A method and system for intelligent coal blending and energy optimization in power plants

Abstract

This application discloses a method and system for intelligent coal blending and energy optimization in power plants. The method includes: real-time acquisition of industrial and elemental analysis data of incoming coal; prediction of key combustion characteristic parameters based on a machine learning model; establishing a multi-objective optimization model with the optimization objectives of lowest total fuel cost, optimal comprehensive emission indicators, and highest boiler combustion stability, and considering coal yard inventory, unit load demand, and equipment constraints; and dynamically recommending optimal boiler key operating parameter settings based on a data-driven proxy model, taking into account the current unit load and boundary conditions. This application's method upgrades from single-cost optimization to dynamic collaborative optimization of multiple objectives including cost, environmental protection, and safety. The solution results are more closely aligned with the actual comprehensive needs of the power plant. By dynamically optimizing combustion parameters, it deeply links coal blending decisions with boiler combustion control, achieving collaborative optimization between the fuel side and the equipment side, and improving overall energy conversion efficiency.

Description

Technical Field

[0001] This application relates to the technical field of power plant coal blending systems, specifically a method and system for intelligent coal blending and energy optimization in power plants. Background Technology

[0002] Fuel costs constitute a significant portion of the total cost of coal-fired power plants, and their combustion efficiency and pollutant emissions directly impact the plant's economic viability and environmental impact. Traditional coal blending methods rely heavily on operator experience, using the received lower heating value of coal as the primary reference indicator for simple mixing to meet the calorific value requirements of the coal entering the furnace. This method has several limitations: First, it cannot dynamically respond to fluctuations in coal quality, load changes, and equipment status, resulting in outdated and rigid coal blending schemes that struggle to achieve a dynamic optimal balance among conflicting objectives such as cost, environmental protection, and stable combustion. Second, coal blending decisions and boiler combustion control are typically two independent processes. Coal blending is based solely on coal inventory and calorific value, failing to fully consider the combustion characteristics (such as ignition and burnout characteristics), slagging tendency, and subsequent impacts on the pulverizing and desulfurization / denitrification systems of specific coal types within a particular boiler. This leads to blended coal that may not be the optimal choice under current boiler operating conditions, resulting in energy conversion efficiency losses and increased pollutant emissions.

[0003] While some existing technologies have incorporated linear programming and other optimization algorithms for cost optimization, these models are mostly linear and static, with a single optimization objective (usually just minimizing cost). This makes them ineffective at handling complex multi-objective optimization problems such as nonlinear coal-quality interactions and combustion stability. More importantly, existing methods fail to achieve closed-loop optimization across the entire process, from fuel blending to combustion parameter setting. They cannot autonomously learn patterns from massive amounts of historical operational data, nor can they automatically correct and adaptively adjust the blending and combustion control models based on feedback from actual combustion effects (such as boiler efficiency and pollutant emissions). The systems have low levels of intelligence and rely heavily on manual experience for adjustments.

[0004] Therefore, there is an urgent need for an intelligent coal blending and energy optimization solution that can sense coal quality in real time, perform multi-objective collaborative optimization, dynamically seek optimization, and have self-learning and evolution capabilities, in order to solve the above-mentioned core problems. Summary of the Invention

[0005] This application provides a method and system for intelligent coal blending and energy optimization in power plants, aiming to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this application provides the following technical solution: A method for intelligent coal blending and energy optimization in power plants includes the following steps: S1. Real-time collection of industrial analysis data and elemental analysis data of coal entering the plant, and prediction of its key combustion characteristic parameters based on machine learning models; S2. With the optimization objectives of the plant’s total fuel cost being the lowest, the comprehensive emission index being the best, and the boiler combustion stability being the highest, a multi-objective optimization model is established with coal yard inventory, unit load demand, and equipment constraints as conditions. S3. Based on the optimal coal blending scheme obtained in step S2, and combined with the current unit load and boundary conditions, the optimal set values ​​of key boiler operating parameters are dynamically recommended through a data-based proxy model. S4. Real-time collection of boiler efficiency, pollutant emissions, and key status parameters after step S3, comparison with predicted values, and generation of feedback signals for online correction of the prediction model in step S1 and the surrogate model in step S3.

