Multi-parameter regulated boiler combustion efficiency dynamic optimization method

Abstract

The application discloses a multi-parameter regulated boiler combustion efficiency dynamic optimization method. The specific operation steps are as follows: real-time acquisition of combustion parameters from a power plant system, pre-processing of the collected original data, and feature selection after data pre-processing; the feature data selected and pre-processed are taken as input according to time sequence to construct an IGWO-transformer prediction model; taking the degree of electricity coal consumption, nitrogen oxide emission concentration and fly ash carbon content as optimization targets, a NSGA-III algorithm is used for multi-objective optimization; aiming at the control requirements of the feature data, an independent PID controller is configured for each control object, the actual value collected is compared with the target set value, the control deviation is calculated as the input signal of the PID controller, and the controller is driven to perform dynamic adjustment. The application has the beneficial effects that the data processing and generalization ability are enhanced, accurate prediction and regulation are realized, and the combustion efficiency is improved; the economy and environmental protection are balanced, and dynamic optimization is realized.

Description

Technical Field

[0001] This invention relates to the field of boiler combustion technology, and in particular to a method for dynamic optimization of boiler combustion efficiency with multi-parameter adjustment. Background Technology

[0002] Boiler combustion is a complex physicochemical process involving multiple stages, including fuel-air mixing, ignition, combustion reaction, and heat transfer. It is characterized by strong nonlinearity, large time lag, and multivariate coupling. Numerous factors influence boiler combustion efficiency, including fuel quality, fuel supply pressure, air volume, initial air temperature, furnace temperature and pressure, pulverizing system performance, burner type, and operator skill.

[0003] Traditional modeling methods (such as Support Vector Machines (SVM) and Backpropagation Neural Networks) are susceptible to noise and missing data when dealing with such problems, resulting in large fluctuations in model performance. They also struggle to analyze the complex nonlinear relationships between multiple parameters and combustion indicators, making it impossible to precisely control the combustion process.

[0004] Algorithms such as NSGA-II are inefficient at handling high-dimensional problems of boiler combustion parameters, are prone to getting trapped in local optima, cannot achieve optimal balance of multiple objectives, and lack coordinated dynamic optimization of fly ash carbon content, coal consumption per kilowatt-hour, and nitrogen oxide emission concentration. Summary of the Invention

[0005] The present invention aims to overcome the aforementioned shortcomings in the prior art and provides a dynamic optimization method for boiler combustion efficiency by multi-parameter adjustment, which can achieve accurate prediction and control and improve combustion efficiency.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The dynamic optimization method for boiler combustion efficiency with multi-parameter adjustment, under the premise of meeting the dynamic load adjustment requirements, achieves the synergistic optimization of coal consumption per kilowatt-hour, NOx emission concentration, and fly ash carbon content through an integrated framework of "precise prediction - multi-objective optimization - closed-loop control". The specific operation steps are as follows:

[0008] (1) Data acquisition and preprocessing: Combustion parameters are collected in real time from the power plant's DCS / SIS system. The collected raw data is preprocessed. After data preprocessing, feature selection is performed, importance thresholds are set, and features that have a significant impact on the prediction target are selected. Finally, primary air flow, secondary air flow, coal feed rate, and initial air temperature are determined as core input features.

[0009] (2) Constructing the IGWO-transformer prediction model: The primary air flow, secondary air flow, coal feed rate and initial air temperature data after feature selection and preprocessing are used as input according to the time series to construct the IGWO-transformer prediction model. The output layer of the model outputs the predicted values ​​of fly ash carbon content, nitrogen oxide emission concentration and coal consumption per kilowatt-hour.

[0010] (3) Multi-objective optimization: The NSGA-III algorithm is used to optimize the coal consumption per kilowatt-hour, nitrogen oxide emission concentration and fly ash carbon content as optimization objectives. The primary air flow rate, secondary air flow rate, coal feed rate and initial air temperature data collected and processed in real time are input into the NSGA-III algorithm for optimization and adjustment to achieve the objectives of minimizing coal consumption per kilowatt-hour, minimizing nitrogen oxide emission concentration and minimizing fly ash carbon content.

[0011] (4) Dynamic control of combustion parameters: For the control requirements of primary air flow, secondary air flow, coal feed rate and initial air temperature, an independent PID controller is configured for each controlled object; each PID controller includes a proportional coefficient. , , Three key parameters: During the system initialization phase, based on the boiler equipment design parameters, historical operating experience, and commissioning data under similar operating conditions, initial parameters are set for each PID controller. The optimal solution determined by the maximum-minimum fuzzy method is used as the target setpoint for each controlled object. The actual operating data of primary air flow, secondary air flow, coal feed rate, and initial air temperature are collected in real time through the DCS / SIS system and used as feedback signals for the PID controller. The collected actual values ​​are compared with the target setpoints to calculate the control deviation e(t). This deviation is used as the input signal for the PID controller to drive the controller to make dynamic adjustments.

[0012] This invention selects primary air, secondary air, coal feed rate, and initial air temperature as the main input parameters, and uses the IGWO-transformer model to optimize the parameters, enhancing data processing and generalization capabilities to achieve accurate prediction and control, and improve combustion efficiency. It employs the NSGA-III algorithm combined with the IGWO-transformer prediction model to efficiently search for the optimal solution, balancing economic efficiency and environmental friendliness, and achieving dynamic optimization.

