A method for constructing a pollutant emission model in the process of municipal solid waste combustion

Abstract

The application provides a kind of urban solid waste combustion process pollutant emission model construction method, belongs to urban solid waste combustion field, including: constructing simulation model, carrying out multiple working condition orthogonal experiment, data preprocessing, data expansion, obtaining IT2FBLS network output, input-output matching, data screening, pollutant emission main model training, obtaining main model output and simulation error, calculating mean square error, determining segmentation variable, updating LRDT algorithm weight matrix, calculating leaf node output and carrying out weighted summation to main model output and leaf node output.The application reduces the time cost of simulation process by constructing simulation model, and improves the accuracy of emission model by using LRDT algorithm to establish compensation model.

Description

Technical Field

[0001] The present invention relates to the field of municipal solid waste combustion, and in particular to a method for constructing a pollutant emission model in a municipal solid waste combustion process. Background Art

[0002] The increasing accumulation of municipal solid waste and the inherent flaws of traditional landfill technology have made municipal solid waste incineration, with its advantages of harmlessness, volume reduction, and resource utilization, a key supporting technology for the green and sustainable development of cities around the world. Currently, most solid waste incineration plants use manual control by domain experts to achieve safe and stable operation. Compared with automatic combustion control systems, manual control modes have disadvantages such as variability in control effects, time lags in control, and suboptimal operation. Therefore, it is difficult for solid waste incineration plants to maintain long-term stable and optimized operating conditions, making it difficult to consistently meet pollution emission standards. Frequent fluctuations in operating conditions pose significant challenges to incineration equipment and operation and maintenance management.

[0003] The foundation for conducting research on intelligent optimization control algorithms is to establish a reliable and interpretable model of the municipal solid waste incineration process. At the industrial site, domain experts use a built-in "brain model" based on multimodal data such as process data, flame videos, work logs, and existing knowledge of incineration mechanisms to predict the current operating conditions and then adopt effective control methods to maintain its safe operation. Obviously, the aforementioned models established using single simulation mechanism data or real process data are difficult to fully characterize the dynamic characteristics of the municipal solid waste incineration process. However, there is currently no research on the establishment of a model for multiple conventional pollutants in the municipal solid waste incineration process using mixed mechanism and process data. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide a method for constructing a pollutant emission model for the municipal solid waste combustion process, reduce the time cost of the simulation process by constructing a simulation model, and use the LRDT algorithm to establish a compensation model to improve the accuracy of the emission model.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for constructing a pollutant emission model for a municipal solid waste combustion process, comprising:

[0007] Perform simulation based on the preset physical model to obtain a full-process numerical simulation model;

[0008] Conducting multi-operating condition orthogonal experiments based on the full-process numerical simulation model to obtain multi-operating condition mechanism data;

[0009] Preprocessing the multi-operating condition mechanism data to obtain preprocessed mechanism data; the preprocessing includes: mean processing and summation processing;

[0010] Expanding the pre-processed mechanism data using an equidistant interpolation technique to obtain virtual sample points;

[0011] Inputting the virtual sample points into a preset first IT2FBLS network to obtain a network output;

[0012] Matching the virtual sample points with the network output to obtain original samples;

[0013] Selecting the original samples based on a preset physical parameter threshold to obtain a virtual process data set;

[0014] Training a preset second IT2FBLS network based on the virtual process data set to obtain a main pollutant emission model;

[0015] Inputting pre-collected real data into the pollutant emission master model to obtain the master model output and simulation error;

[0016] Calculate a mean square error based on the simulation error to obtain a mean square error set;

[0017] Determine the minimum mean square error in the mean square error set as the segmentation variable, and return to the step of "calculating the mean square error based on the simulation error to obtain the mean square error set" until the number of segmentation variables meets the preset minimum number of samples, and output all the segmentation variables;

[0018] Based on the segmentation variables and the preset weight formula, a regularized least squares loss function is used to update the weight matrix of the preset LRDT algorithm;

[0019] Calculate leaf node outputs of the LRDT algorithm based on the weight matrix;

[0020] The main model output and the leaf node output are weightedly summed to obtain a target emission output and a target pollutant emission model.

[0021] Preferably, simulation is performed based on a preset physical model to obtain a full-process numerical simulation model, including:

[0022] The FLIC software is used to perform numerical simulations on the preset water evaporation model, volatile analysis model, volatile combustion model, and coke combustion model to obtain combustion simulation results.

[0023] Based on the combustion simulation results, FLUENT software is used to perform numerical simulations on the preset discrete coordinate radiation model, standard k-ε model, and eddy dissipation conceptual model to obtain thermal radiation distribution data;

[0024] Inputting the thermal radiation distribution data into the FLIC software, and returning to the step of "using the FLIC software to perform numerical simulations on a preset water evaporation model, a volatile analysis model, a volatile combustion model, and a coke combustion model to obtain combustion simulation results" until the temperature in the thermal radiation distribution data satisfies a preset temperature error, and then outputting the temperature field, velocity field, and concentration field;

[0025] Based on the temperature field, the velocity field, and the concentration field, Aspen Plus software is used to simulate the storage and incineration of municipal solid waste, heat exchange in waste heat boilers, and flue gas treatment and emission to obtain the full-process numerical simulation model.