[0007] As a preferred technical solution of this application, in step S1, the key combustion characteristic parameters include, but are not limited to: received lower heating value, dry ash-free volatile matter, softening temperature, and theoretical flue gas volume; the machine learning prediction model is a gradient boosting decision tree model, and its prediction function is as follows:

[0008] in, The model predicted values ​​of the key combustion characteristic parameters for the i-th coal sample; The input feature vector is its industrial analysis and elemental analysis data; K is the total number of decision trees in the model; Let F be the k-th decision tree; F is the function space of all possible decision trees; the model minimizes a regularized objective function that includes prediction error and model complexity. To obtain through training, among which The loss function is used to measure the model's predicted values. and actual observed values The error between them This is a regularization term used to control the k-th decision tree. Model complexity, These are the actual observed values ​​of the key combustion characteristic parameters for the i-th coal sample.

[0009] As a preferred technical solution of this application, in step S2, the specific form of the multi-objective optimization model is as follows: Minimize:

[0010] satisfy:

[0011] Where x is the decision variable vector, i.e. the proportion of various types of coal; R represents the total fuel cost, and R represents the total quantity of different types of coal used in the blending. These represent the unit price and blending quality of the r-th type of coal, respectively. For comprehensive emission indicators, , SO2 and NO respectively x The weighting coefficient of dust. , , SO2 and NO, respectively, are predicted based on coal quality and operating parameters. x Dust emission concentration; It is a negative indicator of combustion stability. This represents the standard deviation function, used to evaluate the uniformity of volatile matter in coal blending. The dry ash-free volatile matter of the r-th type of coal, The blending quality of the r-th type of coal, Total coal quantity; Let j be the j-th inequality constraint function; there are m inequality constraint functions in total. Let p be the p-th equality constraint function; there are q such functions in total. and Decision variables The lower and upper bounds of the value.

[0012] As a preferred technical solution of this application, in step S2, the multi-objective optimization model is solved by a non-dominated sorting genetic algorithm with an elite strategy to obtain the Pareto optimal solution set, and the compromise optimal solution is selected as the coal blending instruction by the membership function method.

[0013] As a preferred technical solution of this application, in step S3, the data-based proxy model is a deep neural network, and its input layer is a vector. Where B is the coal quality characteristic vector of the current optimal coal blending scheme, and L is the current and predicted short-term unit load. The output layer represents the combined state of the coal mills; the output layer is a vector of recommended values ​​for key boiler operating parameters. ,in This refers to the temperature of the boiler feedwater when it enters the economizer or directly into the steam drum / water-cooled wall. The ratio of the actual amount of air fed into the furnace to the theoretical amount of air required for complete fuel combustion is a core parameter for boiler combustion adjustment. The percentage opening of the boiler burnout damper is a key adjustable parameter of the boiler combustion system; the surrogate model is trained using historical operating data, with the goal of minimizing the mean square error between the actual operating energy efficiency and the predicted energy efficiency under recommended parameters.

[0014] As a preferred technical solution of this application, in step S4, the closed-loop feedback calculates the deviation between the actual values ​​of key parameters and the model prediction values. This is achieved by using a recursive least squares method with a forgetting factor to update the weight parameters θ of the machine learning prediction model and the surrogate model online.

[0015]

[0016]

[0017] in, Let be the weight parameter vector of the model at time t; The Kalman gain matrix at time t+1 is used to control the step size / magnitude of parameter updates; These are the actual observed values ​​of the key parameters at time t+1; P represents the model's predicted values ​​for the key parameters at time t+1. t Let be the covariance matrix at time t, representing the uncertainty of the parameter estimation; λ is the model input feature vector at time t+1; λ is the forgetting factor, used to control the weight of historical data; I is the identity matrix.

