A dual-domain cooperation-based coal-saving optimization control system and method for a chemical boiler

Abstract

The application provides a chemical boiler coal-saving optimization control system and method based on double-domain cooperation, and belongs to the technical field of intelligent boiler control. The system comprises a data acquisition and preprocessing module, a feature engineering module, a double-domain cooperative prediction module, a causal intelligent optimization module, a multi-objective cooperative decision module, a closed-loop control execution module and a model self-updating module. The causal intelligent optimization module identifies key variables and their causal paths affecting the coal consumption rate from historical data through causal structure learning, and generates optimization control instructions based on counterfactual reasoning. The multi-objective cooperative decision module integrates the coal consumption target, the thermal efficiency target and the environmental protection target into a unified optimization framework. The model self-updating module realizes adaptive and continuous optimization of the prediction model and the causal structure. The application overcomes the defects of single control target and lack of causal explanation in the optimization of air-coal ratio in the prior art, significantly reduces coal consumption and improves the optimization stability in the whole life cycle.

Description

Technical Field

[0001] This invention relates to the field of intelligent boiler control technology, specifically to a coal-saving optimization control system and method for chemical boilers, which is particularly applicable to energy-saving optimization control of coal-fired / pulverized coal boilers in the chemical industry. Background Technology

[0002] The coal chemical industry is one of the pillar industries of the national economy. As a core power equipment, chemical boilers consume huge amounts of fuel, and even a small improvement in boiler efficiency can bring considerable coal-saving benefits and emission reduction effects. However, the combustion process of chemical boilers has complex characteristics such as large time lag, strong coupling, time variability, and multiple constraints, making it difficult to achieve global optimization using traditional PID control and manual adjustment.

[0003] Existing boiler combustion optimization technologies have the following main shortcomings: First, the control objectives are singular. Most methods only focus on improving thermal efficiency or reducing flue gas temperature, failing to incorporate multiple objectives such as coal consumption rate, pollutant emissions, and furnace safety into a unified optimization framework, making it difficult to achieve a synergy between economic efficiency and environmental protection. Second, data utilization is insufficient. Traditional methods rely on manual experience or offline test data, which cannot respond in real time to fluctuations in coal quality and changes in operating conditions. Furthermore, the potential optimization patterns hidden in massive historical operating data have not been fully explored. Although existing methods have proposed data-driven modeling, they generally ignore the causal structure information implicit in the operating data, causing the model to lose key process mechanism clues in black-box prediction, resulting in insufficient interpretability and generalization of optimization strategies. Third, the air-to-coal ratio adjustment is lagging. The key to boiler efficiency lies in the "air-to-coal ratio"—excessive air intake leads to heat loss due to heat being carried away by the air, while insufficient air intake results in incomplete coal combustion; both cause coal waste. Although existing methods employ algorithms such as multivariate predictive control to optimize the air-to-coal ratio, they fail to trace the causal path between the setpoint and actual coal consumption. The adjustment commands often remain at the level of "knowing what happens but not why." Fourth, insufficient multi-regional coordination. The combustion system exhibits strong coupling characteristics between the furnace zone and the flue zone. Existing technologies have not established a cross-regional collaborative optimization mechanism, often resulting in neglecting one aspect for another. Fifth, model updates lag. Factors such as changes in coal type and equipment aging during boiler operation cause the performance of offline predictive models to gradually decline, making it difficult to meet the needs of long-term online optimization.