[0026] Preferably, the water evaporation model is: ; The volatile analysis model is: ; The volatile matter combustion model is: ; The coke combustion model is: ; is the water evaporation rate; is the particle surface area; is the convective mass transfer coefficient; and are the water concentrations in the solid and gas phases, respectively; It is the sum of convective heat transfer and radiation heat transfer; The heat required to evaporate water in the solid; is the solid temperature; is the volatile analysis rate; is the solid density; is the pre-exponential factor; is the activation energy; is the universal gas constant; is the ambient temperature; is the overall mixing rate of the gas phase in the bed; is the mixing rate constant; is the gas density; is the diffusion coefficient of volatile molecules; is the bed void ratio; is the particle diameter; is the gas flow rate; is the fuel concentration; is the O2 concentration; and is the stoichiometric coefficient in the reaction; is the ratio of CO to CO2.

[0027] Preferably, the discrete coordinate radiation model is: ; The standard k-ε model is: and ; The eddy dissipation conceptual model is: ; is the radiation intensity; and are position vector and direction vector respectively; is the absorption coefficient; is the diffusion coefficient; is the refractive index; is the Boltzmann constant; is the phase function; is a fixed angle; is the gas density; is the turbulent kinetic energy; for speed; and is the coordinate axis; is the turbulent viscosity; For material The quality score of For material The diffusion flux of is the additional generation rate, is the net production rate; and are the turbulent Prandtl coefficients, which are 1.0 and 1.3 respectively; is the terminal flow energy term generated by the velocity gradient; is the terminal flow energy term generated by buoyancy; is the consumption of turbulent pulsation kinetic energy dispersion rate; is the coefficient of fluctuation; and Customize source items for users; and are constants, taking values ​​of 1.44 and 1.92 respectively.

[0028] Preferably, the multi-operating condition mechanism data includes: urban solid waste composition, feed rate, grate speed, primary air temperature, primary air temperature, primary air temperature, primary air flow, primary air flow, primary air flow, primary air flow, secondary air temperature and secondary air flow.

[0029] Preferably, the equal-interval interpolation technique is: ; and Respectively The maximum and minimum values ​​of the input features; For the The number of interpolation points for each input feature; For the The interpolation interval for each input feature.

[0030] Preferably, the calculation formula of the segmentation variable is: ; is the loss function value; For the In the iteration The first sample eigenvalues; The first non-leaf node.

[0031] Preferably, the weight matrix is: ; 、 and Respectively Output, input and weight matrices of non-leaf nodes; The non-leaf node weight vector representing the first input feature; is the regularization coefficient that is not less than 0.

[0032] The present invention discloses the following technical effects:

[0033] The present invention provides a method for constructing a pollutant emission model for a municipal solid waste combustion process. By constructing a main pollutant emission model, the pollutant emission prediction problem is solved and a preliminary prediction of pollutant emissions is achieved. By updating the LRDT algorithm weight matrix to calculate the leaf node output and performing weighted summation of the main model output and the leaf node output, the main model prediction error problem is solved and the pollutant emission model is accurately constructed. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 A flow chart for constructing a pollutant emission model provided in an embodiment of the present invention;

[0036] Figure 2 A flow chart of the municipal solid waste combustion process provided by an embodiment of the present invention;

[0037] Figure 3 A flow chart of the modeling strategy provided by an embodiment of the present invention;

[0038] Figure 4 Aspen Plus software numerical simulation flow chart provided in an embodiment of the present invention;

[0039] Figure 5 This is a diagram of the IT2FBLS network structure provided by an embodiment of the present invention;

[0040] Figure 6 This is a scatter plot of some input features provided by the embodiment of the present invention. Figure 6 (a) is the mechanism data set obtained by multi-software coupling simulation. Figure 6 (b) is the virtual sample data set of mechanism data;

[0041] Figure 7 A PCC calorific value diagram provided by an embodiment of the present invention;

[0042] Figure 8 The test set fitting curve provided by the embodiment of the present invention is Figure 8 (a) is the fitting curve of CO emission concentration, Figure 8 (b) is the fitting curve of CO2 emission concentration, Figure 8 (c) is the fitting curve of SO2 emission concentration. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] The purpose of the present invention is to provide a method for constructing a pollutant emission model for the municipal solid waste combustion process, which reduces the time cost of the simulation process by constructing a simulation model and improves the accuracy of the emission model by establishing a compensation model using the LRDT (Linear Regression Decision Tree) algorithm.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Figure 1 The pollutant emission model construction flow chart provided in the embodiment of the present invention is as follows: Figure 1 As shown, the present invention provides a method for constructing a pollutant emission model in a municipal solid waste combustion process, comprising:

[0047] Perform simulation based on the preset physical model to obtain a full-process numerical simulation model;

[0048] Conducting multi-operating condition orthogonal experiments based on the full-process numerical simulation model to obtain multi-operating condition mechanism data;

[0049] Preprocessing the multi-operating condition mechanism data to obtain preprocessed mechanism data; the preprocessing includes: mean processing and summation processing;

[0050] Expanding the pre-processed mechanism data using an equidistant interpolation technique to obtain virtual sample points;

[0051] Inputting the virtual sample points into a preset first IT2FBLS (Interval type-2 fuzzy broadlearning system) network to obtain a network output;

[0052] Matching the virtual sample points with the network output to obtain original samples;

[0053] Selecting the original samples based on a preset physical parameter threshold to obtain a virtual process data set;

