Thermal power plant coal blending combustion optimization method based on multi-source data and intelligent algorithm

Abstract

The invention relates to a thermal power plant coal blending combustion optimization method based on multi-source data and an intelligent algorithm, and the method comprises the steps: obtaining the multi-source data from a fuel management and control platform, a coal yard digital system, a unit operation monitoring system, and an environmental protection monitoring system in real time, establishing a dynamic database including coal quality data, real-time inventory, unit operation constraints and environmental protection constraints; based on the dynamic database, taking the minimization of the total fire coal cost in the future T hours as a target function, and taking stable operation of a boiler, an environmental protection emission index and equipment processing capacity as constraint conditions, establishing a corresponding fire coal blending optimization model for each load period, and forming a multi-period optimization model; solving the multi-period optimization model by adopting an improved genetic algorithm to generate an optimal blending scheme sequence; through online monitoring, according to the optimal blending scheme, adopting a preset control strategy to realize closed-loop optimization of the combustion process; according to the invention, the comprehensive optimization of the fire coal cost, the combustion efficiency and the environment-friendly emission is realized.

Description

Technical Field

[0001] This invention relates to the field of fuel management technology for thermal power units, and in particular to an optimized method for coal blending and combustion in thermal power plants based on multi-source data and intelligent algorithms. Background Technology

[0002] In thermal power plants, coal costs account for a significant proportion of the total power generation cost. Meanwhile, pollutant emissions from coal combustion are subject to increasingly stringent environmental regulations. To balance economic efficiency and environmental friendliness, coal blending has become a common choice for power plants. However, existing blending methods often rely on the experience of operators or employ simple linear programming for static optimization, making it difficult to cope with multiple complex constraints such as coal quality fluctuations, load changes, equipment status, and environmental requirements.

[0003] Specifically, existing coal blending methods have the following shortcomings: First, coal quality data, inventory information, and unit operation data often exist in isolation, failing to achieve effective integration and dynamic updates, resulting in inaccurate and untimely data foundations for blending schemes. Second, optimization models typically simplify the boiler combustion process into a linear relationship, ignoring the nonlinear characteristics of coal consumption and pollutant generation, leading to discrepancies between theoretical schemes and actual operating results. Third, the generated blending schemes are mostly static plans, lacking a closed-loop mechanism for dynamic feedback adjustments based on real-time changes in coal quality and unit status during execution, thus failing to guarantee that the combustion process is always under optimal operating conditions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an optimization method for coal blending in thermal power plants based on multi-source data and intelligent algorithms. This method solves the problem that existing blending methods are unable to cope with coal quality fluctuations, load changes, equipment status, and environmental protection requirements, and achieves comprehensive optimization of coal cost, combustion efficiency, and environmental emissions.

[0005] The technical solution adopted in this invention is as follows: This invention provides a method for optimizing coal blending in thermal power plants based on multi-source data and intelligent algorithms, comprising: Real-time acquisition of multi-source data from fuel management platform, coal yard digital system, unit operation monitoring system and environmental monitoring system, and establishment of dynamic database including coal quality data, real-time inventory, unit operation constraints and environmental constraints; Based on the dynamic database, the objective function is to minimize the total coal cost in the next T hours, with constraints on boiler stable operation, environmental emission indicators and equipment processing capacity. Different load periods are divided according to the grid load characteristics in the next T hours, and corresponding coal blending optimization models are established for each load period to form a multi-period optimization model. An improved genetic algorithm is used to solve the multi-time period optimization model to generate a sequence of optimal blending schemes for the next T hours with Δt as the time unit. By real-time online monitoring to identify the measured values ​​of coal quality and key combustion parameters entering the furnace, and comparing them with the optimal blending scheme, the blending ratio of each coal mill is adjusted based on the preset control strategy to achieve closed-loop optimization of the combustion process. The step of solving the multi-time-period optimization model using an improved genetic algorithm includes: A hybrid coding approach is used to characterize multi-time-segment blending scheme sequences; The fitness function is used to evaluate the quality of the blending scheme, wherein the fitness function is F = G–P; where G is the economic benefit, which is calculated based on the physical parameters predicted by the boiler combustion neural network model; and P is the penalty for violating calorific value, emission or unit operation constraints. The boiler combustion neural network model is obtained by training the neural network, with the blending scheme and unit load command of each time unit as input and the power generation coal consumption rate and pollutant generation amount under the corresponding operating conditions as output.

[0006] The preferred technical solution is: The calculation of the economic benefit G includes: G= ; In the formula, This represents the cost savings brought about by the blending scheme corresponding to the t-th time unit, where This is a preset benchmark cost, which is the average operating cost under historical loads or the designed operating cost for the coal type. This refers to the actual operating cost; This is a combustion efficiency bonus item; This is the time-of-use weighting coefficient, whose value is adjusted according to different load periods: it is increased during peak grid load periods. By rewarding blending schemes with high combustion efficiency and low coal consumption, the unit's load-carrying capacity is ensured, and during periods of low grid load, the power grid is adjusted downwards. In order to reduce costs.

