Method for reducing total pollutant emission of environmental protection island based on morphological transformation and residual compensation

Abstract

The present application relates to the technical field of coal-fired power plant flue gas pollution control, and discloses an environmental protection island full-pollutant emission reduction method based on form conversion and residual compensation, comprising: obtaining real-time operation data of a coal-fired power plant intelligent environmental protection island, generating a time sequence characteristic vector and setting an entry boundary condition; constructing a pollutant form conversion mechanism model for each environmental protection subsystem, inputting relevant data and the entry boundary condition, and calculating a pollution theoretical benchmark concentration; taking LSTM as an error compensation prediction model, inputting the current working condition time sequence characteristic vector after historical data training, obtaining a concentration prediction residual value, obtaining the deviation degree of the historical training working condition and the current working condition, calculating a confidence factor, and generating a pollutant mixed prediction concentration; finally, taking ammonia injection flow and the like as optimization variables, setting a constraint condition, constructing a target function with the lowest total cost, obtaining an optimal control parameter group, so as to realize the optimization control of the environmental protection subsystem and pollutant emission reduction; the present application significantly improves the pollutant emission reduction effect.

Description

Technical Field

[0001] This invention relates to the field of flue gas pollution control technology in coal-fired power plants, specifically to a method for reducing total pollutant emissions in environmental islands based on morphological transformation and residual compensation. Background Technology

[0002] With the full implementation of national ultra-low emission standards, the environmental protection facilities (denitrification, dust removal, desulfurization, etc.) of coal-fired power plants have gradually shifted from an independent operation mode of single equipment to a systematic and intelligent "smart environmental island" model. The smart environmental island aims to achieve efficient removal of flue gas pollutants and reduce operating costs through the coordinated control of the entire process of environmental subsystems such as selective catalytic reduction (SCR), flue gas cooler (FGC), electrostatic precipitator (ESP), and wet desulfurization (WFGD). Understanding and predicting the real-time concentration and speciation of multiple pollutants in the flue gas of coal-fired power plants is a prerequisite for achieving precise control in the smart environmental island.

[0003] Currently, power plants mainly rely on Continuous Emission Monitoring Systems (CEMS) installed at the end of the flue gas duct to obtain nitrogen oxide (NOx) data. x Emission data for sulfur dioxide (SO2) and particulate matter are available. However, continuous emission monitoring systems (CEMS) suffer from drawbacks such as measurement lag, high maintenance costs, inability to monitor unconventional pollutants (e.g., heavy metal pollutant mercury, trace pollutant SO3, volatile organic compounds (VOCs) online, and inability to reflect internal process parameters. Therefore, predictive control technology based on soft sensing has become a research hotspot in the industry.

[0004] In recent years, although related technologies have made some progress in the construction of smart environmental protection islands, most existing smart environmental protection island control technologies are limited to conventional pollutants (such as nitrogen oxides, NO). x The system lacks independent control of pollutants such as sulfur dioxide (SO2) and particulate matter, and lacks the ability to coordinate the treatment of unconventional pollutants such as mercury (Hg), sulfur trioxide (SO3), and dichloromethane (VOCs). Furthermore, single "pure mechanism" or "pure data" modeling methods are difficult to balance the physical consistency and high accuracy of predictions. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a method for reducing total pollutant emissions in environmental islands based on morphological transformation and residual compensation. This method solves the problem that existing technologies are limited to the control of conventional pollutants, resulting in low pollutant emission reduction effects.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: The method for reducing total pollutant emissions on environmental islands based on morphological transformation and residual compensation includes the following steps: Acquire real-time operation data of the smart environmental protection island of a coal-fired power plant, generate time-series feature vectors, and set inlet boundary conditions; For each environmental protection subsystem of the smart environmental protection island of a coal-fired power plant, a pollutant form transformation mechanism model is constructed, real-time operation data is input, and combined with the inlet boundary conditions, the theoretical removal efficiency or transformation efficiency of each environmental protection island subsystem for each pollutant is generated, so as to calculate the theoretical baseline concentration of each pollutant at the outlet of each environmental protection island subsystem. The Long Short-Term Memory Network model is used as the error compensation prediction model. The model is trained using historical time series feature vectors to learn the nonlinear deviation between the actual monitored values ​​and the theoretical benchmark concentration in the historical time series feature vectors. After generating the trained error compensation prediction model, the time series feature vector under the current operating conditions is input to generate the concentration prediction residual values ​​of each pollutant under the current operating conditions. The deviation between the historical time-series feature vector under the historical training conditions and the time-series feature vector under the current conditions is obtained. The confidence factor of the error compensation prediction model is calculated to perform weighted fusion of the theoretical baseline concentration and the concentration prediction residual value. After generating the fusion result, physical boundary constraints are applied for verification to generate the optimized mixed prediction concentration value of each pollutant. Using the ammonia injection flow rate, outlet flue gas temperature, operating voltage, number of slurry circulation pumps in operation, and slurry pH setpoint of the environmental protection subsystem as optimization variables, an objective function is established to minimize the total cost of electricity consumption, material consumption, and equipment wear and tear risk. Under the constraints of environmental limits for pollutant emission concentrations, equipment safety operation boundaries, and pollutant form transformation correlations, a multi-objective optimization solution is performed to generate the optimal control parameter set for the intelligent environmental protection island of the coal-fired power plant. The optimal control parameter set is sent to the distributed control system of the intelligent environmental protection island of the coal-fired power plant to achieve optimized control of each environmental protection subsystem for treating pollutants and reduce pollutant emissions.

[0007] The present invention has the following beneficial effects: 1. The proposed method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation, through the construction of a dual-drive fusion architecture of pollutant morphological transformation mechanism model and error compensation prediction model, can not only take into account physical consistency and dynamic feature capture, but also significantly improve the prediction accuracy of multiple pollutants under non-steady-state conditions such as load and coal quality fluctuations, and ultimately significantly improve the pollutant reduction effect. 2. Based on the multi-pollutant transformation mechanism, this invention synergistically controls mercury (Hg), sulfur trioxide (SO3), and dichloromethane (VOCs) on the basis of meeting conventional indicators, achieving comprehensive pollutant emission reduction, avoiding the drawbacks of single treatment, and meeting increasingly stringent environmental standards. 3. This invention also reduces energy and material consumption through global optimization, while avoiding the risks of ammonium bisulfate ABS formation and equipment corrosion, thereby extending equipment life and maximizing the comprehensive benefits of cost reduction and equipment life extension. 4. This invention also utilizes soft measurement to complete the source data, and at the same time, it prevents impact on the actuator through instruction smoothing and physical truncation mechanisms, thus ensuring the safe execution of control instructions in the industrial distributed control system (DCS). Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the process for the total pollutant reduction method for environmental islands based on morphological transformation and residual compensation proposed in this invention. Figure 2 This is a schematic diagram illustrating the multi-pollutant speciation transformation mechanism throughout the entire process in the embodiment. Figure 3 This is a schematic diagram of the mechanism-data dual-driven prediction model architecture in the embodiment; Figure 4 This is a schematic diagram of the global multi-objective collaborative optimization and cost control logic in the embodiment.

