A method for synergistically optimizing secondary air volume in coupled combustion of coal and biomass

Abstract

This invention provides a method for synergistic optimization of secondary air volume in coal-biomass coupled combustion, belonging to the field of intelligent power generation technology. This method uses machine vision and online spectral monitoring of the boiler furnace to acquire flame images and emission spectra, analyzing the flame center position, furnace outlet temperature, and potassium alkali metal signal value. The monitoring data is fused with unit operating parameters to construct a deep learning database. Based on this database, a multi-objective model of boiler coupled combustion is built, establishing a mapping relationship between unit operating parameters, combustion state, and flue gas emission parameters. A multi-objective weighted sum function is constructed, and the optimal air distribution parameter combination is obtained through iterative optimization using an optimization algorithm. Finally, the control unit interacts with the distributed control system to generate control commands to complete the closed-loop regulation of secondary air volume. This invention can effectively balance boiler efficiency and NOx levels. X The contradiction between emissions and the release of gaseous alkali metal potassium needs to be addressed to improve boiler efficiency and reduce NO in flue gas. X This addresses emissions and mitigates the problem of ash and slag buildup on heated surfaces.

Description

Technical Field

[0001] This invention relates to the field of intelligent power generation technology, and in particular to a method for synergistic optimization of secondary air volume in coal and biomass coupled combustion. Background Technology

[0002] Currently, there is a large stock of coal-fired power units. To adapt to the development of new power systems, the low-carbon transformation of these units is imperative, and implementing coal-fired power units coupled with biomass power generation is one important approach. However, the current biomass is mainly composed of straw, which is of lower quality and can affect boiler efficiency during coupled combustion. Furthermore, biomass contains higher levels of alkali metals and chlorine impurities, easily causing ash slagging on the boiler's heating surface. In addition, existing large-scale coal-fired power units have already achieved ultra-low emissions, and the impact of coupled biomass combustion on boiler pollutant emissions also requires addressing. Therefore, it is urgent to develop new methods and technologies to solve the aforementioned key problems in the application of coal-biomass coupled combustion technology.

[0003] Boiler combustion optimization is a crucial technical means to achieve safe, efficient, and clean combustion in boilers. However, a boiler is a multi-input, multi-output system, with numerous and complex influencing factors affecting its operation. Traditional optimization control methods struggle to reflect the relationship between inputs and outputs, and boilers exhibit significant delays and inertia, often rendering traditional optimization control methods lagging and thus limiting their effectiveness. With the development of artificial intelligence, data-driven boiler combustion optimization technology has shown promising progress. This technology utilizes machine learning based on long-term operational data to identify combustion operating parameters that improve performance indicators, thereby optimizing the combustion process. Essentially, it is a form of predictive control, allowing for advance adjustments to ensure the boiler is always in an optimal state, thus addressing the issues of significant delays and inertia in optimization control.

[0004] However, combustion optimization of coal-fired boilers is a key process for reducing pollutants (especially NO). X This involves a multi-objective optimization problem of improving boiler efficiency while reducing emissions. The research and development of key technologies for coupled coal and biomass combustion aims to ensure safe, efficient, and clean boiler operation across a wide load range, simultaneously considering combustion efficiency, nitrogen oxide emissions, and slagging corrosion overheating. The main challenge lies in the fact that these three indicators cannot be simultaneously optimized and are contradictory and coupled with each other; traditional optimization control methods cannot achieve multi-objective optimization.

[0005] Therefore, it is necessary to provide a method for the coordinated optimization of secondary air volume in coal and biomass coupled combustion based on deep learning of online monitoring and unit operation data to solve the above problems. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method for synergistic optimization of secondary air volume in coal-biomass coupled combustion. By adding in-furnace combustion machine vision and spectral detection methods, supplementary data such as flame center position, furnace outlet temperature, and alkali metal potassium signal are obtained. Through multi-objective synergistic optimization, boiler efficiency and NO2 can be effectively balanced. X The contradiction between emissions and the release of gaseous alkali metal potassium, while simultaneously improving boiler efficiency and reducing NO in flue gas. X Discharge and mitigation of ash and slag formation on heated surfaces.

[0007] To achieve the above objectives, this invention provides a method for synergistic optimization of secondary air volume in coal and biomass coupled combustion, comprising the following steps: The boiler furnace is monitored online using machine vision and spectral analysis. Visual images and emission spectra of the combustion flames inside the furnace are collected. The flame center position, furnace outlet temperature and potassium alkali metal signal value are determined through image processing and spectral analysis. The monitoring data is fused with the unit operating parameters to build a deep learning database. The deep learning database is input into the deep learning algorithm for model training, a boiler coupled combustion multi-objective model is constructed, and the mapping relationship between unit operating parameters, combustion state parameters, and flue gas emission parameters is established. Based on the boiler coupled combustion multi-objective model, multi-objective collaborative optimization is carried out, a multi-objective weighted sum function is constructed, and intelligent optimization algorithm is used to iteratively optimize the optimal air distribution parameter combination that minimizes the multi-objective weighted sum function. The optimal air distribution parameter combination is input into the control unit for closed-loop control. The control unit interacts with the distributed control system to compare the optimal total secondary air volume with the current secondary air volume and generate control commands to be sent back to the distributed control system, thus completing the closed-loop control of the secondary air volume of coal and biomass coupled combustion.

