A Boiler Combustion Optimization Method and System Based on a Dual-Model Cooperative Architecture Algorithm

Abstract

This invention proposes a boiler combustion optimization method and system based on a dual-model collaborative architecture algorithm, belonging to the field of boiler combustion control technology in thermal power plants. The method includes: collecting furnace operating parameters; preprocessing the parameters, including data cleaning, denoising, normalization, and hierarchical spatiotemporal fusion, to obtain a comprehensive feature vector of the furnace operating parameters; constructing a furnace operating parameter prediction model based on the comprehensive feature vector, outputting predicted values ​​of the furnace operating parameters, wherein the prediction model is formed by fusing a random forest model and a long short-term memory network model, and adjusting the fusion weights according to the rate of change of furnace combustion conditions; adjusting control equipment according to the predicted values; acquiring actual operating data of fly ash carbon content, nitrogen oxide concentration, and oxygen content after adjustment, calculating the deviation from the predicted values, and incrementally updating the prediction model locally; repeating the above steps to continuously optimize the furnace operating parameters. This invention improves the efficiency of boiler combustion optimization through dual-model collaborative optimization.

Description

Technical Field

[0001] This invention belongs to the field of boiler combustion control technology in thermal power plants, specifically relating to a boiler combustion optimization method and system based on a dual-model collaborative architecture algorithm. Background Technology

[0002] As the core equipment for thermal power generation, the combustion efficiency and environmental performance of boilers in thermal power plants directly determine the economic benefits and environmental compliance of the power plant. Currently, traditional boiler combustion control in the industry mainly relies on manual experience to set fixed operating parameters. This mode has significant limitations, as it cannot respond in real time to dynamic factors such as fluctuations in coal quality, changes in unit load, and sudden changes in furnace operating conditions. This leads to combustion states deviating from the optimal range, resulting in problems such as low combustion efficiency and serious combustion waste. At the same time, an unreasonable combustion process will cause nitrogen oxides (NOx) to be produced. x Emissions of pollutants such as carbon monoxide (CO) exceed standards, making it difficult to meet increasingly stringent environmental protection standards.

[0003] Chinese invention patent application document with publication number CN121297039A discloses a combustion optimization control system for boilers in thermal power plants. It achieves combustion parameter adjustment through multi-sensor fusion and optimization algorithm. However, its model lacks an adaptive correction mechanism based on operating deviation and has no feedback update mechanism, resulting in a certain lag in control commands, which reduces the control accuracy and response speed of boiler combustion.

[0004] Chinese invention patent application document with publication number CN121854889A discloses a machine learning-based intelligent optimization control system for boiler combustion. It predicts the combustion state through a time-series coupled model and adjusts the boiler combustion with the goal of maximizing the comprehensive evaluation index of the combustion state. Although a feedback update mechanism is introduced, the overall model parameter adjustment still leads to a large computational overhead for model update and reduces the response speed. Summary of the Invention

[0005] To address the problems in the prior art, this invention proposes a boiler combustion optimization method and system based on a dual-model collaborative architecture algorithm. By combining a random forest model and a long short-term memory network model in a dual-model collaborative architecture with incremental local updates, the response speed of boiler combustion optimization control in thermal power plants is improved.

[0006] The technical solution of the present invention is as follows: In a first aspect, this invention proposes a boiler combustion optimization method based on a dual-model collaborative architecture algorithm, comprising the following steps: Collect furnace operating parameters during the boiler combustion process; The furnace operating parameters are preprocessed, including data cleaning, noise reduction, normalization and hierarchical spatiotemporal fusion, and a comprehensive feature vector of the furnace operating parameters is obtained after preprocessing. A furnace operating parameter prediction model is constructed based on comprehensive feature vectors, and the predicted values ​​of furnace operating parameters are output. The prediction model is formed by weighted fusion of random forest model and long short-term memory network model, and the fusion weight is adjusted according to the rate of change of furnace combustion conditions. Adjust the boiler combustion control equipment according to the predicted values ​​of furnace operating parameters; Obtain actual operating data of adjusted fly ash carbon content, nitrogen oxide concentration and oxygen content, calculate the deviation between the actual operating data and the predicted values ​​of furnace operating parameters, and perform incremental local updates on the furnace operating parameter prediction model. Repeat the above steps to continuously optimize the furnace operating parameters.