[0009] Figure 5 In this embodiment, the dual-drive prediction model is used to predict SO2 and NO in flue gas. x A scatter plot showing the comparison between predicted and actual values ​​of particulate matter emission concentrations; Figure 6 This is a bar chart showing the comparison between the original flue gas purification cost and the cost after multi-objective optimization under different boiler loads in the embodiment. Detailed Implementation

[0010] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0011] The principle of this invention is as follows: To address the issues of lagging data monitoring, weak generalization ability of pure data models, and lack of coordinated treatment of unconventional pollutants in existing intelligent environmental island control systems for coal-fired power plants, this paper proposes a comprehensive pollutant reduction method for environmental islands based on speciation transformation and residual compensation. Through a dual-drive approach of mechanism and data, this method achieves coordinated emission reduction of multiple pollutants. Specifically, firstly, based on the principles of physicochemical reaction kinetics, a multi-pollutant speciation transformation mechanism model is constructed covering the entire process of the Selective Catalytic Reduction (SCR) subsystem, the Flue Gas Cooler (FGC) subsystem, the Electrostatic Precipitator (ESP) subsystem, and the Wet Flue Gas Desulfurization (WFGD) subsystem. The theoretical baseline concentrations of each pollutant are calculated, revealing the physical nature of nitrogen oxides... The study investigates complex processes such as NOx reduction, mercury (Hg) oxidation and transformation, catalytic oxidation of volatile organic compounds (VOCs), and SO3 condensation phase transition. Simultaneously, it introduces an error compensation prediction model based on Long Short-Term Memory (LSTM) deep learning to capture nonlinear deviations caused by equipment aging and coal quality fluctuations during actual operation. By constructing a similarity assessment index for operating conditions and dynamically generating confidence weights, the study adaptively integrates the physical consistency of the mechanistic model with the high sensitivity of the data model. This allows for the acquisition of mixed prediction results for multiple pollutants (conventional pollutants, Hg, SO3, and VOCs) with both high accuracy and robustness even under varying operating conditions.

[0012] Based on accurate prediction, this invention also establishes a global cost objective function that includes power consumption (energy consumption), material consumption, and equipment wear and tear risks (such as catalyst aging and corrosion risks). The optimized mixed predicted concentration values ​​of each pollutant generated in the aforementioned steps are used as key state feedback variables and compared and verified in real time with the environmental protection limits for the emission concentrations of each pollutant, serving as strong constraints for multi-objective optimization. At the same time, combined with the constraints of equipment safe operation boundary (ammonium bisulfate ABS generation temperature) and pollutant form transformation correlation, intelligent optimization methods (Particle Swarm Optimization (PSO) / Differential Evolutionary Algorithm (DE)) are used to find the optimal set of control parameters (such as ammonia injection rate, flue gas temperature, voltage, slurry pH, etc.) in the multi-dimensional solution space. Finally, through a closed-loop feedback control strategy, the smoothed optimal command is sent to the distributed control system (DCS) to maintain the global optimization of the overall system operating cost while ensuring that the smart environmental protection island of the coal-fired power plant meets the environmental protection requirements for ultra-low emissions of multiple pollutants.

[0013] Based on the above principles, the innovation of this invention compared to existing methods for reducing pollutant emissions in smart environmental islands of coal-fired power plants lies in: (1) A hybrid modeling architecture (grey box model) of "mechanism benchmark + data residual" was constructed. Unlike the existing pure mechanism model (which is difficult to characterize equipment aging) or pure data model (which has poor anti-interference and lacks interpretability), this invention establishes the physical boundary with the mechanism model, corrects the nonlinear residual with the data model, and solves the problem of prediction distortion and low robustness of a single model under extreme or unknown working conditions by adaptive weighted fusion based on working condition similarity.

[0014] (2) It applies a multi-pollutant speciation transformation and synergistic removal mechanism throughout the entire process. This breaks through the limitations of existing technologies that only focus on conventional pollutants (NOx). x The limitations of SO2 / particulate matter emission concentrations have been extended to the microscopic morphology level, innovatively establishing a method for detecting elemental mercury (Hg) at the SCR stage. 0 Oxidation and VOCs compete for adsorption; SO3 / semi-volatile organic compounds (SVOCs) condensation phase transition at FGC; gaseous divalent mercury (Hg) at WFGD. 2+ The re-release inhibition and other correlation models have enabled comprehensive and in-depth synergistic treatment of trace and unconventional pollutants.

[0015] (3) A global dynamic cost optimization strategy incorporating "equipment health loss" is proposed. Unlike traditional local optimization that only aims to save on the input of environmentally friendly agents such as ammonia and limestone, this invention quantifies the risk of air preheater blockage caused by ammonium bisulfate and the risk of low-temperature corrosion caused by acidic aerosols into economic costs (equipment risk loss costs), thereby achieving a balance between short-term operating costs and long-term equipment life in the optimization decision.

[0016] Finally, based on the above principles, the specific implementation method adopted by the present invention is as follows: like Figure 1 As shown, the method for reducing total pollutant emissions on an environmental island based on morphological transformation and residual compensation includes the following steps: Step 1: Obtain real-time operation data of the smart environmental protection island of the coal-fired power plant, generate time-series feature vectors, and set entry boundary conditions.

[0017] In this embodiment, this step is a multi-source data acquisition and preprocessing process. Specifically, within the same time period, real-time operating data of the smart environmental protection island of the coal-fired power plant is collected at set time intervals (e.g., every 5 seconds). The purpose is to solve the problem of blind spots or noise from a single data source under the harsh operating environment of a coal-fired power plant. Therefore, the real-time operating data includes: unit load, raw coal composition data (volatile matter and mercury content detected by an online coal quality analyzer based on laser-induced breakdown spectroscopy (LIBS) technology), and flue gas parameters at the furnace outlet (inlet parameters of each environmental protection subsystem, including flow rate, temperature, carbon monoxide (CO) concentration, oxygen (O2) concentration, and nitrogen oxides (NO) concentration) collected by the distributed control system (DCS) of the coal-fired power plant. xThe parameters include: concentration of sulfur dioxide (SO2), particulate matter concentration, combustion intermediate product parameters (including carbon content of fly ash), complete equipment status parameters of each environmental protection subsystem (including selective catalytic reduction (SCR) subsystem, flue gas cooler (FGC) subsystem, electrostatic precipitator (ESP) subsystem, and wet flue gas desulfurization (WFGD) subsystem) (including inlet and outlet flue gas temperature, inlet and outlet water temperature, valve opening, pump current, operating voltage, etc.), and measured data of pollutant concentration at the chimney obtained through the continuous emission monitoring system (CEMS).

[0018] Meanwhile, the real-time operational data collected above undergoes denoising, filtering, and normalization preprocessing. It is then organized into a time-series feature vector based on the time sequence of collection, which includes flame spectral features. The process of acquiring flame spectral features is as follows: In actual engineering, the high-temperature and high-dust environment at the furnace outlet prevents online measurement of dichloromethane VOCs concentration and fly ash carbon content using physical sensors. These two parameters are key inlet boundaries for subsequent calculations of dichloromethane VOCs oxidation efficiency and the dust removal efficiency of the electrostatic precipitator (ESP) subsystem. Based on this, a soft measurement model of the inlet boundary based on furnace combustion conditions can be constructed. Specifically: First, the radiative energy distribution of the furnace combustion region is obtained using a flame spectrometer. Fast Fourier Transform (FFT) or wavelet transform is used to extract the free radical (OH*, CH*, C2*) radiative intensity and spectral entropy in specific bands, which are then used as flame spectral features. Secondly, boiler load, furnace outlet oxygen content, coal mill current, and flame spectral characteristics were selected as auxiliary input variables. A nonlinear mapping function was constructed using the support vector machine (SVM) regression method. The concentration of dichloromethane (VOCs) and fly ash carbon content at the furnace outlet were estimated in real time using the following formula:

[0019] In the formula, The concentration of dichloromethane at the furnace outlet; The carbon content of fly ash at the furnace outlet; It is a nonlinear mapping function; For boiler load; The amount of oxygen at the furnace outlet; This refers to the current of the coal mill. This method utilizes flame spectral characteristics to calculate the concentration of dichloromethane VOCs and fly ash carbon content at the furnace outlet in real time. This provides accurate inlet conditions for subsequent pollutant speciation transformation mechanism models, avoiding calculation errors caused by missing inlet boundaries.

