Boiler combustion carbon reduction and pollution control method, device, equipment, medium and program

Abstract

The application relates to the technical field of energy control, in particular to a boiler combustion carbon reduction and pollution reduction control method, device, equipment, medium and program, wherein the method comprises the following steps: acquiring coal quality data, fly ash images and boiler operation parameters of a boiler; inputting the fly ash images into a carbon content prediction model, and outputting corresponding fly ash carbon contents by the carbon content prediction model; determining boiler efficiency and carbon emission intensity according to the coal quality data, the fly ash carbon contents and the boiler operation parameters; taking maximization of the boiler efficiency, minimization of the carbon emission intensity, minimization of operation coal consumption and minimization of nitrogen oxide emission concentration as targets, and calculating corresponding target control parameters under different working conditions; and controlling the boiler to perform corresponding carbon reduction and pollution reduction control operations according to the target control parameters. Therefore, the problems, such as low combustion efficiency, increased pollutants and carbon emissions, caused by the fact that the artificial experience and PID control strategy in the related art lead to combustion regulation lag and insufficient precision and are difficult to adapt to coal quality fluctuation and load change, are solved.

Description

Technical Field

[0001] This application relates to the field of energy control technology, and in particular to a method, device, equipment, medium and procedure for controlling carbon reduction and pollution reduction in boiler combustion. Background Technology

[0002] As the core equipment in thermal power generation, the combustion efficiency and emission performance of coal-fired boilers directly affect energy utilization efficiency and ecological environmental impact. Achieving efficient, clean, and low-carbon operation of boiler combustion while ensuring power supply security has become a critical issue that the energy and power industry urgently needs to address.

[0003] In related technologies, power plant boilers still mainly rely on the experience of operators combined with traditional PID (proportional-integral-derivative) control strategies for combustion adjustment. However, this approach is slow to respond to disturbances such as changes in coal quality and load fluctuations, has limited control precision, and makes it difficult to achieve dynamic optimization of multivariable coupled systems. Especially under medium and low load operating conditions, problems such as decreased combustion stability and imbalance in the air-coal ratio often lead to increased carbon content in fly ash, increased incomplete combustion losses, and increased nitrogen oxides (NOx). x Pollutants such as carbon dioxide (CO2) and carbon dioxide emissions have increased significantly. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, medium, and program for controlling carbon reduction and pollution reduction in boiler combustion, in order to solve the problems caused by human experience and PID control strategies in related technologies, which result in lag and insufficient precision in combustion regulation, making it difficult to adapt to fluctuations in coal quality and load changes, thereby causing low combustion efficiency and increased pollutant and carbon emissions.

[0005] The first aspect of this application provides a method for controlling carbon reduction and pollution control in boiler combustion, comprising the following steps: acquiring coal quality data, fly ash images, and boiler operating parameters; inputting the fly ash images into a carbon content prediction model, wherein the carbon content prediction model outputs the corresponding fly ash carbon content; determining boiler efficiency and carbon emission intensity based on the coal quality data, the fly ash carbon content, and the boiler operating parameters; calculating target control parameters corresponding to different operating conditions with the objectives of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration; and controlling the boiler to perform corresponding carbon reduction and pollution control operations based on the target control parameters.

[0006] Optionally, the carbon content prediction model includes: an image preprocessing and feature extraction layer, a principal component analysis (PCA) dimensionality reduction layer, a convolutional neural network (CNN) feature enhancement layer, and a support vector regression (SVR) output layer; wherein, the image preprocessing and feature extraction layer is used to preprocess the fly ash image and extract the feature vectors corresponding to the preprocessed fly ash image, wherein the feature vectors include texture feature vectors, shape feature vectors, and color feature vectors; the PCA dimensionality reduction layer is used to perform PCA dimensionality reduction on the feature vectors to obtain dimensionality-reduced features; the CNN feature enhancement layer is used to extract deep spatial features based on the dimensionality-reduced features; and the SVR output layer is used to query a mapping table based on the deep spatial features to determine the corresponding fly ash carbon content.

[0007] Optionally, determining boiler efficiency and carbon emission intensity based on the coal quality data, fly ash carbon content, and boiler operating parameters includes: inputting the coal quality data, fly ash carbon content, and boiler operating parameters into a target model, and the target model outputting boiler efficiency and carbon emission intensity; wherein, the target model includes a first long short-term memory network layer and a second long short-term memory network layer, the first long short-term memory network layer being used to extract the time-dependent feature sequences corresponding to the coal quality data, fly ash carbon content, and boiler operating parameters at each time step; the second long short-term memory network layer being used to calculate boiler efficiency and carbon emission intensity by fusing multi-scale features based on the time-dependent feature sequences.

[0008] Optionally, the calculation formula for the target model is: , ; in, The weight matrix for the efficiency output layer. This is the final temporal dependency feature vector extracted after passing through a two-layer long short-term memory network. For activation function, For the bias term of the efficiency output layer, The weight matrix for the carbon emission intensity output layer. This is the bias term for the carbon emission intensity output layer. For boiler efficiency, Carbon emission intensity; in, ; The temporal features are the output of the first Long Short-Term Memory (LSTM) network layer. For the second long short-term memory network layer in The temporal dependency features of the output at each time step. For the second long short-term memory network layer in Temporal dependency features of output at each time step; in, ; for Input data at any time, For the first long short-term memory network layer in The temporal dependency features of the output at each time step. For the first long short-term memory network layer in Temporal dependency features of output at each time step; Wherein, the loss function is: ;in, This is the predicted value for boiler efficiency. This represents the true value of boiler efficiency. This is the predicted value of carbon emission intensity. This represents the true value of carbon emission intensity. This is the predicted carbon content of fly ash. This represents the true carbon content of fly ash. , , These are the weights for efficiency loss, carbon emission loss, and fly ash carbon content loss, respectively.

[0009] Optionally, the calculation of target control parameters under different operating conditions, with the objectives of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration, includes:

[0010] in, (x) = 1- The reciprocal of boiler efficiency (x) = C(C) ) For carbon emission intensity, (x) =C( ) This refers to the carbon content of fly ash. Nitrogen oxide emission concentration; The constraints are as follows:

[0011] Wherein, O2 represents the actual oxygen content of the flue gas during boiler operation. The minimum oxygen content in the flue gas. P represents the maximum oxygen content in the flue gas and the actual unit load. This is the minimum safe and stable load allowed for the unit. This is the maximum allowable load for the unit. This represents the actual nitrogen oxide emission concentration. These are the limits for nitrogen oxide emissions.

[0012] Optionally, after controlling the boiler to perform the corresponding carbon reduction and pollution reduction control operation according to the target control parameters, the process includes: acquiring coal quality data, fly ash images, and boiler operating parameters after the carbon reduction and pollution reduction control operation; if the difference between the coal quality data, fly ash images, and boiler operating parameters after the carbon reduction and pollution reduction control operation and historical coal quality data, historical fly ash images, and historical boiler operating parameters is greater than a preset threshold, then the current operating condition is determined to have changed, and the target model is trained using the coal quality data, fly ash images, and boiler operating parameters of the changed operating condition until the model converges.

