Data-driven multi-objective optimization control method for municipal solid waste incineration process

Abstract

A data-driven multi-objective optimization control method for a municipal solid waste incineration (MSWI) process. The method comprises: first, establishing an MSWI process whole-process model for pollutant emission, wherein the model includes controlled object models, i.e. a furnace hearth temperature model, a boiler steam flow model and a flue gas oxygen-content model, and CO, CO2 and NOx pollutant emission index models; then, using single-neuron adaptive PID to establish a multi-input multi-output loop controller for key controlled variables, so as to realize stable control over a combustion process; and finally, in view of expert knowledge in the field, realizing adaptive solving of a key controlled variable optimization set value based on a multi-objective particle swarm optimization algorithm. Multi-objective optimization control over an MSWI process can be realized, existing running factories are guided, the pollutant emission concentration can be reduced, and the efficiency of combustion is improved.

Description

A data-driven multi-objective optimization control method for municipal solid waste incineration process Technical Field

[0001] The present invention relates to the technical field of municipal solid waste treatment, and in particular to a data-driven multi-objective optimization control method for a municipal solid waste incineration process. Background Art

[0002] Municipal solid waste (MSW) incineration is a major contributor to my country's increasingly severe urban pollution problem. With the country's increasing emphasis on waste sorting, its calorific value has gradually increased, making MSW incineration (MSWI) a key method for treating MSW. Compared to other treatment methods, MSWI, with its harmless, reduced, and resourceful properties, can conserve significant land resources and offer significant social, economic, and environmental benefits. Due to the complex composition and low calorific value of MSW in my country, imported MSWI technology is ineffective in supporting the normal operation of power plants. Currently, it relies heavily on manual operation by domain experts. However, this variability in expert experience and the delay in addressing issues make it difficult for MSWI power plants in my country to maintain stable operating conditions, resulting in short-term pollutant emissions exceeding standards and incomplete MSW combustion. Therefore, reducing the concentration of these pollutants and improving MSW combustion efficiency are urgent issues that need to be addressed. In particular, major pollutants such as dioxins that cause the "NIMBY effect" of incineration plants, NOx that causes acid rain, and CO2 that is the focus of the current "dual carbon strategy" only need to reduce their emission concentrations.

[0003] Establishing an accurate model of the controlled plant is generally the foundation of intelligent optimal control research. However, due to the complexity of the mechanism, frequent operating condition fluctuations, and uncertainty in disturbances, constructing an accurate mathematical model of the MSWI process is difficult. Currently, researchers often use industrial process data to construct controlled plant models. However, these studies often focus on modeling a single controlled variable, ignoring the existence and inter-coupling of multiple controlled variables in the MSWI process. Achieving stable control of key controlled variables in the MSWI process is a prerequisite for optimizing their setpoints. Currently, PID controllers and model predictive control are commonly used for MSWI process control. There is no research on multi-loop controllers for key controlled variables in the MSWI process that meets industrial requirements. For MSWI process optimization, furnace temperature, boiler steam flow rate, and flue gas oxygen content are key controlled variables, but no research has yet been conducted on their optimal settings. Currently, there are no reports on optimal control of the MSWI process targeting these pollutant emission concentrations and MSW combustion efficiency. Therefore, it is imperative to design a data-driven multi-objective optimal control method for municipal solid waste incineration processes.

[0004] Summary of the Invention

[0005] The purpose of the present invention is to provide a data-driven multi-objective optimization control method for the municipal solid waste incineration process, which can realize multi-objective optimization control of the municipal solid waste incineration process, provide guidance for existing operating factories, reduce pollutant emission concentrations, and improve combustion efficiency.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A data-driven multi-objective optimization control method for a municipal solid waste incineration process comprises the following steps:

[0008] Step 1: Analyze the influencing factors for multi-objective optimization;

[0009] Step 2: Detect the multi-objective optimization control model based on the influencing factors.