[0020] In addition, inlet boundary conditions need to be set so that subsequent steps can solve for the theoretical baseline concentration. The inlet boundary conditions include furnace outlet flue gas parameters, raw coal composition data, and combustion intermediate product parameters. The furnace outlet flue gas parameters include flow rate, temperature, carbon monoxide concentration, oxygen concentration, nitrogen oxide concentration, sulfur dioxide concentration, particulate matter concentration, fly ash carbon content, and dichloromethane concentration.

[0021] Most importantly, all the raw data collected above need to be preprocessed. For example, wavelet transform is used for multi-scale denoising to address electromagnetic interference and various noises that may exist in the field signals. Median filtering is used to remove abrupt pulses. Max-min normalization is used to map data of different dimensions (such as temperatures of several hundred degrees and concentrations of several milligrams) to the [0, 1] interval to eliminate the influence of order-of-magnitude differences on the gradient of the subsequent deep learning model.

[0022] Step 2: For each environmental protection subsystem of the smart environmental protection island in the coal-fired power plant, construct a pollutant form transformation mechanism model, input real-time operating data, and combine inlet boundary conditions to generate the theoretical removal efficiency or transformation efficiency of each environmental protection island subsystem for each pollutant, so as to calculate the theoretical baseline concentration of each pollutant at the outlet of each environmental protection island subsystem.

[0023] In this embodiment, Figure 2 It demonstrates the data processing and connectivity relationships of various environmental protection subsystems, and from... Figure 2 As can be seen, this step aims to construct a "white box" model that conforms to the essence of physicochemical processes. Specifically, based on physicochemical reaction kinetics and mass transfer theory, it builds a full-process pollutant speciation mechanism model covering the Selective Catalytic Reduction (SCR) subsystem, the Flue Gas Cooler (FGC) subsystem, the Electrostatic Precipitator (ESP) subsystem, and the Wet Flue Gas Desulfurization (WFGD) subsystem. This allows the input of real-time operational data processed in step S1 into this mechanism model, enabling the calculation of conventional pollutants (such as NO) at the outlet of each subsystem. x The theoretical baseline concentrations of pollutants (such as Hg, SO2, particulate matter) and unconventional pollutants (such as Hg, SO3, VOCs) are determined, and the operating procedure is as follows: (1) Construction of pollutant speciation transformation mechanism model for selective catalytic reduction (SCR) subsystem, including nitrogen oxide reduction reaction model, mercury oxidation speciation transformation model, and dichloromethane catalytic oxidation model, specifically: 1) Nitrogen oxide reduction reaction model

[0024] The nitrogen oxide reduction model employs an improved pseudo-first-order reaction kinetic equation, comprehensively considering design baselines, ammonia balance constraints, and catalyst aging corrections: First, the denitrification reaction rate constant for selective catalytic reduction (SCR) is calculated based on Arrhenius's law. ,Right now:

[0025] In the formula, The preceding factor is determined by the characteristics of the catalyst material. It is an exponential function; The activation energy is used to characterize the ease with which a reaction proceeds. The ideal gas constant is 8.314 J / (mol·K); The absolute temperature of the flue gas in the reaction zone of the selective catalytic reduction subsystem is denoted as .

[0026] Secondly, based on the standard selective catalytic reduction (SCR) reaction mechanism, a method for constructing nitrogen oxide (NO) reaction was developed. x Theoretical removal rate The calculation formula is as follows:

[0027] In the formula, This is the catalyst activity correction factor, used to characterize the degree of physical aging or poisoning that occurs with increasing operating time; The rated design denitrification efficiency is the baseline for the catalyst; This is the ammonia-nitrogen molar ratio correction factor, and its calculation formula is: , To obtain the minimum value, This refers to the molar amount of ammonia injection flow rate converted to the selective catalytic reduction subsystem. This represents the molar amount of nitrogen oxides at the inlet of the selective catalytic reduction subsystem. This represents the catalyst volume; This represents the flue gas volumetric flow rate.

[0028] Therefore, based on nitrogen oxides NO x Theoretical removal rate Combined with the nitrogen oxide concentration at the inlet of the selective catalytic reduction subsystem The theoretical baseline concentration of nitrogen oxides at the outlet of the selective catalytic reduction subsystem can then be calculated. ,Right now:

[0029] In the formula, The concentration of nitrogen oxides at the inlet of the selective catalytic reduction subsystem is denoted as , and the concentration of nitrogen oxides at the outlet of the furnace is denoted as .

[0030] 2) Mercury oxidation form transformation model

[0031] The mercury oxidation form transformation model is based on the catalyst active site mechanism and calculates the oxidation form transformation of elemental mercury (Hg). 0 Under the action of a catalyst, it is converted into divalent mercury (Hg).2+ oxidation efficiency The calculation formula is as follows:

[0032] In the formula, The rate constant for the oxidation of mercury is affected by temperature according to the Arrhenius law and can be determined experimentally. The concentration of hydrogen chloride in the flue gas is calculated by combining the chlorine content of raw coal detected by the LIBS online coal quality analyzer with the amount of coal burned, based on real-time operating data.

[0033] Therefore, the theoretical baseline concentration of elemental mercury at the outlet of the selective catalytic reduction subsystem is... The calculation formula is as follows:

[0034] In the formula, The concentration of elemental mercury at the inlet of the selective catalytic reduction subsystem is calculated based on the material balance of mercury content in coal during the initial steps described above.

[0035] 3) Catalytic oxidation model of dichloromethane

[0036] The catalytic oxidation model for dichloromethane VOCs is based on the Langmuir-Hinshelwood multi-component competitive adsorption mechanism: First, the oxidation rate equation for dichloromethane VOCs generated in the furnace on the surface of the selective catalytic reduction SCR catalyst is constructed, namely:

[0037] In the formula, The catalytic oxidation rate of dichloromethane; Let be the intrinsic oxidation rate constant of dichloromethane, which follows Arrhenius's law; , , , respectively, are the adsorption equilibrium constants of dichloromethane, ammonia, and mercury on the catalyst surface, which follow the Van't Hoff equation and vary with temperature; , , These are the gaseous concentrations of dichloromethane, ammonia, and mercury, respectively.

[0038] Secondly, based on the catalytic oxidation reaction rate of dichloromethane Combined with the residence time of flue gas in the catalyst bed The conversion efficiency of dichloromethane VOCs was calculated by integration. ,Right now:

[0039] In the formula, This represents the dichloromethane concentration at the furnace outlet, which is also the dichloromethane concentration at the inlet of the selective catalytic reduction subsystem.

[0040] Therefore, the theoretical baseline concentration of dichloromethane at the outlet of the selective catalytic reduction subsystem is... The calculation formula is as follows:

[0041] In the formula, The concentration of dichloromethane at the inlet of the selective catalytic reduction subsystem is denoted as , and the concentration of dichloromethane at the outlet of the furnace is denoted as .

[0042] In addition, a side-reaction constraint model can be established, which is based on sulfur dioxide. The catalytic oxidation mechanism was investigated, and the total sulfur trioxide at the outlet of the selective catalytic reduction (SCR) subsystem was calculated. concentration :

[0043] In the formula, The sulfur trioxide concentration at the inlet of the selective catalytic reduction subsystem is typically taken as the inlet sulfur dioxide concentration. A concentration of 0.5%-1% is used as the baseline for combustion formation; This refers to the sulfur dioxide concentration in the real-time operating data, and also the sulfur dioxide concentration at the inlet of the selective catalytic reduction subsystem. It is the pre-exponential factor for the oxidation reaction of sulfur dioxide and is related to the catalyst formulation; It is the activation energy for the oxidation reaction; Subsequently, the dew point temperature of ammonium bisulfate ABS can be calculated using the Matsuda empirical formula based on the partial pressures of the gas phase components. :

[0044] In the formula, The partial pressure of ammonia escape is given by the formula. Conversion, The set limit for ammonia escape concentration; Total sulfur trioxide at the outlet of a selective catalytic reductant (SCR) system Pressure distribution, as described above The conversion yields the result; and All of these are thermodynamic empirical constants calibrated based on Matsuda experimental data, and in this embodiment, is taken as... , .