[0013] A second aspect of this application provides a boiler combustion carbon reduction and pollution control device, comprising: an acquisition module for acquiring coal quality data, fly ash images, and boiler operating parameters; a processing module for inputting the fly ash images into a carbon content prediction model, wherein the carbon content prediction model outputs the corresponding fly ash carbon content; a determination module for determining boiler efficiency and carbon emission intensity based on the coal quality data, the fly ash carbon content, and the boiler operating parameters; a calculation module for calculating target control parameters corresponding to different operating conditions with the objectives of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration; and an execution module for controlling the boiler to perform corresponding carbon reduction and pollution control operations based on the target control parameters.

[0014] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the boiler combustion carbon reduction and pollution control method as described in the above embodiments.

[0015] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to perform the boiler combustion carbon reduction and pollution control method as described in the above embodiments.

[0016] The fifth aspect of this application provides a computer program product, including a computer program or instructions, which, when executed, implement the boiler combustion carbon reduction and pollution control method as described in the above embodiments.

[0017] Therefore, this application has at least the following beneficial effects: This application embodiment integrates coal quality data, fly ash images, and boiler operating parameters. It uses a carbon content prediction model to obtain the carbon content of fly ash in real time and combines multi-dimensional data to comprehensively evaluate boiler efficiency and carbon emission intensity. Then, with the goal of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration, it dynamically generates and executes optimal control parameters. This effectively improves the intelligence and precision of the combustion process and achieves synergistic optimization of carbon reduction, pollution reduction, and energy saving under complex operating conditions such as medium and low loads. It significantly reduces incomplete combustion losses and pollutant emissions, and improves the economic efficiency and environmental performance of power plant operation.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a boiler combustion carbon reduction and pollution control method according to an embodiment of this application; Figure 2 This is an example diagram illustrating the process of a fusion data-driven intelligent combustion system and method for reducing carbon emissions and pollution in boilers, provided in an embodiment of this application. Figure 3 This is a schematic diagram of a boiler combustion carbon reduction and pollution control device provided according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0020] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0021] The functional positioning of coal-fired power plants is gradually shifting from basic, guaranteed power supply to system-regulating and backup power supply, requiring frequent participation in grid peak shaving and long-term operation at medium and low loads. Under these conditions, boiler combustion stability decreases, leading to increased coal consumption per unit of power generation and higher carbon emission intensity, thus placing higher demands on combustion control.

[0022] Currently, combustion optimization control of coal-fired boilers mainly relies on the experience of operators and traditional proportional-integral-derivative (PI-DE) control strategies, which have several technical limitations: First, the ability to perceive combustion status is insufficient. Existing control systems mostly adjust based on conventional operating parameters such as oxygen content, furnace temperature, and air volume. The carbon content of fly ash is usually obtained through manual sampling and offline analysis in laboratories, with a detection cycle of 4 to 8 hours, making it difficult to reflect changes in combustion efficiency in real time. There is also a lack of rapid online detection methods for the characteristics of coal entering the furnace, making it impossible to respond promptly to the impact of coal quality fluctuations on the combustion process. Although some units have deployed continuous monitoring systems for flue gas carbon dioxide emissions, carbon emission data has not yet been effectively integrated into the combustion control closed loop. Second, there is a lack of multi-objective collaborative optimization mechanisms. Existing methods mostly focus on single performance indicators (such as boiler thermal efficiency or nitrogen oxide emission concentration), failing to establish a collaborative optimization model covering multiple dimensions such as boiler efficiency, carbon emission intensity, nitrogen oxides, sulfur dioxide, and particulate matter emissions. This makes it difficult to achieve a comprehensive balance between energy efficiency improvement, pollutant reduction, and carbon emission reduction under complex operating conditions such as variable coal quality and deep peak shaving. Third, the control strategy is not refined enough. The adjustment of key operating variables such as air-coal ratio, secondary air ratio, and feedwater flow rate is highly dependent on manual experience. There is a lack of a feedforward-feedback composite control architecture based on real-time data, which can easily lead to parameter coupling oscillation and regulation lag. This results in increased fluctuations in fly ash carbon content and increased incomplete combustion losses, thereby reducing boiler efficiency and increasing carbon emission intensity.

[0023] With the gradual maturation and engineering application of rapid coal quality testing devices, online fly ash carbon content monitoring equipment, and continuous carbon emission monitoring systems, real-time acquisition of key three carbon data for boilers (i.e., coal quality, fly ash carbon content, and flue gas CO2 emissions) has become possible. Based on this, there is an urgent need to construct an intelligent combustion optimization control method that integrates multi-source real-time data to achieve coordinated control of efficient, clean, and low-carbon operation of coal-fired boilers under medium and low load conditions.

[0024] The following describes a boiler combustion carbon reduction and pollution control method according to an embodiment of this application, with reference to the accompanying drawings.

[0025] Specifically, Figure 1 This is a schematic flowchart of a boiler combustion carbon reduction and pollution control method provided in an embodiment of this application.

[0026] like Figure 1 As shown, the boiler combustion carbon reduction and pollution control method includes the following steps: In step S101, coal quality data, fly ash images, and boiler operating parameters of the boiler are acquired.

[0027] It is understood that the embodiments of this application can comprehensively perceive key state information of the combustion process by acquiring boiler coal quality data, fly ash images and boiler operating parameters in real time, providing a high-timeliness and high-precision data foundation for subsequent combustion efficiency assessment and emission characteristic analysis, and effectively overcoming the information lag problem caused by traditional offline detection.

[0028] It should be noted that coal quality data can be obtained in real time through online rapid coal quality testing devices (such as near-infrared, laser-induced breakdown spectroscopy (LIBS), microwave or gamma-ray analyzers, etc.), mainly including: lower heating value on received basis, total moisture, volatile matter on dry ash-free basis, ash content on received basis, fixed carbon on received basis, elemental analysis data, Hardgrove grindability index and ash fusion point.

[0029] Fly ash images are acquired in real time by high-definition industrial cameras or optical imaging systems installed in fly ash sampling ports or flues. They serve as input for carbon content prediction models. Fly ash images refer to digital images of fly ash particles collected from the tail flue of the boiler or the inlet of the dust collector. They are used to characterize the state of residual carbon after combustion and typically include: grayscale distribution of fly ash particles, particle morphology (such as sphericity and surface roughness), number and distribution characteristics of black spots or unburned carbon particles, and image texture information (reflecting the carbon content).

[0030] Boiler operating parameters refer to key operating condition data collected in real time by a distributed control system or sensors during boiler operation. These mainly include: load, main steam pressure and temperature, feedwater flow and temperature, total air volume, primary / secondary air volume and temperature, furnace outlet oxygen content (O2 concentration), flue gas temperature (furnace outlet, economizer outlet, etc.), exhaust gas temperature, burner sway angle or nozzle opening, number of coal mills in operation and output, nitrogen oxide (NOx) and sulfur dioxide (SO2) emission concentrations, and boiler efficiency-related heat loss parameters (such as exhaust gas heat loss and estimated values ​​of incomplete mechanical combustion heat loss). No specific limitations are imposed.

[0031] In step S102, the fly ash image is input into the carbon content prediction model, and the carbon content prediction model outputs the corresponding fly ash carbon content.