[0013] Preferably, in step (1), the specific operations of data preprocessing are as follows:

[0014] (11) Coarse value processing and linear interpolation: Remove abnormal data points caused by sensor failure or interference, and supplement the data based on time series and linear interpolation to ensure the continuity of the data series;

[0015] (12) Filtering and noise reduction: Use moving average filtering or wavelet transform to suppress high-frequency noise in the measurement signal;

[0016] (13) Data normalization: Using the min-max normalization method, the data is uniformly mapped to the interval [0, 1] to eliminate the dimensional differences between different parameters.

[0017]

[0018] Where X* represents the normalized data, X represents the original data, and X... max and X min These are the maximum and minimum values ​​in the original data, respectively.

[0019] (14) Feature selection: In order to improve model efficiency and prediction accuracy, feature selection is carried out after data preprocessing. The feature importance assessment method based on random forest is adopted to calculate the contribution of each feature to the prediction of NOx emission concentration, fly ash carbon content and coal consumption per kilowatt-hour. An importance threshold is set to screen out features that have a significant impact on the prediction target. Finally, primary air flow, secondary air flow, coal feed rate and initial air temperature are determined as core input features, and redundant and irrelevant features are removed.

[0020] As a preferred option, in step (14), the specific operation for selecting feature variables is as follows: the preprocessed data is divided into a training set and a test set, a random forest model containing multiple decision trees is constructed on the training set, and after the model is trained, the importance score of each feature is obtained; an importance threshold is set, and features with scores higher than the threshold are selected. These features are identified as core features that have a significant impact on the prediction target; finally, the primary air flow, secondary air flow, coal feed, and initial air temperature are determined as core input features, and redundant and irrelevant features are removed.

[0021] Preferably, in step (2), the specific operations for constructing the IGWO-transformer prediction model are as follows:

[0022] (21) Transformer model design: The data input layer receives multi-dimensional time series data and transforms it into a vector form suitable for model processing. Multiple attention heads capture the correlation between parameters in the input data from different angles, and explore the nonlinear dependence between primary air flow, secondary air flow, coal feed rate, initial air temperature parameters and fly ash carbon content, nitrogen oxide emission concentration and coal consumption per kilowatt-hour. After the data is processed by the multi-head attention mechanism, it enters the feedforward neural network for further feature extraction and integration. The feedforward neural network performs in-depth processing on the features extracted by the attention mechanism through the calculation of multiple layers of neurons, strengthens the role of key features, and weakens irrelevant or interfering information. Finally, the output layer of the model outputs the predicted values ​​of fly ash carbon content, nitrogen oxide emission concentration and coal consumption per kilowatt-hour, respectively, providing reliable data support for subsequent multi-objective optimization.

[0023] (22) Using the improved Grey Wolf Algorithm (IGWO) to search for the optimal combination: The Logistic chaotic mapping is used to generate an initial population with good ergodicity. The equation of the Logistic mapping is as follows:

[0024]

[0025] Where μ is the control parameter; X n The result is a random number within (0,1); to prevent the population from being unable to escape local extrema in the later stages of the algorithm, a non-linear decreasing formula is introduced, and the convergence factor 'a' decreases according to a cosine law:

[0026]

[0027] Where a is the convergence factor. Initial value of convergence factor The convergence factor is the final target value, and t is the current iteration number. Maximum number of iterations; the core mechanism of the Grey Wolf Algorithm is to guide population updates through the positions of α, β, and δ wolves, employing an adaptive weighted average strategy for optimization. Weights are dynamically allocated based on individual fitness, reinforcing the guiding role of high-quality solutions. Weights are dynamically calculated based on the fitness values ​​of α, β, and δ wolves; higher fitness results in greater weights.

[0028]

[0029] Among them, W j X is the adaptive weight for the j-th wolf; j Let X be the position vector of the j-th wolf at the current iteration time; k This represents the position vectors of the α, β, and δ wolves at the current iteration time when the traversal index k takes the values ​​α, β, and δ; k is the traversal index used to traverse the three key wolves α, β, and δ; f(X jLet be the fitness value of the j-th wolf. Wolves with better fitness have a greater impact on position updates; the position update formula is:

[0030]

[0031] Where wα, wβ, and wδ are the adaptive weights of α wolf, β wolf, and δ wolf, respectively; X(t+1) is the new position vector of the population after position update at time t+1; Xα, Xβ, and Xδ are the position vectors of α wolf, β wolf, and δ wolf at the tth iteration, respectively. + + =1.

[0032] As a preferred option, the specific operation of multi-objective optimization using the NSGA-III algorithm in step (3) is as follows:

[0033] (31) Constraints: The primary air flow rate must meet the following requirements. ,in and These are the minimum and maximum flow rates for safe operation of the primary air fan, respectively; the secondary air flow rate must meet... The performance of the secondary air fan and the load-bearing capacity of the pipeline are determined by the performance of the secondary air fan and the load-bearing capacity of the pipeline. and These are the minimum and maximum flow rates for safe operation of the secondary air fan; the coal feed rate must meet... ,in Minimum fuel supply to ensure stable combustion in the boiler. The maximum coal feed rate that the coal feeder and furnace can handle; the oxygen content in the furnace must be maintained within a reasonable range. To ensure complete combustion of the fuel, where y o2min y o2max These are the minimum and maximum oxygen content in the furnace, respectively; a certain proportional relationship must be maintained between the primary air flow rate and the secondary air flow rate. To ensure the stability of the combustion process, where r min r max These are the minimum and maximum values ​​of the ratio of primary airflow to secondary airflow, respectively; the initial air temperature must meet the following requirements. ,in and These are the minimum and maximum initial air temperatures, respectively; the net flue gas NOx concentration must meet national environmental emission standards, i.e. ,in The upper limit for nitrogen oxide emission concentration is set at 100%.

