FIG. 9, FIG. 9 is a schematic diagram illustrating the structure of a computer device provided by an alternative embodiment of the present invention. As shown in FIG. 9, the computer device includes one or more processors 10, memory 20, and interfaces for connecting components, including high-speed and low-speed interfaces. The components are communicatively connected to each other using different buses and may be mounted on a common motherboard or by other means as needed. The processor can process instructions executed within the computer device, including instructions stored in or on a memory, to display graphical information of a GUI on an external input / output device (e.g., a display device coupled to the interface). In some alternative embodiments, multiple processors and / or multiple buses can be used, along with multiple memories and multiple memories, as needed. Similarly, multiple computer devices can be connected, each providing a portion of the required operations (e.g., an array of servers, a blade server, or a multiprocessor system). FIG. 9 illustrates a single processor 10 as an example.

[0098] The processor 10 may be a central processing unit, a network processor, or a combination thereof. Here, the processor 10 may further include a hardware chip. The hardware chip may be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general-purpose array logic, or any combination thereof.

[0099] Here, the memory 20 stores instructions that can be executed by at least one processor 10, thereby causing the at least one processor 10 to perform the methods shown in the above embodiments.

[0100] The memory 20 includes a storage program area and a storage data area, where the storage program area can store an operating system, application programs necessary for at least one function, and data generated based on use of the computer device, etc. Furthermore, the memory 20 may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk memory device, flash memory device, or other non-transitory solid-state memory device. In some alternative embodiments, the memory 20 optionally includes memory located remotely from the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, a corporate intranet, a local area network, a mobile communication network, and combinations thereof.

[0101] Memory 20 may include volatile memory such as random access memory, memory may also include non-volatile memory such as flash memory, a hard drive or solid hard drive, and memory 20 may even include a combination of the above types of memory.

[0102] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means, and Fig. 9 shows an example of connection via a bus.

[0103] The input device 30 can receive input numeric or textual information as well as generate key signal inputs related to user settings and function control of the computing device, such as a touch screen, keypad, mouse, trackpad, touchpad, indicator bar, one or more mouse buttons, trackball, joystick, etc. The output device 40 can include a display device, auxiliary lighting devices (e.g., LEDs), and tactile feedback devices (e.g., vibration motors), etc. The display device can include, but is not limited to, a liquid crystal display, a light-emitting diode display, a plasma display, etc. In some alternative embodiments, the display device can be a touch screen.

[0104] Embodiments of the present invention further include a computer-readable storage medium, and the methods described herein may be implemented in hardware, firmware, or as computer code that can be stored on a storage medium, or may be originally stored on a remote or non-transitory machine-readable storage medium but downloaded over a network and stored on a local storage medium. Thus, the methods described herein can be processed by such software stored on a storage medium using a general-purpose computer, a special-purpose processor, programmable hardware, or dedicated hardware. The storage medium may be a magnetic disk, optical disk, read-only memory, random memory, flash memory, hard drive, solid-state drive, or the like. The storage medium may also include a combination of the above types of memory. It should be understood that a computer, processor, microprocessor controller, or programmable hardware may include a storage component that can store or receive software or computer code, and that the software or computer code, when accessed and executed by the computer, processor, or hardware, implements the methods described in the above embodiments.

[0105] Some aspects of the present invention may be implemented as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the present invention through the operation of the computer. As will be understood by those skilled in the art, the existence of computer program instructions on a computer-readable medium may include, but is not limited to, a source file, an executable file, an installer file, etc. Correspondingly, the manner in which the computer program instructions are executed by a computer may include, but is not limited to, the computer directly executing the instructions, compiling the instructions and then executing a corresponding compiled program, reading and executing the instructions, or reading and installing the instructions and then executing a corresponding installed program. Here, the computer-readable medium may be any computer-readable storage medium or communication medium accessible by a computer.

[0106] Although the embodiments of the present invention have been described in conjunction with the drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and all such modifications and variations are intended to be included within the scope of the appended claims.

Claims

1. A method for training a pulverized coal flame morphology prediction model, comprising: obtaining a sample pulverized coal flame image, sample combustion parameters and sample flame morphology types corresponding to the sample pulverized coal flame image; preprocessing the sample pulverized coal flame image to obtain a preprocessed flame image; constructing a training set and a validation set using the preprocessed flame images, corresponding sample combustion parameters, and sample flame morphology types as samples, wherein the training set includes training flame images, training combustion parameters, and training flame morphologies corresponding to the training flame images, and the validation set includes validation flame images, validation combustion parameters, and validation flame morphologies corresponding to the validation flame images; Inputting the training flame image into a pre-constructed initial pulverized coal flame morphology prediction model to extract key features of the training flame image; Combining the main features with the training combustion parameters to construct a feature vector, and performing an analysis of variance on the feature vector and different training flame forms to select influential parameters, wherein the feature vector is used to characterize each of the sample feature parameters, and the influential parameters are used to characterize a plurality of training combustion parameters that have the greatest impact on distinguishing the training flame forms; training the initial pulverized coal flame morphology prediction model using the influence parameters, the training flame morphology, the verification flame image, and verification combustion parameters and verification flame morphology corresponding to the verification flame image to obtain a target pulverized coal flame morphology prediction model, wherein the target pulverized coal flame morphology prediction model is used to perform pulverized coal flame morphology prediction.