[0007] The actual operating cost The calculations include: = ; In the formula, the first term is the fuel consumption cost, where Let t be the unit load in the t-th time unit. The power generation coal consumption rate predicted by the boiler combustion neural network model. The first item is the weighted average purchase price of the blending scheme for the t-th time unit; the second item is the environmental treatment cost, where... The sulfur oxide and nitrogen oxide production amounts predicted by the boiler combustion neural network model. , All are preset unit pollutant treatment cost coefficients.

[0008] The objective function of the coal blending optimization model is: min J = ∑[ ]; in, The fuel consumption cost, The environmental remediation cost, These are the corresponding weights; The constraints include: Calorific value constraint is used to ensure that the overall calorific value of the coal blending scheme in each time unit is not lower than the minimum requirement of the boiler; Ash content constraints are used to limit the total ash content to within a specified upper limit. Sulfur content constraints are used to ensure that SO2 emissions meet standards. Environmental protection equipment processing capacity constraints restrict desulfurization, dust removal, and denitrification systems from exceeding their maximum processing load; Overall proportion normalization constraint: ,in For the first i The blending ratio of different types of coal.

[0009] The preset control strategy is driven by a time-series prediction model or a reinforcement learning controller trained based on experimental and operational history. It is used to adaptively correct the set values ​​of key combustion parameters and optimize the blending ratio of each coal mill online. Optimization using the aforementioned time-series prediction model includes: The time-series prediction model outputs predicted values ​​of coal quality, emissions, and efficiency for a future period, which are then compared with the corresponding target values ​​to assess the deviation level, which is evaluated as slight deviation or significant deviation. In the event of a slight deviation, the set values ​​of the key combustion parameters are adaptively adjusted based on the prediction results; When there is a significant deviation, the objective function weights of the multi-period coal blending optimization model are updated, and the new blending scheme is quickly solved by the improved genetic algorithm. The linear transition strategy is used to gradually adjust the model, and closed-loop optimization is achieved by combining online monitoring and incremental model training. The target value is the constraint basis for the optimal blending scheme. It is dynamically adjusted according to the grid load period and is consistent with the unit operation constraints and environmental protection constraints.

[0010] The construction of the time series prediction model includes: Collect multi-source time-series data including coal quality, emissions, and efficiency, and construct sliding window samples. Each sliding window sample includes data from N historical time units and one predicted future time unit. Design a multi-layer stacked LSTM architecture with a fully connected layer. Train the system using the mean squared error loss function and Adam as the optimizer. Ensure generalization capability through early stopping and regularization. The system is completed after verification.

[0011] The improved genetic algorithm employs roulette wheel or tournament selection methods for genetic operations and iterative optimization. It prioritizes the retention of chromosomes with high fitness, randomly selects several time periods between parent chromosomes to exchange genes, and retains the coal type ratio information of high fitness time periods in the parent generation to maintain excellent local solutions. Within each time unit, the coal blending ratio is fine-tuned to maintain population diversity and enhance local search capabilities. The algorithm terminates when the population fitness converges for several consecutive generations or reaches the maximum number of iterations, and outputs the chromosome with the highest fitness as the optimal coal blending scheme for hour T.

[0012] The online monitoring includes real-time estimation of coal quality parameters entering the furnace using a near-infrared spectroscopy analyzer and an online calorific value analyzer; the key combustion parameters include pulverized coal concentration, mill speed, air-coal ratio, and flue gas temperature.

[0013] The dynamic database contains coal quality data including calorific value, sulfur content, ash content, volatile matter and ash fusion point of each type of coal; unit operation constraints include maximum processing capacity of desulfurization and denitrification system, maximum output of coal mill, minimum calorific value requirement to ensure stable boiler operation; and environmental constraints include emission concentration limits for SO2 and NOx. The multi-source data also includes real-time meteorological data and historical seasonal coal quality changes, which are used to dynamically correct the constraints of the coal blending optimization model.

[0014] The execution records of the optimal blending scheme and the corresponding combustion effect data are structured and stored in a historical database, which provides a data foundation for tracing operational effects and optimizing and retraining the boiler combustion neural network model.

[0015] The technical solution of the present invention can achieve at least some of the following beneficial effects: This invention employs end-to-end data fusion, precise intelligent optimization algorithms, and online dynamic feedback control. Through end-to-end control from coal procurement, inventory management, blending decisions to combustion regulation, it achieves scientific, economical, and environmentally friendly coal blending. Specifically, it has the following advantages: This invention achieves scientific and intelligent coal blending by dynamically acquiring coal quality and unit operation data, constructing an optimization model, and solving it using an improved genetic algorithm. It optimizes coal costs, improves unit stability, and significantly enhances energy utilization efficiency and economy while ensuring combustion efficiency and emission standards are met. Simultaneously, it enables batch-based and dynamic management of coal quality and inventory, ensuring the accuracy and traceability of original data for blending decisions.

[0016] This invention embeds unit operation constraints and environmental protection facility processing capacity as hard boundary conditions into the optimization model, systematically avoiding the operational risks of unit load limits and exceeding environmental parameters caused by traditional experience-based blending methods due to human negligence or difficulty in coordinating multiple constraints. The improved genetic algorithm used is closely coupled with the boiler combustion neural network prediction model, integrating complex nonlinear combustion and emission characteristics into the optimization process. This makes the generated blending scheme not only mathematically optimal but also closely matches the actual combustion characteristics of the boiler, thereby significantly improving the engineering applicability and actual economic benefits of the scheme in the field.