[0007] Furthermore, the denoising employs an adaptive threshold wavelet thresholding denoising algorithm, specifically as follows: Wavelet decomposition is performed on each data sequence in the furnace operating parameters. The threshold is adjusted according to the energy distribution of the wavelet coefficients of each decomposition layer using wavelet basis functions. An adaptive adjustment coefficient is introduced into the adjustment threshold. The wavelet coefficients processed by the adjusted threshold are reconstructed to obtain the denoised furnace operating parameters. The adaptive adjustment coefficients are determined based on the energy proportion of the wavelet coefficients of the corresponding sub-levels.

[0008] Furthermore, the hierarchical spatiotemporal fusion integrates different types of furnace operating parameters to construct multi-level features, specifically including: The first layer of spatiotemporal alignment performs unified processing of different types of furnace operating parameters in both the temporal and spatial domains, aligning the time axis and three-dimensional spatial coordinates of the furnace operating parameters. The second layer of operating condition feature association is based on the spatiotemporally aligned furnace operating parameters. It extracts furnace combustion operating condition features, flue gas flow field features, and pulverized coal conveying features, and establishes the coupling relationship between furnace combustion operating condition features, flue gas flow field features, and pulverized coal conveying features through correlation analysis. The third layer of comprehensive feature construction uses principal component analysis to extract the combustion efficiency-pollutant generation coupled feature vector, and retains principal components with a preset contribution rate greater than a preset percentage as comprehensive feature vectors.

[0009] Furthermore, the rate of change of the furnace combustion conditions is the average value of the Euclidean distance within a preset time window; When the rate of change of furnace combustion conditions is greater than the preset rate of change threshold, the weight ratio of the random forest model is greater than the preset weight ratio of the long short-term memory network model. When the rate of change of furnace combustion conditions is less than the preset rate of change threshold, the weight ratio of the random forest model is less than the preset weight ratio of the long short-term memory network model.

[0010] Furthermore, the furnace operating parameter prediction model uses a non-dominated sorting genetic algorithm to optimize the furnace operating parameters, introduces adaptive crossover probability and adaptive mutation probability, and uses a sorting strategy based on crowding distance to filter the non-dominated solution set. The adaptive crossover probability gradually decreases from the first crossover probability to the second crossover probability with the number of iterations. The adaptive mutation probability gradually increases from the first mutation probability to the second mutation probability with the number of iterations.

[0011] Furthermore, the furnace operating parameter prediction model aims to maximize combustion efficiency and minimize nitrogen oxide emissions, with optimization constraints including: Primary air volume constraints; Secondary air volume constraints; Constraints on the burner swing angle range; Upper limit constraint on furnace temperature.

[0012] Furthermore, when the deviation between the actual operating data and the predicted values ​​of the furnace operating parameters exceeds a preset deviation value, the random forest model only updates the tree node parameters of the decision tree related to the furnace operating parameters that exceed the preset deviation value; the long short-term memory network model only updates the weight parameters of the output layer.

[0013] Secondly, this invention proposes a boiler combustion optimization system based on a dual-model collaborative architecture algorithm, the system comprising: A multi-sensor monitoring module is used to collect furnace operating parameters during the boiler combustion process; The data fusion and preprocessing unit preprocesses the furnace operating parameters, including data cleaning, noise reduction, normalization, and hierarchical spatiotemporal fusion. The parameter prediction optimization unit constructs a furnace operation parameter prediction model through a dual-model collaborative architecture of random forest model and long short-term memory network model. It adjusts the fusion weights of random forest model and long short-term memory network model according to the rate of change of furnace combustion conditions and outputs the predicted values ​​of furnace operation parameters. The control and execution unit adjusts the boiler combustion control equipment according to the predicted values ​​of furnace operating parameters and receives and executes the optimal combustion parameter command. The feedback execution unit compares the furnace operating parameters with the predicted values. When the deviation exceeds the preset deviation value, it performs an incremental local update on the furnace operating parameter prediction model. The update only applies to the parts of fly ash content, nitrogen oxide concentration, and oxygen content that exceed the deviation.