[0032] It is understood that the embodiments of this application can input fly ash images into the carbon content prediction model, which can automatically identify and quantify the unburned carbon content based on the morphology, grayscale and texture features of ash particles in the image, realize rapid, non-contact online prediction of fly ash carbon content, effectively replace the traditional manual sampling and offline testing methods, significantly improve the timeliness and continuity of detection, and provide key data support for real-time evaluation of combustion efficiency and closed-loop optimization control.

[0033] In this embodiment, the carbon content prediction model includes: an image preprocessing and feature extraction layer, a principal component analysis (PCA) dimensionality reduction layer, a convolutional neural network (CNN) feature enhancement layer, and a support vector regression (SVR) output layer. The image preprocessing and feature extraction layer preprocesses the fly ash image and extracts the corresponding feature vectors, which include texture, shape, and color feature vectors. The PCA dimensionality reduction layer performs PCA to reduce the feature vectors to dimensionality-reduced features. The CNN feature enhancement layer extracts deep spatial features based on the dimensionality-reduced features. The SVR output layer determines the corresponding fly ash carbon content by querying a mapping table based on the deep spatial features.

[0034] It is understood that the carbon content prediction model in this application obtains the texture, shape, and color feature vectors of fly ash images through image preprocessing and feature extraction layers. The principal component analysis dimensionality reduction layer effectively compresses redundant information and retains key discriminative features. Then, the convolutional neural network feature deepening layer mines deep spatial nonlinear correlations. Finally, the corresponding fly ash carbon content is accurately mapped through the support vector regression output layer. This achieves high-precision, end-to-end online prediction of carbon content from the original image, significantly improving the model's generalization ability, computational efficiency, and prediction stability, and providing reliable technical support for real-time perception and intelligent optimization control of boiler combustion status.

[0035] It should be noted that, as Figure 2 As shown, the carbon content prediction model first preprocesses the collected fly ash images, including denoising, enhancement, and standardization operations, to improve image quality and eliminate interference factors such as illumination and angle. Then, it extracts key features reflecting the combustion state from the processed images, including texture features (such as gray-level co-occurrence matrix parameters), shape features (such as particle roundness and contour complexity), and color features (such as gray-level mean and black spot ratio), forming a high-dimensional feature vector. To reduce data redundancy and retain the most discriminative information, this feature vector is further dimensionality-reduced through principal component analysis, generating low-dimensional but information-dense dimensionality-reduced features.

[0036] Building upon this foundation, the model utilizes a convolutional neural network to perform deep spatial feature mining on the dimensionality-reduced features, automatically learning nonlinear patterns and local structural correlations related to unburned carbon content in fly ash particle images. Finally, the extracted deep features are input into a support vector regression module, which, through a pre-trained mapping table or regression function, accurately outputs the corresponding fly ash carbon content value. This entire process achieves end-to-end, high-precision online prediction of carbon content from the original fly ash image, effectively supporting real-time perception and closed-loop optimization control of the combustion state.

[0037] The online monitoring model for fly ash carbon content employs a periodic incremental learning mechanism for updating, triggering model optimization every 15 minutes. During each update, the system collects feature vectors from the most recent group of fly ash images using a sliding window approach and performs dimensionality reduction on high-dimensional features using principal component analysis to reduce computational complexity while retaining key information. The dimensionality-reduced data is then input into a hybrid model consisting of a convolutional neural network and support vector regression for training. To address operational drift or changes in data distribution, the system calculates the KL divergence between the current data distribution and the historical baseline distribution in real time. When this value exceeds a threshold of 0.15, a transfer learning strategy is automatically activated: the parameters of the convolutional layers in the CNN are frozen to retain the learned general feature extraction capabilities, and only the Lagrange multipliers αi, αi*, and the bias term b in the support vector regression module are fine-tuned. The entire online learning process ensures model prediction performance while strictly constraining inference time to below 200 milliseconds and maintaining a coefficient of determination of no less than 0.85, thus balancing accuracy, efficiency, and adaptability.

[0038] Specifically, the process of constructing the online monitoring model for carbon content in fly ash includes: after converting the fly ash image to grayscale, denoising, and performing edge detection, extracting texture features, shape features, and color features to construct a high-dimensional feature vector; using principal component analysis to reduce dimensionality and retain principal components with a cumulative variance contribution rate ≥95%; after extracting spatial features through a convolutional neural network, establishing a regression mapping using support vector regression, with the kernel function being the RBF (Radial Basis Function) kernel, and the hyperparameters being determined through Bayesian optimization.

[0039] The construction process of the online monitoring model for carbon content in fly ash includes the following steps: 1) Preprocessing the raw fly ash images acquired in real time by industrial cameras or optical imaging systems, sequentially performing grayscale conversion, Gaussian or median filtering for noise reduction, and edge detection operations such as Canny or Sobel to eliminate uneven illumination, background interference, and noise effects, enhance particle boundary clarity, and provide a high-quality image foundation for subsequent feature extraction. 2) Systematically extracting multi-dimensional features from the preprocessed images: In terms of texture features, calculating the contrast, correlation, energy, and entropy of the gray-level co-occurrence matrix; in terms of shape features, extracting geometric attributes such as particle area, perimeter, roundness, eccentricity, and convex hull ratio; in terms of color features, based on parameters such as gray-level histogram or spatial statistical mean, variance, and the proportion of black spot pixels. The above features are fused to form a high-dimensional original feature vector, comprehensively representing the distribution state of unburned carbon in fly ash. 3) To reduce feature redundancy and improve model training efficiency, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional feature vector, retaining principal components with a cumulative variance contribution rate of no less than 95%, ensuring that the original information's discriminative ability is preserved to the maximum extent while compressing the dimensionality. 4) The dimensionality-reduced features are input into a convolutional neural network (CNN) for deep spatial feature mining. The CNN automatically learns local structural patterns, texture combinations, and spatial correlation features highly correlated with carbon content in fly ash particle images through multi-layer convolution and pooling operations, generating more abstract and robust high-order feature representations. 5) The deep features output by the CNN are used as input to construct a support vector regression (SVR) model to predict continuous values ​​of fly ash carbon content. This SVR model uses a radial basis function (RBF) kernel because it can effectively handle nonlinear mapping relationships and has good generalization performance. Key hyperparameters of the RBF kernel (such as the penalty coefficient C and kernel width γ) are automatically optimized using Bayesian optimization methods to minimize cross-validation error, avoiding the subjectivity and inefficiency of manual parameter tuning, thereby obtaining optimal regression performance. Therefore, the entire model building process realizes end-to-end, high-precision, non-contact online prediction of fly ash carbon content from the original image, and has strong adaptability, real-time performance and engineering deployability.

[0040] In step S103, the boiler efficiency and carbon emission intensity are determined based on coal quality data, fly ash carbon content, and boiler operating parameters.

[0041] It is understood that the embodiments of this application can comprehensively reflect fuel characteristics, combustion completeness and equipment operating status by integrating coal quality data, fly ash carbon content and boiler operating parameters, thereby accurately calculating boiler thermal efficiency and carbon emission intensity per unit of power generation. This overcomes the evaluation lag and bias caused by traditional methods relying on steady-state assumptions or offline data, and achieves real-time, dynamic and high-precision quantification of boiler energy efficiency and carbon emission levels, providing a reliable decision-making basis for multi-objective collaborative optimization control.