[0004] Therefore, developing a coal-saving optimization control system and method for chemical boilers that can fully utilize operational data, integrate physical mechanisms and data information, achieve multi-objective synergy, and possess causal interpretability has significant engineering application value. Summary of the Invention

[0005] The purpose of this invention is to provide a coal-saving optimization control system and method for chemical boilers based on dual-domain collaboration, in order to solve the problems of single control objectives, lack of causal explanation in air-coal ratio optimization, and insufficient cross-regional collaboration in the prior art, so as to realize all-time, multi-objective, closed-loop intelligent optimization control of boiler operation and significantly reduce coal consumption.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A coal-saving optimization control system for chemical boilers based on dual-domain collaboration includes: Data acquisition and preprocessing module: used to acquire multi-source heterogeneous data of the boiler combustion system in real time, including coal quality parameters, operating condition parameters, combustion state parameters, flue gas composition parameters and coal consumption related parameters, and to clean, filter and standardize the acquired data; Feature engineering module: connected to the data acquisition and preprocessing module, used to extract multi-scale features from the preprocessed data, including time series statistical features, frequency domain features and operating condition identification features, and to construct a standardized feature matrix of the boiler combustion process; Dual-domain collaborative prediction module: connected to the feature engineering module, including furnace combustion sub-prediction unit and flue heat exchange sub-prediction unit; the furnace combustion sub-prediction unit establishes a furnace combustion efficiency prediction model based on deep temporal network, and the flue heat exchange sub-prediction unit establishes a flue exhaust heat loss prediction model; the dual-domain collaborative prediction module outputs furnace combustion efficiency prediction value and flue exhaust heat loss prediction value. The causal intelligent optimization module, connected to the dual-domain collaborative prediction module, includes a causal structure learning unit and an optimization strategy generation unit. The causal structure learning unit learns the causal relationship between the air-coal ratio, coal consumption rate, and related state variables from historical operating data to construct a structural causal model. The optimization strategy generation unit determines the intervenable variables that have a direct causal effect on the coal consumption rate based on the structural causal model, and generates an optimized control instruction set containing the air-coal ratio adjustment amount, the air supply volume correction value, and the induced draft volume correction value with the goal of minimizing total coal consumption. Multi-objective collaborative decision-making module: connected to the causal intelligent optimization module, used to construct a comprehensive cost function that includes coal consumption target, thermal efficiency target and environmental protection target, and generate optimal control decision through multi-objective optimization algorithm; Closed-loop control execution module: connected to the multi-objective collaborative decision-making module, used to convert the optimal control decision into low-level execution instructions, send them to the distributed control system, and adjust the speed of each burner damper, forced draft fan and induced draft fan and coal feed execution mechanism; Model self-updating module: connected to the closed-loop control execution module, used to collect actual operation feedback data after the control command is executed, and trigger incremental updates of the dual-domain collaborative prediction model and causal structure learning model based on the deviation between the feedback data and the prediction data, forming an adaptive closed-loop control link.

[0007] Furthermore, the causal structure learning unit employs a causal discovery algorithm based on score search or a causal inference algorithm based on constraints to infer the causal graph structure between the set value of the air-coal ratio, the air supply volume, the induced draft volume, the calorific value of coal, the furnace temperature, the flue gas temperature, the flue gas oxygen content, the CO content, and the instantaneous coal consumption rate from historical operating data, and to identify the key controllable variables affecting the coal consumption rate and their causal action paths.

[0008] Furthermore, the optimization strategy generation unit performs counterfactual reasoning based on a structural causal model, calculates the expected change in coal consumption rate when a unit intervention is applied to controllable variables such as the wind-coal ratio, and uses a causal reinforcement learning strategy to generate dynamic optimization sequences of the wind-coal ratio by time period and region.

[0009] Furthermore, the comprehensive cost function of the multi-objective collaborative decision-making module is: J = α*F_coal + β*(1-η_boiler) + γ*C_NOx + δ*P_penalty; Where F_coal is the instantaneous coal consumption, η_boiler is the boiler thermal efficiency, C_NOx is the normalized value of NOx emission concentration, and P_penalty is the safety constraint penalty function term.

[0010] Furthermore, the model self-updating module employs an incremental learning algorithm to integrate newly added running data into the existing model through online learning, thereby ensuring that the prediction accuracy does not decay over time.