[0010] Optionally, in step 1, factors influencing multi-objective optimization are analyzed, specifically:

[0011] The constraints of MSWI are:

[0012] Where t is the environment or time variable, and x is the set value of the process operation variable;

[0013] To ensure that environmental protection indicators are met:

[0014] To improve product indicators, there are:

[0015] To improve economic indicators, there are:

[0016] The optimization objectives are processed in the same direction and the optimization objective of MSWI is expressed as:

[0017] The optimized operation of the MSWI process is achieved by solving x, where x includes the furnace temperature, boiler steam flow rate, and flue gas oxygen content.

[0018] Optionally, in step 1, detecting a multi-objective optimization control model based on influencing factors specifically includes the following steps:

[0019] Step 301: Establish a full-process model for pollutant emission reduction as a controlled object model;

[0020] Step 302: Establish a single neuron adaptive PID multi-input multi-output loop controller to perform stable control on the controlled object model;

[0021] Step 303: Establish a controlled variable optimization setting solution model based on multi-objective PSO to achieve adaptive solution of the optimized setting values ​​of key controlled variables, and send the solution results to the multi-input and multi-output loop controller of the single neuron adaptive PID to stabilize the controlled object model.

[0022] Optionally, in step 301, a full-process model for pollutant emission reduction is established, specifically:

[0023] Establish a full-process model for pollutant emission reduction, including a series controlled plant model and a parallel pollutant index model. Connect a single-neuron adaptive PID multi-input multi-output loop controller to the series controlled plant model, which in turn connects to the parallel pollutant index model. The parallel pollutant index model is then connected to a controlled variable optimization solution model based on a multi-objective PSO.

[0024] The series controlled object model includes the furnace temperature model, the boiler steam flow model and the flue gas oxygen content model. The furnace temperature model, the boiler steam flow model and the flue gas oxygen content model are constructed by the linear regression decision tree algorithm based on Tikhonov regularization. The multi-input and multi-output loop controller of the single neuron adaptive PID is connected to the furnace temperature model, the boiler steam flow model and the flue gas oxygen content model, and the furnace temperature model is inputted with u PriAir 、u Feeder and u Dry , input uFeederWater to the boiler steam flow model, and input u to the flue gas oxygen content model SecAirThe furnace temperature model is connected to the boiler steam flow model and the flue gas oxygen content model. The furnace temperature model, boiler steam flow model and flue gas oxygen content model are connected to the parallel pollutant index model. Among them, uFeederWater, u PriAir 、u Feeder 、u Dry and u SecAir They represent water supply, primary air volume, feeder speed, drying grate speed and secondary air volume output respectively;

[0025] The parallel pollutant index model includes the CO model, the CO2 model and the NOx model. The CO model, the CO2 model and the NOx model are all constructed through the linear regression decision tree algorithm based on Tikhonov regularization. The furnace temperature model, the boiler steam flow model and the flue gas oxygen content model are connected to the CO model, the CO2 model and the NOx model respectively. The CO model, the CO2 model and the NOx model are connected to the controlled variable optimization setting solution model based on the multi-objective PSO.

[0026] Optionally, in step 302, a single neuron adaptive PID multi-input multi-output loop controller is established, specifically:

[0027] The multi-input and multi-output loop controller of the single neuron adaptive PID includes a furnace temperature controller with feeder speed, drying grate speed and primary air volume as operating variables, a boiler steam flow controller with water feed rate as operating variable, and a flue gas oxygen content controller with secondary air volume as operating variable. Among them, the furnace temperature controller, boiler steam flow controller and flue gas oxygen content controller are all implemented by the single neuron adaptive PID algorithm. The controlled variable optimization setting solution model based on multi-objective PSO is respectively connected to the furnace temperature controller, boiler steam flow controller and flue gas oxygen content controller, the furnace temperature controller is connected to the furnace temperature model, the boiler steam flow controller is connected to the boiler steam flow model, and the flue gas oxygen content controller is connected to the flue gas oxygen content model.