[0045] Finally, the calculated dew point temperature can be... As a lower limit constraint condition for the temperature in subsequent steps, namely:

[0046] in, To prevent condensation, a safe temperature margin (10℃) is set to prevent ammonium bisulfate ABS from depositing and clogging due to localized low temperatures.

[0047] (2) Construction of the pollutant speciation transformation mechanism model of the electrostatic precipitator subsystem, including the correction of fly ash resistivity of particulate matter and the synergistic dust removal model, specifically: A multi-component condensation phase transition model for the flue gas cooler (FGC) subsystem can also be established to calculate the proportion of gaseous pollutants (such as sulfur trioxide (SO3) and semi-volatile organic compounds (SVOCs)) that condense into droplets or aerosols during the FGC cooling process. Specifically: Based on the Antoine equation, the outlet flue gas temperature of each component in the current flue gas cooler FGC subsystem is calculated. Saturated vapor pressure ,Right now:

[0048] In the formula, It is a logarithmic function. , , These are the Antoine constants for each pollutant component (such as sulfuric acid vapor, specific organic compounds), which can be obtained by consulting the thermodynamic property library; When the inlet partial pressure of pollutants in the flue gas Greater than saturated vapor pressure At this time, a condensation phase transition occurs; and the formula for calculating the condensation ratio is as follows:

[0049] in, The condensation ratio (0~1) is used to characterize the proportion of gaseous pollutants that are converted into liquid / solid particles; The inlet partial pressure of the pollutant is calculated from the inlet concentration; After condensation, the concentration of pollutants maintained in the gaseous state at the outlet of the flue gas cooler FGC subsystem was calculated. for:

[0050] In the formula, The initial mass concentration of pollutants (such as SO3 and SVOCs) entering the flue gas cooler FGC subsystem is the mass concentration of pollutants at the outlet of the selective catalytic reduction (SCR) subsystem. The condensed contaminants ( The particles will be converted into particulate matter and collected by the fly ash resistivity correction and synergistic dust removal model of the subsequent electrostatic precipitator (ESP) subsystem.

[0051] Therefore, the correction of fly ash resistivity of particulate matter in the electrostatic precipitator subsystem and the synergistic dust removal model are necessary. The construction process is as follows: The fly ash resistivity correction and synergistic dust removal model (electrostatic precipitator subsystem) for particulate matter calculates the collection efficiency of particulate matter and associated pollutants by correcting the fly ash resistivity and using the improved Deutsch formula. First, calculate the comprehensive correction factor for resistivity on dust removal performance. ,Right now:

[0052] In the formula, This is a correction factor for dust removal conditions, used to reflect the influence of fly ash properties on the driving speed; is the sensitivity coefficient, and is a constant, used to characterize the degree of influence of carbon content fluctuations on resistivity and dust removal performance; The carbon content of fly ash is calculated using soft measurement in the initial steps described above. The baseline carbon content under design conditions is denoted as , which is a constant.

[0053] Secondly, the simplified Doychi formula is used to calculate the synergistic dust removal efficiency, which is the theoretical particulate matter removal efficiency. ,Right now:

[0054] In the formula, For the dust collection area, The area of ​​the electrode plate. The flue gas flow rate at the inlet of the electrostatic precipitator subsystem; The reference driving speed is determined by the inherent parameters of the equipment; This is a temperature correction term, usually expressed as This is used to reflect the promoting effect of decreased flue gas viscosity on dust settling under low-temperature conditions. Here, is the temperature sensitivity coefficient, and is a constant. Based on the design temperature, This refers to the flue gas temperature at the inlet of the electrostatic precipitator subsystem.

[0055] Therefore, the theoretical baseline concentration of particulate matter at the outlet of the electrostatic precipitator subsystem The calculation formula is as follows:

[0056] In the formula, This represents the particulate matter concentration at the inlet of the electrostatic precipitator subsystem.

[0057] In addition, since some pollutants (Hg and VOCs removed in the previous step) are adsorbed on the surface of fly ash and captured with it, their concentration at the outlet of the electrostatic precipitator (ESP) subsystem is also high. The calculation is as follows:

[0058] In the formula, It is Hg or VOCs, therefore These represent the concentrations of Hg or VOCs at the inlet of the electrostatic precipitator (ESP) subsystem, respectively. As a synergistic trapping factor, it is related to the carbon content of fly ash. It is directly proportional to the strength of the physical adsorption capacity.

[0059] (3) Construction of pollutant speciation transformation mechanism model for wet flue gas desulfurization subsystem, including sulfur dioxide gas-liquid mass transfer removal model and mercury and dichloromethane synergistic removal model, specifically: 1) Sulfur dioxide Gas-liquid mass transfer and removal model Sulfur dioxide was calculated using a simplified mass transfer rate equation. Theoretical removal efficiency ,Right now:

[0060] In the formula, The comprehensive mass transfer constant of sulfur dioxide; The liquid-to-gas ratio is calculated from the operating current and flow rate of the slurry circulation pump collected in the initial steps above. This is the pH correction factor for the slurry; it reflects that the higher the alkalinity of the slurry, the higher the sulfur dioxide content. The greater the driving force absorbed, the better.

[0061] Therefore, the theoretical baseline concentration of sulfur dioxide at the outlet of the wet flue gas desulfurization subsystem is... The calculation formula is as follows:

[0062] In the formula, This represents the sulfur dioxide concentration at the inlet of the wet flue gas desulfurization subsystem, and also the sulfur dioxide concentration at the outlet of the previous electrostatic precipitator subsystem.

[0063] 2) Co-removal model of mercury and dichloromethane ,Right now:

[0064] In the formula, The theoretical removal efficiency is the synergistic removal efficiency of mercury and dichloromethane. This value is typically set between 90% and 95% as a baseline washing efficiency. The re-release rate is affected by the concentration of reducing substances in the slurry, and its simplified calculation formula is as follows: , The re-release coefficient is related to the sulfite concentration in the slurry. The temperature sensitivity coefficient of the slurry. This refers to the temperature of the slurry.

[0065] Therefore, the theoretical baseline concentration of elemental mercury at the outlet of the wet flue gas desulfurization subsystem The calculation formula is as follows:

[0066] In the formula, This represents the concentration of divalent mercury ions at the inlet of the wet flue gas desulfurization subsystem, and also the concentration of divalent mercury ions at the outlet of the previous electrostatic precipitator subsystem. The monovalent mercury concentration at the inlet of the wet flue gas desulfurization subsystem is also the monovalent mercury concentration at the outlet of the previous electrostatic precipitator subsystem.

[0067] Theoretical baseline concentration of dichloromethane at the outlet of the wet flue gas desulfurization subsystem The calculation formula is as follows:

[0068] In the formula, This represents the dichloromethane concentration at the inlet of the wet flue gas desulfurization subsystem, and also the dichloromethane concentration at the outlet of the previous electrostatic precipitator subsystem.

[0069] Step 3: Use the Long Short-Term Memory Network model as the error compensation prediction model, and train the model using historical time-series feature vectors to learn the nonlinear deviation between the actual monitored values ​​and the theoretical benchmark concentration in the historical time-series feature vectors. After generating the trained error compensation prediction model, input the time-series feature vectors under the current operating conditions to generate the concentration prediction residual values ​​of each pollutant under the current operating conditions.