[0018] A smart coal blending and energy optimization system for power plants includes: The coal quality sensing and digitization module is used to execute step S1; The intelligent coal blending optimization decision module is used to execute step S2; The combustion parameter dynamic optimization module is used to execute step S3; The closed-loop learning and adaptive module is used to execute step S4; A unified data platform and model library provide data support and model services for the above modules.

[0019] As a preferred technical solution of this application, the coal quality sensing and digitization module includes an online coal quality analyzer, a digital coal yard management system, and a coal quality prediction model server; the intelligent coal blending optimization decision-making module is deployed at the plant-level monitoring information system layer and communicates with the fuel management system and the unit load scheduling system.

[0020] As a preferred technical solution of this application, the combustion parameter dynamic optimization module communicates with the distributed control system and sends the recommended key operating parameter settings to the boiler main control loop in the form of guidance values ​​or feedforward signals.

[0021] As a preferred technical solution of this application, the closed-loop learning and adaptive module continuously collects feedback data during operation and is equipped with a model performance evaluation and early warning unit. When the model prediction deviation continues to exceed the set threshold, the model retraining process is automatically triggered or a manual intervention alarm is issued.

[0022] The above-mentioned technical solution of this application has the following beneficial technical effects: The intelligent coal blending and energy optimization method for power plants in this application effectively solves the core problems of traditional coal blending methods, such as isolation, lag, and difficulty in collaborative optimization, by constructing a complete closed-loop process that includes coal quality prediction, multi-objective dynamic optimization, collaborative recommendation of combustion parameters, and closed-loop learning. This method upgrades from a single cost-oriented approach to a multi-objective dynamic collaborative optimization that considers cost, environmental protection, and stability. Its solution is more closely aligned with the comprehensive needs of actual power plant operation. Simultaneously, by introducing a data-based surrogate model to dynamically recommend optimal operating parameters, it deeply correlates coal blending decisions with combustion control, achieving collaborative optimization between the fuel side and the equipment side. This improves the overall energy conversion efficiency of the boiler and ensures the system can continuously learn and adapt to changes through a closed-loop feedback mechanism, significantly improving the power plant's intelligence level and operational economy, environmental friendliness, and safety. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. These drawings are incorporated in and constitute a part of this specification. They illustrate embodiments conforming to this application and, together with the specification, serve to explain the technical solutions of this application. It should be understood that the following drawings only show some embodiments of this application and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a smart coal blending and energy optimization method for power plants. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] refer to Figure 1 This application provides a method for intelligent coal blending and energy optimization in power plants, including the following steps: S1. Real-time collection of industrial analysis data and elemental analysis data of coal entering the plant, and prediction of its key combustion characteristic parameters based on machine learning models; S2. With the optimization objectives of the plant’s total fuel cost being the lowest, the comprehensive emission index being the best, and the boiler combustion stability being the highest, a multi-objective optimization model is established with coal yard inventory, unit load demand, and equipment constraints as conditions. S3. Based on the optimal coal blending scheme obtained in step S2, and combined with the current unit load and boundary conditions, the optimal set values ​​of key boiler operating parameters are dynamically recommended through a data-based proxy model. S4. Real-time acquisition of actual boiler operating parameters after step S3, including boiler efficiency and pollutant emission concentration, comparison with predicted values, and generation of feedback signals for online correction of the prediction model in step S1 and the surrogate model in step S3.

[0027] The aforementioned method upgrades from single-cost optimization to multi-objective dynamic collaborative optimization encompassing cost, environmental protection, and safety. The solution results are more aligned with the actual comprehensive needs of power plants. By dynamically optimizing combustion parameters, it deeply links coal blending decisions with boiler combustion control, achieving collaborative optimization between the fuel side and the equipment side, and improving overall energy conversion efficiency. The system's closed-loop learning mechanism enables it to accumulate operational experience, automatically track and adapt to changes in internal and external conditions, maintain optimization effects over the long term, and reduce reliance on external expert experience and frequent manual adjustments. From coal quality prediction and coal blending calculation to combustion recommendation, the entire process is highly automated, providing operators with clear and quantitative decision support, and effectively improving the intelligent operation level of power plants.