[0017] This invention, relying on online monitoring and dynamic feedback mechanisms, constructs a closed-loop control system encompassing perception, decision-making, execution, and verification. It can respond in real-time to fluctuations in the quality of coal fed into the boiler and changes in unit load, dynamically and precisely correcting the coal blending scheme. This continuously maintains the combustion process within an efficient and stable optimal operating range, effectively improving the overall operating efficiency and long-term reliability of the unit while ensuring strict emission compliance. It not only achieves real-time compliance with boiler combustion calorific value and emission standards but also responds to fluctuations in coal quality and changes in grid load, realizing synergistic optimization of combustion efficiency, economy, and emission control.

[0018] Other features and advantages of the invention will be set forth in the following description or may be learned by practicing the invention. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the method in an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the solution process of the improved genetic algorithm according to an embodiment of the present invention.

[0021] Figure 3 This is a structural block diagram of the optimized system formed based on the method of the embodiments of the present invention.

[0022] Figure 4 This is a structural diagram of a thermal power unit for a specific example in an embodiment of the present invention.

[0023] In the diagram: 1. Raw coal; 2. Generator; 3. Steam turbine; 4. Coal bunker; 5. Coal mill; 6. Boiler; 7. Condenser; 8. Blower; 9. Condensation tower; 10. Dust collector; 11. Exhaust fan; 12. Desulfurization booster fan; 13. Desulfurization device. Detailed Implementation

[0024] The specific embodiments of the present invention are described below with reference to the accompanying drawings.

[0025] See Figure 1 This embodiment provides a method for optimizing coal blending in thermal power plants based on multi-source data and intelligent algorithms, including the following steps: S1. Acquire multi-source data in real time from the fuel management platform, coal yard digital system, unit operation monitoring system and environmental monitoring system, and establish a dynamic database that includes coal quality data, real-time inventory, unit operation constraints and environmental constraints.

[0026] As a specific method, a systematic coal source scanning and batch sampling test is implemented for various coal sources (including high-sulfur coal, low-ash coal, lean coal, lignite, etc.) to acquire and record coal quality data for each batch, including key physical properties such as calorific value, sulfur content, ash content, volatile matter, moisture, and ash fusion point. The test results are then correlated with real-time inventory levels, coal flow sensor data, and incoming procurement records in the coal yard's three-dimensional digital system. A dynamic coal quality and inventory database is established using batch management to achieve coal quality traceability and batch-level data traceability. Simultaneously, high-precision field sensors are used to collect real-time unit operating parameters, including unit load, boiler pressure, main steam temperature, coal mill output, air-coal ratio, and flue gas temperature, as well as environmental monitoring data, including SO2 and NO. x The system will establish emission concentrations and incorporate unit operation constraints, including the maximum processing capacity of the desulfurization and denitrification system, the upper limit of the dust removal system, the maximum output of the coal mill, and the minimum allowable calorific value of the boiler (the minimum calorific value to ensure stable boiler operation), as well as environmental constraints, including emission concentration limits for SO2 and NOx, into the database system in the form of an operational boundary model.

[0027] As a preferred approach, real-time meteorological data such as ambient temperature, humidity, and air pressure, as well as historical seasonal coal quality changes, are also collected synchronously and incorporated into a dynamic database. This database is used to dynamically correct the constraint boundaries of the subsequently constructed model, thereby improving its adaptability and reliability under different environments.

[0028] S2. Based on the dynamic database, with the objective function of minimizing the total coal cost in the next T hours, and with the constraints of stable boiler operation, environmental emission indicators and equipment processing capacity, and according to the grid load characteristics in the next T hours, different load periods are divided, and corresponding coal blending optimization models are established for each load period to form a multi-period optimization model.

[0029] Specifically, in order to take into account the characteristics of the power grid load, the system divides the next T hours into several different load periods such as peak, normal, and off-peak according to the power grid load forecast, and sets optimization objectives for different load periods: during peak hours, the focus is on combustion efficiency and emission control, and during off-peak hours, the focus is on cost reduction and combustion stability, thus forming a multi-period, multi-objective constraint strategy.

[0030] Specifically, the objective function of the coal blending optimization model is: min J = ∑[ ]; in, For fuel consumption costs, For environmental governance costs, These are the corresponding weights; The constraints include: Calorific value constraint is used to ensure that the overall calorific value of the coal blending scheme in each time unit is not lower than the minimum requirement of the boiler; Ash content constraints are used to limit the total ash content to within a specified upper limit. Sulfur content constraints are used to ensure that SO2 emissions meet standards. Environmental protection equipment processing capacity constraints restrict desulfurization, dust removal, and denitrification systems from exceeding their maximum processing load; Overall proportion normalization constraint: ,in For the first i The blending ratio of different types of coal.

[0031] Specifically, the constraints consist of a set of linear inequalities or equations, with the general form being: ∑(aji * x i ) ≤ (or =, ≥) bj; In the above formula, aji is the content or characteristic coefficient of the j-th component (such as sulfur or ash) of the i-th type of coal; bj is the target value (limit) corresponding to the j-th constraint (such as environmental emission limits, equipment processing capacity, or minimum calorific value requirements of boilers).