[0014] Furthermore, the sensors of the multi-sensor monitoring module are deployed in a distributed and hierarchical manner, with sensors configured according to three major areas: the furnace combustion zone, the flue zone, and the pulverized coal conveying zone. These sensors include at least a furnace temperature sensor, an oxygen content sensor, a carbon monoxide concentration sensor, a pulverized coal concentration sensor, a feedwater temperature sensor, a steam pressure sensor, and a fly ash carbon content sensor. The furnace temperature sensors are arranged in layers along the height of the furnace. The oxygen content sensor is arranged at the furnace outlet and in each stage of the flue to monitor the degree of combustion completeness; The coal powder concentration sensor is placed at the inlet of the coal powder pipeline to monitor the mass of coal powder contained in a unit volume of airflow.

[0015] Furthermore, the system also includes a remote monitoring and diagnostic module, which builds a remote interactive interface through a cloud platform to display boiler operating parameters, optimization parameter curves and system working status in real time, perform fault diagnosis and early warning for the system, and issue intervention commands to the control execution unit.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: By performing hierarchical spatiotemporal fusion processing on multi-source furnace operating parameters, unified modeling of data in the time dimension, spatial dimension, and operating condition correlation dimension can be achieved, which can effectively characterize the multi-physical field coupling relationship between the furnace, flue gas flow field, and pulverized coal conveying during boiler combustion, thereby improving the accuracy of feature expression. A dual-model collaborative architecture of random forest model and long short-term memory network model is adopted, and the model fusion weight is adaptively adjusted according to the rate of change of furnace combustion conditions. This enables the model to have a stronger rapid response capability when the operating conditions change drastically and a better time series prediction capability when the operating conditions are stable, thereby improving the overall prediction accuracy and stability. By introducing a bias-based feedback adjustment mechanism, when the deviation between the predicted and actual values ​​exceeds a threshold, only the local parameters of the model are incrementally updated without overall retraining, which can improve the model's update efficiency and combustion control response speed. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the steps of a boiler combustion optimization method based on a dual-model collaborative architecture algorithm. Figure 2 A flowchart for optimizing predictions using a dual-model collaborative architecture algorithm; Figure 3 This is a structural diagram of a boiler combustion optimization system based on a dual-model collaborative architecture algorithm. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0019] Example 1 This embodiment provides a boiler combustion optimization method based on a dual-model collaborative architecture algorithm, such as... Figure 1 As shown, it includes the following steps: S1. Collect furnace operating parameters during boiler combustion, including at least furnace temperature, oxygen content, carbon monoxide concentration, nitrogen oxide concentration, pulverized coal concentration, feedwater temperature, steam pressure, and fly ash carbon content. S2. Preprocess the furnace operating parameters, including data cleaning, noise reduction, normalization and hierarchical spatiotemporal fusion; The data cleaning process involves outlier detection and missing value completion for the original furnace operating parameters. The 3σ criterion is used to identify outliers. Specifically, if a data point exceeds three times the standard deviation of the parameter's mean under the current operating conditions, the data point is determined to be an outlier and removed. The sliding window mean method is used to complete the missing data, with the sliding window size set to 5 sampling points. The denoising process employs an adaptive threshold wavelet thresholding algorithm to denoise the furnace operating parameters. Specifically, the adaptive threshold wavelet thresholding algorithm involves performing a 5-level wavelet decomposition on the data sequences of different types of data within the furnace operating parameters, using the sym8 wavelet basis function, and adaptively adjusting the threshold based on the energy distribution of the wavelet coefficients at each level. The formula for calculating the threshold is as follows: ; in, is the noise standard deviation of the wavelet coefficients at the j-th layer; N is the length of the data sequence; Let be the adaptive adjustment coefficient for the j-th layer; The adaptive adjustment coefficient is determined based on the energy proportion of the wavelet coefficients in the corresponding layer, and the value range is [0.8, 1.2]. The wavelet coefficients after adaptive adjustment threshold processing are reconstructed to obtain the denoised furnace operating parameters. The normalization adopts the maximum-minimum normalization method, mapping the furnace operating parameters to the [0, 1] interval according to different furnace operating parameter types, as expressed by the formula: ; in, Normalized data; x represents the raw data of furnace operating parameters; x min This represents the minimum furnace operating parameter value for the corresponding type; xmax This represents the maximum furnace operating parameter value for the corresponding type; The hierarchical spatiotemporal fusion constructs multi-level features for multi-source furnace operating parameters, fusing different types of furnace operating parameters, specifically including: The first layer of spatiotemporal alignment unifies the temporal and spatial processing of different types of furnace operating parameters. In the time domain, linear interpolation is used to resample data at different sampling frequencies, aligning all data to a unified time axis with a time resolution of 1 second. In the spatial domain, based on the actual installation location of the sensors that collect different types of furnace operating parameters, the corresponding data is mapped to the three-dimensional spatial coordinate system of the boiler furnace, achieving spatial alignment of multi-source data. The second layer of operating condition feature correlation involves extracting furnace combustion characteristics, flue gas flow field characteristics, and pulverized coal conveying characteristics based on spatiotemporally aligned data. Furnace combustion characteristics include furnace temperature gradient and temperature distribution uniformity; flue gas flow field characteristics include flue gas velocity and flue gas temperature distribution; and pulverized coal conveying characteristics include pulverized coal concentration uniformity and pulverized coal flow rate. Subsequently, Pearson correlation coefficient analysis is used to establish the correlation between these characteristics, in order to characterize the boiler's multi-physics coupling properties. The third layer of comprehensive feature construction uses principal component analysis (PCA) to extract the combustion efficiency-pollutant generation coupled feature vector, and retains the top k principal components with a cumulative contribution rate greater than 95% as the comprehensive feature vector.