[0034] (32) Objective function: The objective function is to minimize NOx emission concentration, coal consumption per kilowatt-hour, and carbon content in fly ash. The Pareto front solution is obtained by applying the NSGA-III method. The following settings are provided:

[0035]

[0036]

[0037]

[0038]

[0039] (33) When determining the final optimal solution from the set of optimal solutions, a dynamic weight adjustment mechanism is introduced and combined with the maximum-minimum fuzzy method for comprehensive decision-making; according to the actual operating requirements of the boiler, dynamic weight adjustment rules are set: when the load rate is greater than 70%, considering the increased environmental pressure, the focus is on NOx emission optimization and the weight of NOx emission index in the comprehensive evaluation is increased; when the load rate is less than 40%, for the purpose of energy saving and consumption reduction, the focus is on coal consumption optimization and the weight of the coal consumption index per unit of electricity is increased, and the weight coefficients α and β are automatically updated with the load rate; in the load rate range of 40%-70%, the weight is smoothly transitioned based on historical data and operating experience.

[0040] As a preferred embodiment, in step (33), a fuzzy membership function is constructed based on dynamic weights; for the three optimization objectives of fly ash carbon content, nitrogen oxide emission concentration and coal consumption per kilowatt-hour, the membership degree of each objective is determined in different value ranges according to the actual working conditions and optimization requirements, in combination with the weights corresponding to the current load rate; the membership degree of each objective is comprehensively processed using the maximum-minimum fuzzy synthesis operator; for each solution in the optimal solution set, its membership degree value under the three optimization objectives is calculated in combination with the weights corresponding to the current load rate, and the fuzzy value of the solution under the comprehensive evaluation is determined by taking the minimum value; finally, the solution corresponding to the maximum value is selected from the fuzzy values ​​of all solutions and determined as the final optimal solution.

[0041] Preferably, in step (4), the dynamic adjustment of the PID controller is as follows:

[0042] Deviation formula:

[0043] Where r(t) is the target set value, and y(t) is the actual measured value;

[0044] The PID controller outputs μ(t) according to the control deviation e(t) using the following formula:

[0045]

[0046] Among them, the proportion element The control action is generated quickly based on the current deviation magnitude, causing the controlled parameter to change in the direction of reducing the deviation; integral element. Accumulate historical deviations to eliminate the system's steady-state error and ensure that the final controlled parameter can stabilize at the target setpoint; differential and integral... Adjusting the control input in advance based on the trend of deviation changes can suppress system overshoot and improve the dynamic response performance of the system.

[0047] As a preferred method, to prevent the PID controller output from being too large or too small, which could lead to abnormal equipment operation, the calculated control output μ(t) is limited. The limited PID control output signal is then converted into a corresponding control command and sent to the actuator in the boiler automatic control system. During system operation, the actual operating parameters and control effects of each controlled object are continuously monitored. If a problem is found with the controlled parameters, the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller are adjusted and optimized.

[0048] The beneficial effects of this invention are: selecting primary air, secondary air, coal feed rate, and initial air temperature as the main input parameters, using the IGWO-transformer model to optimize the parameters, enhancing data processing and generalization capabilities, achieving accurate prediction and control, and improving combustion efficiency; and employing the NSGA-III algorithm combined with the IGWO-transformer prediction model to efficiently search for the optimal solution, balancing economic efficiency and environmental protection, and achieving dynamic optimization. Attached Figure Description

[0049] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0050] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0051] like Figure 1 In the described embodiment, the multi-parameter-controlled dynamic optimization method for boiler combustion efficiency, while meeting the requirements of dynamic load adjustment, achieves optimal synergistic optimization of coal consumption per kilowatt-hour, NOx emission concentration, and fly ash carbon content through an integrated framework of "precise prediction - multi-objective optimization - closed-loop control," ensuring the best economic and environmental performance. The specific operation steps are as follows:

[0052] (1) Data Acquisition and Preprocessing: Combustion parameters are collected in real time from the power plant's DCS / SIS system, including but not limited to: actual load, total fuel quantity, coal feeder fuel quantity, primary air volume of the coal mill, secondary damper opening, total air volume, flue gas oxygen content, initial air temperature, NOx concentration at both inlets / outlets, and net flue gas NOx concentration. The collected raw data is preprocessed, and feature selection is performed after data preprocessing. An importance threshold is set to screen out features that have a significant impact on the prediction target. Finally, primary air flow, secondary air flow, coal feed rate, and initial air temperature are determined as the core input features. The specific operations for preprocessing the collected raw data are as follows:

[0053] (11) Coarse value processing and linear interpolation: Remove abnormal data points caused by sensor failure or interference, and supplement the data based on time series and linear interpolation to ensure the continuity of the data series.

[0054] (12) Filtering and noise reduction: Use moving average filtering or wavelet transform to suppress high-frequency noise in the measurement signal.

[0055] (13) Data normalization: Using the min-max normalization method, the data is uniformly mapped to the interval [0, 1] to eliminate the dimensional differences between different parameters.

[0056]

[0057] Where X* represents the normalized data, X represents the original data, and X... max and X min These are the maximum and minimum values ​​in the original data, respectively.