2. The initial pulverized coal flame morphology prediction model includes a convolution layer, a pooling layer, and a fully connected layer; Inputting the training flame image into a pre-constructed initial pulverized coal flame shape prediction model and extracting key features of the training flame image, inputting the training flame image into the convolutional layer and performing a nonlinear transformation using an activation function to obtain a first feature map; inputting the first feature map into the pooling layer and downsampling the first feature map to obtain a second feature map; and inputting the second feature map into the fully connected layer for learning to extract key features of the training flame images.

3. The above-mentioned main features are combined with the training combustion parameters to construct a feature vector, and a variance analysis is performed on the feature vector and different training flame forms to select influential parameters.

2. The method of claim 1, further comprising calculating variances between the feature vectors and different training flame configurations, and using the training combustion parameters with the greatest variances as influence parameters.

4. The above-mentioned influence parameters, the training flame shape, the verification flame image, and the verification combustion parameters and verification flame shape corresponding to the verification flame image are used to train the initial pulverized coal flame shape prediction model, and to obtain a target pulverized coal flame shape prediction model. training a vector machine classifier using the influence parameters and the training flame morphology, and performing cross-validation on the vector machine classifier by combining the validation flame image, the validation combustion parameters corresponding to the validation flame image, and the validation flame morphology, and training to obtain a target vector machine classifier; and adjusting parameters of the initial pulverized coal flame morphology prediction model using the target vector machine classifier to obtain the target pulverized coal flame morphology prediction model.

5. A pulverized coal flame shape prediction method, Acquiring a pulverized coal flame image to be measured and combustion parameters to be measured corresponding to the pulverized coal flame image to be measured; the pulverized coal flame image to be measured and combustion parameters to be measured corresponding to the pulverized coal flame image to be measured are input into a pulverized coal flame morphology prediction model to obtain a flame morphology type corresponding to the pulverized coal flame image to be measured, wherein the pulverized coal flame morphology prediction model is obtained by training using the method for training a pulverized coal flame morphology prediction model according to any one of claims 1 to 4.

6. A training device for a pulverized coal flame morphology prediction model, a first data acquisition module used to acquire a sample pulverized coal flame image, sample combustion parameters and sample flame morphology types corresponding to the sample pulverized coal flame image; a data pre-processing module used to pre-process the sample pulverized coal flame image to obtain a pre-processed flame image; a sample set construction module used to construct a training set and a validation set using the preprocessed flame images, corresponding sample combustion parameters, and sample flame morphology types as samples, wherein the training set includes training flame images, training combustion parameters, and training flame morphologies corresponding to the training flame images, and the validation set includes validation flame images, validation combustion parameters, and validation flame morphologies corresponding to the validation flame images; and a feature extraction module used to input the training flame image into a pre-constructed initial pulverized coal flame morphology prediction model and extract key features of the training flame image; an influence parameter selection module that combines the main features with the training combustion parameters to construct a feature vector, and performs variance analysis on the feature vector for different training flame forms to select influence parameters; and an influence parameter selection module that uses the feature vector to characterize each of the sample feature parameters, and uses the influence parameters to characterize multiple training combustion parameters that have the greatest impact on distinguishing the training flame forms. a model training module used to train the initial pulverized coal flame morphology prediction model using the influence parameters, the training flame morphology, the verification flame image, and verification combustion parameters and verification flame morphology corresponding to the verification flame image, to obtain a target pulverized coal flame morphology prediction model, wherein the target pulverized coal flame morphology prediction model is used by the model training module to perform pulverized coal flame morphology prediction.

7. A pulverized coal flame shape prediction device, a second data acquisition module used to acquire a to-be-measured pulverized coal flame image and a to-be-measured combustion parameter corresponding to the to-be-measured pulverized coal flame image; 7. The pulverized coal flame morphology prediction device according to claim 6, further comprising: a flame morphology prediction module used to input the pulverized coal flame image to be measured and combustion parameters to be measured corresponding to the pulverized coal flame image to be measured into a pulverized coal flame morphology prediction model, and to obtain a flame morphology type corresponding to the pulverized coal flame image to be measured, wherein the pulverized coal flame morphology prediction model is obtained by training using the training device for a pulverized coal flame morphology prediction model according to claim 6.

8. A computer device comprising: The computer device includes a memory and a processor, the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the computer instructions to perform the method for training a pulverized coal flame morphology prediction model according to any one of claims 1 to 4 or the method for predicting a pulverized coal flame morphology according to claim 5.

9. 1. A computer-readable storage medium, comprising: The computer-readable storage medium has computer instructions stored thereon, the computer instructions being used to cause a computer to execute the method for training a pulverized coal flame morphology prediction model according to any one of claims 1 to 4 or the method for predicting a pulverized coal flame morphology according to claim 5.

10. 1. A computer program product comprising: The computer program product includes computer instructions, the computer instructions being used to cause a computer to execute the method for training a pulverized coal flame morphology prediction model according to any one of claims 1 to 4 or the method for predicting pulverized coal flame morphology according to claim 5.

Images