[0020] S3. Optimized prediction of the comprehensive feature vector, such as... Figure 2 As shown, a furnace operating parameter prediction model is constructed using a dual-model collaborative architecture of a random forest model and a long short-term memory network model. The fusion weights of the random forest model and the long short-term memory network model are adjusted according to the rate of change of the furnace combustion conditions, and the predicted values ​​of the furnace operating parameters are output. The random forest model takes the comprehensive feature vector as input and outputs control parameters for primary air volume, secondary air volume, pulverized coal conveying volume and burner swing angle. The random forest model contains 100 decision trees, and the maximum depth of the decision trees is set to 10 layers. The long short-term memory network model includes an input layer, an LSTM layer, and an output layer. The LSTM layer contains at least two long short-term memory network layers, each containing 64 neurons. The long short-term memory network model takes a sequence of integrated feature vectors from 20 consecutive time steps as input and outputs the burning parameters at the predicted time step. The rate of change of the furnace combustion conditions is the average Euclidean distance of the comprehensive feature vector over the past 10 moments. When the rate of change is greater than the preset rate of change threshold, the fusion weight of the random forest model is set to 0.7 and the fusion weight of the long short-term memory network model is set to 0.3. When the rate of change is less than the preset rate of change threshold, the fusion weight of the random forest model is set to 0.3 and the fusion weight of the long short-term memory network model is set to 0.7. The furnace operation parameter prediction model aims to maximize combustion efficiency and minimize nitrogen oxide emissions, and imposes constraints on the range of control parameters, specifically including: The primary air volume range is [1.8, 2.2] kg (air) / kg (coal); The secondary air volume range is [0.8, 1.2] kg (air) / kg (coal); The burner swing angle range is [-30°, +30°]; Furnace temperature ≤1250℃.

[0021] S4. Based on the predicted optimal furnace operating parameters, control the boiler combustion control equipment and adjust the combustion state; S5. Calculate the deviation between the actual operating data and the predicted values ​​of fly ash carbon content, nitrogen oxide concentration and oxygen content. When the deviation exceeds the preset deviation value, perform incremental local updates to the furnace operating parameter prediction model. The deviation Expressed as a formula: ; in, This is actual operational data; This is a predicted value; The incremental local update only updates the furnace operating parameters whose deviation exceeds the preset deviation value, without retraining the entire model parameters. For the random forest model, only the tree node parameters of the decision tree that are related to the furnace operating parameters are updated; For Long Short-Term Memory (LSTM) network models, only the weight parameters of the output layer are updated.

[0022] S6. Repeat steps S1 to S5 to achieve continuous dynamic optimization of the boiler combustion process.