[0008] Preferably, the boiler furnace is monitored online using machine vision and spectral analysis to acquire visual images and emission spectra of the combustion flames inside the furnace, including: First, at least two machine vision measurement points are set up in the furnace area above the boiler burnout air and below the flame deflector angle so that the vertical field of view of the machine vision equipment completely covers the furnace outlet area. Then, at least two spectral measurement points are set up in the furnace outlet area to collect visual images of the flame using machine vision measurement points, and at the same time, the flame emission spectrum is collected using spectral measurement points.

[0009] Preferably, the flame center position, furnace outlet temperature, and potassium alkali metal signal value are determined through image processing and spectral analysis, including: First, the flame visual images at at least two height positions are intersected by the region of maximum brightness to determine the flame center position in the furnace along the height direction. Then, the emission spectrum data of potassium alkali metal at at least two furnace outlet measuring points are used to extract the emission line intensity of potassium alkali metal and calculate the average value to obtain the potassium alkali metal signal value at the furnace outlet. Next, the temperature at each spectral measurement point is calculated using the multi-wavelength method, and the average value is obtained to obtain the furnace outlet temperature.

[0010] Preferably, the monitoring data is fused with the unit operating parameters to construct a deep learning database, including: First, the flame center position, furnace outlet temperature, and potassium alkali metal signal value are transmitted to the distributed control system. Then, the flame center position offset and furnace outlet temperature offset are calculated by combining the boiler design parameters. Then, the above offset, potassium alkali metal signal value, CO content, and NO... X Concentration is fused with data from unit load, main steam flow, feedwater flow, feedwater temperature, total coal volume in the coal mill, coal mill temperature, total secondary air volume, secondary air damper, and burnout air damper. The merged dataset is then divided into training and validation sets according to a set ratio, thus forming a deep learning database that can be used for model training.

[0011] Preferably, the deep learning database is input into a deep learning algorithm for model training to construct a boiler coupled combustion multi-objective model, including: First, a multi-layer perceptron neural network is used to build a multi-objective model of boiler coupled combustion. Then, the unit load, main steam flow, feedwater flow, feedwater temperature, total coal flow in the coal mill, coal mill temperature, total secondary air flow, secondary air damper, and burnout air damper are set as the model input parameters; Next, the flame center position offset, furnace outlet temperature offset, alkali metal potassium signal value, CO content, and NO were measured. X The concentration is set as a model output parameter.

[0012] Preferably, establishing a mapping relationship between unit operating parameters and combustion state parameters and flue gas emission parameters includes: First, the training set data is input into the multilayer perceptron neural network. After nonlinear calculation by the hidden layer, the output result is reconstructed to obtain the furnace temperature field. Then, the weights of the hidden layer neurons are iteratively updated using the backpropagation algorithm, thereby completing the training of the multilayer perceptron neural network; The validation set data is then input into the trained model, and the model output value is compared with the actual operating conditions on site, thereby solidifying the mapping relationship between the unit's operating parameters and output parameters.

[0013] Preferably, based on the boiler coupled combustion multi-objective model, multi-objective collaborative optimization is carried out to construct a multi-objective weighted sum function, including: First, select the flame center position offset, furnace outlet temperature offset, alkali metal potassium signal value, CO content, and NO. X Concentration as a multi-objective optimization variable; Then, weight coefficients are assigned to each optimization variable. , , , , And make the sum of the weight coefficients satisfy + + + + =1; Then, the multi-objective weighted sum function is constructed according to the formula, which is: ; In the formula, This represents the offset of the flame center position. This represents the temperature offset at the furnace outlet. The signal value is for potassium alkali metal, CO is the CO content at the furnace outlet, and NO is... X NO at furnace outlet X concentration.

[0014] Preferably, the optimal combination of wind distribution parameters that minimizes the multi-objective weighted sum function is obtained through iterative optimization using an intelligent optimization algorithm, including: First, a multi-dimensional particle swarm optimization algorithm with dynamically adjusted weights is used to generate a multi-dimensional particle swarm, so that each particle corresponds to a set of parameters for total secondary air volume, secondary damper, and burnout damper. Next, the parameter combination corresponding to each particle is input into the boiler coupled combustion multi-objective model to obtain the parameter prediction values ​​output by the model; Based on this, the predicted values ​​are substituted into a multi-objective weighted sum function for calculation, and the parameter combination that minimizes the function value is selected as the optimal air distribution parameter combination.