[0070] In this embodiment, to address the problem that the aforementioned standalone pollutant speciation transformation mechanism models are insufficient to characterize equipment aging (such as catalyst activity decay) and complex nonlinear disturbances, this step constructs an error-compensated prediction model based on a Long Short-Term Memory (LSTM) network model. During training, the input layer of this model is used to receive historical time-series feature vectors. Then, through the model's special gating structure—the forget gate, input gate, and output gate—the large lag and time-varying characteristics in the historical time-series feature vector are captured (i.e., the nonlinear deviation between the actual monitored values ​​and the theoretical baseline concentration in the historical time-series feature vector is learned). Finally, using the output layer, the nonlinear residual value that the above mechanism model failed to represent is output, defined as:

[0071] in, For nonlinear residual values, These are actual monitored values. This is the theoretical baseline concentration.

[0072] Therefore, the model is trained using historical time-series feature vectors, enabling it to dynamically correct the biases of the mechanistic model. Meanwhile, if the time-series feature vectors under the current operating conditions are input, the concentration prediction residuals of each pollutant under the current operating conditions can be generated.

[0073] Step 4: Obtain the deviation between the historical time-series feature vector under the historical training conditions and the time-series feature vector under the current conditions, calculate the confidence factor of the error compensation prediction model, and perform weighted fusion of the theoretical baseline concentration and the concentration prediction residual value. After generating the fusion result, apply physical boundary constraints for verification, and generate the optimized mixed prediction concentration value of each pollutant, thereby providing key environmental verification basis for the global optimization process in subsequent steps.

[0074] In this embodiment, to address the poor generalization ability of a single data-driven model under unknown operating conditions (such as drastic fluctuations in coal quality), this step employs an adaptive weighted fusion strategy based on operating condition similarity. The model structure and principle are as follows: Figure 3 As shown: First, the similarity of operating conditions is calculated; the arithmetic mean of all historical time-series feature vectors under the historical training conditions in step three is used as the feature center vector. Calculate the time-series feature vector under the current operating condition. With feature center vector European distance This allows for the quantification of the deviation between the current operating conditions and historical training conditions.

[0075] Secondly, an adaptive confidence factor is constructed; based on the deviation, an exponential decay function is used to construct the confidence factor of the error compensation prediction model. The calculation formula is as follows:

[0076] In the formula, It is an exponential function. This is the preset bandwidth parameter.

[0077] Furthermore, when the deviation exceeds the preset deviation threshold (it is determined to be an unknown working condition). The value adaptively decreases, making the prediction results more dependent on the physical mechanism; conversely, it increases. To fully utilize the high precision of the data model; Then, weighted fusion and boundary verification are performed. The fusion result is generated by weighting and fusing the theoretical baseline concentration and the concentration prediction residuals according to the following formula. ,Right now:

[0078] In the formula, The theoretical baseline concentration, i.e. Figure 3 Mechanism baseline concentration, For concentration prediction residuals; Finally, physical boundary constraints are applied to verify the fusion results, specifically as follows: Determine if the fusion result is less than 0. If so, force the output to 0 and use it as the optimized mixed predicted concentration value of each pollutant. Otherwise, output the fusion result normally and use it as the optimized mixed predicted concentration value of each pollutant.

[0079] Step 5: Using the ammonia injection flow rate, outlet flue gas temperature, operating voltage, number of slurry circulation pumps in operation, and slurry pH setpoint of the environmental protection subsystem as optimization variables, establish an objective function that minimizes the total cost of electricity consumption, material consumption, and equipment wear and tear risk. Under the constraints of environmental limits for pollutant emission concentrations, equipment safety operation boundaries, and pollutant form transformation correlations, perform multi-objective optimization to generate the optimal control parameter set for the intelligent environmental protection island of the coal-fired power plant.

[0080] In this embodiment, the process of this step is as follows: Figure 4 As shown, with the goal of achieving a win-win situation for both the economy and environmental protection of the smart environmental island, an objective function and constraints for minimizing the total global cost are established: First, the objective function Taking into account indicators including energy consumption, material consumption, and equipment wear and tear, as shown in the following formula:

[0081] In the formula, To obtain the minimum value, This is the total cost, including electricity consumption cost, material consumption cost, and equipment wear and tear risk cost. For electricity consumption costs, For material consumption costs, Costs associated with equipment wear and tear risks.

[0082] The formula for calculating electricity consumption cost is as follows:

[0083] In the formula, The unit on-grid electricity price; This refers to the flue gas volumetric flow rate; The overall system resistance pressure drop is related to the outlet flue gas temperature. Negative correlation; For the efficiency of the induced draft fan; The number of slurry circulation pumps in operation for the wet flue gas desulfurization subsystem; This refers to the rated power of a single circulating pump. This refers to the operating voltage of the electrostatic precipitator subsystem. This refers to the operating current of the electrostatic precipitator. The modulation period.

[0084] The formula for calculating material consumption costs is as follows:

[0085] In the formula, , , These are the unit prices for liquid ammonia, limestone, and process water, respectively. This refers to the ammonia injection flow rate at the outlet of the selective catalytic reduction subsystem. , These are the limestone slurry supply flow rate and process water flow rate of the wet flue gas desulfurization subsystem, respectively. , Related to the sulfur dioxide concentration at the inlet of the wet flue gas desulfurization subsystem; This is the pH setting value for the slurry in the wet flue gas desulfurization subsystem.

[0086] The formula for calculating the risk cost of equipment loss is as follows:

[0087] In the formula, This refers to the equipment corrosion depreciation factor. It is an exponential function; Risk-sensitive factors; This is the dew point temperature of ammonium bisulfate; The operating temperature of the selective catalytic reduction subsystem; and when the operating temperature When approaching or falling below the safety boundary, It grows exponentially, thus limiting unsafe, low-cost operating strategies through high-cost penalties during the optimization process.

[0088] Therefore, the established equipment wear and tear risk cost, based on the ammonium bisulfate formation potential and acid dew point corrosion risk predicted in the above steps, quantifies the equipment life loss into economic cost, thus avoiding damage to equipment life by simply pursuing low material consumption.

[0089] Secondly, for the aforementioned global cost objective function, a corresponding set of multidimensional inequality constraints is established. The core of this constraint set lies in using the optimized mixed predicted concentration values ​​of each pollutant generated in step four as state feedback variables, comparing them in real-time with the specified environmental protection limits for pollutant emission concentrations to construct a strong constraint boundary. Therefore, the constraints cover: 1) Environmental protection limit constraints for the emission concentrations of conventional pollutants and three types of unconventional pollutants: Hg, SO3, and VOCs, expressed as: , For the current optimized control variable vector, For control variable vector Next, the Optimized mixed predicted concentration values ​​of the pollutants. For the first 1) Environmental limits for the emission concentration of various pollutants; 2) Equipment safety operation boundary constraints, including the requirement that the SCR operating temperature must be higher than the ABS generation temperature range and that the set values ​​of each control actuator must be within the upper and lower limits of the corresponding physical stroke of the control actuator. The corresponding expressions are as follows: , , For control variable vector Below, the dew point temperature of ammonium bisulfate ABS , For the first The setpoints of each control actuator (such as valve switching, pump frequency). , The first The upper and lower limits of the physical travel of each control actuator; 3) Pollutant morphology transformation correlation constraints, namely the complete pollutant morphology transformation mechanism model constructed in step two.

[0090] Finally, the objective function is optimized using either Particle Swarm Optimization (PSO) or Differential Evolution (DE) algorithm, based on the ammonia injection flow rate. FGC outlet flue gas temperature ESP secondary voltage (Also refers to operating voltage), number of WFGD slurry circulation pumps in operation slurry pH setting value To optimize the variables, the optimal set of control parameters can be obtained by solving in a multidimensional solution space.

[0091] Step Six: Send the optimal control parameter set to the distributed control system of the intelligent environmental protection island of the coal-fired power plant to achieve optimized control of each environmental protection subsystem for treating pollutants and reduce pollutant emissions.