[0023] Preferably, in the optimization process of the furnace operating parameter prediction model in step S3, a non-dominated sorting genetic algorithm is used to optimize the furnace operating parameters, and adaptive crossover probability and mutation probability are introduced. The crossover probability gradually decreases linearly from 0.9 to 0.6 with the number of iterations, and the mutation probability gradually increases linearly from 0.1 to 0.3 with the number of iterations, so as to enhance the global search capability in the early stage of optimization and improve the local search accuracy in the later stage of optimization. Meanwhile, a sorting strategy based on crowding distance is adopted to screen the non-dominated solution set. Specifically, the non-dominated solution sets of the same level are sorted according to the crowding distance, and a preset number of solutions with larger crowding distances are retained according to the sorting results to maintain the diversity of population distribution and thus obtain a uniformly distributed Pareto optimal solution set.

[0024] Example 2 This embodiment provides a boiler combustion optimization system based on a dual-model collaborative architecture algorithm, such as... Figure 3 As shown, the system includes: The multi-sensor monitoring module is used to collect furnace operating parameters during the boiler combustion process. The sensors adopt a distributed and hierarchical deployment method, and are configured according to three major areas: furnace combustion zone, flue zone, and pulverized coal conveying zone. The sensors include at least furnace temperature sensor, oxygen content sensor, carbon monoxide concentration sensor, pulverized coal concentration sensor, feedwater temperature sensor, steam pressure sensor, and fly ash carbon content sensor. The data fusion and preprocessing unit preprocesses the furnace operating parameters, including data cleaning, noise reduction, normalization, and hierarchical spatiotemporal fusion. The parameter prediction optimization unit constructs a furnace operation parameter prediction model through a dual-model collaborative architecture of random forest model and long short-term memory network model. It adjusts the fusion weights of random forest model and long short-term memory network model according to the rate of change of furnace combustion conditions and outputs the predicted values ​​of furnace operation parameters. The control execution unit adjusts the boiler combustion control equipment, including damper regulator, pulverized coal conveying device and burner regulating device, according to the predicted values ​​of furnace operating parameters, and receives and executes the optimal combustion parameter command; The feedback adjustment module compares the furnace operating parameters with the predicted values. When the deviation exceeds the preset deviation value, it performs incremental local updates to the furnace operating parameter prediction model. The updates only target the out-of-range parts of fly ash content, nitrogen oxide concentration, and oxygen content, without performing full retraining.

[0025] Preferably, the system also includes a remote monitoring and diagnostic module, which builds a remote interactive interface through a cloud platform, allowing technicians to view boiler operating parameters in real time, optimize parameter curves and system operating status, and has fault diagnosis and early warning functions. It can remotely issue boiler combustion optimization strategy adjustment intervention commands to the control execution unit, thereby improving the convenience of system operation and maintenance.

[0026] Preferably, the furnace temperature sensor is arranged in 8 to 12 monitoring positions along the height of the furnace; the oxygen content sensor is arranged at the furnace outlet and each level of flue to monitor the degree of combustion; and the pulverized coal concentration sensor is arranged at the inlet of the pulverized coal pipeline to monitor the mass of pulverized coal contained in a unit volume of airflow.

[0027] Preferably, the hierarchical spatiotemporal fusion constructs multi-level features for the furnace operating parameters collected by each sensor in the multi-sensor monitoring module, fusing different types of furnace operating parameters, specifically including: The first layer of spatiotemporal alignment unifies the temporal and spatial processing of different types of furnace operating parameters. In the time domain, linear interpolation is used to resample data at different sampling frequencies, aligning all data to a unified time axis with a time resolution of 1 second. In the spatial domain, based on the actual installation location of the sensors that collect different types of furnace operating parameters, the corresponding data is mapped to the three-dimensional spatial coordinate system of the boiler furnace, achieving spatial alignment of multi-source data. The second layer of operating condition feature correlation involves extracting furnace combustion operating condition features, flue gas flow field features, and pulverized coal conveying features based on spatiotemporally aligned data. Furnace combustion operating condition features include furnace temperature gradient and temperature distribution uniformity; flue gas flow field features include flue gas velocity and flue gas temperature distribution; and pulverized coal conveying features include pulverized coal concentration uniformity and pulverized coal flow rate. Subsequently, Pearson correlation coefficient analysis is used to establish the correlation between furnace combustion operating condition features, delayed flow features, and pulverized coal conveying features to characterize the boiler's multiphysics coupling characteristics. The third layer of comprehensive feature construction uses principal component analysis (PCA) to extract the combustion efficiency-pollutant generation coupled feature vector, and retains the top k principal components with a cumulative contribution rate greater than 95% as the comprehensive feature vector.