[0015] Preferably, the optimal air distribution parameter combination is input into the control unit for closed-loop control, and the control unit interacts with the distributed control system, including: The control unit is an external module built around a high-performance computing server. First, the external module establishes a communication connection with the distributed control system, and then reads the unit's operating parameters in real time; When the boiler operating conditions change, the unit load, coal feed rate, and water feed rate are input into the boiler coupled combustion multi-objective model. The optimal total secondary air volume is obtained by iterative optimization based on the particle swarm optimization algorithm to adapt to the current operating conditions.

[0016] Preferably, the optimal total secondary air volume is compared with the current secondary air volume, and a control command is generated and sent back to the distributed control system to complete the closed-loop control of the secondary air volume for coal and biomass coupled combustion, including: First, the external module determines whether the boiler is in a steady-state load condition or a variable load condition, and then dynamically adjusts the multi-objective optimization weights and parameter adjustment modes. Meanwhile, quantitative expert knowledge is embedded in the control logic as an optimization constraint, and the parameters of secondary air volume, secondary damper, and burnout damper are adjusted synchronously. Finally, the optimal total secondary air volume is compared with the current secondary air volume to generate control optimization commands and send them back to the distributed control system to complete closed-loop control.

[0017] Compared with the prior art, the present invention discloses at least the following technical effects: This invention acquires key combustion data such as flame center position, furnace outlet temperature, and potassium alkali metal signal through in-furnace machine vision and online spectral monitoring. It then integrates unit operating parameters to construct a deep learning model, establishing a mapping relationship between operating parameters, combustion state, and flue gas emissions. Based on multi-objective collaborative optimization, it obtains the optimal air distribution parameters and achieves real-time optimization of secondary air volume through closed-loop control. This simultaneously improves boiler combustion efficiency and reduces NOx emissions. X It effectively solves the technical problems of lagging regulation and difficulty in coordinating multiple objectives in traditional methods by emitting and mitigating ash slag on the heated surface, and is suitable for safe and stable operation under wide loads in coal and biomass coupled combustion. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to the present invention; Figure 2 This is a diagram showing the arrangement of machine vision measurement points and spectral measurement points provided in an embodiment of the present invention. Figure 3 The schematic diagram of the multi-objective optimization control principle of boiler secondary air based on particle swarm optimization algorithm is provided for the embodiments of the present invention. Detailed Implementation

[0020] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, this invention provides a method for synergistic optimization of secondary air volume in coupled combustion of coal and biomass, comprising the following steps: Step 100: Perform machine vision and spectral online monitoring of the boiler furnace, collect visual images and emission spectra of the combustion flames inside the furnace, determine the flame center position, furnace outlet temperature and potassium alkali metal signal value through image processing and spectral analysis, and integrate the monitoring data with the unit operating parameters to construct a deep learning database.

[0023] Specifically, this step first involves establishing a boiler furnace machine vision and spectral online monitoring system, such as... Figure 2 As shown, the system symmetrically installs multiple sets of machine vision devices at different heights in the furnace area above the burnout air and below the flame deflector, and correspondingly deploys spectral monitoring devices near the flame deflector at the furnace outlet. The field of view lines of the machine vision devices intersect and point towards the central axis of the furnace along the height direction, and the vertical field of view completely covers the furnace outlet area. The spectral monitoring devices are directly aimed at the flue gas area at the furnace outlet to achieve all-round acquisition of combustion parameters.

[0024] First, visual image signals and emission spectrum signals of the combustion flame inside the furnace are acquired using machine vision and spectral acquisition equipment at each measuring point. Second, based on the principles of thermal radiation and photoelectric conversion, the pixel position corresponding to the high-temperature combustion zone in the visual image is the area of ​​maximum brightness. By taking visual images from at least two different height positions, the area of ​​maximum brightness in the vertical field of view is determined. By intersecting this area of ​​maximum brightness, the flame center position along the height direction of the furnace can be determined. Simultaneously, the emission spectral intensity of alkali metal potassium is extracted from the emission spectrum data of at least two measuring points, and the average value is taken as the alkali metal potassium signal at the furnace outlet. The temperature at each measuring point is calculated using a multi-wavelength method, and the average value is taken to obtain the furnace outlet temperature.

[0025] Furthermore, the flame center position, furnace outlet temperature, and alkali metal potassium signal value obtained from online monitoring are output to the unit's distributed control system. Combined with boiler design parameters, the flame center position offset and furnace outlet temperature offset are calculated. Then, these offsets and alkali metal potassium signal values ​​are compared with unit load, main steam flow, feedwater flow, feedwater temperature, total coal consumption in the coal mill, coal mill temperature, total secondary air volume, secondary air damper, burnout air damper, CO content, and NO content. X Key operating parameters of the unit, such as concentration, are fused together to construct a dataset for deep learning, which is then divided into training and validation sets according to a set ratio.