[0011] This invention also provides a coal-saving optimization control method for chemical boilers based on dual-domain collaboration, comprising the following steps: S1: Real-time acquisition of multi-source heterogeneous data from the boiler combustion system, followed by cleaning, filtering, and standardization processing; S2: Extract multi-scale features from the preprocessed data and construct a standardized feature matrix; S3: Input the standardized feature matrix into the furnace combustion sub-prediction unit and the flue heat exchange sub-prediction unit respectively to obtain the predicted values ​​of furnace combustion efficiency and flue heat loss. S4: Learn the causal structure of the wind-coal ratio and coal consumption rate from historical operating data, and construct a structural causal model; S5: Based on the aforementioned structural causal model, perform counterfactual reasoning to calculate the expected change in coal consumption rate when intervention is applied to the wind-coal ratio, and generate an optimized control instruction set; S6: Construct a comprehensive cost function that includes coal consumption targets, thermal efficiency targets, and environmental protection targets, and generate the optimal control decision through a multi-objective optimization algorithm; S7: Converts the optimal control decision into low-level execution instructions, sends them to the distributed control system, and adjusts each actuator; S8: Collect actual operation feedback data after the control command is executed. Based on the deviation between the feedback data and the predicted data, trigger incremental model update and return to step S1 to form a closed-loop control link.

[0012] Furthermore, in step S4, the method for constructing the structural causal model includes: initializing a variable set based on historical operating data, learning the causal skeleton using an equivalence class search algorithm, determining the direction of the edges through conditional independence testing, and outputting a directed acyclic graph containing causal directions.

[0013] Furthermore, in step S6, the multi-objective optimization algorithm adopts the fast non-dominated sorting genetic algorithm NSGA-II with an elitist strategy, and selects the compromise optimal solution on the Pareto front by minimizing the turning distance criterion.

[0014] Furthermore, in step S8, the triggering condition for incremental model update is: the cumulative relative deviation between the actual coal consumption rate and the predicted coal consumption rate exceeds 3% and lasts for more than 10 minutes, or the confidence interval width of the causal effect estimation of the wind-coal ratio exceeds a preset threshold.

[0015] Furthermore, in step S3, the furnace combustion prediction unit adopts a hybrid deep network structure combining convolutional neural networks and bidirectional gated recurrent units, and the flue heat exchange prediction unit establishes a fast prediction model for flue gas heat loss based on support vector regression or extreme learning machine.

[0016] Compared with the prior art, the beneficial effects of the present invention are reflected in the following aspects: 1. Significantly reduce coal consumption: By using causal intelligent optimization to perform counterfactual reasoning and precise control of the air-coal ratio, compared with traditional PID control and manual experience adjustment, the coal consumption per ton of steam can be reduced by 2.0%-3.5%, and the annual coal saving of a single boiler can reach more than 1,500-3,000 tons.

[0017] 2. Strong causal interpretability: By introducing causal structure learning units, key variables affecting coal consumption rate and their causal paths are automatically discovered from the data, making the optimization strategy interpretable and overcoming the shortcomings of traditional black box models that "know what but not why".

[0018] 3. Multi-objective synergistic optimization: Coal consumption target, thermal efficiency target and environmental protection target are integrated into the comprehensive cost function. The optimal balance point is found on the Pareto frontier through multi-objective optimization, so as to both ensure coal consumption and control emissions.

[0019] 4. More accurate dual-domain collaboration: Dedicated prediction models are established for the furnace combustion zone and the flue heat exchange zone, which fully consider the heat transfer characteristics and time scale differences of different regions, improving the prediction accuracy by 15%-25% compared with a single black box model.

[0020] 5. Model self-updating ensures long-term performance: Incremental learning technology is used to enable the model to adaptively update as operating conditions change, ensuring the stability and reliability of the optimization system throughout its entire lifecycle. Attached Figure Description

[0021] Figure 1 The overall architecture diagram of the dual-domain collaborative coal-saving optimization control system for chemical boilers provided by this invention is shown.

[0022] Figure 2 The flowchart shows the coal-saving optimization control method for chemical boilers based on dual-domain collaboration provided by this invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments.