[0028] Optionally, in step 303, a controlled variable optimization setting solution model based on multi-objective PSO is established, specifically:

[0029] The output values ​​of the CO model, CO2 model, and NOx model, that is, the sum of the emission values ​​of CO, CO2, and NOx, are used as the comprehensive pollutant emission concentration to be minimized. At the same time, the MSW combustion efficiency is maximized and the controlled variable optimization setting solution model based on the multi-objective PSO is established as follows:

[0030] Where, and Represents the CO, CO2 and NOx index model outputs, r FT 、r BSF and rOX They represent the set values ​​of furnace temperature, boiler steam flow and flue gas oxygen content in the multi-loop control system respectively. The controlled variables, namely the optimized set values ​​of furnace temperature, boiler steam flow and flue gas oxygen content, are optimized based on the improved multi-objective particle swarm algorithm. Among them, the improved multi-objective particle swarm algorithm includes population initialization, updating individual optimal particles, global optimal particles and archives, updating particle speed and position, mutating the population, detecting boundary limits, calculating fitness values, judging termination conditions and outputting the optimal set values ​​based on expert experience.

[0031] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the data-driven multi-objective optimization control method for the municipal solid waste incineration process provided by the present invention, based on the MSWI process analysis for multi-objective optimization, establishes a multi-objective optimization model with key controlled variables as decision variables, minimizing the comprehensive pollutant emission concentration and maximizing the MSW combustion efficiency, and proposes a corresponding multi-objective optimization control strategy. Specifically, first, a full-process model of the MSWI process for pollutant emissions is established, including the controlled object models of furnace temperature, boiler steam flow and flue gas oxygen content, and the CO, CO2 and NOx pollutant emission index models; then, a single neuron adaptive PID is used to establish a multi-input and multi-output loop controller for key controlled variables to achieve stable control of the combustion process; finally, combined with the knowledge of domain experts, an adaptive solution of the optimal set values ​​of key controlled variables based on the multi-objective particle swarm optimization algorithm is realized. Based on the field data of a certain MSWI power plant, it is verified that the proposed method can obtain the optimal set values ​​of three key controlled variables, reduce the pollutant emission concentration, and improve the combustion efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] 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.

[0033] Figure 1 is the process flow diagram of the MSWI power plant;

[0034] FIG2 is a flow chart of a data-driven multi-objective optimization control method for a municipal solid waste incineration process according to an embodiment of the present invention;

[0035] FIG3 is a schematic diagram of the structure of a multi-objective optimization control model;

[0036] FIG4 is a schematic diagram of the input-output relationship structure of the combustion process model;

[0037] Figure 5 is a schematic diagram of the SNAPID controller structure;

[0038] Figure 6a is a schematic diagram of the fitting curve of the FT model test set;

[0039] Figure 6b is a schematic diagram of the fitting curve of the BSF model test set;

[0040] Figure 6c is a schematic diagram of the fitting curve of the OX model test set;

[0041] Figure 7a is a schematic diagram of the fitting curve of the NOx model test set;

[0042] Figure 7b is a schematic diagram of the fitting curve of the CO model test set;

[0043] Figure 7c is a schematic diagram of the fitting curve of the CO2 model test set;

[0044] Figure 8a is a schematic diagram of the furnace temperature tracking curve;

[0045] Figure 8b is a schematic diagram of a boiler steam flow tracking curve;

[0046] Figure 8c is a schematic diagram of the flue gas oxygen content tracking curve;

[0047] Figure 9a is a schematic diagram of the furnace temperature error curve;

[0048] Figure 9b is a schematic diagram of a boiler steam flow error curve;

[0049] Figure 9c is a schematic diagram of the flue gas oxygen content error curve;

[0050] Figure 10a is a schematic diagram of the IAE index change curve of the furnace temperature;

[0051] Figure 10b is a schematic diagram of the IAE index change curve of boiler steam flow;

[0052] Figure 10c is a schematic diagram of the flue gas oxygen content IAE index change curve;