[0054] Training a preset second IT2FBLS network based on the virtual process data set to obtain a main pollutant emission model;

[0055] Inputting pre-collected real data into the pollutant emission master model to obtain the master model output and simulation error;

[0056] Calculate a mean square error based on the simulation error to obtain a mean square error set;

[0057] Determine the minimum mean square error in the mean square error set as the segmentation variable, and return to the step of "calculating the mean square error based on the simulation error to obtain the mean square error set" until the number of segmentation variables meets the preset minimum number of samples, and output all the segmentation variables;

[0058] Based on the segmentation variables and the preset weight formula, a regularized least squares loss function is used to update the weight matrix of the preset LRDT algorithm;

[0059] Calculate leaf node outputs of the LRDT algorithm based on the weight matrix;

[0060] The main model output and the leaf node output are weightedly summed to obtain a target emission output and a target pollutant emission model.

[0061] Furthermore, simulation is performed based on the preset physical model to obtain a full-process numerical simulation model, including:

[0062] The FLIC software is used to perform numerical simulations on the preset water evaporation model, volatile analysis model, volatile combustion model, and coke combustion model to obtain combustion simulation results.

[0063] Based on the combustion simulation results, FLUENT software is used to perform numerical simulations on the preset discrete coordinate radiation model, standard k-ε model, and eddy dissipation conceptual model to obtain thermal radiation distribution data;

[0064] Inputting the thermal radiation distribution data into the FLIC software, and returning to the step of "using the FLIC software to perform numerical simulations on a preset water evaporation model, a volatile analysis model, a volatile combustion model, and a coke combustion model to obtain combustion simulation results" until the temperature in the thermal radiation distribution data satisfies a preset temperature error, and then outputting the temperature field, velocity field, and concentration field;

[0065] Based on the temperature field, the velocity field, and the concentration field, Aspen Plus software is used to simulate the storage and incineration of municipal solid waste, heat exchange in waste heat boilers, and flue gas treatment and emission to obtain the full-process numerical simulation model.

[0066] Specifically, the water evaporation model is: ; The volatile analysis model is: ; The volatile matter combustion model is: ; The coke combustion model is: ; is the water evaporation rate; is the particle surface area; is the convective mass transfer coefficient; and are the water concentrations in the solid and gas phases, respectively; It is the sum of convective heat transfer and radiation heat transfer; The heat required to evaporate water in the solid; is the solid temperature; is the volatile analysis rate; is the solid density; is the amount of volatile analysis; is the pre-exponential factor; is the activation energy; is the universal gas constant; is the ambient temperature; is the overall mixing rate of the gas phase in the bed; is the mixing rate constant; is the gas density; is the diffusion coefficient of volatile molecules; is the bed void ratio; is the particle diameter; is the gas flow rate; is the fuel concentration; is the O2 concentration; and is the stoichiometric coefficient in the reaction; is the ratio of CO to CO2.

[0067] Specifically, the discrete coordinate radiation model is: ; The standard k-ε model is: and ; The eddy dissipation conceptual model is: ; is the radiation intensity; and are position vector and direction vector respectively; is the absorption coefficient; is the diffusion coefficient; is the refractive index; is the Boltzmann constant; is the phase function; is a fixed angle; is the gas density; is the turbulent kinetic energy; for speed; and is the coordinate axis; is the turbulent viscosity; For material The quality score of For material The diffusion flux of is the additional generation rate, is the net production rate; and are the turbulent Prandtl coefficients, which are 1.0 and 1.3 respectively; is the terminal flow energy term generated by the velocity gradient; is the terminal flow energy term generated by buoyancy; is the consumption of turbulent pulsation kinetic energy dispersion rate; is the coefficient of fluctuation; and Customize source items for users; and are constants, taking values ​​of 1.44 and 1.92 respectively.

[0068] Specifically, the multi-operating condition mechanism data include: urban solid waste composition, feed rate, grate speed, primary air temperature, primary air temperature, primary air temperature, primary air flow, primary air flow, primary air flow, primary air flow, secondary air temperature and secondary air flow.

[0069] Optionally, the equal-interval interpolation technique is: ; and Respectively The maximum and minimum values ​​of the input features; For the The number of interpolation points for each input feature; For the The interpolation interval for each input feature.

[0070] Specifically, the calculation formula of the segmentation variable is: ; is the loss function value; For the In the iteration The first sample eigenvalues; The first non-leaf node.

[0071] Specifically, the weight matrix is: ; 、 and Respectively Output, input and weight matrices of non-leaf nodes; The non-leaf node weight vector representing the first input feature; is the regularization coefficient that is not less than 0.