[0026] Step 200: Input the deep learning database into the deep learning algorithm for model training, construct a boiler coupled combustion multi-objective model, and establish the mapping relationship between unit operating parameters, combustion state parameters, and flue gas emission parameters.

[0027] Specifically, such as Figure 3 As shown, this invention constructs a multi-objective model for coupled combustion in a boiler and a full-process optimized control system. The core of this system is a multi-layer perceptron neural network structure for the multi-objective model of coupled combustion in a boiler. First, the input parameters of this model are divided into two categories: one category is non-adjustable input parameters, including fixed operating parameters such as unit load, main steam flow, feedwater flow, feedwater temperature, total coal quantity in the pulverizer, and pulverizer temperature; the other category is adjustable input parameters, including optimizable controllable parameters such as total secondary air volume, secondary damper, and burnout damper. Both types of parameters are input into the input layer of the model. The output parameters of the model are the flame center position offset, furnace outlet temperature offset, alkali metal potassium signal value, CO content, and NO. X Combustion state and flue gas emission indicators such as concentration.

[0028] Secondly, the multi-objective model is trained using the training set constructed in step 100. The preprocessed input parameters are input into the input layer of the model. After nonlinear calculations in multiple hidden layers, the calculation results are obtained in the output layer. The output vector is reconstructed to obtain the furnace temperature field. Subsequently, the weights of all neurons in the hidden layer are iteratively updated through the backpropagation process to complete the training of the multilayer perceptron neural network model and establish the mapping relationship between the key operating parameters of the unit and the output parameters of the model.

[0029] Furthermore, the model trained with validation set data is used to obtain the model output value and compare it with the actual value of the corresponding working condition to verify the model's accuracy and generalization ability.

[0030] Step 300: Based on the boiler coupled combustion multi-objective model, carry out multi-objective collaborative optimization, construct a multi-objective weighted sum function, and iteratively optimize through intelligent optimization algorithm to obtain the optimal air distribution parameter combination that minimizes the multi-objective weighted sum function.

[0031] Specifically, such as Figure 3 As shown on the right, this invention employs a particle swarm optimization algorithm based on dynamic weight adjustment to achieve multi-objective collaborative optimization. First, weight coefficients are assigned to each output parameter of the coupled combustion multi-objective model. , , , , And the weighting coefficients satisfy + + + + =1, through the formula Construct a multi-objective weighted sum function.

[0032] In the formula, This represents the offset of the flame center position. This represents the temperature offset at the furnace outlet. The signal value is for potassium alkali metal, CO is the CO content at the furnace outlet, and NO is... X NO at furnace outlet X concentration.

[0033] In the above, the flame center position offset, furnace outlet temperature offset, and CO content are parameters characterizing boiler efficiency; the furnace outlet temperature offset and alkali metal potassium signal value are parameters characterizing slagging on the heating surface; NO... X Concentration is a parameter characterizing pollutant emissions. This weighted sum function is used as a performance index for multi-objective collaborative optimization. When the function value is minimized, boiler efficiency can be improved, pollutant emissions can be reduced, and ash slagging on the heated surface can be alleviated simultaneously.

[0034] Secondly, a multi-objective optimization algorithm with dynamically adjusted weights is used to implement multi-objective optimization. Multiple multi-dimensional particles are randomly generated, and each particle corresponds to a set of adjustable combustion control quantities under the current operating conditions, namely the parameter combination of total secondary air volume, secondary damper, and burnout damper. The parameter combination corresponding to each particle is input into the coupled combustion multi-objective model trained in step 200 to obtain the predicted value of the model output parameters. The quality of the particles is judged according to the multi-objective weighted sum function. After iterative optimization, the parameter combination that minimizes the multi-objective weighted sum function value is selected as the optimal air distribution parameter combination.

[0035] Step 400: Input the optimal air distribution parameter combination into the control unit for closed-loop control. The control unit interacts with the distributed control system to compare the optimal total secondary air volume with the current secondary air volume and generate control commands to be sent back to the distributed control system, thus completing the closed-loop control of the secondary air volume of coal and biomass coupled combustion.

[0036] Specifically, such as Figure 3As shown, the optimal air distribution parameter combination obtained through optimization is input into the control unit to form a complete closed-loop control process. First, an external module with a high-performance computing server as its core is built as the control unit, establishing a communication connection with the distributed control system of the thermal power unit to read key operating parameters of the unit in real time. When the boiler operating conditions change, the combustion variables such as unit load, coal feed, and water feed are input into the coupled combustion multi-objective model trained in step 200. The multi-objective optimization process in step 300 is invoked to iteratively optimize the particle swarm. Each particle updates its parameters based on its own historical best value and the global best value. After multiple iterations, the particle swarm converges to the optimal position, obtaining the optimal total secondary air volume suitable for the current operating conditions. The optimal total secondary air volume is compared with the current secondary air volume to generate control optimization commands, which are then sent back to the distributed control system to achieve closed-loop control.