[0032] S3. An improved genetic algorithm is used to solve the multi-time period optimization model to generate a sequence of optimal blending schemes for the next T hours with Δt as the time unit.

[0033] As a preferred method, the step of solving the multi-time-period optimization model using an improved genetic algorithm includes: A hybrid coding method is used to characterize the multi-time period blending scheme sequence; specifically, one chromosome can represent a complete T-hour coal blending scheme sequence, the chromosome gene loci correspond to the blending ratio of different coal types in each Δt time unit, the ratio value is represented by real number coding, and the blending amount is represented by integer coding, thus forming a hybrid coding method; The fitness function is used to evaluate the quality of the blending scheme, wherein the fitness function is F = G–P; where G is the economic benefit, which is calculated by the cost conversion model based on the physical parameters (including coal consumption for power generation and pollutant generation) predicted by the boiler combustion neural network model; and P is the penalty for violating the constraints on calorific value, emissions or unit operation.

[0034] Preferably, the economic benefits are calculated using a cost conversion model based on the physical parameters (power generation coal consumption, pollutant generation) predicted by the boiler combustion neural network model, including: Fuel consumption costs are calculated based on the predicted power generation coal consumption rate, unit load, and weighted average coal price. The cost of environmental remediation is calculated based on the predicted amount of pollutants generated and the preset unit treatment cost coefficient. The difference between the sum of the above costs and the preset benchmark cost (or the negative value of the sum of the above costs) is taken as the G value.

[0035] The actual operating fuel cost item is calculated based on the weighted unit price of the coal blending scheme and the power generation coal consumption rate predicted by the boiler combustion neural network model; the environmental treatment cost item is calculated based on the pollutant generation amount predicted by the boiler combustion neural network model and the preset unit treatment cost coefficient, so as to minimize the total operating cost in the next T hours under the premise of meeting the constraints of stable boiler operation, environmental emissions and equipment treatment capacity.

[0036] Specifically, the calculation of the economic benefit G includes: G= ; In the formula, This represents the cost savings brought about by the blending scheme corresponding to the t-th time unit, where This is a preset benchmark cost, which is the average operating cost under historical loads or the designed operating cost for the coal type. This refers to the actual operating cost; This is a combustion efficiency bonus item; This is the time-of-use weighting coefficient, whose value is adjusted according to different load periods: it is increased during peak grid load periods. By rewarding blending schemes with high combustion efficiency and low coal consumption, the unit's load-carrying capacity is ensured, and during periods of low grid load, the power grid is adjusted downwards. To reduce costs. Therefore, the time-sharing weighting coefficient. The difference is one of the core differences in solving the coal blending optimization model under different load periods.

[0037] Specifically, the actual operating cost The calculations include: = ; In the above formula, the first term is the fuel consumption cost, which is the objective function S2. ,in Let t be the unit load in the t-th time unit. The power generation coal consumption rate predicted by the boiler combustion neural network model. The first term is the weighted average purchase price of the blending scheme in the t-th time unit; the second term is the environmental treatment cost, i.e., the cost in the S2 objective function. ,in The sulfur oxide and nitrogen oxide production amounts predicted by the boiler combustion neural network model. , All are preset unit pollutant treatment cost coefficients.

[0038] Specifically, the boiler combustion neural network model takes the blending scheme and unit load command of each time unit as input, and the power generation coal consumption rate and pollutant generation amount under the corresponding operating conditions as output, and is obtained through neural network training.

[0039] Specifically, P represents the penalty for violating constraints, including situations such as calorific value falling below the boiler's minimum requirements, excessive ash or sulfur content, and overloading of environmental protection equipment. If any constraint is violated, P takes a large value, causing the fitness F to drop sharply, thus eliminating the violator in the selection operation of the genetic algorithm.

[0040] Through step S3, this embodiment successfully transforms the nonlinear physical quantities (coal consumption, emissions) of the boiler combustion neural network model into quantifiable economic indicators—economic benefits G—ensuring that the generated coal blending scheme not only meets environmental and safety constraints on the engineering site but also achieves real operating cost reduction. The boiler combustion neural network model is invoked in real-time during algorithm iteration to predict each scheme, ensuring that the fitness calculation simultaneously considers economic efficiency, combustion efficiency, and emission constraints, avoiding infeasible solutions. It can automatically adjust weights for different load periods, prioritizing combustion efficiency and emissions optimization during peak periods and economic efficiency and stability optimization during off-peak periods.

[0041] The flowchart illustrating the solution process of the improved genetic algorithm in step S3 of this embodiment can be found in [link to flowchart]. Figure 2 .

[0042] Specifically, the improved genetic algorithm generates its initial population by combining randomly generated feasible solutions with historical operating experience, ensuring that the initial solution space has broad coverage and meets engineering constraints.

[0043] Specifically, the improved genetic algorithm employs roulette wheel or tournament selection methods for genetic operations and iterative optimization. This aims to retain excellent blending schemes from the parent generation while introducing local fine-tuning to maintain population diversity and improve search efficiency. Specifically, chromosomes with high fitness are prioritized for retention. Genes are exchanged randomly between parent chromosomes over several time periods, while retaining coal type proportion information from high-fitness time periods in the parent generation to maintain excellent local solutions. Within each time unit, the coal blending ratio is fine-tuned, preferably by ±1-3%, thereby maintaining population diversity and enhancing local search capabilities. The algorithm terminates when the population fitness converges after several consecutive generations or reaches the maximum number of iterations, outputting the chromosome with the highest fitness as the optimal coal blending scheme for time T.