[0058] (14) Feature Variable Selection: To improve model efficiency and prediction accuracy, feature selection was performed after data preprocessing. A feature importance assessment method based on random forest was adopted to calculate the contribution of each feature to the prediction of NOx emission concentration, fly ash carbon content, and coal consumption per kilowatt-hour. An importance threshold was set to screen out features that have a significant impact on the prediction target. Finally, primary air flow, secondary air flow, coal feed rate, and initial air temperature were determined as core input features. Redundant and irrelevant features were removed to reduce data dimensionality and model computational complexity.

[0059] The specific steps for feature selection are as follows: The preprocessed data is divided into training and testing sets. A random forest model containing multiple decision trees is constructed on the training set. After model training, the importance score of each feature is obtained. An importance threshold is set, and features with scores higher than the threshold are selected as core features that significantly impact the prediction target. Finally, primary airflow, secondary airflow, coal feed rate, and initial air temperature are determined as core input features. Redundant and irrelevant features are removed, effectively reducing data dimensionality and lowering the computational complexity of subsequent models.

[0060] (2) Constructing the IGWO-transformer prediction model: Using the primary air flow rate, secondary air flow rate, coal feed rate, and initial air temperature data after feature selection and preprocessing, and taking them as time series data as input, the IGWO-transformer prediction model is constructed. The primary air flow rate, secondary air flow rate, coal feed rate, and initial air temperature are used as core inputs, supplemented by auxiliary parameters such as load and burnout air opening. The model's output layer outputs predicted values ​​for fly ash carbon content, nitrogen oxide emission concentration, and coal consumption per kilowatt-hour. The specific steps for constructing the IGWO-transformer prediction model are as follows:

[0061] (21) Transformer model design:

[0062] The primary airflow, secondary airflow, coal feed rate, and initial air temperature data, after feature selection and preprocessing, are used as time series inputs to construct the IGWO-transformer prediction model. The data input layer receives multi-dimensional time series data and transforms it into a vector form suitable for model processing. Multiple attention heads capture the correlations between parameters in the input data from different perspectives, uncovering the nonlinear dependencies between parameters such as primary airflow, secondary airflow, coal feed rate, and initial air temperature and fly ash carbon content, nitrogen oxide emission concentration, and coal consumption per kilowatt-hour. For example, the attention heads focus on how the airflow-to-coal feed rate ratio affects combustion completeness under different initial air temperatures, thus affecting fly ash carbon content; and the formation pattern of nitrogen oxides during combustion under specific initial air temperature and airflow combinations, thereby establishing a link with nitrogen oxide emission concentration. The data processed by the multi-head attention mechanism enters a feedforward neural network for further feature extraction and integration. The feedforward neural network, through the computation of multiple layers of neurons, deeply processes the features extracted by the attention mechanism, strengthening the role of key features and weakening irrelevant or interfering information. Finally, the model's output layer outputs predicted values ​​for fly ash carbon content, nitrogen oxide emission concentration, and coal consumption per kilowatt-hour, providing reliable data support for subsequent multi-objective optimization.

[0063] (22) The optimal combination is searched using the improved Grey Wolf algorithm IGWO:

[0064] Chaotic initialization: A Logistic chaotic mapping is used to generate an initial population with good ergodicity, covering the solution space of parameters such as oxygen content and fly ash carbon content, thus solving the problem of insufficient population diversity caused by random initialization. The equation of the Logistic mapping is as follows:

[0065]

[0066] Where the control parameter μ is set to 4, the system is in a completely chaotic state; Xn Use a random number within (0,1) (avoid using fixed points such as 0.25, 0.5, 0.75, etc.).

[0067] Nonlinear convergence factor: To prevent the population from being unable to escape local optima in the later stages of the algorithm, a nonlinear decreasing formula is introduced, where the convergence factor 'a' decreases according to a cosine law.

[0068]

[0069] Where a is the convergence factor. Initial value of convergence factor The convergence factor is the final target value, and t is the current iteration number. Maximum number of iterations.

[0070] Adaptive Position Strategy: The core mechanism of the Grey Wolf Algorithm (GWO) is to guide population updates through the positions of the α, β, and δ wolves (the three currently optimal solutions). However, the original GWO position update strategy is prone to insufficient population diversity and premature convergence. To overcome these shortcomings, an adaptive weighted average strategy is adopted for optimization, dynamically allocating weights based on individual fitness to strengthen the guiding role of high-quality solutions. The weights are dynamically calculated based on the fitness values ​​of the α, β, and δ wolves, with higher fitness resulting in greater weights.

[0071]

[0072] Among them, W j X is the adaptive weight for the j-th wolf; j Let X be the position vector of the j-th wolf at the current iteration time; k This represents the position vector of the wolf (α wolf, β wolf, or δ wolf) at the current iteration time when the traversal index k takes the values ​​α, β, or δ; k is the traversal index used to traverse the three key wolves (α wolf, β wolf, δ wolf); f(X) j Let be the fitness value (objective function value) of the j-th wolf. Wolves with better fitness have a greater impact on position updates. The position update formula is:

[0073]

[0074] Where wα, wβ, and wδ are the adaptive weights of α wolf, β wolf, and δ wolf, respectively; X(t+1) is the new position vector of the population after position update at time t+1; Xα, Xβ, and Xδ are the position vectors of α wolf, β wolf, and δ wolf at the tth iteration, respectively. + + =1.

[0075] (3) Multi-objective optimization: The NSGA-III algorithm is used to optimize the coal consumption per kilowatt-hour, nitrogen oxide emission concentration and fly ash carbon content as optimization objectives. The primary air flow rate, secondary air flow rate, coal feed rate and initial air temperature data collected and processed in real time are input into the NSGA-III algorithm for optimization and adjustment to achieve the objectives of minimizing coal consumption per kilowatt-hour, minimizing nitrogen oxide emission concentration and minimizing fly ash carbon content.