[0037] Secondly, such as Figure 3 As shown on the right, the external module integrates load judgment and control loop switching functions. Depending on whether the boiler is operating under steady-state or variable load conditions, it dynamically adjusts the multi-objective optimization weights and parameter adjustment modes: under steady-state load conditions, the control logic prioritizes improving boiler efficiency; under variable load conditions, the control logic prioritizes reducing NO. X The generation rate is increased to prevent emissions from exceeding standards at the source. At the same time, quantitative expert knowledge is embedded in the control logic as optimization constraints to simultaneously adjust multiple variables such as secondary air volume, secondary damper opening, and burnout damper, quickly converging to the optimal air distribution mode. This enables dynamic optimization of the boiler at full load, ensuring the safe, stable, economical, and efficient operation of the unit.

[0038] The above content will be further described below through specific implementation methods.

[0039] Example 1 To verify the practical engineering application effect of the optimization method of this invention, a 660MW supercritical coal-fired boiler in a power plant was selected as the test object. This unit has been carrying out straw-based biomass co-firing operations for a long time. In response to the problems of unstable combustion, large slagging tendency, and pollutant fluctuations in the biomass co-firing process, the method of this invention was used to carry out field test applications. The specific test implementation process is as follows: During the experiment, an integrated online monitoring system combining machine vision and spectral analysis was installed inside the boiler furnace. The entire system comprised six machine vision devices, two spectral detection devices, a signal transmission module, and a data preprocessing terminal. The system was integrated with the unit's DCS system to collect raw combustion data from the furnace. The six machine vision monitoring points were evenly distributed along the front and rear walls of the boiler, with three monitoring points on each wall at elevations of 33m, 40m, and 47m respectively. All monitoring points were embedded in the furnace sidewall using high-temperature resistant protective sleeves, with the lenses vertically facing the furnace interior. The vertical field of view was set to approximately 65° to ensure overlapping monitoring views and complete coverage of the furnace outlet area. Simultaneously, two spectral monitoring points were symmetrically arranged at the furnace outlet, with the spectral detection devices connected to these points via high-temperature resistant optical fibers. After commissioning, the system simultaneously acquired flame images from inside the furnace and spectral signals from the furnace outlet. The raw signals underwent multiple filtering and smoothing processes to remove outliers, and the original data was completely backed up to ensure accuracy.

[0040] During the experiment, based on the principles of thermal radiation and photoelectric conversion, a self-developed image processing software on the MATLAB platform was used to perform grayscale conversion and image enhancement processing on flame images at six measuring points. The highest-brightness high-temperature combustion region in each image was selected and mapped to the actual height of the furnace using coordinate conversion relationships. The center heights of the high-temperature regions at multiple measuring points were intersected to accurately determine the flame center position along the height direction of the furnace. The emission spectral signals at two spectral measuring points were analyzed to extract the characteristic emission lines of potassium alkali metal, and the average value of the two measuring points was taken as the potassium alkali metal signal at the furnace outlet. Simultaneously, based on Planck's law of thermal radiation, the temperature at the two measuring points was calculated using a multi-wavelength method, and the average value was taken as the furnace outlet temperature. The monitored flame center, furnace outlet temperature, and alkali metal potassium signals were transmitted to the unit's DCS system via a data interface. The flame center offset was then calculated using the boiler's inherent design parameters. Furnace outlet temperature deviation Simultaneously, data are collected from the DCS system including unit load, main steam flow, feedwater flow, feedwater temperature, total coal consumption in the coal mill, coal mill outlet temperature, total secondary air volume, secondary air damper opening, burnout damper opening, furnace outlet CO content, and furnace outlet NO content. X Key operating parameters such as concentration were integrated with monitoring characteristic parameters, offsets, and unit operating parameters to form a raw dataset. Outlier data was removed from the raw dataset, missing data was filled using linear interpolation, and all parameters were normalized. Finally, a deep learning database of 10,000 sets was constructed, strictly divided in an 8:2 ratio, with 8,000 sets for training and 2,000 sets for validation, providing a data foundation for building and training neural network models.