[0044] S4. By real-time monitoring and identification of the measured values ​​of coal quality and key combustion parameters entering the furnace, and comparing them with the optimal blending scheme, the blending ratio of each coal mill is adjusted based on the preset control strategy to achieve closed-loop optimization of the combustion process.

[0045] As a preferred method, the online monitoring includes real-time estimation of coal quality parameters during system operation using near-infrared spectroscopy analyzers, online calorific value analyzers, and other means. Combined with the online acquisition of key combustion parameters such as pulverized coal concentration, pulverizer output, air-coal ratio, and flue gas temperature, the actual combustion state during execution is gradually compared with the optimal blending scheme generated in step S3.

[0046] As a preferred approach, the preset control strategy is driven by a time-series prediction model or a reinforcement learning controller trained based on experimental and operational history, in order to adaptively correct the set values ​​of the key combustion parameters and optimize the blending ratio of each coal mill online, so as to ensure that the coal blending ratio remains in the optimal state under different loads and coal quality fluctuations.

[0047] As a preferred approach, optimization is performed using the aforementioned time-series prediction model, including: The time-series prediction model outputs predicted values ​​of coal quality, emissions, and efficiency for a future period, which are then compared with the corresponding target values ​​to assess the deviation level, which is evaluated as slight deviation or significant deviation. In the event of a slight deviation, the set values ​​of the key combustion parameters are adaptively adjusted based on the prediction results; When there is a significant deviation, the objective function weights of the multi-period coal blending optimization model are updated, and the new blending scheme is quickly solved by the improved genetic algorithm. The linear transition strategy is used to gradually adjust the model, and closed-loop optimization is achieved by combining online monitoring and incremental model training.

[0048] Specifically, the target value serves as the constraint basis for the optimal blending scheme. It is dynamically adjusted according to the grid load period and is consistent with the unit operation constraints and environmental constraints. The target value needs to be dynamically set in conjunction with multiple constraints, as shown in Table 1 below.

[0049] Table 1 Target values ​​for coal quality, emissions, and efficiency

[0050] It can be understood that the relationship between the target value and the optimal blending scheme in this embodiment is that of "constraint basis" and "optimal solution that satisfies the constraints": the target value is the core constraint of the optimal blending scheme. When solving the problem, the optimization model needs to convert the target value into mathematical constraints (such as "comprehensive calorific value of blended coal ≥ target value" and "SO2 emissions ≤ target value") to ensure that the generated scheme does not exceed the target boundary. The optimal blending scheme is the means to achieve the target value. The scheme adjusts the blending ratio of each type of coal to make the actual coal quality, emissions, and efficiency parameters as close as possible to or better than the target value (such as under the constraint of "emission target value", by reasonably blending low-sulfur coal and high-sulfur coal, SO2 emissions just meet the standard and the cost is minimized). There is a dynamic linkage between the target value and the optimal blending scheme. If the target value is adjusted, such as the upgrade of environmental standards leading to a decrease in the NOx target value, the optimization model will re-solve and generate a new optimal blending scheme; conversely, if the parameters deviate from the target value during the execution of the optimal scheme, dynamic regulation will be triggered, such as increasing the proportion of low-sulfur coal to reduce SO2 emissions.

[0051] Specifically, the objective function weights are initially set to w1=1 and w2=1 by default. When the weights need to be updated, the ratio of w1 and w2 will be adjusted according to the actual operating conditions: during peak grid load periods (emphasizing combustion efficiency and emission control), the environmental cost weight w2 can be increased, and the model will preferentially reduce it. Reducing SO2 / NOx generation is achieved even with a slight increase in fuel costs. During off-peak periods, the focus is on reducing coal costs, and the fuel cost weight w1 is increased. The model will prioritize lower-priced coal types, as long as environmental and safety constraints are met. After updating w1 and w2, the improved genetic algorithm, in solving new blending schemes, introduces a time-sharing weight coefficient α for combustion efficiency rewards in the calculation of the fitness function F and the economic benefit G. For models operating at different load periods, the adjustment of α is as described above, and its adjustment directly relates to the optimization direction of the objective function. The penalty term P for violating constraints is essentially the weight of the constraint conditions: when environmental requirements tighten (e.g., SO2 emission limits are reduced) or equipment approaches full load (e.g., coal mill output is nearing its limit), the weight of P can be increased, meaning the penalty for violating constraints is heavier. In this case, the model will prioritize satisfying the constraints, even if the total cost of the objective function increases slightly. If the measured sulfur content of the coal entering the furnace far exceeds the predicted value (significant deviation), to avoid exceeding environmental standards, the penalty weight P corresponding to the "sulfur content constraint" will be increased, forcing the model to adjust the blending ratio (reducing the proportion of high-sulfur coal). In this case, the priority of environmental constraints is higher than simply minimizing costs. In summary, by adjusting the relative priorities w1 and w2 of fuel costs and environmental costs, the time-sharing weight α, and the penalty intensity P for constraint violations, different operating conditions are dynamically adapted to ensure that the objective function, under the core premise of cost minimization, also considers constraints such as combustion stability and environmental compliance, ultimately achieving comprehensive optimization.