[0092] In this embodiment, this step involves sending the optimal control parameter set obtained in step five to the distributed control system (DCS) of the smart environmental protection island to adjust the actuators of each environmental protection subsystem. To ensure the safe execution of the control strategy in the industrial field, this step includes two major safety execution strategies: instruction smoothing and actuator safety verification. The operation is as follows: Determine whether the deviation between the target values ​​of each control parameter under the optimal control parameter group and the target values ​​of the control parameters currently being executed by the distributed control system exceeds the corresponding preset threshold. If so, abandon the direct issuance of the optimal control parameter target values ​​and use a weighted iterative method to generate the execution instructions for the current control cycle, i.e.:

[0093] in, The execution instruction for the current control cycle. The values ​​of the control parameters currently being executed by the distributed control system. The preset smoothing coefficient, The target value of the control parameter in the optimal control parameter set; Simultaneously, it determines whether the execution instructions for the current control cycle are within the set execution instruction range corresponding to each environmental protection subsystem. If so, the execution instructions for the current control cycle are executed; otherwise, the boundary truncation strategy is executed, generating truncation execution instructions for the current control cycle and triggering an over-limit alarm.

[0094] in, The instruction to be executed is the truncation instruction for the current control cycle. , Take the minimum value and the maximum value respectively. , These are the upper and lower limits of the execution commands corresponding to each environmental protection subsystem; Otherwise, directly execute the target value instructions for each control parameter under the optimal control parameters.

[0095] To verify the effectiveness of the proposed method for reducing total pollutant emissions in environmental islands based on morphological transformation and residual compensation, the following experiments were conducted: Taking a 660MW ultra-low emission coal-fired power unit in Zhejiang Province as an example, this unit is configured with a tangential combustion / opposed combustion boiler and is equipped with denitrification, low-temperature economizer, electrostatic precipitator, and limestone-gypsum wet desulfurization device. Based on this coal-fired power unit, a comprehensive pollutant emission reduction system for the environmental island was built based on the morphological transformation mechanism and data residual compensation.

[0096] First, data acquisition and preprocessing were performed: A total of 30 consecutive days of DCS operation data and CEMS pollutant emission data for the unit were collected, with a sampling period of 5 seconds, resulting in approximately 518,000 data entries. Collected variables included unit load, total coal consumption, pulverizer current, furnace oxygen content, and SCR inlet / outlet NO. x A total of 156 measurement points were used, including concentration, ammonia injection flow rate, flue gas temperature at the inlet and outlet of the air preheater, water temperature at the inlet and outlet of the FGC, secondary voltage / current of the ESP, pH value of the WFGD slurry, current of the circulating pump, and SO2 and dust concentration at the outlet.

[0097] To address the challenge of online measurement of VOCs and fly ash carbon content at the furnace outlet, a training dataset was first constructed using offline fly ash sampling and testing data, along with short-term measured data from a portable VOCs analyzer. A soft sensor model was then established using the Support Vector Machine (SVM) regression method. The input variables were boiler load, oxygen content, and flame spectral signals. The model output estimated initial VOCs concentrations and fly ash carbon content over the entire time period, thus supplementing the inlet boundary data of the mechanistic model.

[0098] Subsequently, Daubechies wavelet transform was used to perform multi-scale denoising on the original signal to remove high-frequency noise generated by on-site electromagnetic interference; the max-min normalization method was used to map all input features to the [0, 1] interval.

[0099] Next, the dual-drive prediction model is constructed and trained. The cleaned data is input into the full-process mechanism model. For the SCR subsystem, NO is calculated using the Arrhenius law. x The theoretical removal rate; based on multi-component competitive adsorption kinetics, the removal rates of VOCs and elemental mercury (Hg) were calculated. 0 The oxidation efficiency on the catalyst surface was assessed; for the FGC and ESP subsystems, the condensation phase change ratio of SO3 and semi-volatile organic compounds was calculated based on thermodynamic equilibrium; a correction coefficient for fly ash resistivity due to reduced flue gas temperature was introduced, and the synergistic capture efficiency of ESP for particulate matter, particulate mercury, and condensed aerosols was calculated using the corrected Deutsch formula; for the WFGD subsystem, the gas-liquid mass transfer rate of SO2 was calculated based on the two-film theory, and a model of divalent mercury (Hg) in the slurry was constructed. 2+ A competition model for the reduction and re-release of NO was developed. This yielded the results for conventional pollutants (NOx) at various time points. x The theoretical baseline concentration of all pollutants (Hg, SO2, particulate matter) and unconventional pollutants (Hg, SO3, VOCs) at the outlet of the environmental protection subsystem.

[0100] An LSTM residual compensation model was built using the Python (V3.8) and PyTorch (V1.10) deep learning frameworks. The model structure is as follows: input layer (time step = 12, i.e., data from the past minute), 2 LSTM hidden layers (128 neurons per layer, Tanh activation function), and 1 fully connected layer (output dimension = 6, corresponding to NO). x (Residuals of SO2, particulate matter, SO3, Hg, and VOCs), output layer.

[0101] The training, validation, and test sets were divided in a 7:2:1 ratio. The Adam optimizer was used with an initial learning rate of 0.001, and the mean squared error (MSE) of the predicted residuals to the actual residuals was employed as the loss function. The model converged after 100 epochs of training.

[0102] During the testing phase, the similarity of operating conditions (Euclidean distance) is calculated in real time, the confidence factor is dynamically adjusted, the mechanism baseline value and the residual value predicted by LSTM are weighted and fused, and the final result is output after physical boundary verification.

[0103] Furthermore, to verify the accuracy of the dual-drive model, data from different load conditions in the test set were selected for verification. This embodiment uses the dual-drive prediction model to predict SO2 and NO in flue gas. x Taking the verification of the predicted emission concentrations of three types of pollutants, namely particulate matter, as an example, by Figure 5 The scatter plot comparing the predicted and actual values ​​shows a high degree of agreement between the predicted and actual CEMS monitoring values ​​for the three pollutants. The model is able to capture the trend of drastic fluctuations in pollutant levels, demonstrating good prediction accuracy and generalization ability. Statistical results show that the mean square error of the predicted values ​​for the three pollutants is 0.78 mg / m³. 3 3.6 mg / m 3 and 0.12 mg / m 3 This fully demonstrates the robustness and accuracy of the "mechanism + data" dual-drive mode under complex working conditions.

[0104] Then, global collaborative optimization and cost control are performed. Based on the high-precision prediction results, the particle swarm optimization (PSO) algorithm or differential evolution algorithm (DE) is used to globally optimize the system operating parameters. The objective function is: .in, The potential for ammonium bisulfate (ABS) formation and the risk of FGC acid dew point corrosion are converted into equipment depreciation costs (for example, a 1% increase in ABS formation risk per hour corresponds to an increase of 50 yuan in equipment lifespan cost). Constraints include: NO x Emissions <40 mg / Nm 3 SO2 < 30 mg / Nm 3 Particulate matter <2 mg / Nm3 Hg < 0.05 mg / Nm 3 SO3 < 5 mg / Nm 3 VOCs < 2 mg / Nm 3 Simultaneously, the SCR reaction temperature is constrained to be at least 10°C higher than the critical ABS formation temperature. Optimization variables include ammonia injection flow rate, FGC outlet flue gas temperature setpoint, secondary voltages of each electric field in the ESP, WFGD slurry pH value, and the combination of circulating pump operation.

[0105] The optimized control commands were sent to the DCS system for closed-loop control testing. To visually demonstrate the dual environmental and economic benefits brought about by the multi-objective optimization control method proposed in this invention, the emission concentrations of multiple pollutants under different unit loads (500MW, 600MW, 700MW, 800MW, 900MW) before and after optimization, as well as the unit hour flue gas treatment cost before and after optimization, were statistically compared.