[0053] Figure 11a is a schematic diagram of the ISE index change curve of the furnace temperature;

[0054] Figure 11b is a schematic diagram of the boiler steam flow ISE index change curve;

[0055] Figure 11c is a schematic diagram of the flue gas oxygen content ISE index change curve;

[0056] Figure 12a is a schematic diagram of the solution results of the IMOPSO algorithm;

[0057] Figure 12b is a schematic diagram of the solution results of the MOPSO algorithm. DETAILED DESCRIPTION

[0058] The purpose of the present invention is to provide a data-driven multi-objective optimization control method for the municipal solid waste incineration process, which can realize multi-objective optimization control of the municipal solid waste incineration process, provide guidance for existing operating factories, reduce pollutant emission concentrations, and improve combustion efficiency.

[0059] 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.

[0060] The process flow of the MSWI power plant is shown in Figure 1.

[0061] As shown in FIG2 , the data-driven multi-objective optimization control method for the municipal solid waste incineration process provided by an embodiment of the present invention includes the following steps:

[0062] Step 1: Analyze the influencing factors for multi-objective optimization;

[0063] Step 2: Detect the multi-objective optimization control model based on the influencing factors.

[0064] In step 1, the factors affecting multi-objective optimization are analyzed, specifically:

[0065] The constraints of MSWI are:

[0066] Where t is the environment or time variable, and x is the set value of the process operation variable;

[0067] To ensure that environmental protection indicators (reducing the concentration of solid, liquid, gaseous pollutants and greenhouse gas emissions, etc.) meet the standards, there are:

[0068] In order to improve product indicators (reduce slag thermal loss rate and fly ash production, improve combustion efficiency and organic matter removal rate, etc.), there are:

[0069] To improve economic indicators (reducing MSW processing costs, increasing grid-connected power generation, etc.), there are:

[0070] The optimization objectives are processed in the same direction and the optimization objective of MSWI is expressed as:

[0071] The optimized operation of the MSWI process is achieved by solving x, where x includes the furnace temperature, boiler steam flow rate, and flue gas oxygen content.

[0072] Analysis shows that in order to achieve emission control of major pollutants (CO, CO2 and NOx) and maximize the combustion efficiency of MSW, it is necessary to solve the optimal set values ​​of key controlled variables (furnace temperature, boiler steam flow rate and flue gas oxygen content);

[0073] As shown in FIG3 , in step 1, the multi-objective optimization control model is detected according to the influencing factors, which specifically includes the following steps:

[0074] Step 301: Establish a full-process model for pollutant emission reduction as a controlled object model;

[0075] Step 302: Establish a single neuron adaptive PID multi-input multi-output loop controller to perform stable control on the controlled object model;

[0076] Step 303: Establish a controlled variable optimization setting solution model based on multi-objective PSO to achieve adaptive solution of the optimized setting values ​​of key controlled variables, and send the solution results to the multi-input and multi-output loop controller of the single neuron adaptive PID to stabilize the controlled object model.

[0077] The functions of the above three modules are as follows:

[0078] 1) Multi-objective PSO-based controlled variable optimization setting solution: This module uses the multi-objective PSO algorithm to solve the Pareto frontier of the multi-objective optimization model and outputs the optimal setting value based on the experience of domain experts;

[0079] 2) Multi-input and multi-output loop control based on single neuron adaptive PID: Based on the characteristics analysis mentioned above, this module establishes a multi-loop controller for key controlled variables of the MSWI process and effectively tracks the controlled object by adjusting the controller output;

[0080] 3) Full-process model for pollutant emission reduction: This module serves as the basis for the research of intelligent optimization algorithms, providing pollutant emission index model outputs and controlled object models for multi-input and multi-output loop control for multi-objective optimization models.