[0052] Preferably, the construction of the time series prediction model includes: Multi-source time-series data including coal quality, emissions and efficiency are collected. After data processing such as cleaning, alignment and normalization, a sliding window sample is constructed. Each sliding window sample includes data from N historical time units and one predicted future time unit. Design a multi-layer stacked LSTM (Long Short-Term Memory) network architecture, with a fully connected layer. Train the network using the mean squared error loss function and Adam as the optimizer. Ensure generalization ability through early stopping and regularization. After verification, the network is completed.

[0053] Specifically, the set values ​​of the key combustion parameters serve as instructions and the basis for execution. Each key combustion parameter has a corresponding set value, such as an ideal air-coal ratio of 1.2 and a mill speed of 1500 r / min. The corresponding measured values ​​are objective monitoring data collected in real time during combustion, reflecting the combustion state. For example, the actual air-coal ratio is 1.18, the actual mill speed is 1480 r / min, and the actual flue gas temperature is 1050℃. The measured values ​​are feedback data to verify whether the set values ​​are implemented. In this embodiment, when there is a slight deviation, the set values ​​of air-coal ratio, mill speed, etc., are adaptively corrected according to the predicted results. Essentially, the set values ​​are adjusted in reverse by the deviation between the measured values ​​and the set values ​​to ensure that the key combustion parameters are close to the ideal state.

[0054] In the specific engineering implementation process, this embodiment, through the linkage of the fuel management platform, the coal yard 3D digital system, and the unit's DCS / PLC system, automatically sends the optimized blending plan for the next T hours to each coal mill. Each coal mill then executes the plan precisely according to the blending ratio for each time unit. Simultaneously, the online monitoring system continuously collects actual combustion status data and feeds this data back to the optimization system for prediction and adjustment of the next time unit. The optimized coal blending instructions are automatically sent to the fuel management and unit distributed control system (DCS), driving the coal yard blending device and the coal mills to work collaboratively, achieving seamless decision-making and execution. Complete plan execution records and corresponding combustion effect data are structured and stored in a historical database, providing a data foundation for operational performance tracking and model retraining optimization. The structure of the optimization system formed based on the method of this embodiment is described in [reference needed]. Figure 3 As shown.

[0055] To further verify the effectiveness of the method in this embodiment, a specific calculation example is used for verification below.

[0056] This example demonstrates an optimization method for coal blending in thermal power plants based on multi-source data and intelligent algorithms. The structure of the thermal power unit targeted is described in [reference needed]. Figure 4 ,like Figure 4As shown, the core equipment of the thermal power unit system targeted in this example includes fuel storage and transportation equipment, combustion system equipment, steam-water circulation equipment, environmental protection equipment, power generation and transmission equipment, and monitoring and control equipment. Raw coal 1 from the coal yard is transported to coal bunker 4 via a coal conveying pipeline. The outlet of coal bunker 4 is connected to the input end of five coal mills 5. The output end of coal mills 5 is connected to the furnace of boiler 6 via a pulverized coal pipeline, realizing the transportation of raw coal 1 from storage, grinding to combustion. The outlet of blower 8 is connected to an air preheater, and the preheated air is introduced into the furnace of boiler 6. The flue gas outlet of boiler 6 is connected in sequence to the air preheater, dust collector 10, desulfurization booster fan 12, and desulfurization device 13, and is discharged from the system by induced draft fan 11. Condensation tower 9 is connected to condenser 7 via a pipeline. The outlet of condenser is connected to feedwater heater via condensate return pipeline. The output end of feedwater heater is connected to the water system of boiler 6. The steam generated by boiler 6 is transported to steam turbine 3 via a steam pipeline. The exhaust steam of steam turbine 3 is connected to condenser 7 for condensation. The output shaft of steam turbine 3 is connected to generator 2 and exciter. Generator 2 is connected to main transformer via transmission line to realize power output. Each device is connected to the unit operation monitoring system (DCS / PLC system), and the monitoring data is uploaded in real time, and the control commands are issued accurately. Raw coal is stored in the coal yard and, according to the optimal blending scheme, is proportionally transported to five coal mills via coal conveying pipelines. The ground coal powder is then fed into the furnace of boiler 6, where it is fully mixed with preheated air supplied by blower 8 before combustion. The coal ash produced after combustion is collected and treated through an ash removal system. The waste flue gas produced in boiler 6 is preheated by an air preheater to recover waste heat (preheating the combustion air), and then sequentially enters a dust collector 10 to remove dust and a desulfurization device to remove sulfur oxides. It is then pressurized by a desulfurization booster fan and extracted by an induced draft fan, finally meeting emission standards. Condensate is preheated in the feedwater heater via a condensate return pipeline and then sent to the water system of boiler 6. In boiler 6, it absorbs the heat energy generated by combustion and converts it into steam. The steam drives the turbine 3 to rotate through pipelines. The steam that has done work enters the condenser 7 and condenses into condensate under the cooling effect of the condensate tower 9. The condensate then returns to the feedwater heater via the condensate return pipeline, completing the cycle.