[0041] This experiment used a multilayer perceptron neural network (MLP) coupled combustion multi-objective model built on the TensorFlow framework. The model input variables included nine key unit operating parameters: unit load, main steam flow, feedwater flow, feedwater temperature, total coal output to the pulverizer, pulverizer outlet temperature, total secondary air volume, secondary damper opening, and burnout damper opening. The output variables were set as five multi-objective evaluation parameters, namely, flame center position offset. Furnace outlet temperature deviation Alkali metal potassium signal CO content at furnace outlet, NO content at furnace outlet X Concentration. The model's internal structure is precisely configured as follows: the input layer has 9 neurons to match 9 input parameters, using the ReLU activation function; the hidden layer has 3 consecutive layers with 64, 32, and 16 neurons respectively, all using the ReLU activation function to complete non-linear mapping, and a Dropout layer with a coefficient of 0.2 is added after each hidden layer to suppress overfitting during model training; the output layer has 5 neurons to match 5 output parameters, using a linear activation function to ensure that the dimensions of the output physical quantities are consistent with the true values.

[0042] After the model is built, the training set data is normalized to eliminate dimensional differences. The preprocessed data is then input into the model for iterative training. After nonlinear operations through multiple hidden layers, the results are output, and the output vector is reconstructed to obtain the furnace temperature field distribution, which intuitively represents the temperature change pattern inside the furnace. The mean squared error is used as the loss function during training, and the Adam optimizer is used to update the weights. The learning rate is set to 0.001, the maximum number of iterations is 1000, and the batch size for each training iteration is 64. During training, the model accuracy is continuously verified using the validation set. When the validation set loss function shows no decreasing trend after 50 consecutive iterations, an early stopping mechanism is triggered to terminate training and avoid the risk of overfitting.

[0043] Once the training model has converged, it can accurately establish the nonlinear mapping relationship between unit operating parameters and combustion and emission indicators. Inputting compliant operating parameters will quickly output the corresponding prediction results. To quantify the model's accuracy, 2000 sets of validation samples were imported one by one into the trained model. Predicted values ​​were compared with measured values, and the mean absolute error (MAE) and coefficient of determination (R²) were calculated. 2 The test results are as follows: MAE = 0.25m, R 2 =0.93; MAE = 7.2℃, R 2 =0.94 MAE = 0.028au, R 2 =0.91; MAE of CO content = 18ppm, R 2 =0.91; NOX MAE concentration = 27 mg / m 3 R 2 =0.92, R² = 0.92, the coefficient of determination for all parameters 2 All values ​​are greater than 0.90, indicating that the model fitting accuracy and engineering generalization ability meet the requirements for field use.

[0044] Based on the trained coupled combustion multi-objective model, a weighted particle swarm optimization (PSO) algorithm is used to collaboratively optimize the air distribution parameters. Taking into account the characteristics of the boiler's coupled combustion operation, fixed initial weights are assigned to the five output indicators, among which… =0.3、 =0.2、 =0.25、 =0.15、 =0.1, the total weight is 1, and the weight values ​​can be dynamically adjusted according to the real-time operating conditions of the unit. Based on this, a multi-objective weighted sum function is constructed, and the calculation formula is shown in the following formula: ; In the formula, the offset of the flame center position is... Furnace outlet temperature deviation CO content is used to characterize boiler combustion efficiency; the lower the value, the better the combustion heat exchange effect; furnace outlet temperature deviation. With alkali metal potassium signal value Used to determine the risk of ash and slag formation on heated surfaces; the lower the value, the lower the tendency for corrosion and slag formation. X Concentration is an environmental emission evaluation indicator; the lower the concentration, the better the emission performance. Therefore, the objective function is... Minimization is the optimization criterion, which can simultaneously achieve the comprehensive optimization goals of efficient combustion, low emissions, and low slagging.

[0045] The algorithm's operating parameters were calibrated and adjusted based on the actual operating conditions of the unit. The particle swarm size was set to 50, the particle dimension to 3, and the maximum number of iterations to 100. The learning factors c1=c2=2.0, and the initial inertia weight ω was set to 0.9, decreasing linearly to 0.4 with each iteration, balancing global search capability with local optimization accuracy. During algorithm execution, multi-dimensional particles were randomly generated, with each particle representing a set of air distribution schemes. The safe range for adjusting parameters was strictly limited: the total secondary air volume was controlled between 320,000 and 480,000 m³. 3 / h, the secondary air damper opening is controlled at 35%~85%, and the burnout air damper opening is controlled at 25%~75%. The control parameters corresponding to each particle are imported into a multi-objective model, and the predicted values ​​of the five output parameters are obtained. These predicted values ​​are then substituted into a weighted sum function to calculate the value of a single particle. P Value, in P The minimum value is used as the screening criterion, and the optimal combination of air distribution parameters is selected through iterative optimization.