[0024] Example 1: Coal-saving optimization control system for chemical boilers based on dual-domain collaboration This embodiment provides a coal-saving optimization control system for chemical boilers based on dual-domain collaboration, such as... Figure 1 As shown, it includes a data acquisition and preprocessing module, a feature engineering module, a dual-domain collaborative prediction module, a causal intelligent optimization module, a multi-objective collaborative decision-making module, a closed-loop control execution module, and a model self-updating module, and is used for combustion optimization control of a 240 t / h pulverized coal boiler in a chemical plant.

[0025] I. Data Acquisition and Preprocessing Module A sensor network is deployed at key locations on the boiler body: furnace temperature is measured using a thermocouple matrix (4 layers each on the front and rear walls, 3 points per layer); flue gas temperature is measured at the inlet and outlet of the tail flue; an oxygen analyzer is installed in the economizer outlet flue; flue gas CO content is monitored online using a laser gas analyzer; and coal quality parameters (ash content, volatile matter, sulfur content, and calorific value) are obtained in real time using an online elemental analyzer (LIBS technology). Instantaneous coal feed rate is calculated by integrating the feeder speed and tare weight; forced draft and induced draft rates are measured by the inlet flow meters of each fan. All sensor data are collected in a real-time database via the OPC protocol at a sampling frequency of 1 second per point.

[0026] The data preprocessing steps are as follows: missing values ​​are detected in the original time series data of each measurement point, and linear interpolation is used to fill in missing values ​​with a missing rate of less than 5%; a Hampel filter is used to remove abrupt noise (window length is 50 sampling points, threshold coefficient is 3); then, noise reduction and smoothing are completed by sliding window mean filtering (window size is 10 seconds); finally, Z-score standardization is performed on the feature data.

[0027] II. Feature Engineering Module Three types of features are extracted from the preprocessed data: Time series statistical characteristics: The mean, standard deviation, maximum value, minimum value, and slope change rate of each operating parameter were extracted at three time scales: 60 seconds, 300 seconds, and 1800 seconds. Frequency domain characteristics: Fourier transform is performed on the furnace temperature and flue gas composition sequence to extract the energy proportion of the dominant frequency component; Operating condition identification features: Based on historical data clustering, operating condition labels such as boiler load, coal type, and ambient temperature are generated.

[0028] Each feature is normalized and then concatenated into a standardized feature matrix.

[0029] III. Dual-Domain Collaborative Prediction Module Furnace Combustion Prediction Unit: A hybrid CNN-BiGRU deep network was established. The CNN layer consists of two convolutional layers and two pooling layers, used to extract local correlation patterns and spatial structure from the feature matrix; the BiGRU layer contains a bidirectional gated recurrent unit with 64 hidden units to capture the forward and backward temporal dependencies of the combustion process. Input features include coal feed rate, air supply rate, induced draft rate, furnace temperature and oxygen content, and CO content. Outputs are predicted combustion efficiency and predicted NOx concentration. The model was trained using the Adam optimizer with an initial learning rate of 0.001, a batch size of 128, and the MSE loss function.

[0030] Flue heat exchanger prediction unit: Establish an SVR regression model with input features including flue gas temperature, flue gas oxygen content, flue gas flow rate, and economizer inlet and outlet water temperatures, and output as the predicted value of flue gas heat loss.

[0031] IV. Causal Intelligence Optimization Module The causal structure learning unit employs a scoring-based equivalence class search algorithm to learn the causal structure between variables from the boiler's historical operating data over the past three months. The set of variables to be learned includes: coal feed rate Q_coal, air-coal ratio R_ac, forced draft rate Q_air, induced draft rate Q_exh, furnace temperature T_furn, flue gas temperature T_exh, flue gas oxygen content O2_exh, CO content C_CO, coal calorific value H_coal, and instantaneous coal consumption rate F_coal. Through greedy equivalence class search and Bayesian information criterion scoring, a directed acyclic causal graph is inferred among the variables, identifying controllable variables directly affecting the coal consumption rate and their causal path lengths.