[0057] The optimization method described in this example includes: Initial forecast and scheme generation: The load plan for the next 4 hours, real-time inventory and coal quality data of each coal type (from the coal yard digital system), and current boiler status are used to predict the main combustion trends (such as theoretical calorific value, sulfur content, and volatile matter) and the estimated concentrations of key emissions (SO2, NOx) of the boiler under different blending schemes in the next 4 hours. The controller uses total fuel cost as a negative reward and violates constraints such as exceeding emission standards and large fluctuations in calorific value as penalties. It explores in a simulation environment of millions of times and outputs the recommended blending ratio scheme of 5 coal mills (corresponding to 4 types of coal) for the next 4 hours, with each hour as a time unit.

[0058] Scheme Implementation and Real-time Monitoring: The optimized blending ratio scheme is automatically distributed to the unit's DCS system through the fuel management platform. Before entering the pulverizing system, the blended coal is analyzed by an online near-infrared / calorific value analyzer to obtain the actual coal quality entering the furnace: calorific value 20.5 MJ / kg, sulfur content 0.99%, and volatile matter 27.1%. The data has slight drift but is generally reliable and is recorded in the central database.

[0059] Dynamic feedback and rolling optimization are implemented, with the DCS system collecting real-time combustion data: boiler main steam temperature / pressure, real-time power generation, and SO2 (75 mg / Nm³) data transmitted from the online environmental monitoring system (CEMS). 3 NOx (85 mg / Nm3) 3 The data compares predicted coal quality, measured coal quality, and actual combustion / emission effects. Deviation data is used for online fine-tuning (incremental learning) of the LSTM model, making its next round of coal quality and combustion trend predictions more accurate. At the end of the first hour, the reinforcement learning controller, based on the actual load response, emissions, and coal consumption, combined with the updated LSTM predictions, performs rolling re-optimization of the coal blending scheme for the second hour (load 600MW). For example, because the measured sulfur content of coal C is slightly higher than the prediction, to avoid SO2 approaching the upper limit, the system automatically adjusts the scheme for the second hour to: coal A: 20%, coal B: 55%, coal C: 15%, coal D: 10%. Specific parameters involved in the optimization method of this example are shown in Table 2.

[0060] Table 2 Specific Parameters

[0061] To evaluate the effectiveness of this scheme, a one-week comparative experiment was conducted between the closed-loop optimization scheme of this example and the fixed-ratio co-firing experience scheme previously used by the power plant. The comparison results are shown in Table 3.

[0062] Table 3 Comparison of the optimization results of the closed-loop optimization scheme in this example with the empirical scheme of fixed-ratio co-firing previously used by the power plant.

[0063] As can be seen from the above examples, the method in this embodiment shows significant engineering advantages in practical applications. By collecting coal quality and unit operation data in real time and combining it with historical data and neural network prediction models, it can generate the optimal blending scheme that meets the boiler calorific value requirements, ash and sulfur limits, and environmental emission constraints, thereby achieving the optimal economic efficiency of coal types.

[0064] In summary, the optimization method presented in this application, with its online monitoring and dynamic control mechanism, can respond promptly to changes in coal quality and load fluctuations, ensuring a stable and reliable combustion process while guaranteeing that emissions continuously meet national and local environmental standards. The time-segmented blending strategy optimizes combustion efficiency and economy for different load periods, focusing on improving combustion performance and emission reduction capabilities during peak load periods, and reducing fuel costs and maintaining boiler stability during off-peak load periods, thus improving the overall energy utilization efficiency of the power plant. Closed-loop control and data recording form a complete closed loop for coal type management, coal blending optimization, execution control, and combustion performance, improving operational efficiency and extending equipment lifespan. It possesses good engineering scalability and provides a smart coal blending solution that can be engineered and promoted for thermal power plants. It can simultaneously optimize combustion economy and environmental protection while ensuring the safe and stable operation of the boiler.

[0065] It will be understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing coal blending in thermal power plants based on multi-source data and intelligent algorithms, characterized in that, include: Real-time acquisition of multi-source data from fuel management platform, coal yard digital system, unit operation monitoring system and environmental monitoring system, and establishment of dynamic database including coal quality data, real-time inventory, unit operation constraints and environmental constraints; Based on the dynamic database, the objective function is to minimize the total coal cost in the next T hours, with constraints on boiler stable operation, environmental emission indicators and equipment processing capacity. Different load periods are divided according to the grid load characteristics in the next T hours, and corresponding coal blending optimization models are established for each load period to form a multi-period optimization model. An improved genetic algorithm is used to solve the multi-time period optimization model to generate a sequence of optimal blending schemes for the next T hours with Δt as the time unit. By real-time online monitoring to identify the measured values ​​of coal quality and key combustion parameters entering the furnace, and comparing them with the optimal blending scheme, the blending ratio of each coal mill is adjusted based on the preset control strategy to achieve closed-loop optimization of the combustion process. The step of solving the multi-time-period optimization model using an improved genetic algorithm includes: A hybrid coding approach is used to characterize multi-time-segment blending scheme sequences; The fitness function is used to evaluate the quality of the blending scheme, wherein the fitness function is F = G–P; where G is the economic benefit, which is calculated based on the physical parameters predicted by the boiler combustion neural network model; and P is the penalty for violating calorific value, emission or unit operation constraints. The boiler combustion neural network model is obtained by training the neural network, with the blending scheme and unit load command of each time unit as input and the power generation coal consumption rate and pollutant generation amount under the corresponding operating conditions as output.