[0072] Preferably, the process flow of MSWI (Municipal Solid Waste Incineration) plant is as follows Figure 2 As shown, MSW (municipal solid waste) is transported by vehicles, weighed on a scale and discharged into a solid waste pool. After 3 to 7 days of biological fermentation and dehydration, the MSW in the solid waste pool is put into the hopper by a grab bucket; then, the feeder pushes the MSW to the grate, and it goes through three stages of drying, combustion, and incineration. In order to ensure that harmful substances in the high-temperature flue gas are fully decomposed and burned, the flue gas temperature should be controlled above 850°C, the flue gas stays for more than 2 seconds and has sufficient flue gas turbulence; then, the high-temperature flue gas enters the waste heat boiler, and the high-temperature steam generated by heat exchange drives the steam turbine generator set to generate electricity; then, the flue gas mixed with slaked lime and activated carbon enters the deacidification reactor, where a neutralization reaction occurs and dioxins and heavy metals in the flue gas are adsorbed; then, the flue gas particles, neutralization reactants and activated carbon adsorbents are removed in the bag filter; finally, the flue gas containing dust, CO, NO xWaste gases containing other substances, such as SO2, HCl, HF, Hg, Cd, and DXN, are discharged into the atmosphere through the chimney. Ash produced by incineration is transported to a slag pit by a slag lift and then transported by vehicle to a designated landfill. Specifically, 1) Storage, Transportation, and Fermentation Stage: Sanitation vehicles transport MSW from various urban collection points to the MSWI plant. After weighing and recording, the MSW is dumped from the unloading platform into the unfermented area of ​​the solid waste storage tank. A solid waste grab then mixes and stirs the MSW before transporting it to the fermentation area. The ash undergoes fermentation and dehydration for 3-7 days to ensure the low calorific value of the MSW during incineration. 2) Solid Waste Combustion Stage: A solid waste grabber deposits fermented MSW into the feed hopper, where it is pushed into the incinerator via a feeder. After drying, combustion 1, combustion 2, and the burnout grate, the combustible components in the MSW are completely burned. The required combustion air is injected from below the grate and into the center of the furnace by primary and secondary fans. The resulting ash falls from the end of the burnout grate into a slag collector, where it is water-cooled and then transferred to the slag pool. This stage requires strict control of flue gas temperature above 850°C, a high-temperature flue gas residence time in the furnace exceeding 2 seconds, and sufficient flue gas turbulence. 3) Waste Heat Exchange Stage: The high-temperature flue gas generated in the furnace is drawn into the waste heat boiler system by an induced draft fan. It passes through the superheater, evaporator, and economizer, where it undergoes heat exchange with the liquid water in the boiler drum to generate high-temperature steam. This process is then cooled to below 200°C at the waste heat boiler outlet (i.e., flue gas G1). 4) Steam Generation Stage: The high-temperature steam generated by the waste heat boiler drives the turbine generator, converting mechanical energy into electrical energy. This achieves self-sufficiency in plant-level electricity consumption and allows surplus electricity to be supplied to the grid, thereby generating economic benefits. 5) Flue Gas Purification Stage: The MSWI process involves a series of flue gas purification processes, including denitrification, desulfurization, heavy metal removal, dioxin adsorption, and dust (particulate matter) removal, ensuring that the resulting flue gas pollutant emissions meet national standards. 6) Flue Gas Emission Stage: After cooling and purification, the incineration flue gas (i.e., flue gas G2) is drawn by an induced draft fan and discharged into the atmosphere through a chimney.

[0073] refer to Figure 3 ,The modeling strategy proposed in this embodiment includes a simulation mechanism data ,acquisition module, a virtual process data generation module and a virtual-real data driven ,model construction module.

[0074] Furthermore, the simulation mechanism data acquisition module. This embodiment uses a multi-software coupling strategy to construct a full-process numerical simulation model of the MSWI. This involves using FLIC software to simulate the combustion process of solid MSW on the grate, FLUENT software to simulate the combustion process of gaseous components in the furnace, and Aspen Plus software to simulate other processes. Experimental design and implementation are performed for this numerical simulation model under baseline operating conditions to obtain simulation mechanism data under multiple operating conditions.

[0075] Specifically, the reference working condition full-process numerical simulation model construction submodule. The following assumptions and simplifications are made for the solid-phase MSW combustion process on the grate: the fuel on the grate is considered as a homogeneous porous medium with the same porosity during the combustion process; the flow phenomenon of particles in the furnace is not considered; the grate bed is considered as a constant speed forward movement; MSW is composed of moisture, volatile matter, fixed carbon and ash components; gas phase components only include CO, CO2, CH4, H2, H2O, NO, HCN, NH3, N2 and O2; the solid-phase MSW combustion process on the grate includes moisture evaporation, volatile matter release and combustion and coke combustion models:

[0076] The moisture evaporation model is as follows:

[0077]

[0078] wherein, is the moisture evaporation rate; is the particle surface area; is the convective mass transfer coefficient; and are the moisture concentrations in the solid phase and the gas phase, respectively; is the sum of convective heat transfer and radiation heat transfer; is the heat absorbed by the solid for moisture evaporation; is the solid temperature.

[0079] The volatile matter release model is as follows:

[0080]

[0081] wherein, is the volatile matter release rate; is the solid density; is the pre-exponential factor; is the activation energy; is the universal gas constant; is the ambient temperature.

[0082] The volatile matter combustion model is as follows:

[0083]

[0084] wherein, is the total gas phase bulk mixing rate in the bed; is the mixing rate constant; is the gas density; is the volatile molecule diffusion coefficient; is the bed porosity; is the particle diameter; is the gas flow rate; is the fuel concentration; is the O2 concentration; and are stoichiometric coefficients in the reaction.

[0085]

[0086] where, is the CH4 kinetic reaction rate, is the gas temperature.

[0087]

[0088] where, is the CO kinetic reaction rate; is the CO concentration, is the H2O concentration.