[0081] Explain Figure 3, where γ emission represents the exhaust emission model output, and Represent the output of NOx model, CO model and CO2 model respectively; CE represents combustion efficiency; r FT 、r BSF and r OX represent the FT, BSF and OX particle dimensions respectively; and represent the optimal setting values ​​of FT, BSF and OX respectively; and Represent the outputs of FT model, BSF model and OX model respectively; e FT 、e BSF and e OX represent the error values ​​of FT, BSF and OX respectively; and Represents the error state transition output of the FT controller; and Represents the error state transition output of the BSF controller; and Indicates the error state conversion output of the OX controller; uFeederWater, u PriAir 、u Feeder 、u Dry and u SecAir They represent water feed rate, primary air volume, feeder speed, drying grate speed and secondary air volume output respectively.

[0082] In step 301, a full-process model for pollutant emission reduction is established, specifically:

[0083] To eliminate the different noises in process data, the present invention uses wavelet denoising data to establish a full-process model for pollutant emission reduction, including a series controlled plant model and a parallel pollutant index model. A single neuron adaptive PID multi-input multi-output loop controller is connected to the series controlled plant model, which is then connected to the parallel pollutant index model. The parallel pollutant index model is then connected to a controlled variable optimization setting solution model based on a multi-objective PSO.

[0084] As shown in Figure 4, the series controlled object model includes a furnace temperature model, a boiler steam flow model, and a flue gas oxygen content model. The furnace temperature model, the boiler steam flow model, and the flue gas oxygen content model are all constructed using a linear regression decision tree algorithm based on Tikhonov regularization. The single neuron adaptive PID multi-input multi-output loop controller is connected to the furnace temperature model, the boiler steam flow model, and the flue gas oxygen content model, and inputs u into the furnace temperature model. PriAir 、u Feeder and u Dry , input uFeederWater to the boiler steam flow model, and input u to the flue gas oxygen content model SecAir The furnace temperature model is connected to the boiler steam flow model and the flue gas oxygen content model. The furnace temperature model, boiler steam flow model and flue gas oxygen content model are connected to the parallel pollutant index model. Among them, uFeederWater, u PriAir 、u Feeder 、u Dry and u SecAir They represent water supply, primary air volume, feeder speed, drying grate speed and secondary air volume output respectively;

[0085] To explain, the present invention denotes the furnace temperature, boiler steam flow and flue gas oxygen content model as f FT (·),f BSF (·) and fOX (·) are constructed using the Tikhonov regularized linear regression decision tree (TR-LRDT) algorithm, which has the advantage of strong interpretability. FT (·) as an example, the modeling process is described as follows:

[0086] Combined with industrial practice, f FT The input of (·) is recorded as

[0087] First, we search for the optimal segmentation variables and segmentation points by traversing until the number of leaf node samples is less than the empirically set threshold θ. FT The following criterion is used to divide the input feature space into K FT Regions get:

[0088] Where, and Representing regions and True value of the sample; and Representing regions and The true mean value of the samples in ;

[0089] The leaf node prediction value of the CART tree only considers the sample average value and ignores the relationship between the process variable and the true value. The present invention improves it by using a linear regression method to calculate the predicted output of the CART tree leaf node as follows:

[0090] Where, and The predicted output, input and weight vector of the leaf nodes respectively;

[0091] Considering the fact that the number of samples is greater than the feature dimension, obtaining the weight vector can be classified as solving a class of overdetermined matrix equations. To ensure the convergence of the weight vector, the regularized least squares loss function is used here, as follows:

[0092] Where λ FT Indicates the regularization coefficient with positive value;

[0093] Furthermore, the above loss function is The gradient of can be expressed as:

[0094] make get:

[0095] Finally, according to the weight vector obtained from the above formula, the leaf node prediction value is calculated based on formula (10), considering all leaf nodes, so the furnace temperature model can be expressed as:

[0096] Where I(·) is the indicator function. The function value is 1 if it exists, otherwise it is 0.