[0046] To facilitate the engineering implementation of the optimization algorithm, an external high-performance computing server was built as a control module. The hardware utilizes an 11th-generation Intel Core i-series processor, a motherboard with dual Gigabit Ethernet ports and four USB 3.2 Gen2 ports, dual-channel DDR4 memory expandable up to 64GB, and a reserved M.2 2280 solid-state drive slot, compatible with the NVMe high-speed protocol, ensuring high-speed processing of large volumes of data. The software incorporates a trained multi-objective model, a dynamically adjusted weighted PSO algorithm, and a DCS communication interface program. Bidirectional data transmission between the unit is achieved via the OPC UA protocol, with communication latency consistently controlled within 0.5 seconds, meeting the real-time boiler control requirements.

[0047] During coupled combustion operation, the external module synchronously collects operational data such as unit load, medium flow rate, coal quantity, and damper opening, maintaining the same collection frequency as the DCS system. When fluctuations in operating conditions are detected, the module automatically imports the changed parameters and initiates an optimization process. Particles dynamically update their position and velocity based on their historical and global best values, and after iteration, convergence yields the optimal total secondary air volume for the current operating condition. The external module compares the optimal air volume with the actual air volume on site in real time, setting a 3000m³ / h value. 3 / h is the action threshold. When the air volume deviation exceeds the threshold, an optimization control command is issued. If the deviation is below the threshold, the current operating state is maintained to avoid frequent disturbances that could affect the boiler combustion stability.

[0048] This add-in module has a built-in load discrimination program that uses real-time computer groups to differentiate between steady-state and variable load conditions: it increases the boiler efficiency-related weights during steady-state operation and increases NO during variable load operation. X The system prioritizes the control of pollutant generation; at the same time, it embeds quantitative expert constraints to define safety boundaries such as furnace outlet temperature of 980~1180℃ and CO content not exceeding 400ppm. If the optimization parameters exceed the limit range, the module automatically constrains the particle values ​​and re-iterates the calculation, ultimately achieving fully automatic, high-precision, and safe closed-loop optimization and control of secondary air volume under the coupled combustion of coal and biomass.

[0049] Therefore, by employing the aforementioned method for co-optimizing secondary air volume in coal-biomass coupled combustion, and by adding in-furnace combustion machine vision and spectral detection methods to obtain supplementary data such as flame center position, furnace outlet temperature, and alkali metal potassium signals, multi-objective co-optimization can effectively balance boiler efficiency and NOx levels. X The contradiction between emissions and the release of gaseous alkali metal potassium, while simultaneously improving boiler efficiency and reducing NO in flue gas. X Discharge and mitigation of ash and slag formation on heated surfaces.

[0050] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for synergistic optimization of secondary air volume in coupled combustion of coal and biomass, characterized in that, Includes the following steps: Machine vision and spectral online monitoring of boiler furnace are carried out to collect visual images and emission spectra of combustion flames in the furnace. The flame center position, furnace outlet temperature and potassium alkali metal signal value are determined through image processing and spectral analysis. The monitoring data is fused with unit operating parameters to build a deep learning database. The deep learning database is input into the deep learning algorithm for model training, a multi-objective model of boiler coupled combustion is constructed, and the mapping relationship between unit operating parameters, combustion state parameters, and flue gas emission parameters is established. Based on the boiler coupled combustion multi-objective model, multi-objective collaborative optimization is carried out, a multi-objective weighted sum function is constructed, and intelligent optimization algorithm is used to iteratively optimize the optimal air distribution parameter combination that minimizes the multi-objective weighted sum function. The optimal air distribution parameter combination is input into the control unit for closed-loop control. The control unit interacts with the distributed control system to compare the optimal total secondary air volume with the current secondary air volume and generate control commands to be sent back to the distributed control system, thus completing the closed-loop control of the secondary air volume of coal and biomass coupled combustion.

2. The method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to claim 1, characterized in that, Online monitoring of the boiler furnace using machine vision and spectral analysis is performed to acquire visual images and emission spectra of the combustion flames inside the furnace, including: First, at least two machine vision measurement points are set up in the furnace area above the boiler burnout air and below the flame deflector angle so that the vertical field of view of the machine vision equipment completely covers the furnace outlet area. Then, at least two spectral measurement points are set up in the furnace outlet area to collect visual images of the flame using machine vision measurement points, and at the same time, the flame emission spectrum is collected using spectral measurement points.

3. The method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to claim 1, characterized in that, The flame center location, furnace outlet temperature, and potassium alkali metal signal value were determined through image processing and spectral analysis, including: First, the flame visual images at at least two height positions are intersected by the region of maximum brightness to determine the flame center position in the furnace along the height direction. Then, the emission spectrum data of potassium alkali metal at at least two furnace outlet measuring points are used to extract the emission line intensity of potassium alkali metal and calculate the average value to obtain the potassium alkali metal signal value at the furnace outlet. Next, the temperature at each spectral measurement point is calculated using the multi-wavelength method, and the average value is obtained to obtain the furnace outlet temperature.