[0089] The coke combustion model is shown as follows:

[0090]

[0091] where, is the absorption coefficient, i.e., the ratio of CO and CO2:

[0092]

[0093]

[0094] where, is the coke reaction rate; is the oxygen partial pressure; is the chemical kinetic rate; is the diffusion reaction rate.

[0095] Further, the FLIC software is used to simulate the above process to obtain the flue gas temperature, velocity and components and other outputs, thereby supporting the subsequent simulation. The FLIC software output is regarded as the boundary condition of the numerical simulation of the FLUENT software, and the secondary air injection position is designed based on the MSWI plant process structure to perfect the simulation of the gas phase component combustion process in the furnace. The process is assumed and simplified as follows: the second-order upwind format is used for equation discretization, and then the SIMPLE (semi-implicit method for pressure-linked equations) algorithm is used for solution; the C m H n is replaced by CH4; the incinerator inlet and outlet are set as "velocity inlet" and "pressure-outlet" respectively; the gas phase component combustion process in the furnace is composed of the discrete ordinate radiation model describing thermal radiation, the standard k-ε model solving turbulent gas flow and the vortex dissipation concept model solving the interaction of gas flow and combustion chemical reaction:

[0096] The discrete ordinate radiation model is shown as follows:

[0097]

[0098] in, is the radiation intensity; and are position vector and direction vector respectively; is the absorption coefficient; is the diffusion coefficient; is the refractive index; is the Boltzmann constant; is the phase function; is a fixed angle.

[0099] The standard k-ε model is shown below:

[0100]

[0101]

[0102] in, is the gas density; is the turbulent kinetic energy; for speed; and is the coordinate axis; is the turbulent viscosity, which can be expressed as:

[0103]

[0104] in, is a constant, with a value of 0.09; and are the turbulent Prandtl coefficients, which are 1.0 and 1.3 respectively; is the terminal flow energy term generated by the velocity gradient; is the terminal flow energy term generated by buoyancy; is the consumption of turbulent pulsation kinetic energy dispersion rate; is the coefficient of fluctuation; and Customize source items for users; and are constants, taking values ​​of 1.44 and 1.92 respectively.

[0105] The conceptual model of eddy dissipation is shown below:

[0106]

[0107] in, For material The quality score of For material The diffusion flux of is the additional generation rate, is the net production rate, which is calculated as follows:

[0108]

[0109] in, is the time scale; For experienced time Substance after quality score.

[0110] Specifically, FLUENT software simulates the above process to obtain thermal radiation distribution data, which is used as input to the FLIC software for iterative coupling until the output temperatures of the two meet the specified tolerance. Visualized furnace temperature, velocity, and concentration fields are then output. The flue gas temperature and composition output by FLIC software, the furnace temperature output by FLUENT software, and actual industrial field data (MSW ash content, secondary air flow rate, secondary air temperature, urea solution, economizer feed water, calcium hydroxide solution, activated carbon, and reclaimed water) are used as input to simulate other processes, including MSW storage and incineration, waste heat boiler heat exchange, and flue gas treatment and emission, using Aspen Plus software. The following assumptions and simplifications are made: the furnace combustion process is simplified to a gaseous component combustion process; the effect of MSW particle size on the combustion reaction is ignored; the MSWI process is always in a stable state, and the furnace temperature and pressure are constant; all reactions occurring in the furnace can reach equilibrium; MSW ash is an inert component and does not participate in any reactions; MSW and air are fully mixed and evenly distributed during the incineration process; and pressure and gas losses and leakage are not considered.

[0111] Furthermore, based on the above assumptions and simplifications, numerical simulation models of MSW storage and incineration, waste heat boiler heat exchange, flue gas treatment and emission processes were established (see Figure 4 ). Figure 4 Refer to Table 1 for the functions of each module.

[0112] Table 1

[0113]

[0114]

[0115] Furthermore, for the MSW storage and incineration process, this example utilizes FLIC software to simulate the solid-phase MSW combustion process on the grate. For this reason, the yield reactor pyrolysis process can be omitted during the incineration process. Inputs to this process include the flue gas temperature and components (H2O, N2, O2, H2, CO, CO2, CH4, and S) output by the FLIC software, the furnace temperature output by FLUETN, the MSW ash content obtained through laboratory testing, and the secondary air temperature, secondary air flow rate, and urea solution dosage collected from actual industrial field data. Furthermore, the RDibbs module within Aspen Plus software was used to simulate the flue gas combustion process within the furnace, and the Rstoic module within Aspen Plus software was used to simulate the denitrification reaction. The relevant chemical equations are as follows:

[0116]

[0117]

[0118] Specifically, for the heat exchange process of the waste heat boiler, this embodiment uses HeatX1-6 and Heater1, 2 to simulate the heat exchange process between high-temperature flue gas and economizer feed water, and finally obtains the furnace outlet flue gas G1 after cooling.