[0032] The optimization strategy generation unit performs counterfactual reasoning based on the above structural causal model: when the air-coal ratio R_ac is intervened from its current value r0 to r0+Δr, the expected change ΔF in the coal consumption rate F_coal is calculated. The counterfactual reasoning formula is: ΔF = E[F_coal | do(R_ac = r0+Δr)] - E[F_coal | do(R_ac = r0)] The do operator represents the intervention operation. Based on the causal effect strength of each controllable variable, a dynamic optimization sequence of wind-coal ratio is generated by time period and region. With a 15-minute optimization window, the wind-coal ratio optimization sequence is decomposed into three components within the window: forward adjustment (first 5 minutes), stable maintenance (middle 5 minutes), and tracking correction (last 5 minutes).

[0033] V. Multi-objective collaborative decision-making module Construct the comprehensive cost function: J = α*F_coal + β*(1-η_boiler) + γ*C_NOx + δ*P_penalty; The weighting coefficients are set as follows: α=0.50, β=0.20, γ=0.20, δ=0.10. The safety penalty function P_penalty is activated when the furnace temperature exceeds the limit or the CO concentration exceeds the threshold. The penalty value increases quadratically with the degree of exceedance.

[0034] A multi-objective optimization framework based on NSGA-II was adopted, with an initial population size of 200, a crossover probability of 0.9, and a mutation probability of 0.1. After 200 iterations, the final compromise optimal solution was selected on the Pareto front using the criterion of minimizing the turning distance. This yielded the control parameter combination that minimized the overall cost function, from which the wind-coal ratio adjustment ΔR_ac, the forced air volume correction ΔQ_air, and the induced draft volume correction ΔQ_exh were extracted.

[0035] VI. Closed-loop control execution module The optimized control command set is converted into low-level protocol commands (Modbus TCP / RTU) and sent to the boiler DCS system via OPC communication. Execution strategies include: smooth adjustment of the air-coal ratio at a rate of 1 minute per step, with a total adjustment time not exceeding 5 minutes; damper opening correction executed via a proportional-integral controller; and frequency conversion adjustment of the forced and induced draft fan speeds according to a preset slope. Seamless switching between the system and the DCS is achieved via a hard-wired switch, ensuring that the operator can manually reset to the original DCS control under any circumstances.

[0036] VII. Model Self-Update Module A dual-threshold trigger mechanism is implemented: Model updates are triggered when the cumulative relative deviation between the actual and predicted coal consumption rates exceeds 3% for more than 10 minutes, or when the confidence interval width of the estimated causal effect of the wind-coal ratio exceeds a preset threshold. Incremental update strategy: The baseline model is retrained using a sliding window (data from the most recent 3 months) on historical data. New data is sampled with time weighting, with higher weights for more recent data. The dual-domain collaborative prediction model performs partial network fine-tuning every 7 days, and the causal structure learning model performs a global causal graph re-push every 30 days.

[0037] like Figure 2 As shown, the present invention also provides a coal-saving optimization control method for chemical boilers based on dual-domain collaboration, comprising the following steps: S1: Real-time acquisition of multi-source heterogeneous data from the boiler combustion system, followed by cleaning, filtering, and standardization processing; S2: Extract multi-scale features from the preprocessed data and construct a standardized feature matrix; S3: Input the standardized feature matrix into the furnace combustion sub-prediction unit and the flue heat exchange sub-prediction unit respectively to obtain the predicted values ​​of furnace combustion efficiency and flue heat loss. S4: Learn the causal structure of the wind-coal ratio and coal consumption rate from historical operating data, and construct a structural causal model; S5: Based on the aforementioned structural causal model, perform counterfactual reasoning to calculate the expected change in coal consumption rate when intervention is applied to the wind-coal ratio, and generate an optimized control instruction set; S6: Construct a comprehensive cost function that includes coal consumption targets, thermal efficiency targets, and environmental protection targets, and generate the optimal control decision through a multi-objective optimization algorithm; S7: Converts the optimal control decision into low-level execution instructions, sends them to the distributed control system, and adjusts each actuator; S8: Collect actual operation feedback data after the control command is executed. Based on the deviation between the feedback data and the predicted data, trigger incremental model update and return to step S1 to form a closed-loop control link.