[0076] The NSGA-III algorithm is used for multi-objective optimization, with adjustable key combustion parameters as decision variables, including primary air flow rate, secondary air flow rate, coal feed rate, and initial air temperature. These decision variables directly affect the boiler's combustion process. The NSGA-III algorithm is used to optimize and adjust these parameters to achieve the goals of minimizing coal consumption per kilowatt-hour, minimizing nitrogen oxide emission concentration, and minimizing fly ash carbon content. The specific operation is as follows:

[0077] (31) Constraints:

[0078] Equipment operating limitations: Primary air flow rate must meet the following requirements. ,in and These are the minimum and maximum flow rates for safe operation of the primary air fan, respectively; the secondary air flow rate must meet... The performance of the secondary air fan and the load-bearing capacity of the pipeline are determined by the performance of the secondary air fan and the load-bearing capacity of the pipeline. and These are the minimum and maximum flow rates for safe operation of the secondary air fan; the coal feed rate must meet... ,in Minimum fuel supply to ensure stable combustion in the boiler. This is the maximum coal feed rate that the coal feeder and furnace can handle.

[0079] Process requirements: The oxygen content in the furnace must be maintained within a reasonable range. To ensure complete combustion of the fuel, where y o2min y o2max These are the minimum and maximum oxygen content in the furnace, respectively; a certain proportional relationship must be maintained between the primary air flow rate and the secondary air flow rate. To ensure the stability of the combustion process, where r min r max These are the minimum and maximum values ​​of the ratio of primary airflow to secondary airflow, respectively; the initial air temperature must meet the following requirements. ,in and These are the minimum and maximum initial wind temperatures, respectively.

[0080] Environmental standards require that the NOx concentration in the net flue gas must meet national environmental emission standards. ,in This refers to the upper limit of the prescribed nitrogen oxide emission concentration.

[0081] (32) Objective function: The objective function is to minimize NOx emission concentration, coal consumption per kilowatt-hour and carbon content in fly ash. The Pareto front solution is obtained by applying the NSGA-III method.

[0082] set up:

[0083]

[0084]

[0085]

[0086]

[0087] Among them, f Nox f ash f coal The objective functions are NOx emission concentration, coal consumption per kilowatt-hour, and fly ash carbon content, respectively. The four parameters in the objective function are as described above: x1 primary air flow rate, x2 secondary air flow rate, x3 coal feed rate, and x4 initial air temperature. F(x) is the Pareto front solution (optimal solution) obtained using the NSGA-III method. The optimal solution set is obtained through the NSGA-III method. The NSGA-III process generates a set of solutions, each corresponding to a set of combustion parameters (such as primary air flow rate, secondary air flow rate, coal feed rate, initial air temperature, etc.), and finds the optimal equilibrium point among the three parameters through non-dominated sorting and reference point projection. The NSGA-III method process is as follows:

[0088] (1) Initialize the population;

[0089] (2) Fitness assessment;

[0090] (3) Non-dominated sorting;

[0091] (4) Reference point setting and association;

[0092] (5) Genetic operations (selection, crossover, mutation);

[0093] (6) The iteration converges to obtain the Pareto front solution (i.e. the optimal solution).

[0094] Finally, the optimal solution that best suits the current working conditions is selected through the following steps (33) using fuzzy decision-making or dynamic weighting (e.g., focusing on NOx emission reduction during high load and coal consumption optimization during low load).

[0095] (33) When determining the final optimal solution from the set of optimal solutions, a dynamic weight adjustment mechanism is introduced, and a comprehensive decision is made in combination with the maximum-minimum fuzzy method. According to the actual operating requirements of the boiler, the dynamic weight adjustment rules are set: when the load rate is greater than 70%, considering the increased environmental pressure, the focus is on NOx emission optimization, and the weight of NOx emission index in the comprehensive evaluation is increased; when the load rate is less than 40%, for the purpose of energy saving and consumption reduction, the focus is on coal consumption optimization, and the weight of the coal consumption per unit of electricity index is increased, and the weight coefficients α and β are automatically updated with the load rate; in the load rate range of 40%-70%, the weight is smoothly transitioned based on historical data and operating experience.

[0096] Based on dynamic weights, a fuzzy membership function is constructed. For the three optimization objectives—fly ash carbon content, nitrogen oxide emission concentration, and coal consumption per kilowatt-hour—the membership degree of each objective is determined within different value ranges, taking into account the weights corresponding to the current load rate and the actual operating conditions and optimization requirements. For example, when the load rate is greater than 70%, for nitrogen oxide emission concentration, a higher membership degree is assigned to lower emission concentrations, and the weight percentage is increased; for coal consumption per kilowatt-hour, the membership degree construction is also appropriately adjusted based on the weights.

[0097] The membership degrees of each objective are comprehensively processed using a maximum-minimum fuzzy synthesis operator. For each solution in the optimal solution set, its membership value under the three optimization objectives is calculated by combining the weight corresponding to the current load rate. The fuzzy value of the solution under the comprehensive evaluation is determined by taking the minimum value. Finally, the solution with the maximum value among all solutions is selected as the final optimal solution. This optimal solution not only comprehensively considers the fuzziness of multiple optimization objectives, but also makes the adjustment of combustion parameters more in line with the actual complex needs under different load conditions through dynamic weight adjustment.