[0042] In this embodiment, determining boiler efficiency and carbon emission intensity based on coal quality data, fly ash carbon content, and boiler operating parameters includes: inputting coal quality data, fly ash carbon content, and boiler operating parameters into a target model, and the target model outputting boiler efficiency and carbon emission intensity; wherein, the target model includes a first long short-term memory network layer and a second long short-term memory network layer, the first long short-term memory network layer is used to extract the time-dependent feature sequences corresponding to coal quality data, fly ash carbon content, and boiler operating parameters in each time step; the second long short-term memory network layer is used to calculate boiler efficiency and carbon emission intensity by fusing multi-scale features based on the time-dependent feature sequences.

[0043] It is understood that the embodiments of this application can input coal quality data, fly ash carbon content and boiler operating parameters into a target model containing a two-layer long short-term memory network. The first layer effectively captures the temporal dependency features of multi-source input data at each time step, and the second layer further integrates dynamic features at different time scales, thereby achieving high-precision, real-time joint estimation of boiler efficiency and carbon emission intensity. This significantly improves the accuracy and timeliness of energy efficiency and emission assessment under variable load and coal quality fluctuation conditions, and provides a reliable and dynamic quantitative basis for subsequent multi-objective optimization control.

[0044] In this embodiment of the application, the calculation formula for the target model is: , ; in, The weight matrix for the efficiency output layer. This is the final temporal dependency feature vector extracted after passing through a two-layer long short-term memory network. For activation function, For the bias term of the efficiency output layer, The weight matrix for the carbon emission intensity output layer. This is the bias term for the carbon emission intensity output layer. For boiler efficiency, Carbon emission intensity; in, ; The temporal features are the output of the first Long Short-Term Memory (LSTM) network layer. For the second long short-term memory network layer in The temporal dependency features of the output at each time step. For the second long short-term memory network layer in Temporal dependency features of output at each time step; in, ; for Input data at any time, For the first long short-term memory network layer in The temporal dependency features of the output at each time step. For the first long short-term memory network layer in Temporal dependency features of output at each time step; Wherein, the loss function is: ; in, This is the predicted value for boiler efficiency. This represents the true value of boiler efficiency. This is the predicted value of carbon emission intensity. This represents the true value of carbon emission intensity. This is the predicted carbon content of fly ash. This represents the true carbon content of fly ash. , , These are the weights for efficiency loss, carbon emission loss, and fly ash carbon content loss, respectively.

[0045] It is understood that the embodiments of this application can extract and fuse the temporal dynamic features of coal quality data, fly ash carbon content, and boiler operating parameters step by step through a two-layer long short-term memory network structure. The first layer captures the local dependencies of input variables in the time dimension, and the second layer further integrates multi-scale temporal information to generate a high-dimensional representation vector. Boiler efficiency and carbon emission intensity are output through linear mapping with activation functions, respectively, to achieve simultaneous and accurate prediction of both. At the same time, the model uses a weighted joint loss function that includes three indicators: boiler efficiency, carbon emission intensity, and fly ash carbon content, for end-to-end training, which effectively improves the prediction consistency and generalization ability of each output variable. Under complex and variable operating conditions, it significantly enhances the real-time perception accuracy of boiler energy efficiency and carbon emission status, providing a reliable and coordinated quantitative basis for intelligent combustion optimization.

[0046] It should be noted that, as Figure 2 As shown, the target model first organizes the real-time collected coal quality data, fly ash carbon content, and boiler operating parameters into an input vector according to time series, and then inputs it into the first Long Short-Term Memory (LSTM) network layer. This layer processes the input data step by step through a gating mechanism, effectively capturing the dynamic changes and short-term dependencies of each variable in the time dimension, and outputs a hidden state vector containing time-series features. Subsequently, this vector is passed as input to the second long short-term memory network layer, which further integrates historical information from different time scales to uncover long-term coupling relationships and deep temporal patterns among multiple variables, generating a more representative final temporal-dependent feature vector. .

[0047] Based on these high-dimensional temporal features, the model calculates boiler efficiency and carbon emission intensity through two independent output mapping structures: boiler efficiency is output through a fully connected layer with an activation function to ensure its physical rationality and nonlinear fitting ability; carbon emission intensity is directly output through a linear mapping, balancing computational efficiency and accuracy. The entire model is trained end-to-end using a weighted joint loss function that includes prediction errors for boiler efficiency, carbon emission intensity, and fly ash carbon content. This allows the network to simultaneously optimize multiple related tasks during the learning process, thereby improving the consistency, robustness, and generalization ability of the prediction results, and achieving high-precision, real-time joint estimation of boiler energy efficiency and carbon emission status.

[0048] Specifically, the boiler energy efficiency-carbon efficiency dynamic calculation model adopts a two-layer long short-term memory (LSTM) network architecture, aiming to fully explore the temporal dynamic characteristics and multi-scale coupling relationships of multi-source input data during boiler operation. The model's input includes real-time collected coal quality data, fly ash carbon content, and boiler operating parameters, which are organized according to time series and first fed into the first-layer LSTM network. This layer focuses on capturing the local temporal dependencies of each variable within a short time window, such as the immediate impact of load fluctuations, air-coal ratio adjustments, or coal quality changes on the combustion state, and outputs the hidden state vector for each time step.

[0049] Subsequently, the output of the first layer serves as the input to the second LSTM layer, which further integrates long-term dynamic information across time steps, achieving deep fusion of features at different time scales (such as minute-level adjustment response and hour-level operating condition drift). Through this hierarchical time series modeling, the model can more comprehensively depict the nonlinear evolution of the boiler system under complex operating conditions such as variable load and coal quality disturbance, generating a comprehensive time series feature vector with high representational capabilities.

[0050] In the output phase, the model employs a parallel structure to predict two core performance indicators: boiler efficiency and carbon emission intensity. The boiler efficiency prediction branch typically incorporates activation functions such as Sigmoid or Softplus to ensure the output value falls within a physically reasonable range (e.g., 0.8–0.95), while the carbon emission intensity branch uses linear output to maintain numerical accuracy and interpretability. Furthermore, although fly ash carbon content is not the primary output objective of this model, it is closely related to combustion efficiency and coal consumption, and is included as an auxiliary task in the loss function during training, forming a multi-task learning mechanism.

[0051] The overall loss function is designed as a weighted multi-task loss, reflecting equal emphasis on boiler efficiency and carbon emission intensity. Simultaneously, fly ash carbon content is introduced as an auxiliary monitoring signal to enhance the model's ability to perceive the degree of combustion completeness. This design not only improves the prediction accuracy of the main task but also enhances the model's robustness in the face of data noise or partial sensor failure. Through the above structure and training strategy, this dynamic computation model can achieve high-precision, real-time, and collaborative estimation of boiler energy efficiency and carbon efficiency under typical operating scenarios such as low to medium loads and variable coal quality.

[0052] In step S104, with the objectives of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration, the target control parameters corresponding to different operating conditions are calculated.