[0106] With SO2, NO x Taking particulate matter and other pollutants as examples, Table 1 compares the emission concentrations of multiple pollutants before and after optimization using different algorithms under different unit loads: Table 1 Comparison of pollutant emission concentrations before and after optimization using different methods

[0107] As shown in Table 1, regardless of whether the particle swarm optimization algorithm or the differential evolution algorithm is used, the SO2 emission concentration is controlled at 20 mg / m³. 3 The following is far superior to 30 mg / m². 3 The emission limits. NO2 within the full load range. x The emission concentration has remained relatively stable at 30-40 mg / m³ 3 Within the specified range, the dust concentration remained consistently at 1.1 mg / m³. 3 The following are extremely low levels. This indicates that although the method adjusted the control parameters to reduce costs (such as reducing ammonia injection and reducing fan power consumption), all adjustments were strictly limited within environmental constraints. There were no cases of emissions exceeding standards due to excessive pursuit of cost savings, and both methods demonstrated excellent multi-pollutant synergistic removal capabilities.

[0108] In addition, such as Figure 6 As shown, within the full load range of 500MW to 900MW, the operating costs after adopting the optimization method (PSO or DE) of this invention are significantly lower than the original operating costs, and the cost reduction ratio is more than 10%, with a significant overall cost reduction effect.

[0109] The cost reduction mainly stems from three aspects: (1) Reduced material consumption: By precisely controlling the amount of ammonia injected and the pH value of the slurry, "excessive ammonia injection" and "excessive slurry supply" are avoided, significantly reducing the consumption of ammonia water and limestone. (2) Reduced energy consumption: The combination of flue gas temperature and circulating pump at the FGC outlet is optimized, reducing the power consumption of fans and water pumps while ensuring compliance with standards. (3) Reduced hidden costs: The optimization strategy avoids the ABS generation area and acid dew point corrosion area, effectively extending the cleaning and replacement cycle of the air preheater and heat exchanger. This long-cycle hidden benefit is more considerable.

[0110] In summary, the proposed method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation not only utilizes a mechanism-data dual-driven architecture to achieve accurate and coordinated prediction of all conventional and unconventional pollutants in flue gas from coal-fired power plants, but also, based on the constructed full-process prediction model and coupled with a global collaborative optimization method, achieves effective intelligent control of each subsystem of the environmental island. Ultimately, it not only ensures that the emissions of various pollutants meet national ultra-low emission standards, but also minimizes the full life-cycle operating costs of the smart environmental island by quantifying equipment wear and tear risks. Furthermore, compared to traditional methods relying on CEM (Conditional Emission Model), the proposed method offers significant advantages. The method of monitoring data with lag and adjusting manually or simply using PID control only for a single pollutant has advantages such as strong physical consistency, fast response speed, strong generalization ability, and multi-system coordination. It can effectively overcome the shortcomings of CEMS in monitoring unconventional pollutants online, and at the same time greatly reduce material waste and equipment damage caused by control lag. In addition, by equipping coal-fired units with the intelligent environmental protection island collaborative emission reduction system proposed in this invention, the digitalization level and cost-effectiveness of power plant environmental protection facilities can be significantly improved, and the collaborative emission reduction and intelligent control of the full spectrum of pollutants in flue gas under complex and variable operating conditions can be achieved safely and reliably.

[0111] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0112] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for reducing total pollutant emissions in environmental islands based on morphological transformation and residual compensation, characterized in that: Includes the following steps: Acquire real-time operation data of the smart environmental protection island of a coal-fired power plant, generate time-series feature vectors, and set inlet boundary conditions; For each environmental protection subsystem of the smart environmental protection island of a coal-fired power plant, a pollutant form transformation mechanism model is constructed, real-time operation data is input, and combined with the inlet boundary conditions, the theoretical removal efficiency or transformation efficiency of each environmental protection island subsystem for each pollutant is generated, so as to calculate the theoretical baseline concentration of each pollutant at the outlet of each environmental protection island subsystem. The Long Short-Term Memory Network model is used as the error compensation prediction model. The model is trained using historical time series feature vectors to learn the nonlinear deviation between the actual monitored values ​​and the theoretical benchmark concentration in the historical time series feature vectors. After generating the trained error compensation prediction model, the time series feature vector under the current operating conditions is input to generate the concentration prediction residual values ​​of each pollutant under the current operating conditions. The deviation between the historical time-series feature vector under the historical training conditions and the time-series feature vector under the current conditions is obtained. The confidence factor of the error compensation prediction model is calculated to perform weighted fusion of the theoretical baseline concentration and the concentration prediction residual value. After generating the fusion result, physical boundary constraints are applied for verification to generate the optimized mixed prediction concentration value of each pollutant. Using the ammonia injection flow rate, outlet flue gas temperature, operating voltage, number of slurry circulation pumps in operation, and slurry pH setpoint of the environmental protection subsystem as optimization variables, an objective function is established to minimize the total cost of electricity consumption, material consumption, and equipment wear and tear risk. Under the constraints of environmental limits for pollutant emission concentrations, equipment safety operation boundaries, and pollutant form transformation correlations, a multi-objective optimization solution is performed to generate the optimal control parameter set for the intelligent environmental protection island of the coal-fired power plant. The optimal control parameter set is sent to the distributed control system of the intelligent environmental protection island of the coal-fired power plant to achieve optimized control of each environmental protection subsystem for treating pollutants and reduce pollutant emissions.

2. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 1, characterized in that, The temporal feature vector includes flame spectral features, which are free radical radiation intensity and spectral entropy extracted in a specific band based on the radiation energy distribution in the combustion zone of the furnace using fast Fourier transform or wavelet transform.

3. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 2, characterized in that, The inlet boundary conditions include furnace outlet flue gas parameters, raw coal composition data, and combustion intermediate product parameters. Among them, furnace outlet flue gas parameters include flow rate, temperature, carbon monoxide concentration, oxygen concentration, nitrogen oxide concentration, sulfur dioxide concentration, particulate matter concentration, fly ash carbon content, and dichloromethane concentration.

4. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 3, characterized in that, The formula for calculating the carbon content or dichloromethane concentration in fly ash is: in, The concentration of dichloromethane at the furnace outlet. The carbon content of fly ash at the furnace outlet. It is a nonlinear mapping function. For boiler load, This refers to the amount of oxygen at the furnace outlet. For the coal mill current, These are flame spectral characteristics.

5. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 1, characterized in that, The pollutant speciation transformation mechanism models include a nitrogen oxide reduction reaction model, a mercury oxidation speciation transformation model, a dichloromethane catalytic oxidation model, a fly ash resistivity correction and synergistic dust removal model for particulate matter, a sulfur dioxide gas-liquid mass transfer removal model, and a mercury and dichloromethane synergistic removal model, and their calculation formulas are as follows: in, This is a model for the reduction of nitrogen oxides in a selective catalytic reductant subsystem. The denitrification reaction rate constant is As the pre-factor, It is an exponential function. The activation energy of the reaction. Let be the ideal gas constant. The absolute temperature of the flue gas in the reaction zone of the selective catalytic reduction subsystem. This represents the theoretical removal efficiency of nitrogen oxides. This is the catalyst activity correction factor. The standard is the catalyst's rated design denitrification efficiency. This is the correction factor for the ammonia-nitrogen molar ratio. For catalyst volume, The flue gas volumetric flow rate, This is a model for the mercury oxidation form transformation in a selective catalytic reduction subsystem. The oxidation efficiency of elemental mercury. Here is the rate constant for the mercury oxidation reaction. The concentration of hydrogen chloride in the flue gas. This is a model for the catalytic oxidation of dichloromethane in a selective catalytic reductant subsystem. The catalytic oxidation rate of dichloromethane is given. Let be the intrinsic oxidation rate constant of dichloromethane. , , , respectively, are the adsorption equilibrium constants of dichloromethane, ammonia, and mercury on the catalyst surface. , , These represent the gaseous concentrations of dichloromethane, ammonia, and mercury, respectively. The conversion efficiency of dichloromethane, The concentration of dichloromethane at the furnace outlet. A model for correcting the fly ash resistivity of particulate matter and for collaborative dust removal in an electrostatic precipitator subsystem. This is a correction factor for dust removal operating conditions. The sensitivity coefficient, This refers to the carbon content of fly ash. Based on carbon content, This represents the theoretical removal efficiency of particulate matter. The area of ​​the electrode plate is... This refers to the flue gas flow rate at the inlet of the electrostatic precipitator subsystem. As the reference driving speed, For temperature correction, For temperature sensitivity coefficient, Based on the design temperature, The flue gas temperature at the inlet of the electrostatic precipitator subsystem. This is a model for the gas-liquid mass transfer removal of sulfur dioxide in a wet flue gas desulfurization subsystem. This represents the theoretical removal efficiency of sulfur dioxide. The comprehensive mass transfer constant of sulfur dioxide is... The liquid-to-gas ratio, This is the pH correction factor for the slurry. This is a model for the synergistic removal of mercury and dichloromethane in a wet flue gas desulfurization subsystem. This represents the theoretical removal efficiency of the synergistic removal of mercury and dichloromethane. Based on the washing efficiency, For re-release rate, The re-release coefficient, The temperature sensitivity coefficient of the slurry. This refers to the temperature of the slurry.

6. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 5, characterized in that, The theoretical baseline concentrations of each pollutant at the outlet of each environmental protection island subsystem include the theoretical baseline concentrations of nitrogen oxides, elemental mercury, and dichloromethane at the outlets of the selective catalytic reduction (SCR) subsystem, the electrostatic precipitator subsystem, the wet flue gas desulfurization (WGD) subsystem, and the wet flue gas desulfurization (WGD) subsystem. The calculation formulas are as follows: in, This represents the theoretical baseline concentration of nitrogen oxides at the outlet of the selective catalytic reduction subsystem. The concentration of nitrogen oxides at the inlet of the selective catalytic reduction subsystem. This represents the theoretical baseline concentration of elemental mercury at the outlet of the selective catalytic reduction subsystem. The concentration of elemental mercury at the inlet of the selective catalytic reduction subsystem. This represents the theoretical baseline concentration of dichloromethane at the outlet of the selective catalytic reduction subsystem. The concentration of dichloromethane at the inlet of the selective catalytic reduction subsystem. This represents the theoretical baseline concentration of particulate matter at the outlet of the electrostatic precipitator subsystem. The particulate matter concentration at the inlet of the electrostatic precipitator subsystem. This represents the theoretical baseline concentration of sulfur dioxide at the outlet of the wet flue gas desulfurization subsystem. This refers to the sulfur dioxide concentration at the inlet of the wet flue gas desulfurization subsystem. This represents the theoretical baseline concentration of elemental mercury at the outlet of the wet flue gas desulfurization subsystem. The concentration of divalent mercury at the inlet of the wet flue gas desulfurization subsystem. The concentration of monovalent mercury at the inlet of the wet flue gas desulfurization subsystem. This represents the theoretical baseline concentration of dichloromethane at the outlet of the wet flue gas desulfurization subsystem. This refers to the concentration of dichloromethane at the inlet of the wet flue gas desulfurization subsystem.

7. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 1, characterized in that, The process of using a long short-term memory network model as an error compensation prediction model and training it with historical time-series feature vectors to learn the nonlinear deviation between actual monitored values ​​and theoretical baseline concentrations in the historical time-series feature vectors is as follows: The historical time-series feature vector is input into the input layer of the Long Short-Term Memory (LSTM) network model. Through forgetting gates, input gates, and output gates, the large lag and time-varying characteristics of the historical time-series feature vector are captured. Finally, the output layer outputs the nonlinear residual value after learning the nonlinear residual relationship between the actual monitored value and the theoretical baseline concentration in the historical time-series feature vector. in, For nonlinear residual values, These are actual monitored values. This is the theoretical baseline concentration.

8. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 1, characterized in that, The process of obtaining the deviation between the historical time-series feature vector under historical training conditions and the time-series feature vector under current conditions, calculating the confidence factor of the error compensation prediction model, and then weighting and fusing the theoretical baseline concentration and the concentration prediction residual values ​​to generate the fusion result, followed by applying physical boundary constraints for verification, and finally generating the optimized mixed predicted concentration values ​​for each pollutant, is as follows: The arithmetic mean of the historical time series feature vectors under the historical training conditions is used as the feature center vector. The Euclidean distance between the time series feature vector under the current condition and the feature center vector is calculated and used as the deviation between the current condition and the historical training conditions. Based on the deviation, the confidence factor of the error compensation prediction model is calculated, namely: in, As the confidence factor, It is an exponential function. For deviation, For bandwidth parameters; Based on the confidence factor, the theoretical baseline concentration and the concentration prediction residuals are weighted and fused to generate the fusion result, namely: in, For the fusion result, The theoretical baseline concentration, For concentration prediction residuals; Determine if the fusion result is less than 0. If so, force the output to 0 and use it as the optimized mixed predicted concentration value of each pollutant. Otherwise, output the fusion result normally and use it as the optimized mixed predicted concentration value of each pollutant.

9. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 1, characterized in that, The formula for calculating the objective function that minimizes the total cost of electricity consumption, material consumption, and equipment depreciation risk is as follows: in, To obtain the minimum value, This is the total cost, including electricity consumption cost, material consumption cost, and equipment wear and tear risk cost. For electricity consumption costs, For material consumption costs, Costs associated with equipment wear and tear risks. For unit grid connection electricity price, The flue gas volumetric flow rate, The outlet flue gas temperature of the flue gas cooler subsystem. For the overall system resistance pressure drop, For the efficiency of the induced draft fan, This refers to the number of slurry circulation pumps in operation for the wet flue gas desulfurization subsystem. This refers to the rated power of a single circulating pump. This refers to the operating voltage of the electrostatic precipitator subsystem. This is the operating current of the electrostatic precipitator. For the modulation period, , , These are the unit prices for liquid ammonia, limestone, and process water, respectively. This refers to the ammonia injection flow rate at the outlet of the selective catalytic reduction subsystem. , These are the limestone slurry supply flow rate and process water flow rate of the wet flue gas desulfurization subsystem, respectively. The pH setpoint for the slurry in the wet flue gas desulfurization subsystem. This is the equipment corrosion depreciation factor. It is an exponential function. As a risk-sensitive factor, This refers to the dew point temperature of ammonium bisulfate. To prevent condensation, a safe temperature margin is required. This refers to the operating temperature of the selective catalytic reduction subsystem.

10. The method for total pollutant reduction in environmental islands based on morphological transformation and residual compensation according to claim 1, characterized in that, The optimal control parameter set is distributed to the distributed control system of the intelligent environmental protection island in the coal-fired power plant to achieve optimized control of various environmental protection subsystems for pollutant treatment, thereby reducing pollutant emissions. Determine whether the deviation between the target values ​​of each control parameter under the optimal control parameter group and the target values ​​of the control parameters currently being executed by the distributed control system exceeds the corresponding preset threshold. If so, abandon the direct issuance of the optimal control parameter target values ​​and use a weighted iterative method to generate the execution instructions for the current control cycle, i.e.: in, The execution instruction for the current control cycle. The values ​​of the control parameters currently being executed by the distributed control system. The preset smoothing coefficient, The target value of the control parameter in the optimal control parameter set; Simultaneously, it determines whether the execution instructions for the current control cycle are within the set execution instruction range corresponding to each environmental protection subsystem. If so, the execution instructions for the current control cycle are executed; otherwise, the boundary truncation strategy is executed, generating truncation execution instructions for the current control cycle and triggering an over-limit alarm. in, The instruction to be executed is the truncation instruction for the current control cycle. , Take the minimum value and the maximum value respectively. , These are the upper and lower limits of the execution commands corresponding to each environmental protection subsystem; Otherwise, directly execute the target value instructions for each control parameter under the optimal control parameters.

Images