[0098] (4) Dynamic control of combustion parameters: For the control requirements of primary air flow, secondary air flow, coal feed rate, and initial air temperature, an independent PID controller is configured for each controlled object. Each PID controller includes a proportional coefficient. , , Three key parameters. During the system initialization phase, initial parameters are set for each PID controller based on the boiler equipment's design parameters, historical operating experience, and commissioning data under similar operating conditions.

[0099] The optimal solution determined by the maximum-minimum fuzzy method is used as the target setpoint for each controlled object. Real-time operating data of primary airflow, secondary airflow, coal feed rate, and initial air temperature are collected in real time through the DCS / SIS system and used as feedback signals for the PID controller. The data acquisition frequency is set to once per minute to ensure timely capture of parameter changes. The collected actual values ​​are compared with the target setpoint to calculate the control deviation e(t). This deviation serves as the input signal for the PID controller, driving the controller to make dynamic adjustments.

[0100] Deviation formula:

[0101] Where r(t) is the target set value and y(t) is the actual measured value.

[0102] The PID controller outputs μ(t) according to the control deviation e(t) using the following formula:

[0103]

[0104] Among them, the control outputs are the primary fan frequency, secondary fan frequency, coal feeder speed, or initial air temperature; the controller has three parameters. (proportion coefficient) (Integral coefficient) and (Differential coefficients) It is an integral variable; a proportional element. The control action is generated quickly based on the current deviation magnitude, causing the controlled parameter to change in the direction of reducing the deviation; integral element. Accumulate historical deviations to eliminate the system's steady-state error and ensure that the final controlled parameter can stabilize at the target setpoint; differential and integral... Adjusting the control input in advance based on the trend of deviation changes can suppress system overshoot and improve the dynamic response performance of the system.

[0105] To prevent the PID controller output from being too large or too small, which could lead to abnormal equipment operation, the calculated control output μ(t) is limited. For example, the output range of the primary fan speed control signal is limited to 20% - 100% of the fan's rated speed; the coal feed rate control signal is limited to the range from the minimum to the maximum coal feed rate of the coal feeder.

[0106] The PID control output signal, after amplitude limiting, is converted into corresponding control commands and sent to the actuators in the boiler automatic control system. For the blower, the control commands adjust the blower speed via the frequency converter, thereby changing the primary and secondary air volumes. For the coal feed rate, the control commands control the motor speed of the coal feeder to regulate the coal feed rate. During system operation, the actual operating parameters and control effects of each controlled object are continuously monitored. If problems such as large fluctuations in controlled parameters, excessively long adjustment times, or significant steady-state errors are found, the proportional, integral, and derivative coefficients of the PID controller are adjusted and optimized.

[0107] The key technical design of this invention lies in:

[0108] Multivariate feature selection: Considering the multivariate and strongly coupled characteristics of the boiler combustion system, primary air, secondary air, coal feed rate, and initial air temperature are selected as core input parameters to reduce model complexity and improve prediction accuracy.

[0109] The IGWO-Transformer hybrid model combines the improved Grey Wolf Optimization Algorithm (IGWO) with the Transformer model to address the nonlinear and time-varying characteristics of combustion scenarios, enabling high-precision prediction of fly ash carbon content, NOx emission concentration, and coal consumption per kilowatt-hour.

[0110] Multi-objective collaborative optimization: The NSGA-III algorithm is used to collaboratively optimize the coal consumption per kilowatt-hour, nitrogen oxide emission concentration and fly ash carbon content, so as to balance the economy and environmental protection of boiler operation.

[0111] Dynamic control mechanism: By dynamically adjusting combustion parameters based on real-time data feedback and model prediction results, the boiler combustion system can achieve real-time optimized control.

[0112] In summary, the improved Grey Wolf optimization algorithm effectively avoids local optima by dynamically adjusting the search strategy, thus finding a better boiler operating point. By uniformly quantifying economic and environmental indicators and employing a dynamic weighting strategy, the optimization objective can intelligently adapt to different operating conditions, maximizing overall benefits.

[0113] To address the problems mentioned in the background art, the present invention provides the following solutions:

[0114] The problem of strong coupling among multiple variables is that traditional PID control cannot effectively handle the coupling relationship between multiple parameters such as primary air, secondary air, coal feed rate, and initial air temperature, leading to control lag or error accumulation. This invention extracts key parameters through feature selection and combines them with the IGWO-Transformer model to capture the complex relationships between multiple variables, thereby improving prediction accuracy.

[0115] Due to its nonlinear and time-varying characteristics, single models (such as LSTM) have poor adaptability to nonlinear changes in combustion scenarios, resulting in unstable prediction results. This invention utilizes the IGWO algorithm to optimize the Transformer model parameters, enhancing the model's adaptability to time-varying characteristics and improving prediction robustness.

[0116] The problem of balancing environmental protection and economic efficiency is addressed by single-objective optimization algorithms that only optimize either coal consumption per kilowatt-hour or emission concentration, failing to consider multiple factors simultaneously, leading to substandard operating costs or environmental indicators. This invention, the NSGA-III algorithm, simultaneously optimizes coal consumption per kilowatt-hour, nitrogen oxide emission concentration, and fly ash carbon content, generating a Pareto optimal solution set and achieving multi-objective collaborative optimization.