[0119] Specifically, with respect to the flue gas treatment and emission process, this embodiment utilizes the Rstoic module within Aspen Plus software to simulate the deacidification process of the cooled furnace outlet flue gas G1. Mixer1 is then used to simulate the adsorption of heavy metals, dioxins, and other pollutants in the flue gas by activated carbon. Sep is then used to simulate the bag filter. Finally, SSplit2 and Compr are used to simulate the process of fly ash entering the ash silo and being discharged into the atmosphere, respectively. The chemical reaction equations involved in the deacidification process are as follows:

[0120]

[0121] Furthermore, a numerical simulation model of the entire MSWI process under the benchmark working conditions was constructed based on a multi-software coupling strategy, and then an orthogonal experiment was designed and implemented. A total of 12 factors, including one non-operated variable (MSW component) and 11 operated variables (feed rate, grate speed, primary air temperature 1, primary air temperature 2, primary air temperature 3, primary air flow 1, primary air flow 2, primary air flow 3, primary air flow 4, secondary air temperature and secondary air flow), were selected to conduct a four-level orthogonal experimental design to obtain 64 sets of mechanism data under different working conditions. .

[0122] Furthermore, a virtual process data generation module is used. Mechanism data has certain limitations in terms of operating condition coverage and data dimensionality. However, using CFD software to perform comprehensive numerical simulations of the MSWI process consumes significant computing resources and is time-consuming. Considering the similarities in the inherent physical and chemical reaction mechanisms under different operating conditions, this embodiment utilizes VSG (virtual synchronous generator) technology to enhance the simulation mechanism data.

[0123] Specifically, the virtual sample input data generation submodule. To simplify the difficulty of virtual sample generation and match the actual industrial process data, this embodiment performs mean processing on the primary air temperature, primary air temperature, and primary air temperature in the orthogonal experimental design to obtain the primary air temperature value; and performs sum processing on the primary air flow, primary air flow, primary air flow, and primary air flow to obtain the primary air flow value, and finally obtains the pre-processed mechanism data. .

[0124] Furthermore, this embodiment uses an equidistant interpolation technique to expand the preprocessed mechanism data input space, thereby generating virtual sample points with different input features. The formula is as follows:

[0125]

[0126] in, and Respectively The maximum and minimum values ​​of the input features; For the The number of interpolation points for each input feature; For the The interpolation interval for each input feature.

[0127] Furthermore, the input features are interpolated at equal intervals to obtain virtual sample inputs. ,in, represents the initial number of virtual samples.

[0128] Specifically, the virtual sample output data generation submodule. This embodiment uses IT2FBLS to establish a mapping model to obtain virtual sample output, and its network structure is as follows: As shown, the IT2FBLS network consists of an input layer, an IT2FNN layer, an enhancement layer, and an output layer:

[0129] 1) Input layer:

[0130] This layer maps the input data into the formula:

[0131]

[0132] in, .

[0133] 2) IT2FNN layer:

[0134] This layer contains The output of an IT2FNN subsystem is the feature mapping result, which can be expressed as:

[0135]

[0136]

[0137] in, Indicates the Feature map output of each IT2FNN subsystem; Indicates in The first IT2FNN subsystem Sample feature map output:

[0138]

[0139] in, is the lower bound ratio value; and Respectively represent The lower and upper bounds of the IT2FNN subsystem feature map are calculated as follows:

[0140]

[0141]

[0142] in, is the number of rules; For the Rule weights; and Respectively The lower and upper bounds of the activation strength of the rule can be expressed as:

[0143]

[0144]

[0145] in, and Respectively represent The input corresponding to the The lower and upper bounds of the membership values ​​of the rules are given. In this embodiment, a Gaussian membership function with uncertain mean is used to calculate the membership of the fuzzy set:

[0146]

[0147]

[0148] in, For the The first sample input variables; Respectively The input variable corresponds to the The membership function has lower and upper bounds on the uncertainty center; For the The input variable corresponds to the The width of the membership function.

[0149] 3) Enhancement layer:

[0150] This layer contains Enhanced nodes, this embodiment uses The function performs nonlinear changes on the output of the IT2FNN layer, which can be expressed as:

[0151]

[0152]

[0153] in, and Represents the IT2FNN layer and the enhanced layer The connection weights and biases are randomly generated for each enhanced node.

[0154] 4) Output layer:

[0155] This layer contains nodes, and its output can be expressed as:

[0156]

[0157] in, Output for IT2FBLS network; and Represent the connection weights between the IT2FNN layer and the enhancement layer and the output layer respectively; is the network connection weight, which is calculated as follows:

[0158]

[0159] in, is the regularization coefficient, and ; is the identity matrix; Represents the true value of the model.

[0160] The virtual sample output based on the trained IT2FBLS network can be expressed as:

[0161]

[0162] Furthermore, the virtual sample mixing and selection submodule matches the above virtual sample input and output data to obtain the original virtual sample of the simulation mechanism data. Considering that the virtual sample input is generated based on the equal interval interpolation method and the output is generated based on the IT2FBLS mapping model, there may be abnormal values ​​in the generated virtual sample output or sample redundancy for input features with low sensitivity. Therefore, the original virtual sample obtained above is The selection is based on the preset physical parameter threshold, and the final virtual process data set is recorded as ;in, Indicates the number of virtual samples after selection.

[0163] Furthermore, the virtual-real data drives the model construction module. This embodiment uses the virtual process data set The IT2FBLS network is trained to construct the main pollutant emission model, and its output can be expressed as:

[0164]

[0165] in, Represents the main model of pollutant emissions from the MSWI process.

[0166] Specifically, the real process data drives the LRDT compensation model submodule. The real process data is input into the trained IT2FBLS main model to obtain the residual between the main model output and the actual true value. , the calculation formula is as follows:

[0167]

[0168] in, Represents the input feature data collected during the actual process; and They represent the output characteristic data collected from the actual process and the output of the IT2FBLS main model respectively.