[0038] As an optional embodiment, in step S4, the method for constructing the structural causal model includes: initializing a variable set based on historical running data, learning the causal skeleton using an equivalence class search algorithm, determining the direction of the edges through conditional independence testing, and outputting a directed acyclic graph containing causal directions.

[0039] As an optional embodiment, in step S6, the multi-objective optimization algorithm adopts the fast non-dominated sorting genetic algorithm NSGA-II with an elitist strategy, and selects the compromise optimal solution on the Pareto front by minimizing the turning distance criterion.

[0040] As an optional embodiment, in step S8, the triggering condition for incremental model update is: the cumulative relative deviation between the actual coal consumption rate and the predicted coal consumption rate exceeds 3% and lasts for more than 10 minutes, or the confidence interval width of the causal effect estimation of the wind-coal ratio exceeds a preset threshold.

[0041] As an optional embodiment, in step S3, the furnace combustion sub-prediction unit adopts a hybrid deep network structure combining convolutional neural networks and bidirectional gated recurrent units, and the flue heat exchange sub-prediction unit establishes a fast prediction model for flue heat loss based on support vector regression or extreme learning machine.

[0042] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Those skilled in the art can make various modifications and variations to the present invention without departing from its spirit and scope, and all such modifications and variations fall within the protection scope of the present invention.

Claims

1. A coal-saving optimization control system for chemical boilers based on dual-domain collaboration, characterized in that, include: The data acquisition and preprocessing module is used to acquire multi-source heterogeneous data from the boiler combustion system in real time, and to clean, filter and standardize the acquired data. The feature engineering module, connected to the data acquisition and preprocessing module, is used to extract multi-scale features from the preprocessed data and construct a standardized feature matrix. The dual-domain collaborative prediction module, connected to the feature engineering module, includes a furnace combustion sub-prediction unit and a flue heat exchange sub-prediction unit; the furnace combustion sub-prediction unit establishes a furnace combustion efficiency prediction model, and the flue heat exchange prediction unit establishes a flue gas heat loss prediction model. The causal intelligent optimization module, connected to the dual-domain collaborative prediction module, includes a causal structure learning unit and an optimization strategy generation unit. The causal structure learning unit learns the causal structure between the wind-coal ratio and the coal consumption rate from historical operating data. The optimization strategy generation unit determines the intervenable variables affecting the coal consumption rate based on the causal structure and generates an optimization control instruction set. The multi-objective collaborative decision-making module is used to construct a comprehensive cost function that includes coal consumption targets, thermal efficiency targets, and environmental protection targets, and to generate the optimal control decision through a multi-objective optimization algorithm; The closed-loop control execution module is used to convert the optimal control decision into low-level execution instructions and send them to the distributed control system. The model self-update module is used to collect actual operation feedback data and trigger incremental updates of the model based on the deviation between the feedback data and the predicted data.

2. The dual-domain collaborative coal-saving optimization control system for chemical boilers according to claim 1, characterized in that, The causal structure learning unit employs a score-based search-based causal discovery algorithm or a constraint-based causal inference algorithm to infer the causal graph structure between the setpoint of the air-coal ratio, the air supply volume, the induced draft volume, the calorific value of the coal, the furnace temperature, the flue gas temperature, the flue gas oxygen content, the CO content, and the instantaneous coal consumption rate from historical operating data, and to identify the key controllable variables affecting the coal consumption rate and their causal paths.