[0053] It is understood that the embodiments of this application can take the optimization of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption and minimizing nitrogen oxide emission concentration as the multi-objective optimization guide, comprehensively consider the coupling constraint relationship under different load and coal quality conditions, and use intelligent optimization algorithms to solve the optimal combination of control parameters under the corresponding operating conditions. This effectively achieves a synergistic balance between energy efficiency improvement, low-carbon operation, economic improvement and pollutant emission reduction in the combustion process, and significantly enhances the boiler's adaptive regulation capability and comprehensive operating performance under medium and low loads and variable operating conditions.

[0054] In this embodiment of the application, with the objectives of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration, the target control parameters corresponding to different operating conditions are calculated, including:

[0055] in, (x) = 1- The reciprocal of boiler efficiency (x) = C(C) ) For carbon emission intensity, (x) =C( ) This refers to the carbon content of fly ash. Nitrogen oxide emission concentration; The constraints are as follows:

[0056] Wherein, O2 represents the actual oxygen content of the flue gas during boiler operation. The minimum oxygen content in the flue gas. P represents the maximum oxygen content in the flue gas and the actual unit load. This is the minimum safe and stable load allowed for the unit. This is the maximum allowable load for the unit. This represents the actual nitrogen oxide emission concentration. These are the limits for nitrogen oxide emissions.

[0057] It is understood that the embodiments of this application can construct a multi-objective optimization model with the objectives of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration. The reciprocal of boiler efficiency, carbon emission intensity, fly ash carbon content, and nitrogen oxide emission concentration are uniformly incorporated into the optimization function. Combined with actual operating constraints such as flue gas oxygen content range, unit load upper and lower limits, and nitrogen oxide emission limits, the multi-objective optimization algorithm is used to solve for the optimal control parameters under different operating conditions. This effectively balances combustion efficiency, economy, low carbon emissions, and environmental protection, and achieves synergistic improvement of multi-dimensional performance indicators while ensuring the safe and stable operation of the boiler. This significantly enhances the system's intelligent control capability and comprehensive operating efficiency under complex and variable operating conditions.

[0058] It should be noted that, in the embodiments of this application, the optimal compromise solution can also be selected using a fuzzy membership function through multi-objective intelligent optimization: ,in and The first The maximum and minimum values ​​of the objective function at the Pareto front. The number of objective functions; the capacity of the dynamic knowledge base. Records, supports K-nearest neighbor fast search, search response time .

[0059] In the multi-objective intelligent optimization process, a fuzzy membership function is used to evaluate the solution set in the Pareto front to select the compromise solution with the best overall performance. The membership function is calculated as follows: for each candidate solution, its value on each objective function is first obtained, and then a normalized satisfaction index is constructed by combining the maximum and minimum values ​​of that objective in the entire Pareto front; subsequently, the normalized terms of all objectives are multiplied together, and the k-th root is taken (where k is the total number of optimization objectives), thus obtaining the overall fuzzy membership degree of the solution. The closer the membership degree is to 1, the better the balance achieved by the solution among multiple objectives.

[0060] Meanwhile, the system is equipped with a dynamic knowledge base with a capacity of no less than 5,000 historical optimization records, used to store past high-quality solutions and their corresponding operating condition characteristics. This knowledge base supports a fast retrieval mechanism based on the K-nearest neighbor algorithm. When a new operating condition is received, it can return a reference solution for a similar scenario within 100 milliseconds, providing an initial solution or decision support for the current optimization, significantly improving the solution efficiency and practicality.

[0061] Specifically, such as Figure 2As shown, the multi-objective intelligent optimization process focuses on minimizing the objective function vector, the reciprocal of boiler efficiency, which is equivalent to maximizing boiler thermal efficiency by minimizing this value. The carbon dioxide emission intensity per unit of electricity generation reflects the level of low-carbon operation; fly ash carbon content serves as a proxy indicator of incomplete combustion loss, indirectly characterizing operating coal consumption, and its reduction signifies more efficient fuel utilization; nitrogen oxide emission concentration in flue gas is used to measure environmental performance. The optimization variables are adjustable combustion operation parameters, including but not limited to total air volume, primary / secondary air ratio, burnout air opening, burner tilt angle, coal feed rate, and furnace oxygen setpoint, to cover the main degrees of freedom of the boiler combustion system.

[0062] The optimization process strictly meets the actual operational safety and environmental constraints, including: maintaining the flue gas oxygen content within a first set range to ensure a balance between combustion stability and economy; keeping the unit's load within a second set range to meet the equipment's safe and stable operation boundaries; and ensuring that nitrogen oxide emission concentrations do not exceed the prescribed emission limits. To solve this high-dimensional, nonlinear, multi-objective constrained optimization problem, a multi-objective particle swarm optimization algorithm is employed. This algorithm initializes a population of 100 particles, each representing a feasible combination of control parameters. During 200 iterations, particles dynamically update their flight speed and position based on their individual historical best solutions and the group's non-dominated frontier. The algorithm sets an inertia weight that decreases linearly within a set interval, initially emphasizing global exploration capabilities and later enhancing local convergence accuracy. Through evolutionary search using MOPSO (Multi-Objective Particle Swarm Optimization), a uniformly distributed and well-converged optimal solution set is ultimately obtained, comprehensively reflecting the boiler's trade-offs between energy efficiency, low carbon emissions, energy saving, and emission reduction. Operators or the upper-level control system can select the most suitable compromise solution based on the current scheduling needs (such as prioritizing carbon reduction or prioritizing stable combustion) and generate corresponding target control parameter instructions to achieve intelligent and refined combustion control oriented towards the "dual carbon" target.

[0063] In step S105, the boiler is controlled to perform corresponding carbon reduction and pollution control operations according to the target control parameters.

[0064] It is understood that the embodiments of this application can adjust key operating variables such as the air-coal ratio, secondary air distribution, burner angle, coal feed rate and oxygen content of the boiler in real time according to the target control parameters obtained by optimization calculation, so as to drive the boiler to perform precise carbon reduction and pollution reduction control operations, effectively reduce the carbon content of fly ash and incomplete combustion loss, suppress the generation of nitrogen oxides, improve combustion efficiency and reduce carbon emissions per unit of power generation, and achieve synergistic optimization of energy efficiency, environmental protection and economy while ensuring the safe and stable operation of the unit.

[0065] In this embodiment of the application, after controlling the boiler to perform corresponding carbon reduction and pollution reduction control operations according to the target control parameters, the process includes: acquiring coal quality data, fly ash images, and boiler operating parameters after the carbon reduction and pollution reduction control operations; if the difference between the coal quality data, fly ash images, and boiler operating parameters after the carbon reduction and pollution reduction control operations and the historical coal quality data, historical fly ash images, and historical boiler operating parameters is greater than a preset threshold, then the current operating condition is determined to have changed, and the target model is trained using the coal quality data, fly ash images, and boiler operating parameters of the changed operating condition until the model converges.

[0066] The preset threshold can be set according to actual needs without specific limitations.