[0028] Preferably, the rate of change of the furnace combustion conditions is the average Euclidean distance of the comprehensive feature vector over the past 10 moments. When the rate of change is greater than a preset rate of change threshold, the fusion weight of the random forest model is set to 0.7 and the fusion weight of the long short-term memory network model is set to 0.3; when the rate of change is less than the preset rate of change threshold, the fusion weight of the random forest model is set to 0.3 and the fusion weight of the long short-term memory network model is set to 0.7.

[0029] Preferably, the furnace operating parameter prediction model aims to maximize combustion efficiency and minimize nitrogen oxide emissions, and constrains the range of control parameters, specifically including: The primary air volume range is [1.8, 2.2] kg (air) / kg (coal); The secondary air volume range is [0.8, 1.2] kg (air) / kg (coal); The burner swing angle range is [-30°, +30°]; Furnace temperature ≤1250℃.

[0030] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure made using the contents of the present invention specification and drawings, or directly or indirectly applied to other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A boiler combustion optimization method based on a dual-model collaborative architecture algorithm, characterized in that, Includes the following steps: Collect furnace operating parameters during the boiler combustion process; The furnace operating parameters are preprocessed, including data cleaning, noise reduction, normalization and hierarchical spatiotemporal fusion, and a comprehensive feature vector of the furnace operating parameters is obtained after preprocessing. A furnace operating parameter prediction model is constructed based on comprehensive feature vectors, and the predicted values ​​of furnace operating parameters are output. The prediction model is formed by weighted fusion of random forest model and long short-term memory network model, and the fusion weight is adjusted according to the rate of change of furnace combustion conditions. Adjust the boiler combustion control equipment according to the predicted values ​​of furnace operating parameters; Obtain actual operating data of adjusted fly ash carbon content, nitrogen oxide concentration and oxygen content, calculate the deviation between the actual operating data and the predicted values ​​of furnace operating parameters, and perform incremental local updates on the furnace operating parameter prediction model. Repeat the above steps to continuously optimize the furnace operating parameters.

2. The boiler combustion optimization method based on a dual-model collaborative architecture algorithm according to claim 1, characterized in that, The denoising employs an adaptive threshold wavelet threshold denoising algorithm, which specifically includes: Wavelet decomposition is performed on each data sequence in the furnace operating parameters. The threshold is adjusted according to the energy distribution of the wavelet coefficients of each decomposition layer using wavelet basis functions. An adaptive adjustment coefficient is introduced into the adjustment threshold. The wavelet coefficients processed by the adjusted threshold are reconstructed to obtain the denoised furnace operating parameters. The adaptive adjustment coefficients are determined based on the energy proportion of the wavelet coefficients of the corresponding sub-levels.

3. The boiler combustion optimization method based on a dual-model collaborative architecture algorithm according to claim 1, characterized in that, The hierarchical spatiotemporal fusion integrates different types of furnace operating parameters to construct multi-level features, specifically including: The first layer of spatiotemporal alignment performs unified processing of different types of furnace operating parameters in both the temporal and spatial domains, aligning the time axis and three-dimensional spatial coordinates of the furnace operating parameters. The second layer of operating condition feature association is based on the spatiotemporally aligned furnace operating parameters. It extracts furnace combustion operating condition features, flue gas flow field features, and pulverized coal conveying features, and establishes the coupling relationship between furnace combustion operating condition features, flue gas flow field features, and pulverized coal conveying features through correlation analysis. The third layer of comprehensive feature construction uses principal component analysis to extract the combustion efficiency-pollutant generation coupled feature vector, and retains principal components with a preset contribution rate greater than a preset percentage as comprehensive feature vectors.