[0028] In this embodiment, the key combustion characteristic parameters in step S1 include, but are not limited to: received lower heating value, dry ash-free volatile matter, softening temperature, and theoretical flue gas volume; the machine learning prediction model is a gradient boosting decision tree model, and its prediction function is as follows:

[0029] in, The model predicted values ​​of the key combustion characteristic parameters for the i-th coal sample; The input feature vector is its industrial analysis and elemental analysis data; K is the total number of decision trees in the model; Let F be the k-th decision tree; F is the function space of all possible decision trees; the model minimizes a regularized objective function that includes prediction error and model complexity. To obtain through training, among which The loss function is used to measure the model's predicted values. and actual observed values The error between them This is a regularization term used to control the k-th decision tree. Model complexity, These are the actual observed values ​​of the key combustion characteristic parameters for the i-th coal sample.

[0030] In this embodiment, the specific form of the multi-objective optimization model in step S2 is as follows: Minimize:

[0031] satisfy:

[0032] Where x is the decision variable vector, i.e. the proportion of various types of coal; R represents the total fuel cost, and R represents the total quantity of different types of coal used in the blending. These represent the unit price and blending quality of the r-th type of coal, respectively. For comprehensive emission indicators, , SO2 and NO respectively x The weighting coefficient of dust. , , SO2 and NO, respectively, are predicted based on coal quality and operating parameters. x Dust emission concentration; It is a negative indicator of combustion stability. This represents the standard deviation function, used to evaluate the uniformity of volatile matter in coal blending. The dry ash-free volatile matter of the r-th type of coal, The blending quality of the r-th type of coal, Total coal quantity; Let j be the j-th inequality constraint function; there are m inequality constraint functions in total. Let p be the p-th equality constraint function; there are q such functions in total. and Decision variables The lower and upper bounds of the value.

[0033] In this embodiment, in step S2, the multi-objective optimization model is solved using a non-dominated sorting genetic algorithm with an elite strategy to obtain the Pareto optimal solution set, and the compromise optimal solution is selected as the coal blending instruction by the membership function method.

[0034] In this embodiment, in step S3, the data-based proxy model is a deep neural network, and its input layer is a vector. Where B is the coal quality characteristic vector of the current optimal coal blending scheme, and L is the current and predicted short-term unit load. The output layer represents the combined state of the coal mills; the output layer is a vector of recommended values ​​for key boiler operating parameters. ,in This refers to the temperature of the boiler feedwater when it enters the economizer or directly into the steam drum / water-cooled wall. The ratio of the actual amount of air fed into the furnace to the theoretical amount of air required for complete fuel combustion is a core parameter for boiler combustion adjustment. The percentage opening of the boiler burnout damper is a key adjustable parameter of the boiler combustion system; the surrogate model is trained using historical operating data, with the goal of minimizing the mean square error between the actual operating energy efficiency and the predicted energy efficiency under recommended parameters.

[0035] In this embodiment, in step S4, the closed-loop feedback calculates the deviation between the actual values ​​of key parameters and the model prediction values. This is achieved by using a recursive least squares method with a forgetting factor to update the weight parameters θ of the machine learning prediction model and the surrogate model online.

[0036]

[0037]

[0038] in, Let be the weight parameter vector of the model at time t; The Kalman gain matrix at time t+1 is used to control the step size / magnitude of parameter updates; These are the actual observed values ​​of the key parameters at time t+1; P represents the model's predicted values ​​for the key parameters at time t+1. t Let be the covariance matrix at time t, representing the uncertainty of the parameter estimation; λ is the model input feature vector at time t+1; λ is the forgetting factor, used to control the weight of historical data; I is the identity matrix.