[0067] It is understood that, after performing carbon reduction and pollution control operations, this application embodiment can continuously collect and update coal quality data, fly ash images, and boiler operating parameters, and compare them with historical operating condition data. When the difference exceeds a preset threshold, it is automatically identified as a significant change in operating conditions. Then, the target model is retrained online using multi-source data under the current new operating conditions until the model converges, thereby realizing dynamic adaptive updating of model parameters. This effectively improves the generalization ability, prediction accuracy, and control robustness of the boiler combustion optimization system under complex scenarios such as coal quality fluctuations and load adjustments, ensuring the long-term stability and continuous optimization of carbon reduction and pollution control effects.

[0068] It should be noted that the embodiments of this application can also establish a dynamic matrix model of multivariate coupling of wind, coal, water, and heat through a multivariate predictive control model to achieve the prediction of key parameters. Specifically, by conducting step or impulse response experiments under typical boiler operating conditions, the dynamic influence of key input variables such as primary / secondary air, coal feed rate, feedwater flow rate, and desuperheating water flow rate on output variables such as main steam temperature, furnace outlet oxygen content, flue gas temperature, and boiler efficiency is identified. A high-dimensional dynamic matrix is ​​constructed to characterize the strong coupling, large inertia, and nonlinear dynamic characteristics among wind, coal, water, and heat. This matrix serves as the core kernel of model predictive control and is used to predict the evolution trend of key operating parameters in the future prediction time domain. This achieves advanced perception and accurate prediction of the boiler's multivariate system, providing a reliable dynamic model foundation for subsequent optimization decisions, overcoming the problem that traditional single-loop control cannot handle mutual interference between variables, and improving the overall coordination and response capability of the system.

[0069] The instruction smoothing layer of the two-layer collaborative control architecture adopts fuzzy rules: when hour, sign ,when hour, ,in, The adjustment range of the coal feeder's feed rate is set according to the characteristics of the actuator. min, fan frequency adjustment range Hz / min; The dynamic compensation layer constructs an MPC optimization problem, which is solved once every 1 minute using a quadratic programming solver. This represents the ideal control command value calculated by the upper-level multi-objective optimization module at the current moment. The value of the control command actually issued and executed in the previous control cycle. This represents the change (adjustment magnitude) of the current optimization instruction relative to the previous time step. This refers to the smoothed control command that is actually transmitted to the lower-level dynamic compensation controller after smoothing.

[0070] The dual-layer collaborative control architecture of this application consists of an instruction smoothing layer and a dynamic compensation layer. Specifically, the upper instruction smoothing layer receives ideal control instructions generated by the multi-objective optimization module and dynamically conditions them using a fuzzy logic controller. Based on the current load state, equipment operation rate limits, and safety boundaries, it intelligently constrains the instruction change rate and amplitude to avoid frequent actuator movements or system oscillations caused by sudden changes in optimized instructions.

[0071] The lower dynamic compensation layer, based on the aforementioned multivariable dynamic matrix model, employs a model predictive control strategy. It combines real-time feedback deviations and measurable disturbances (such as coal quality changes) to solve the optimal control increment sequence within a finite time domain online, implementing only the first control step. Simultaneously, it integrates feedforward (to address known disturbances) and feedback (to correct model mismatch) mechanisms to achieve high-precision dynamic tracking and anti-interference adjustment of the combustion process. The upper layer ensures the safety, executability, and equipment friendliness of the optimization commands, while the lower layer guarantees the dynamic accuracy and robustness of the closed-loop control. Together, they achieve an organic unity between optimized decision-making and stable execution, balancing economy and safety.

[0072] The model parameters are updated online through a sliding window and the transfer learning is used to adapt to drastic changes in operating conditions: The sliding time window mechanism is used to continuously collect the latest operating data (such as the most recent 2-4 hours), and the dynamic matrix model or state space parameters are periodically identified or recursively updated online to keep the model close to the current operating point and adapt to the gradual characteristics such as slow coal type drift or equipment aging.

[0073] When a drastic change in operating conditions is detected (such as a sudden 30% load drop or a change in coal type), a transfer learning mechanism is triggered: a pre-trained model under similar historical operating conditions is used as prior knowledge, and fine-tuned with a small amount of current new operating condition data. This quickly reconstructs a predictive model suitable for the new operating conditions, avoiding the performance gap caused by training from scratch. This enhances the adaptive capability of the control model in long-term operation, enabling it to cope with both regular slow-changing disturbances and rapid responses to sudden, large-scale operating condition changes. It significantly improves the system's control continuity, stability, and generalization performance in complex scenarios such as deep peak shaving and multi-coal blending. Specifically, the system maintains a sliding time window of length L=1440 (corresponding to 24 hours, sampled at the minute level) to store the most recent historical operating data (including coal quality, fly ash image features, boiler parameters, actual efficiency, and emission values). The window slides forward in steps ΔL=60 (i.e., every hour), continuously removing the oldest data and incorporating the latest data to ensure that model training is always based on the most representative recent operating conditions.

[0074] The Kullback-Leibler divergence (KL divergence) is used as a measure of distribution difference to compare the deviation between the current input data distribution and the training data distribution of the benchmark model in real time. When the KL divergence exceeds a preset threshold θ=0.15, it is determined that the system has experienced significant operating condition drift (such as coal type switching, normalization of deep peak shaving, sensor offset, etc.), triggering the model update process.

[0075] To avoid the problems of high computational cost, slow convergence, and forgetting of historical knowledge caused by training from scratch, a transfer learning strategy is adopted: the parameters of the underlying network of the model (such as the CNN feature extraction layer or the LSTM temporal coding layer) are frozen, and only the top output layer (such as the regression head of boiler efficiency and carbon emission intensity) is fine-tuned. This preserves the learned general combustion feature representation ability and quickly adapts the output mapping relationship to the new operating conditions.

[0076] It should be noted that the KL divergence metric for data drift detection is used to quantify the difference between the distribution of newly acquired data and the distribution of model training data. Specifically, the system continuously calculates the KL divergence between the current data distribution Q(x) and the reference distribution P(x) (i.e., the data distribution during model training). The calculation formula is D. KL (P||Q)= .in, Based on the empirical distribution of current actual data, As a benchmark distribution, To represent the additional information brought by encoding P with distribution Q when event x occurs, when the KL divergence exceeds the preset threshold of 0.15, it indicates that the data distribution has shifted significantly, triggering the incremental learning mechanism.

[0077] In incremental learning, the system employs a transfer learning strategy: freezing the parameters of the bottom convolutional feature extraction layers of the deep neural network (including convolutional kernel weights and biases) to retain the model's learned general fly ash image feature extraction capabilities; and fine-tuning only the top output layer (including fully connected layers and regression layers) by updating the Lagrange multiplier α. i α i The method, along with a bias b, adapts the model to the new data distribution. This approach effectively avoids forgetting, preserves the model's adaptability to historical conditions, and requires only a small amount of new data to converge quickly, ensuring model adaptation is completed within 1-2 control cycles and meeting real-time control requirements.