4. The boiler combustion optimization method based on a dual-model collaborative architecture algorithm according to claim 1, characterized in that, The rate of change of the furnace combustion conditions is the average value of the Euclidean distance within a preset time window; When the rate of change of furnace combustion conditions is greater than the preset rate of change threshold, the weight ratio of the random forest model is greater than the preset weight ratio of the long short-term memory network model. When the rate of change of furnace combustion conditions is less than the preset rate of change threshold, the weight ratio of the random forest model is less than the preset weight ratio of the long short-term memory network model.

5. The boiler combustion optimization method based on a dual-model collaborative architecture algorithm according to claim 1, characterized in that, The furnace operating parameter prediction model uses a non-dominated sorting genetic algorithm to optimize the furnace operating parameters, introduces adaptive crossover probability and adaptive mutation probability, and uses a sorting strategy based on crowding distance to filter the non-dominated solution set. The adaptive crossover probability gradually decreases from the first crossover probability to the second crossover probability with the number of iterations. The adaptive mutation probability gradually increases from the first mutation probability to the second mutation probability with the number of iterations.

6. The boiler combustion optimization method based on a dual-model collaborative architecture algorithm according to claim 1, characterized in that, The furnace operating parameter prediction model aims to maximize combustion efficiency and minimize nitrogen oxide emissions. The optimization constraints include: Primary air volume constraints; Secondary air volume constraints; Constraints on the burner swing angle range; Upper limit constraint on furnace temperature.

7. The boiler combustion optimization method based on a dual-model collaborative architecture algorithm according to claim 1, characterized in that, When the deviation between the actual operating data and the predicted values ​​of the furnace operating parameters exceeds a preset deviation value, the random forest model only updates the tree node parameters of the decision tree related to the furnace operating parameters that exceed the preset deviation value. Long Short-Term Memory (LSTM) network models only update the weight parameters of the output layer.

8. A boiler combustion optimization system based on a dual-model collaborative architecture algorithm, characterized in that, The system includes: A multi-sensor monitoring module is used to collect furnace operating parameters during the boiler combustion process; The data fusion and preprocessing unit preprocesses the furnace operating parameters, including data cleaning, noise reduction, normalization, and hierarchical spatiotemporal fusion. The parameter prediction optimization unit constructs a furnace operation parameter prediction model through a dual-model collaborative architecture of random forest model and long short-term memory network model. It adjusts the fusion weights of random forest model and long short-term memory network model according to the rate of change of furnace combustion conditions and outputs the predicted values ​​of furnace operation parameters. The control and execution unit adjusts the boiler combustion control equipment according to the predicted values ​​of furnace operating parameters and receives and executes the optimal combustion parameter command. The feedback execution unit compares the furnace operating parameters with the predicted values. When the deviation exceeds the preset deviation value, it performs an incremental local update on the furnace operating parameter prediction model. The update only applies to the parts of fly ash content, nitrogen oxide concentration, and oxygen content that exceed the deviation.

9. A boiler combustion optimization system based on a dual-model collaborative architecture algorithm according to claim 8, characterized in that, The sensors in the multi-sensor monitoring module are deployed in a distributed, hierarchical manner, with sensors configured according to three major areas: the furnace combustion zone, the flue zone, and the pulverized coal conveying zone. These sensors include at least a furnace temperature sensor, an oxygen content sensor, a carbon monoxide concentration sensor, a pulverized coal concentration sensor, a feedwater temperature sensor, a steam pressure sensor, and a fly ash carbon content sensor. The furnace temperature sensors are arranged in layers along the height of the furnace. The oxygen content sensor is arranged at the furnace outlet and in each stage of the flue to monitor the degree of combustion completeness; The coal powder concentration sensor is placed at the inlet of the coal powder pipeline to monitor the mass of coal powder contained in a unit volume of airflow.

10. A boiler combustion optimization system based on a dual-model collaborative architecture algorithm according to claim 8, characterized in that, The system also includes a remote monitoring and diagnostic module, which builds a remote interactive interface through a cloud platform to display boiler operating parameters, optimization parameter curves and system working status in real time, perform fault diagnosis and early warning for the system, and issue intervention commands to the control execution unit.

Images