[0097] The parallel pollutant index model includes CO model, CO2 model and NOx model. Among them, CO model, CO2 model and NOx model are constructed by linear regression decision tree algorithm based on Tikhonov regularization. The furnace temperature model, boiler steam flow model and flue gas oxygen content model are connected to CO model, CO2 model and NOx model respectively. CO model, CO2 model and NOx model are connected to the controlled variable optimization setting solution model based on multi-objective PSO. Among them, CO model, CO2 model and NOx model can be expressed as f CO (·), and f NOx (·),as follows:

[0098] In step 302, a multi-input multi-output loop control model of a single neuron adaptive PID is established, specifically:

[0099] Based on the experience of experts in the field and the control characteristics analysis in the previous article, a multi-input and multi-output loop controller of a single neuron adaptive PID is constructed. The multi-input and multi-output loop controller of the single neuron adaptive PID includes a furnace temperature controller with feeder speed, drying grate speed and primary air volume as operating variables, a boiler steam flow controller with water supply as operating variable, and a flue gas oxygen content controller with secondary air volume as operating variable. Among them, the furnace temperature controller, boiler steam flow controller and flue gas oxygen content controller are all implemented by the single neuron adaptive PID (Single neuron adaptive PID, SNA-PID) algorithm. The controlled variable optimization setting solution model based on multi-objective PSO is connected to the furnace temperature controller, boiler steam flow controller and flue gas oxygen content controller respectively. The furnace temperature controller is connected to the furnace temperature model, the boiler steam flow controller is connected to the boiler steam flow model, and the flue gas oxygen content controller is connected to the flue gas oxygen content model.

[0100] Take the jth operating variable of FT Taking the example (j=1, 2, 3) for description, the structure of the SNA-PID controller is shown in Figure 5 for explanation;

[0101] First, the error signal is converted into a state, which can be expressed as:

[0102] Where ε(k) represents the error value; and They represent the furnace temperature setting value and the controlled object feedback value respectively; k represents the number of iterations;

[0103] Therefore, the output of the SNA-PID controller for the jth manipulated variable can be expressed as:

[0104] Where, represents the increment of the j-th operating variable; Represents the neuron gain coefficient of the jth manipulated variable; It represents the weight coefficient of the i-th neuron of the j-th operating variable at time k, which is expressed as:

[0105] The SNA-PID controller can track the set value under the supervised Hebb learning algorithm by continuously adjusting the neuron weight coefficients online, as follows:

[0106] Where, and Represent the weight coefficients of the proportional, integral and differential neurons in the j-th operating variable, and Represent the proportional, integral and differential neuron learning coefficients in the j-th manipulated variable respectively.

[0107] In step 303, a controlled variable optimization setting solution model based on multi-objective PSO is established, specifically:

[0108] From the previous analysis, we know that CO, CO2, and NOx are pollutant emission concentrations that must be minimized. The output values ​​of the CO model, CO2 model, and NOx model, that is, the sum of the emission values ​​of CO, CO2, and NOx, are used as the comprehensive pollutant emission concentration to be minimized. At the same time, the controlled variable optimization setting solution model based on multi-objective PSO is established to maximize the MSW combustion efficiency:

[0109] Where, and Represents the CO, CO2 and NOx index model outputs, r FT 、r BSF and r OXThey represent the set values ​​of furnace temperature, boiler steam flow and flue gas oxygen content in the multi-loop control system respectively. The optimized set values ​​of furnace temperature, boiler steam flow and flue gas oxygen content are solved based on the improved multi-objective particle swarm algorithm. The improved multi-objective particle swarm algorithm includes population initialization, updating individual particle optimality, global optimality and archives, updating particle speed and position, mutating the population, detecting boundary limits, calculating fitness values, judging termination conditions and outputting the optimal set values ​​based on expert experience.