[0078] Bayesian update is used to dynamically correct the covariance matrix of the model's prediction error, improving the robustness of prediction accuracy. Specifically, the system employs a Bayesian framework, treating the model's prediction error covariance matrix as a random variable and updating its posterior distribution based on new data. When N=100 new samples have accumulated, the system performs a Bayesian update, calculated using the following formula: In the formula, Σold is the current covariance matrix, and σ² is the variance estimate of the new data. This update mechanism dynamically adjusts the model's confidence in the new data: when the new data is highly consistent with historical data, Σnew is close to Σold; when the new data differs significantly from historical data, Σnew is closer to σ². In this way, the system can adaptively adjust the response intensity to new data while maintaining model stability, effectively preventing overfitting and ensuring a prediction accuracy R² ≥ 0.85 during online learning. This method works in conjunction with the sliding window incremental learning mechanism, enabling the model to continuously adapt to slow shifts in coal quality and load, while rapidly responding to drastic changes in operating conditions, providing reliable data support for combustion optimization control.

[0079] In summary, this application, through image feature fusion and machine learning models, shortens the fly ash carbon content detection cycle from 4-8 hours to minutes, with a relative error of no more than 5% compared to traditional fly ash carbon content detection results, providing real-time status feedback for combustion optimization. For the first time, it integrates rapid coal quality testing, fly ash carbon content, and online carbon emission monitoring data to construct a collaborative optimization model for four objectives: boiler efficiency, operating coal consumption, carbon emissions, and pollutant emissions. This model achieves improved boiler efficiency and reduced operating coal consumption, carbon emission intensity, and pollutant emission concentration under low-to-medium load conditions. The dual-layer control architecture effectively solves the multi-variable coupling problem, achieving smooth parameter adjustments under varying operating conditions and significantly improving boiler operational stability.

[0080] The boiler combustion carbon reduction and pollution control method proposed in this application integrates coal quality data, fly ash images, and boiler operating parameters. It uses a carbon content prediction model to obtain the carbon content of fly ash in real time and combines multi-dimensional data to comprehensively evaluate boiler efficiency and carbon emission intensity. Then, with the goal of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration, it dynamically generates and executes optimal control parameters. This effectively improves the intelligence and precision of the combustion process and achieves synergistic optimization of carbon reduction, pollution reduction, and energy saving under complex operating conditions such as low and medium loads. It significantly reduces incomplete combustion losses and pollutant emissions, and improves the economic efficiency and environmental performance of power plant operation.

[0081] The following will combine Figure 2 The boiler combustion carbon reduction and pollution control method of this application is described in detail below: Step 1: Real-time acquisition and preprocessing of C3 data The system uses a rapid coal quality testing device, an online fly ash carbon content monitoring device, and an online carbon emission monitoring device to collect high-frequency coal quality parameters (such as calorific value, volatile matter, and carbon content), fly ash image features, and flue gas carbon emission data. Missing value imputation (KNN method), outlier removal (3σ criterion), and noise smoothing (Savitzky-Golay filtering) are applied to the raw data to ensure a data validity rate exceeding 99%.

[0082] Step 2: Construction of an online monitoring model for fly ash carbon content After extracting 48-dimensional high-dimensional features from fly ash images, principal component analysis (PCA) was used to reduce the dimensionality to 12, and a CNN-SVR hybrid model was constructed: CNN was used to extract spatial features, and SVR was based on the RBF kernel function to establish a regression mapping. Through Bayesian optimization and automatic parameter tuning, the model was deployed at edge nodes, and the inference time was less than 200 milliseconds, achieving high-precision real-time prediction of the carbon content of fly ash.

[0083] Step 3: Constructing a dynamic calculation model for boiler energy efficiency and carbon efficiency Using over 20 dimensions of features, including coal quality, operating parameters, and fly ash carbon content, as input, a two-layer LSTM network is designed to simultaneously predict boiler efficiency and carbon emission intensity. A weighted multi-task loss function is used to jointly optimize the three indicators of energy efficiency, carbon emission, and fly ash carbon content, and the network is trained until the coefficient of determination R² on the validation set is no less than 0.88.

[0084] Step 4: Multi-objective intelligent optimization and knowledge base construction A multi-objective optimization problem is established with the objectives of minimizing the reciprocal of efficiency, carbon emission intensity, fly ash carbon content, and NOx concentration. Under constraints such as oxygen content, load, and emission limits, a Pareto front solution set is generated using a multi-objective particle swarm optimization algorithm. The optimal compromise solution is selected through a fuzzy membership function, and the solution and its operating conditions are stored in a dynamic knowledge base with a capacity of no less than 5000 entries, supporting fast K-nearest neighbor retrieval with a response time of less than 100 milliseconds.

[0085] Step 5: Construction of Multivariate Predictive Control Model Define control variables (such as coal feed rate, damper opening, etc.) and state variables (efficiency, emissions, etc.), identify the dynamic matrix through step response experiments, establish a state-space model with disturbance compensation, and use Kalman filtering for state estimation to improve control robustness.

[0086] Step 6: Implementation of the Two-Layer Cooperative Control Architecture The upper layer sets up an instruction smoothing layer, which limits the optimized instructions based on the physical limitations of the actuator (such as the upper limit of the coal feeding rate and the fan frequency variation). The lower layer adopts model predictive control, which solves a quadratic programming problem once per minute. Under the premise of satisfying multivariate constraints, it outputs the optimal control increment and sends it to the DCS system to achieve safe and stable regulation.

[0087] Step 7: Closed-loop self-learning and model update The system continuously accumulates running data and uses a sliding window mechanism to manage historical samples. When a significant shift in the data distribution is detected (KL divergence exceeding 0.15), incremental learning is triggered: the underlying parameters of the model are frozen, only the top-level output is fine-tuned, and the error covariance is updated using a Bayesian method. The system ensures a prediction accuracy of R² ≥ 0.85 throughout the process; if performance degrades, a full model retraining is initiated.

[0088] Next, referring to the accompanying drawings, a boiler combustion carbon reduction and pollution control device according to an embodiment of this application is described.

[0089] Figure 3 This is a block diagram of a boiler combustion carbon reduction and pollution control device according to an embodiment of this application.

[0090] like Figure 3 As shown, the boiler combustion carbon reduction and pollution control device 10 includes: an acquisition module 100, a processing module 200, a determination module 300, a calculation module 400, and an execution module 500.

[0091] The module 100 is used to acquire coal quality data, fly ash images, and boiler operating parameters of the boiler; the processing module 200 is used to input the fly ash images into the carbon content prediction model, and the carbon content prediction model outputs the corresponding fly ash carbon content; the determination module 300 is used to determine the boiler efficiency and carbon emission intensity based on the coal quality data, fly ash carbon content, and boiler operating parameters; the calculation module 400 is used to calculate the target control parameters corresponding to different operating conditions with the objectives of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration; and the execution module 500 is used to control the boiler to perform corresponding carbon reduction and pollution reduction control operations based on the target control parameters.

[0092] The boiler combustion carbon reduction and pollution control device proposed in this application integrates coal quality data, fly ash images, and boiler operating parameters. It uses a carbon content prediction model to obtain the carbon content of fly ash in real time and combines multi-dimensional data to comprehensively evaluate boiler efficiency and carbon emission intensity. Then, with the goal of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration, it dynamically generates and executes optimal control parameters. This effectively improves the intelligence and precision of the combustion process and achieves synergistic optimization of carbon reduction, pollution reduction, and energy saving under complex operating conditions such as low and medium loads. It significantly reduces incomplete combustion losses and pollutant emissions, and improves the economic efficiency and environmental performance of power plant operation.