[0039] The intelligent coal blending and energy optimization method for power plants in this application effectively solves the core problems of traditional coal blending methods, such as isolation, lag, and difficulty in collaborative optimization, by constructing a complete closed-loop process that includes coal quality prediction, multi-objective dynamic optimization, collaborative recommendation of combustion parameters, and closed-loop learning. This method upgrades from a single cost-oriented approach to a multi-objective dynamic collaborative optimization that considers cost, environmental protection, and stability. Its solution is more closely aligned with the comprehensive needs of actual power plant operation. Simultaneously, by introducing a data-based surrogate model to dynamically recommend optimal operating parameters, it deeply correlates coal blending decisions with combustion control, achieving collaborative optimization between the fuel side and the equipment side. This improves the overall energy conversion efficiency of the boiler and ensures the system can continuously learn and adapt to changes through a closed-loop feedback mechanism, significantly improving the power plant's intelligence level and operational economy, environmental friendliness, and safety.

[0040] This application also provides a smart coal blending and energy optimization system for power plants to implement the above method, comprising: The coal quality sensing and digitization module is used to execute step S1; The intelligent coal blending optimization decision module is used to execute step S2; The combustion parameter dynamic optimization module is used to execute step S3; The closed-loop learning and adaptive module is used to execute step S4; A unified data platform and model library provide data support and model services for the above modules.

[0041] In this embodiment, the coal quality sensing and digitization module includes an online coal quality analyzer, a digital coal yard management system, and a coal quality prediction model server; the intelligent coal blending optimization decision-making module is deployed at the plant-level monitoring information system layer and communicates with the fuel management system and the unit load scheduling system.

[0042] In this embodiment, the combustion parameter dynamic optimization module communicates with the distributed control system and sends the recommended key operating parameter settings to the boiler main control loop in the form of guidance values ​​or feedforward signals.

[0043] In this embodiment, the closed-loop learning and adaptive module continuously collects feedback data during operation and is equipped with a model performance evaluation and early warning unit. When the model prediction deviation continues to exceed the set threshold, the model retraining process is automatically triggered or a manual intervention alarm is issued.

[0044] The following is an example: Step S1: A new batch of bituminous coal arrives on the same day, and the online coal quality analyzer quickly provides its industrial analysis and elemental analysis results. =2.1%, =20.5%, =28.0%, =0.8%, =55.2%...), the system inputs it into the pre-trained XGBoost model and instantly predicts its key parameters: =21.5 MJ / kg, =38.5%, ST=1250℃.

[0045] The percentage of moisture content in coal is measured based on coal in an air-dried state (naturally air-dried under laboratory conditions). The percentage of inorganic mineral residue remaining in the coal sample mass after complete combustion of an air-dried coal sample. The percentage of organic gases (excluding moisture) that volatilize from an air-dried coal sample after heating it at 900°C for 7 minutes in the absence of air, based on the mass of the coal sample. : Percentage of total sulfur content in all forms (organic sulfur, iron sulfide sulfur, sulfate sulfur) in air-dried coal samples; The percentage of carbon in the coal is measured based on the state of the raw coal (including total moisture) received by the power plant and actually fed into the furnace. The effective calorific value remaining after the complete combustion of the received base coal sample, after deducting the latent heat of vaporization of the water generated during combustion, is expressed in megajoules per kilogram. The percentage of volatile matter content is measured based on pure combustible matter from which moisture and ash have been completely removed. ST: The temperature at which the tip of the ash cone bends and touches the bottom plate during the heating process. The higher the ST, the higher the melting temperature of the ash, and the less likely it is to slag or contaminate the heating surface in the furnace. The lower the ST, the easier the ash melts, which can easily cause boiler coking and ash blockage. In severe cases, it can lead to unplanned shutdowns. When blending coal, the ST of the coal fed into the furnace must be strictly controlled to avoid the risk of slag formation.