2. The method according to claim 1, characterized in that, The calculation of the economic benefit G includes: G= ； In the formula, This represents the cost savings brought about by the blending scheme corresponding to the t-th time unit, where This is a preset benchmark cost, which is the average operating cost under historical loads or the designed operating cost for the coal type. This refers to the actual operating cost; This is a combustion efficiency bonus item; This is the time-of-use weighting coefficient, whose value is adjusted according to different load periods: it is increased during peak grid load periods. By rewarding blending schemes with high combustion efficiency and low coal consumption, the unit's load-carrying capacity is ensured, and during periods of low grid load, the power grid is adjusted downwards. In order to reduce costs.

3. The method according to claim 2, characterized in that, The actual operating cost The calculations include: = ； In the formula, the first term is the fuel consumption cost, where Let t be the unit load in the t-th time unit. The power generation coal consumption rate predicted by the boiler combustion neural network model. The first item is the weighted average purchase price of the blending scheme for the t-th time unit; the second item is the environmental treatment cost, where... The sulfur oxide and nitrogen oxide production amounts predicted by the boiler combustion neural network model. , All are preset unit pollutant treatment cost coefficients.

4. The method according to claim 2, characterized in that, The objective function of the coal blending optimization model is: min J = ∑[ ]; in, The fuel consumption cost, The environmental remediation cost, These are the corresponding weights; The constraints include: Calorific value constraint is used to ensure that the overall calorific value of the coal blending scheme in each time unit is not lower than the minimum requirement of the boiler; Ash content constraints are used to limit the total ash content to within a specified upper limit. Sulfur content constraints are used to ensure that SO2 emissions meet standards. Environmental protection equipment processing capacity constraints restrict desulfurization, dust removal, and denitrification systems from exceeding their maximum processing load; Overall proportion normalization constraint: ,in For the first i The blending ratio of different types of coal.

5. The method according to claim 4, characterized in that, The preset control strategy is driven by a time-series prediction model or a reinforcement learning controller trained based on experimental and operational history. It is used to adaptively correct the set values ​​of key combustion parameters and optimize the blending ratio of each coal mill online. Optimization using the aforementioned time-series prediction model includes: The time-series prediction model outputs predicted values ​​of coal quality, emissions, and efficiency for a future period, which are then compared with the corresponding target values ​​to assess the deviation level, which is evaluated as slight deviation or significant deviation. In the event of a slight deviation, the set values ​​of the key combustion parameters are adaptively adjusted based on the prediction results; When there is a significant deviation, the objective function weights of the multi-period coal blending optimization model are updated, and the new blending scheme is quickly solved by the improved genetic algorithm. The linear transition strategy is used to gradually adjust the scheme, and closed-loop optimization is achieved by combining online monitoring and incremental model training. The target value is the constraint basis for the optimal blending scheme. It is dynamically adjusted according to the grid load period and is consistent with the unit operation constraints and environmental protection constraints.

6. The method according to claim 5, characterized in that, The construction of the time series prediction model includes: Collect multi-source time-series data including coal quality, emissions, and efficiency, and construct sliding window samples. Each sliding window sample includes data from N historical time units and one predicted future time unit. Design a multi-layer stacked LSTM architecture with a fully connected layer. Train the system using the mean squared error loss function and Adam as the optimizer. Ensure generalization capability through early stopping and regularization. The system is completed after verification.

7. The method according to claim 1, characterized in that, The improved genetic algorithm employs roulette wheel or tournament selection methods for genetic operations and iterative optimization. It prioritizes the retention of chromosomes with high fitness, randomly selects several time periods between parent chromosomes to exchange genes, and retains the coal type ratio information of high fitness time periods in the parent generation to maintain excellent local solutions. Within each time unit, the coal blending ratio is fine-tuned to maintain population diversity and enhance local search capabilities. The algorithm terminates when the population fitness converges for several consecutive generations or reaches the maximum number of iterations, and outputs the chromosome with the highest fitness as the optimal coal blending scheme for hour T.

8. The method according to claim 1, characterized in that, The online monitoring includes real-time estimation of coal quality parameters entering the furnace using a near-infrared spectroscopy analyzer and an online calorific value analyzer; the key combustion parameters include pulverized coal concentration, mill speed, air-coal ratio, and flue gas temperature.

9. The method according to claim 1, characterized in that, The dynamic database contains coal quality data including calorific value, sulfur content, ash content, volatile matter and ash fusion point of each type of coal; unit operation constraints include maximum processing capacity of desulfurization and denitrification system, maximum output of coal mill, minimum calorific value requirement to ensure stable boiler operation; and environmental constraints include emission concentration limits for SO2 and NOx. The multi-source data also includes real-time meteorological data and historical seasonal coal quality changes, which are used to dynamically correct the constraints of the coal blending optimization model.

10. The method according to claim 1, characterized in that, The execution records of the optimal blending scheme and the corresponding combustion effect data are structured and stored in a historical database, which provides a data foundation for tracing operational effects and optimizing and retraining the boiler combustion neural network model.

Images