[0093] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 401, the processor 402, and the computer program stored on the memory 401 and capable of running on the processor 402.

[0094] When the processor 402 executes the program, it implements the boiler combustion carbon reduction and pollution control method provided in the above embodiments.

[0095] Furthermore, electronic devices also include: Communication interface 403 is used for communication between memory 401 and processor 402.

[0096] The memory 401 is used to store computer programs that can run on the processor 402.

[0097] Memory 401 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0098] If the memory 401, processor 402, and communication interface 403 are implemented independently, then the communication interface 403, memory 401, and processor 402 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0099] Optionally, in a specific implementation, if the memory 401, processor 402, and communication interface 403 are integrated on a single chip, then the memory 401, processor 402, and communication interface 403 can communicate with each other through an internal interface.

[0100] Processor 402 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0101] This application also provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed by a processor, implements the above-described boiler combustion carbon reduction and pollution control method.

[0102] This application also provides a computer program product, including a computer program or instructions, which, when executed, implement the above-mentioned boiler combustion carbon reduction and pollution control method.

[0103] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0104] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0105] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0106] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0107] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A method for controlling carbon reduction and pollution control in boiler combustion, characterized in that, Includes the following steps: Acquire coal quality data, fly ash images, and boiler operating parameters for the boiler; The fly ash image is input into the carbon content prediction model, and the carbon content prediction model outputs the corresponding fly ash carbon content. Boiler efficiency and carbon emission intensity are determined based on the coal quality data, the carbon content of fly ash, and the boiler operating parameters. With the objectives of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration, the target control parameters corresponding to different operating conditions are calculated. The boiler is controlled to perform corresponding carbon reduction and pollution control operations based on the target control parameters.

2. The boiler combustion carbon reduction and pollution control method according to claim 1, characterized in that, The carbon content prediction model includes: an image preprocessing and feature extraction layer, a principal component analysis dimensionality reduction layer, a convolutional neural network feature enhancement layer, and a support vector regression output layer; The image preprocessing and feature extraction layer is used to preprocess the fly ash image and extract the feature vector corresponding to the preprocessed fly ash image. The feature vector includes texture feature vector, shape feature vector and color feature vector. The principal component analysis dimensionality reduction layer is used to perform principal component analysis on the feature vector to obtain dimensionality-reduced features; The convolutional neural network feature enhancement layer is used to extract deep spatial features based on the dimensionality reduction features; The support vector regression output layer is used to determine the corresponding carbon content of fly ash by querying the mapping relationship table based on the deep spatial features.

3. The boiler combustion carbon reduction and pollution control method according to claim 2, characterized in that, The process of determining boiler efficiency and carbon emission intensity based on the coal quality data, the fly ash carbon content, and the boiler operating parameters includes: The coal quality data, the fly ash carbon content, and the boiler operating parameters are input into the target model, and the target model outputs the boiler efficiency and carbon emission intensity. The target model includes a first long short-term memory network layer and a second long short-term memory network layer. The first long short-term memory network layer is used to extract the time-dependent feature sequences corresponding to the coal quality data, the fly ash carbon content and the boiler operating parameters in each time step. The second long short-term memory network layer is used to calculate the boiler efficiency and carbon emission intensity by fusing multi-scale features based on the time-dependent feature sequences.

4. The boiler combustion carbon reduction and pollution control method according to claim 3, characterized in that, The calculation formula for the target model is: ， ； in, The weight matrix for the efficiency output layer. This is the final temporal dependency feature vector extracted after passing through a two-layer long short-term memory network. For activation function, For the bias term of the efficiency output layer, The weight matrix for the carbon emission intensity output layer. This is the bias term for the carbon emission intensity output layer. For boiler efficiency, Carbon emission intensity; in, ; The temporal features are the output of the first Long Short-Term Memory (LSTM) network layer. For the second long short-term memory network layer in The temporal dependency features of the output at each time step. For the second long short-term memory network layer in Temporal dependency features of output at each time step; in, ; for Input data at any time, For the first long short-term memory network layer in The temporal dependency features of the output at each time step. For the first long short-term memory network layer in Temporal dependency features of output at each time step; Wherein, the loss function is: ; in, This is the predicted value for boiler efficiency. This represents the true value of boiler efficiency. This is the predicted value of carbon emission intensity. This represents the true value of carbon emission intensity. This is the predicted carbon content of fly ash. This represents the true carbon content of fly ash. , , These are the weights for efficiency loss, carbon emission loss, and fly ash carbon content loss, respectively.

5. The boiler combustion carbon reduction and pollution control method according to claim 4, characterized in that, The objective is to maximize boiler efficiency, minimize carbon emission intensity, minimize operating coal consumption, and minimize nitrogen oxide emission concentration. The target control parameters for different operating conditions are calculated, including: in, (x) = 1- The reciprocal of boiler efficiency (x) = C(C) ) For carbon emission intensity, (x) =C( ) This refers to the carbon content of fly ash. Nitrogen oxide emission concentration; The constraints are as follows: Wherein, O2 represents the actual oxygen content of the flue gas during boiler operation. The minimum oxygen content in the flue gas. P represents the maximum oxygen content in the flue gas and the actual unit load. This is the minimum safe and stable load allowed for the unit. This is the maximum allowable load for the unit. This represents the actual nitrogen oxide emission concentration. These are the limits for nitrogen oxide emissions.

6. The boiler combustion carbon reduction and pollution control method according to claim 5, characterized in that, After controlling the boiler to perform corresponding carbon reduction and pollution control operations according to the target control parameters, the following steps are included: Acquire coal quality data, fly ash images, and boiler operating parameters after carbon reduction and pollution control operations; If the difference between the coal quality data, fly ash image, and boiler operating parameters after the carbon reduction and pollution reduction control operation and the historical coal quality data, historical fly ash image, and historical boiler operating parameters is greater than a preset threshold, then the current operating condition is determined to have changed. The target model is then trained using the coal quality data, fly ash image, and boiler operating parameters of the changed operating condition until the model converges.

7. A boiler combustion carbon reduction and pollution control device, characterized in that, include: The acquisition module is used to acquire coal quality data, fly ash images, and boiler operating parameters from the boiler. The processing module is used to input the fly ash image into the carbon content prediction model, and the carbon content prediction model outputs the corresponding fly ash carbon content. The determination module is used to determine the boiler efficiency and carbon emission intensity based on the coal quality data, the fly ash carbon content, and the boiler operating parameters. The calculation module is used to calculate the target control parameters under different operating conditions with the goals of maximizing boiler efficiency, minimizing carbon emission intensity, minimizing operating coal consumption, and minimizing nitrogen oxide emission concentration. The execution module is used to control the boiler to perform corresponding carbon reduction and pollution control operations according to the target control parameters.

8. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the boiler combustion carbon reduction and pollution control method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by the processor, they are used to implement the boiler combustion carbon reduction and pollution control method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed, they implement the boiler combustion carbon reduction and pollution control method as described in any one of claims 1-6.

Images