[0046] Step S2: The intelligent coal blending optimization decision module is activated. Currently, the coal yard contains three types of coal (A: high-cost, low-sulfur coal; B: medium-cost, medium-sulfur coal; C: low-cost, high-sulfur, high-ash coal). The planned unit load for the next four hours is 550MW. Based on the model in claim 3, the module sets constraints (e.g., the proportion of C coal blended ≤30%, and the total calorific value must meet the 550MW requirement) and calls the NSGA-II algorithm to solve the problem. After calculation, a set of Pareto optimal solutions is obtained. Finally, a compromise solution balancing economy and environmental protection is selected: A coal: 40%, B coal: 35%, C coal: 25%. The system automatically generates coal blending instructions and sends them to the coal conveying control system.

[0047] Step S3: The combustion parameter dynamic optimization module calculates the average coal quality characteristics based on the above coal blending scheme (as per...). =35.2%, =0.95%), combined with the current 550MW load and the operating coal mill combination, is input into the DNN surrogate model described in claim 5. The model outputs recommended operating parameters: excess air coefficient in the main combustion zone. =1.18, Burnout wind opening =65%, feedwater temperature T_{fw}=295℃. These setpoints are used as optimization target values ​​and are issued to the boiler main control and related subsystems of the DCS in the form of guidance curves.

[0048] Step S4: After running for 2 hours, the closed-loop learning and adaptive module collects actual operating data from the SIS: boiler efficiency. =93.8%, NOx concentration in flue gas [NOx]_{act}=85 mg / Nm³. Compare this with the model prediction ( }=94.1%, By comparing the coal quality prediction model with a value of 80 mg / Nm³, the deviation was calculated. The RLS algorithm with a forgetting factor was used to update the weight parameters of relevant nodes in the coal quality prediction model and the combustion proxy model with a smaller step size, so that the model's prediction of similar future working conditions is closer to the actual situation.

[0049] By continuously operating this system, the power plant has achieved a significant improvement in the stability of the calorific value of the coal fed into the furnace, a reduction in the overall fuel cost under the average coal consumption for power supply, a narrowing of the fluctuation range of NOx emission concentration, and an increase in the automation rate, thus achieving a comprehensive improvement in the level of safe, economical and environmentally friendly operation.

[0050] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in this application, based on the technical solution and application concept of this application, should be included within the scope of protection of this application.

Claims

1. A method for intelligent coal blending and energy optimization in power plants, characterized in that, Includes the following steps: S1. Real-time collection of industrial analysis data and elemental analysis data of coal entering the plant, and prediction of its key combustion characteristic parameters based on machine learning models; S2. With the optimization objectives of the plant’s total fuel cost being the lowest, the comprehensive emission index being the best, and the boiler combustion stability being the highest, a multi-objective optimization model is established with coal yard inventory, unit load demand, and equipment constraints as conditions. S3. Based on the optimal coal blending scheme obtained in step S2, and combined with the current unit load and boundary conditions, the optimal set values ​​of key boiler operating parameters are dynamically recommended through a data-based proxy model. S4. Real-time acquisition of actual boiler operating parameters after step S3, including boiler efficiency and pollutant emission concentration, comparison with predicted values, and generation of feedback signals for online correction of the prediction model in step S1 and the surrogate model in step S3.

2. The method according to claim 1, characterized in that, In step S1, the key combustion characteristic parameters include, but are not limited to: received lower heating value, dry ash-free volatile matter, softening temperature, and theoretical flue gas volume; the machine learning prediction model is a gradient boosting decision tree model, and its prediction function is as follows: in, The model predicted values ​​of the key combustion characteristic parameters for the i-th coal sample; The input feature vector is its industrial analysis and elemental analysis data; K is the total number of decision trees in the model; Let F be the k-th decision tree; F is the function space of all possible decision trees; the model minimizes a regularized objective function that includes prediction error and model complexity. To obtain through training, among which The loss function is used to measure the model's predicted values. and actual observed values The error between them This is a regularization term used to control the k-th decision tree. Model complexity, These are the actual observed values ​​of the key combustion characteristic parameters for the i-th coal sample.