Claims

1. A dynamic optimization method for boiler combustion efficiency based on multi-parameter adjustment, characterized by: Under the premise of meeting the dynamic load adjustment requirements, the integrated framework of "precise prediction - multi-objective optimization - closed-loop control" is used to achieve the optimal synergy between coal consumption per kilowatt-hour, NOx emission concentration and fly ash carbon content. The specific operation steps are as follows: (1) Data acquisition and preprocessing: Combustion parameters are collected in real time from the power plant's DCS / SIS system. The collected raw data is preprocessed. After data preprocessing, feature selection is performed, importance thresholds are set, and features that have a significant impact on the prediction target are selected. Finally, primary air flow, secondary air flow, coal feed rate, and initial air temperature are determined as core input features. (2) Constructing the IGWO-transformer prediction model: The primary air flow, secondary air flow, coal feed rate and initial air temperature data after feature selection and preprocessing are used as input according to the time series to construct the IGWO-transformer prediction model. The output layer of the model outputs the predicted values ​​of fly ash carbon content, nitrogen oxide emission concentration and coal consumption per kilowatt-hour. (3) Multi-objective optimization: The NSGA-III algorithm is used to optimize the coal consumption per kilowatt-hour, nitrogen oxide emission concentration and fly ash carbon content as optimization objectives. The primary air flow rate, secondary air flow rate, coal feed rate and initial air temperature data collected and processed in real time are input into the NSGA-III algorithm for optimization and adjustment to achieve the objectives of minimizing coal consumption per kilowatt-hour, minimizing nitrogen oxide emission concentration and minimizing fly ash carbon content. (4) Dynamic control of combustion parameters: For the control requirements of primary air flow, secondary air flow, coal feed rate and initial air temperature, an independent PID controller is configured for each controlled object; each PID controller includes a proportional coefficient. , Three key parameters; During the system initialization phase, initial parameters are set for each PID controller based on the boiler equipment design parameters, historical operating experience, and commissioning data under similar operating conditions. The optimal solution determined by the maximum-minimum fuzzy method is used as the target setpoint for each controlled object. The actual operating data of primary air flow, secondary air flow, coal feed rate, and initial air temperature are collected in real time through the DCS / SIS system and used as feedback signals for the PID controller. The collected actual values ​​are compared with the target setpoints to calculate the control deviation e(t). This deviation is used as the input signal for the PID controller to drive the controller to make dynamic adjustments.

2. The method for dynamic optimization of boiler combustion efficiency with multi-parameter adjustment according to claim 1, characterized in that, In step (1), the specific operations for data preprocessing are as follows: (11) Coarse value processing and linear interpolation: Remove abnormal data points caused by sensor failure or interference, and supplement the data based on time series and linear interpolation to ensure the continuity of the data series; (12) Filtering and noise reduction: Use moving average filtering or wavelet transform to suppress high-frequency noise in the measurement signal; (13) Data normalization: Using the min-max normalization method, the data is uniformly mapped to the interval [0, 1] to eliminate the dimensional differences between different parameters. Where X* represents the normalized data, X represents the original data, and X... max and X min These are the maximum and minimum values ​​in the original data, respectively. (14) Feature selection: In order to improve model efficiency and prediction accuracy, feature selection is carried out after data preprocessing. The feature importance assessment method based on random forest is adopted to calculate the contribution of each feature to the prediction of NOx emission concentration, fly ash carbon content and coal consumption per kilowatt-hour. An importance threshold is set to screen out features that have a significant impact on the prediction target. Finally, primary air flow, secondary air flow, coal feed rate and initial air temperature are determined as core input features, and redundant and irrelevant features are removed.

3. The method for dynamic optimization of boiler combustion efficiency with multi-parameter adjustment according to claim 2, characterized in that, In step (14), the specific operation for selecting feature variables is as follows: the preprocessed data is divided into training set and test set, a random forest model containing multiple decision trees is constructed on the training set, and after the model is trained, the importance score of each feature is obtained. An importance threshold is set, and features with scores higher than the threshold are selected. These features are identified as core features that have a significant impact on the prediction target. Finally, primary air flow, secondary air flow, coal feed rate, and initial air temperature are determined as core input features, and redundant and irrelevant features are removed.

4. The method for dynamic optimization of boiler combustion efficiency with multi-parameter adjustment according to claim 1, characterized in that, In step (2), the specific operations for constructing the IGWO-transformer prediction model are as follows: (21) Transformer model design: The data input layer receives multi-dimensional time series data and transforms it into a vector form suitable for model processing. Multiple attention heads capture the correlation between parameters in the input data from different angles, and explore the nonlinear dependence between primary air flow, secondary air flow, coal feed rate, initial air temperature parameters and fly ash carbon content, nitrogen oxide emission concentration and coal consumption per kilowatt-hour. After the data is processed by the multi-head attention mechanism, it enters the feedforward neural network for further feature extraction and integration. The feedforward neural network performs in-depth processing on the features extracted by the attention mechanism through the calculation of multiple layers of neurons, strengthens the role of key features, and weakens irrelevant or interfering information. Finally, the output layer of the model outputs the predicted values ​​of fly ash carbon content, nitrogen oxide emission concentration and coal consumption per kilowatt-hour, respectively, providing reliable data support for subsequent multi-objective optimization. (22) Using the improved Grey Wolf Algorithm (IGWO) to search for the optimal combination: The Logistic chaotic mapping is used to generate an initial population with good ergodicity. The equation of the Logistic mapping is as follows: Where μ is the control parameter; X n The result is a random number within (0,1); to prevent the population from being unable to escape local extrema in the later stages of the algorithm, a non-linear decreasing formula is introduced, and the convergence factor 'a' decreases according to a cosine law: Where a is the convergence factor. Initial value of convergence factor The convergence factor is the final target value, and t is the current iteration number. Maximum number of iterations; the core mechanism of the Grey Wolf Algorithm is to guide population updates through the positions of α, β, and δ wolves, employing an adaptive weighted average strategy for optimization. Weights are dynamically allocated based on individual fitness, reinforcing the guiding role of high-quality solutions. Weights are dynamically calculated based on the fitness values ​​of α, β, and δ wolves; higher fitness results in greater weights. Among them, W j X is the adaptive weight for the j-th wolf; j Let X be the position vector of the j-th wolf at the current iteration time; k This represents the position vectors of the α, β, and δ wolves at the current iteration time when the traversal index k takes the values ​​α, β, and δ; k is the traversal index used to traverse the three key wolves α, β, and δ; f(X j Let be the fitness value of the j-th wolf. Wolves with better fitness have a greater impact on position updates; the position update formula is: Where wα, wβ, and wδ are the adaptive weights of α wolf, β wolf, and δ wolf, respectively; X(t+1) is the new position vector of the population after position update at time t+1; Xα, Xβ, and Xδ are the position vectors of α wolf, β wolf, and δ wolf at the tth iteration, respectively. + + =1.