[0110] Expand the description of the improved multi-objective particle swarm algorithm as follows:

[0111] When the population is initialized, first, the number of particles N Pos , Number of file particles Number of iterations N iter , speed update weight ω, number of grids N grid , mutation rate The relevant parameters such as the upper and lower limits of the set value decision variables are set; then, a population is randomly generated based on the set parameters and the particle fitness value is calculated; then, the global optimal individual is determined based on the particle fitness value using the grid method; finally, the above population is stored in the external file A;

[0112] After the population is initialized, the algorithm enters the iterative optimization phase. First, the particle speed and position are updated according to the individual optimal and global optimal values ​​after population initialization. The formula is as follows:

[0113] Where, and are the velocity and position of the d-th dimension of the i-th particle respectively; c1 and c2 are learning factors, generally set c1=c2=2; r1 and r2 are random numbers between (0,1); and are the individual optimal and global optimal positions of the dth dimension of the i-th particle respectively;

[0114] Then, in order to ensure that the algorithm has a stronger global search capability during the optimization process and avoid falling into the local optimum, the MOPSO algorithm introduces a mutation operation in the iterative process: 1) the particles are divided into three parts according to the total number of particles in the population; 2) the number of required mutated particles is determined according to the mutation rate, among which the first part of particles does not undergo mutation operation, and the mutation rate of the second part of particles is 1. Fixed, the third part of the particle mutation rate is calculated as follows:

[0115] Where, is the particle mutation rate of the third part; n iter and N iterRepresent the current number of iterations and the maximum number of iterations respectively; D represents the particle dimension. 3) After determining the number of particles in each part of the population that need to be mutated, select random particles in the corresponding part for mutation operation as follows:

[0116] Where v and x are the particle velocity information and position information respectively; v max and v min are the upper and lower limits of particle velocity respectively; x max and x min are the upper and lower limits of the particle position, respectively; μ mut represents the particle mutation rate, φ mut (·) is the process of randomly selecting particles for mutation operation;

[0117] Then, the boundary limits are detected for the updated and mutated population position information. The limits of the decision variables in the present invention are as follows:

[0118] 900℃≤r FT ≤950℃

[0119] 76t / h≤r BSF ≤78t / h

[0120] 6.5%≤r OX ≤8.5% (24)

[0121] In order to prevent particles that have exceeded the boundary from exceeding the boundary again in the next iteration, the particle velocity information is inverted to change the particle's flight direction so that it is away from the boundary, thereby ensuring effective iteration.

[0122] Then, the fitness value of the particles in the current population is recalculated, and the external file A is updated and maintained. The present invention determines the dominance relationship of the particles in the external file A and deletes the dominance solution to ensure the uniqueness of the external file A. The dominance relationship is determined as follows:

[0123] From the above formula, we can see that when the combustion efficiency of the i-th particle is greater than that of the j-th particle, and the comprehensive pollutant emission concentration is less than , it means that the j-th particle is dominated by the i-th particle, then the j-th particle in the external file is deleted;

[0124] At the same time, in order to improve the optimization speed of the MOPSO algorithm, the total number of particles in the external file A needs to be capped. If the number of particles in external file A exceeds the upper limit Then: 1) Determine the number of particles N to be deleted Del 2) Sort the particles in the external file A according to their fitness values ​​and calculate the crowding distance using the following formula:

[0125] Where Ω represents the crowding distance of particles in the external archive; i represents the i-th particle; 3) Delete N particles according to the crowding distance Del particles;

[0126] Then, based on the external file A, the individual optimum and the global optimum of the particle are updated. The individual optimum is updated as follows: the dominance relationship between the current particle and the individual optimum particle is determined; if the current particle dominates the individual optimum, the individual optimum is updated; if there is no dominance relationship between the two, whether to update the individual optimum is determined randomly;

[0127] The global optimal method based on the grid method is as follows: 1) Divide the minimum and maximum fitness values ​​of the particles in the external file A into N equal parts. grid , forming N grid ×N grid 1) The grid is generated and the position of particle A in the external file is determined; 2) the grid quality is evaluated according to the following formula:

[0128] Where Q grid (z) represents the zth grid quality, z=1,2,...,N grid ×N grid num(z) represents the number of particles in the zth grid. 3) Select the particle with higher grid quality to determine the global optimal individual. If there are multiple particles in the same grid or multiple grids with the same quality, randomly select the particle that meets the conditions as the global optimal individual.