3. The method according to claim 2, characterized in that, In step S2, the specific form of the multi-objective optimization model is as follows: Minimize: satisfy: Where x is the decision variable vector, i.e. the proportion of various types of coal; R represents the total fuel cost, and R represents the total quantity of different types of coal used in the blending. These represent the unit price and blending quality of the r-th type of coal, respectively. For comprehensive emission indicators, , SO2 and NO respectively x The weighting coefficient of dust. , , SO2 and NO, respectively, are predicted based on coal quality and operating parameters. x Dust emission concentration; It is a negative indicator of combustion stability. This represents the standard deviation function, used to evaluate the uniformity of volatile matter in coal blending. The dry ash-free volatile matter of the r-th type of coal, The blending quality of the r-th type of coal, Total coal quantity; Let j be the j-th inequality constraint function; there are m inequality constraint functions in total. Let p be the p-th equality constraint function; there are q such functions in total. and Decision variables The lower and upper bounds of the value.

4. The method according to claim 3, characterized in that, In step S2, the multi-objective optimization model is solved using a non-dominated sorting genetic algorithm with an elite strategy to obtain the Pareto optimal solution set, and the compromise optimal solution is selected as the coal blending instruction by the membership function method.

5. The method according to claim 1, characterized in that, In step S3, the data-based proxy model is a deep neural network, and its input layer is a vector. Where B is the coal quality characteristic vector of the current optimal coal blending scheme, and L is the current and predicted short-term unit load. The output layer represents the combined state of the coal mill; the output layer is a vector of recommended values ​​for key boiler operating parameters. ,in This refers to the temperature of the boiler feedwater when it enters the economizer or directly into the steam drum / water-cooled wall. The ratio of the actual amount of air fed into the furnace to the theoretical amount of air required for complete fuel combustion is a core parameter for boiler combustion adjustment. The percentage opening of the boiler burnout damper is a key adjustable parameter of the boiler combustion system; the surrogate model is trained using historical operating data, with the goal of minimizing the mean square error between the actual operating energy efficiency and the predicted energy efficiency under recommended parameters.

6. The method according to claim 1, characterized in that, In step S4, the closed-loop feedback calculates the deviation between the actual values ​​of key parameters and the model predictions. This is achieved by using a recursive least squares method with a forgetting factor to update the weight parameters θ of the machine learning prediction model and the surrogate model online. in, Let be the weight parameter vector of the model at time t; The Kalman gain matrix at time t+1 is used to control the step size / magnitude of parameter updates; These are the actual observed values ​​of the key parameters at time t+1; P represents the model's predicted values ​​for the key parameters at time t+1. t Let be the covariance matrix at time t, representing the uncertainty of the parameter estimation; λ is the model input feature vector at time t+1; λ is the forgetting factor, used to control the weight of historical data; I is the identity matrix.

7. A smart coal blending and energy optimization system for power plants for implementing the method of any one of claims 1-6, characterized in that, include: The coal quality sensing and digitization module is used to execute step S1; The intelligent coal blending optimization decision module is used to execute step S2; The combustion parameter dynamic optimization module is used to execute step S3; The closed-loop learning and adaptive module is used to execute step S4; A unified data platform and model library provide data support and model services for the above modules.

8. The system according to claim 7, characterized in that, The coal quality sensing and digitization module includes an online coal quality analyzer, a digital coal yard management system, and a coal quality prediction model server; the intelligent coal blending optimization decision-making module is deployed at the plant-level monitoring information system layer and communicates with the fuel management system and the unit load scheduling system.

9. The system according to claim 7, characterized in that, The combustion parameter dynamic optimization module communicates with the distributed control system and sends recommended key operating parameter settings to the boiler main control loop in the form of guidance values ​​or feedforward signals.

10. The system according to claim 7, characterized in that, The closed-loop learning and adaptive module continuously collects feedback data during operation and is equipped with a model performance evaluation and early warning unit. When the model prediction deviation continues to exceed the set threshold, it automatically triggers the model retraining process or issues a manual intervention alarm.

Images