3. The coal-saving optimization control system for chemical boilers based on dual-domain collaboration according to claim 1, characterized in that, The optimization strategy generation unit performs counterfactual reasoning based on a causal structure model, calculates the expected change in coal consumption rate when a unit intervention is applied to controllable variables such as the wind-coal ratio, and uses a causal reinforcement learning strategy to generate dynamic optimization sequences of the wind-coal ratio by time period and region.

4. The coal-saving optimization control system for chemical boilers based on dual-domain collaboration according to claim 1, characterized in that, The comprehensive cost function of the multi-objective collaborative decision-making module is: J = α*F_coal + β*(1-η_boiler) + γ*C_NOx + δ*P_penalty; Where F_coal is the instantaneous coal consumption, η_boiler is the boiler thermal efficiency, C_NOx is the normalized value of NOx emission concentration, P_penalty is the safety constraint penalty function term, and α, β, γ, and δ are the corresponding weighting coefficients.

5. The dual-domain collaborative coal-saving optimization control system for chemical boilers according to claim 1, characterized in that, The model self-updating module adopts an incremental learning algorithm to integrate newly added running data into the existing model through online learning, enabling the dual-domain collaborative prediction model and the causal structure learning model to achieve adaptive continuous optimization.

6. A coal-saving optimization control method for chemical boilers based on dual-domain collaboration, applied to the system described in any one of claims 1 to 5, characterized in that, Includes the following steps: S1: Real-time acquisition of multi-source heterogeneous data from the boiler combustion system, followed by cleaning, filtering, and standardization processing; S2: Extract multi-scale features from the preprocessed data and construct a standardized feature matrix; S3: Input the standardized feature matrix into the furnace combustion sub-prediction unit and the flue heat exchange sub-prediction unit respectively to obtain the predicted values ​​of furnace combustion efficiency and flue heat loss. S4: Learn the causal structure of the wind-coal ratio and coal consumption rate from historical operating data, and construct a structural causal model; S5: Based on the aforementioned structural causal model, perform counterfactual reasoning to calculate the expected change in coal consumption rate when intervention is applied to the wind-coal ratio, and generate an optimized control instruction set; S6: Construct a comprehensive cost function that includes coal consumption targets, thermal efficiency targets, and environmental protection targets, and generate the optimal control decision through a multi-objective optimization algorithm; S7: Converts the optimal control decision into low-level execution instructions, sends them to the distributed control system, and adjusts each actuator; S8: Collect actual operation feedback data after the control command is executed. Based on the deviation between the feedback data and the predicted data, trigger incremental model update and return to step S1 to form a closed-loop control link.

7. The method for coal-saving optimization control of chemical boilers based on dual-domain collaboration according to claim 6, characterized in that, In step S4, the method for constructing the structural causal model includes: initializing a variable set based on historical running data, learning the causal skeleton using an equivalence class search algorithm, determining the direction of the edges through conditional independence testing, and outputting a directed acyclic graph containing causal directions.

8. The method for coal-saving optimization control of chemical boilers based on dual-domain collaboration according to claim 6, characterized in that, In step S6, the multi-objective optimization algorithm adopts the fast non-dominated sorting genetic algorithm NSGA-II with elitist strategy, and selects the compromise optimal solution on the Pareto front by minimizing the turning distance criterion.

9. The method for coal-saving optimization control of chemical boilers based on dual-domain collaboration according to claim 6, characterized in that, In step S8, the triggering conditions for incremental model updates are: the cumulative relative deviation between the actual coal consumption rate and the predicted coal consumption rate exceeds 3% and lasts for more than 10 minutes, or the confidence interval width of the causal effect estimation of the wind-coal ratio exceeds a preset threshold.

10. The method for coal-saving optimization control of chemical boilers based on dual-domain collaboration according to claim 6, characterized in that, In step S3, the furnace combustion prediction unit adopts a hybrid deep network structure combining convolutional neural networks and bidirectional gated recurrent units, and the flue heat exchange prediction unit establishes a fast prediction model for flue heat loss based on support vector regression or extreme learning machine.

Images