[0169] Furthermore, the LRDT algorithm is used to establish a residual compensation model to improve the overall modeling accuracy. The modeling process is as follows:

[0170] The residual data set is denoted as ;

[0171] Traverse the residual data set to calculate the mean squared error (MSE), the calculation formula is as follows:

[0172]

[0173] in, Represents the loss function value of MSE, Indicates the In the iteration The first sample eigenvalues.

[0174] Select the first non-leaf node with the smallest MSE As a splitting variable:

[0175]

[0176] Repeat the above process until the number of node samples meets the minimum number of samples set based on experience , and finally obtain Intermediate nodes .

[0177] The LRDT algorithm used in this embodiment improves the progress of the model through the linear regression method, and its calculation formula is as follows:

[0178]

[0179] in, 、 and Respectively The output, input and weight matrix of each leaf node.

[0180] The regularized least squares loss function is used to update the weight matrix. The calculation formula is as follows:

[0181]

[0182] in, The value is the regularization coefficient, and .

[0183] The above loss function is The gradient of is:

[0184]

[0185] make , we can get:

[0186]

[0187] The leaf node output is calculated based on the calculation formula of the LRDT algorithm:

[0188]

[0189] in, Represents the pollutant emission compensation model of the MSWI process.

[0190] Furthermore, the pollutant concentration output fusion submodule performs weighted summation on the outputs of the main model built based on IT2FBLS and the compensation model built based on LRDT, and finally obtains the output of the MSWI process pollutant emission model:

[0191]

[0192] Specifically, this embodiment collects real process data for eight hours from 16:00 to 24:00 on a certain day from a certain municipal solid waste incineration plant (operating parameters are shown in Table 2), uses the quartile method to eliminate outliers in the data, and uses Gaussian noise to increase the randomness of the data.

[0193] Table 2

[0194]

[0195]

[0196] Furthermore, orthogonal experiments were conducted based on the solid-phase combustion process on the grate simulated by FLIC software, the gas-phase combustion process in the furnace simulated by FLUENT software, and other processes simulated by Aspen Plus software to obtain 64 sets of experimental data.

[0197] refer to Figure 6 (a) and Figure 6 (b), where T is the furnace input; PAF is the primary air flow rate; and PAT is the primary air temperature. Virtual sample inputs generated using equal-interval interpolation techniques fully fill the gaps in the input dimensions of the original data, improving the operating condition coverage of the simulation mechanism data. The generated virtual samples are input into the IT2FBLS mapping model to generate virtual sample outputs. The Pearson correlation coefficient (PCC) is used to evaluate the quality of the generated virtual sample outputs.

[0198] Specifically, the PCC heat value diagram between input and output before and after mixing virtual samples is as follows: It is worth noting that in order to facilitate the display of the overall correlation between input features and output features before and after generating virtual samples, the color mapping heat takes the same range. It can be seen that except for the first output feature PCC, which is different from the original mechanism simulation data, the PCC values ​​of the input and output features of the other mechanism data virtual sample data sets are similar to those of the mechanism data sets, verifying the effectiveness of using the IT2FBLS mapping model to generate virtual sample outputs in this embodiment.

[0199] Furthermore, in order to verify the effectiveness of the modeling strategy proposed in this embodiment, this section uses IT2FBLS and LRDT algorithms to establish the main model and compensation model respectively. The experimental comparison (divided into M1 group, M2 group, M3 group, and M4 group) is set as follows: As shown in the figure, the model parameters are set as follows: the number of subsystems is 50, the number of rules is 20, the lower bound ratio is 0.5, the number of enhancement nodes is 50, the shrinkage factor is 0.5, and the regularization coefficient is 0.01 in the IT2FBLS algorithm; the minimum number of samples is 20, and the regularization coefficient is 0.01 in the LRDT algorithm. The parameter settings of the same algorithm model are the same in different experimental comparison schemes.

[0200] Table 3

[0201]

[0202] Specifically, the root mean square error (RMSE) indicator is used to evaluate the performance of the above model combination strategy. The formula is as follows:

[0203]

[0204] in, and represent the model predicted value and the true value respectively.

[0205] Further, based on In the experimental comparison setting scheme, the main model uses the mechanism data virtual sample data set for training, 70% of the process data set is used to train the compensation model, and 30% is used for testing. The statistical results of the test set RMSE are as follows: As shown, the fitting curve is (a) Figure 8 (b) and Figure 8 (c). It can be seen that the solution proposed in this embodiment, which uses the IT2FBLS algorithm to establish the main model based on the virtual sample data set of the mechanism data and the LRDT algorithm to establish the compensation model, has a higher accuracy. It can be seen that different schemes all have some degree of outliers, especially in the SO2 modeling task.

[0206] Table 4

[0207]

[0208] The beneficial effects of the present invention are as follows:

[0209] The present invention reduces the time cost of the simulation process by constructing a simulation model, and improves the accuracy of the emission model by establishing a compensation model using the LRDT algorithm.