5. The method for dynamic optimization of boiler combustion efficiency with multi-parameter adjustment according to claim 1, characterized in that, in In step (3), the specific operations for multi-objective optimization using the NSGA-III algorithm are as follows: (31) Constraints: The primary air flow rate must meet the following requirements. ,in and These are the minimum and maximum flow rates for safe operation of the primary air fan, respectively; the secondary air flow rate must meet... The performance of the secondary air fan and the load-bearing capacity of the pipeline are determined by the performance of the secondary air fan and the load-bearing capacity of the pipeline. and These are the minimum and maximum flow rates for safe operation of the secondary air fan; the coal feed rate must meet... ,in Minimum fuel supply to ensure stable combustion in the boiler. The maximum coal feed rate that the coal feeder and furnace can handle; the oxygen content in the furnace must be maintained within a reasonable range. To ensure complete combustion of fuel, where y o2min y o2max These are the minimum and maximum oxygen content in the furnace, respectively; a certain proportional relationship must be maintained between the primary air flow rate and the secondary air flow rate. To ensure the stability of the combustion process, where r min r max These are the minimum and maximum values ​​of the ratio of primary airflow to secondary airflow, respectively; the initial air temperature must meet the following requirements. ,in and These are the minimum and maximum initial air temperatures, respectively; the net flue gas NOx concentration must meet national environmental emission standards, i.e. ,in The upper limit for nitrogen oxide emission concentration is set at the specified level. (32) Objective function: The objective function is to minimize NOx emission concentration, coal consumption per kilowatt-hour and carbon content in fly ash. The Pareto front solution is obtained by applying the NSGA-III method. set up: (33) When determining the final optimal solution from the set of optimal solutions, a dynamic weight adjustment mechanism is introduced and combined with the maximum-minimum fuzzy method for comprehensive decision-making; according to the actual operating requirements of the boiler, dynamic weight adjustment rules are set: when the load rate is greater than 70%, considering the increased environmental pressure, the focus is on NOx emission optimization and the weight of NOx emission index in the comprehensive evaluation is increased; when the load rate is less than 40%, for the purpose of energy saving and consumption reduction, the focus is on coal consumption optimization and the weight of the coal consumption index per unit of electricity is increased, and the weight coefficients α and β are automatically updated with the load rate; in the load rate range of 40%-70%, the weight is smoothly transitioned based on historical data and operating experience.

6. The method for dynamic optimization of boiler combustion efficiency with multi-parameter adjustment according to claim 5, characterized in that, In step (33), a fuzzy membership function is constructed based on dynamic weights; for the three optimization objectives of fly ash carbon content, nitrogen oxide emission concentration and coal consumption per kilowatt-hour, the membership degree of each objective is determined in different value ranges based on the weights corresponding to the current load rate and the actual working conditions and optimization requirements. The membership degree of each objective is comprehensively processed using the maximum-minimum fuzzy synthesis operator. For each solution in the optimal solution set, its membership degree value under the three optimization objectives is calculated by combining the weight corresponding to the current load rate. The fuzzy value of the solution under the comprehensive evaluation is determined by taking the minimum value. Finally, the solution corresponding to the maximum value among all solutions is selected as the final optimal solution.

7. The method for dynamic optimization of boiler combustion efficiency with multi-parameter adjustment according to claim 1, characterized in that, in In step (4), the dynamic adjustment of the PID controller is as follows: Deviation formula: Where r(t) is the target set value, and y(t) is the actual measured value; The PID controller outputs μ(t) according to the control deviation e(t) using the following formula: Among them, the proportion element The control action is generated quickly based on the current deviation magnitude, causing the controlled parameter to change in the direction of reducing the deviation; integral element. Accumulate historical deviations to eliminate the system's steady-state error and ensure that the final controlled parameter can stabilize at the target setpoint; differential and integral... Adjusting the control input in advance based on the trend of deviation changes can suppress system overshoot and improve the dynamic response performance of the system.

8. The method for dynamic optimization of boiler combustion efficiency with multi-parameter adjustment according to claim 7, characterized in that, To prevent the PID controller output from being too large or too small, which could lead to abnormal equipment operation, the calculated control output μ(t) is limited. The limited PID control output signal is then converted into a corresponding control command and sent to the actuator in the boiler automatic control system. During system operation, the actual operating parameters and control effects of each controlled object are continuously monitored. If a problem is found with the controlled parameters, the proportional coefficient, integral coefficient, and derivative coefficient of the PID controller are adjusted and optimized.

Images