4. The method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to claim 1, characterized in that, By integrating monitoring data with unit operating parameters, a deep learning database is constructed, including: First, the flame center position, furnace outlet temperature, and potassium alkali metal signal value are transmitted to the distributed control system. Then, the flame center position offset and furnace outlet temperature offset are calculated by combining the boiler design parameters. Then, the above offset, potassium alkali metal signal value, CO content, and NO... X Concentration is fused with data from unit load, main steam flow, feedwater flow, feedwater temperature, total coal volume in the coal mill, coal mill temperature, total secondary air volume, secondary air damper, and burnout air damper. The merged dataset is then divided into training and validation sets according to a set ratio, thus forming a deep learning database that can be used for model training.

5. The method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to claim 1, characterized in that, The deep learning database is input into a deep learning algorithm for model training to construct a multi-objective model of boiler coupled combustion, including: First, a multi-layer perceptron neural network is used to build a multi-objective model of boiler coupled combustion. Then, the unit load, main steam flow, feedwater flow, feedwater temperature, total coal flow in the coal mill, coal mill temperature, total secondary air flow, secondary air damper, and burnout air damper are set as the model input parameters; Next, the flame center position offset, furnace outlet temperature offset, alkali metal potassium signal value, CO content, and NO were measured. X The concentration is set as a model output parameter.

6. The method for synergistic optimization of secondary air volume in coupled coal and biomass combustion according to claim 5, characterized in that, Establish a mapping relationship between unit operating parameters and combustion state parameters and flue gas emission parameters, including: First, the training set data is input into the multilayer perceptron neural network. After nonlinear calculation by the hidden layer, the output result is reconstructed to obtain the furnace temperature field. Then, the weights of the hidden layer neurons are iteratively updated using the backpropagation algorithm, thereby completing the training of the multilayer perceptron neural network; The validation set data is then input into the trained model, and the model output value is compared with the actual operating conditions on site, thereby solidifying the mapping relationship between the unit's operating parameters and output parameters.

7. The method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to claim 1, characterized in that, Based on the aforementioned boiler coupled combustion multi-objective model, multi-objective collaborative optimization is carried out, and a multi-objective weighted sum function is constructed, including: First, select the flame center position offset, furnace outlet temperature offset, alkali metal potassium signal value, CO content, and NO. X Concentration as a multi-objective optimization variable; Then, weight coefficients are assigned to each optimization variable. , , , , And make the sum of the weight coefficients satisfy + + + + =1; Then, the multi-objective weighted sum function is constructed according to the formula, which is: ； In the formula, This represents the offset of the flame center position. This represents the temperature offset at the furnace outlet. The signal value is for potassium alkali metal, CO is the CO content at the furnace outlet, and NO is... X NO at furnace outlet X concentration.

8. The method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to claim 1, characterized in that, The optimal combination of wind distribution parameters that minimizes the multi-objective weighted sum function is obtained through iterative optimization using intelligent optimization algorithms, including: First, a multi-dimensional particle swarm optimization algorithm with dynamically adjusted weights is used to generate a multi-dimensional particle swarm, so that each particle corresponds to a set of parameters for total secondary air volume, secondary damper, and burnout damper. Next, the parameter combination corresponding to each particle is input into the boiler coupled combustion multi-objective model to obtain the parameter prediction values ​​output by the model; Based on this, the predicted values ​​are substituted into a multi-objective weighted sum function for calculation, and the parameter combination that minimizes the function value is selected as the optimal air distribution parameter combination.

9. The method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to claim 1, characterized in that, The optimal air distribution parameter combination is input into the control unit for closed-loop control. The control unit interacts with the distributed control system, including: The control unit is an external module built around a high-performance computing server. First, the external module establishes a communication connection with the distributed control system, and then reads the unit's operating parameters in real time; When the boiler operating conditions change, the unit load, coal feed rate, and water feed rate are input into the boiler coupled combustion multi-objective model. The optimal total secondary air volume is obtained by iterative optimization based on the particle swarm optimization algorithm to adapt to the current operating conditions.

10. The method for synergistic optimization of secondary air volume in coal and biomass coupled combustion according to claim 9, characterized in that, The optimal total secondary air volume is compared with the current secondary air volume, and control commands are generated and sent back to the distributed control system to complete the closed-loop control of the secondary air volume for coal and biomass coupled combustion, including: First, the external module determines whether the boiler is in a steady-state load condition or a variable load condition, and then dynamically adjusts the multi-objective optimization weights and parameter adjustment modes. Meanwhile, quantitative expert knowledge is embedded in the control logic as an optimization constraint, and the parameters of secondary air volume, secondary damper, and burnout damper are adjusted synchronously. Finally, the optimal total secondary air volume is compared with the current secondary air volume to generate control optimization commands and send them back to the distributed control system to complete closed-loop control.

Images