[0210] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0211] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for constructing a pollutant emission model for a municipal solid waste combustion process, characterized in that: include: Perform simulation based on the preset physical model to obtain a full-process numerical simulation model; Conducting multi-operating condition orthogonal experiments based on the full-process numerical simulation model to obtain multi-operating condition mechanism data; The multi-operating condition mechanism data include: municipal solid waste composition, feed rate, grate speed, primary air temperature, primary air temperature, primary air temperature, primary air flow rate, primary air flow rate, primary air flow rate, primary air flow rate, secondary air temperature and secondary air flow rate; Preprocessing the multi-operating condition mechanism data to obtain preprocessed mechanism data; the preprocessing includes: mean processing and summation processing; Expanding the pre-processed mechanism data using an equidistant interpolation technique to obtain virtual sample points; Inputting the virtual sample points into a preset first IT2FBLS network to obtain a network output; Matching the virtual sample points with the network output to obtain original samples; Selecting the original samples based on a preset physical parameter threshold to obtain a virtual process data set; Training a preset second IT2FBLS network based on the virtual process data set to obtain a main pollutant emission model; Inputting pre-collected real data into the pollutant emission master model to obtain the master model output and simulation error; Calculate a mean square error based on the simulation error to obtain a mean square error set; Determine the minimum mean square error in the mean square error set as the segmentation variable, and return to the step of "calculating the mean square error based on the simulation error to obtain the mean square error set" until the number of segmentation variables meets the preset minimum number of samples, and output all the segmentation variables; Based on the segmentation variables and the preset weight formula, a regularized least squares loss function is used to update the weight matrix of the preset LRDT algorithm; Calculate leaf node outputs of the LRDT algorithm based on the weight matrix; Performing weighted summation on the main model output and the leaf node output to obtain a target emission output and a target pollutant emission model; Simulation is performed based on the preset physical model to obtain a full-process numerical simulation model, including: The FLIC software is used to perform numerical simulations on the preset water evaporation model, volatile analysis model, volatile combustion model, and coke combustion model to obtain combustion simulation results. Based on the combustion simulation results, FLUENT software is used to perform numerical simulations on the preset discrete coordinate radiation model, standard k-ε model, and eddy dissipation conceptual model to obtain thermal radiation distribution data; Inputting the thermal radiation distribution data into the FLIC software, and returning to step "using the FLIC software to perform numerical simulations on a preset water evaporation model, a volatile analysis model, a volatile combustion model, and a coke combustion model to obtain combustion simulation results" until the temperature in the thermal radiation distribution data satisfies a preset temperature error, and then outputting the temperature field, velocity field, and concentration field; Based on the temperature field, the velocity field, and the concentration field, Aspen Plus software is used to simulate the storage and incineration of municipal solid waste, heat exchange in waste heat boilers, and flue gas treatment and emission to obtain the full-process numerical simulation model.

2. The method for constructing a pollutant emission model for a municipal solid waste combustion process according to claim 1, characterized in that: The water evaporation model is: ; The volatile analysis model is: ; The volatile matter combustion model is: ; The coke combustion model is: ; is the water evaporation rate; is the particle surface area; is the convective mass transfer coefficient; and are the water concentrations in the solid and gas phases, respectively; It is the sum of convective heat transfer and radiation heat transfer; The heat required to evaporate water in the solid; is the solid temperature; is the volatile analysis rate; is the solid density; is the pre-exponential factor; is the activation energy; is the universal gas constant; is the ambient temperature; is the overall mixing rate of the gas phase in the bed; is the mixing rate constant; is the gas density; is the diffusion coefficient of volatile molecules; is the bed void ratio; is the particle diameter; is the gas flow rate; is the fuel concentration; is the O2 concentration; and is the stoichiometric coefficient in the reaction; is the ratio of CO to CO2.

3. The method for constructing a pollutant emission model for a municipal solid waste combustion process according to claim 1, characterized in that: The discrete coordinate radiation model is: ; The standard k-ε model is: and ; The eddy dissipation conceptual model is: ; is the radiation intensity; and are position vector and direction vector respectively; is the absorption coefficient; is the diffusion coefficient; is the refractive index; is the Boltzmann constant; is the phase function; is a fixed angle; is the gas density; is the turbulent kinetic energy; for speed; and is the coordinate axis; is the turbulent viscosity; For material The quality score of For material The diffusion flux of is the additional generation rate, is the net production rate; and are the turbulent Prandtl coefficients, which are 1.0 and 1.3 respectively; is the terminal flow energy term generated by the velocity gradient; is the terminal flow energy term generated by buoyancy; is the consumption of turbulent pulsation kinetic energy dispersion rate; is the coefficient of fluctuation; and Customize source items for users; and are constants, taking values ​​of 1.44 and 1.92 respectively.

4. The method for constructing a pollutant emission model for a municipal solid waste combustion process according to claim 1, characterized in that: The equal-interval interpolation technique is: ; and Respectively The maximum and minimum values ​​of the input features; For the The number of interpolation points for each input feature; For the The interpolation interval for each input feature.

5. The method for constructing a pollutant emission model for a municipal solid waste combustion process according to claim 1, characterized in that: The calculation formula of the segmentation variable is: ; is the loss function value; For the In the iteration The first sample eigenvalues; The first non-leaf node.

6. The method for constructing a pollutant emission model for a municipal solid waste combustion process according to claim 1, characterized in that: The weight matrix is: ; 、 and Respectively Output, input and weight matrices of non-leaf nodes; The non-leaf node weight vector representing the first input feature; is the regularization coefficient that is not less than 0.

Images