[0129] Furthermore, by continuously repeating the above steps to iterate and optimize, the present invention introduces the termination condition of the comprehensive pollutant emission concentration difference as the target on the basis of the MOPSO algorithm, when the absolute value of the difference between the mean comprehensive pollutant emission concentration of the particles to be stored in the external archive and the mean comprehensive pollutant emission concentration of the particles in the external archive is less than the set value e stop When , the algorithm stops iterating and outputs the Pareto frontier;

[0130] Finally, based on the Pareto frontier and expert experience, the set values ​​that meet the actual industrial site economic indicators, safety performance and other related requirements are selected as the optimal set values ​​as the multi-input multi-output loop control set values. This paper only provides basic expert experience, as follows:

[0131] Rule 1: When CO concentration increases, the Pareto solution with higher combustion efficiency is preferred.

[0132] Rule 2: When NOx concentration increases, the Pareto solution with smaller pollutant emission index is preferred.

[0133] Rule 3: When the CO2 concentration increases, the MSW combustion efficiency increases accordingly. In the current situation, any solution in the Pareto solution set can be selected based on the current incineration demand.

[0134] This invention provides an embodiment that uses process data from a power plant running continuously for eight hours from 4:00 PM to midnight on a certain day in 2021 to verify the accuracy of the multi-objective optimization control method. To verify the effectiveness of the proposed model, CART tree, RF, and BPNN are used to compare with the TR-LRDT algorithm. The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the performance of the constructed model, as shown below:

[0135] Where, y i and are the model prediction value, true value and mean respectively, and N is the number of samples. The series controlled object model is verified, and the model parameter settings are shown in Table 1;

[0136] Table 1 Model parameter settings

[0137] Where θ TR-LRDT ,θ CART and θ RF represents the minimum number of samples in the TR-LRDT, CART, and RF algorithms, respectively; λ represents the regularization term coefficient; Tn represents the number of decision trees in the RF algorithm; Hidden, epoch, error, and η represent the number of hidden layer neurons, number of iterations, convergence error, and learning rate in the BPNN algorithm, respectively. The fitting curves of the test sets of each model are shown in Figures 6a-6c, and the performance evaluation comparison results are shown in Table 2.

[0138] Table 2 Model performance evaluation results

[0139] The pollutant index model was verified, and the parameter settings of each model are shown in Table 3;

[0140] Table 3 Model parameter settings

[0141] Where θ TR-LRDT ,θ CART and θ RFrepresents the minimum number of samples in the TR-LRDT, CART, and RF algorithms, respectively; λ represents the regularization term coefficient; Tn represents the number of decision trees in the RF algorithm; Hidden, epoch, error, and η represent the number of hidden layer neurons, number of iterations, convergence error, and learning rate in the BPNN algorithm, respectively. The fitting curves of the test sets of each model are shown in Figures 7a-7c, and the performance evaluation comparison results are shown in Table 4.

[0142] Table 4 Model performance evaluation results

[0143] From the modeling comparison results, it can be seen that the TR-LRDT algorithm has better fitting effect and superior modeling accuracy than other algorithms. The reason is that the TR-LRDT algorithm introduces weights into the tree structure model, corrects the defect of the mean output of the leaf nodes of the tree structure model, and thus enhances the model smoothing performance.

[0144] The MIMO loop control results were verified. The relevant parameter settings of the SNA-PID controller are shown in Table 5, and the relevant parameters of the PID controller are shown in Table 6.

[0145] Table 5 SNA-PID controller parameter settings

[0146] Table 6 PID controller parameter settings

[0147] The controller tracking results and error curves are shown in Figures 8a-8c and 9a-9c. As shown in Figures 8a-8c and 9a-9c, both controllers can effectively track the setpoint under variable setpoint conditions, but the SNA-PID controller has a faster adjustment time than the PID controller.

[0148] The present invention uses the integrated square error (ISE), the integrated absolute error (IAE) and the maximum deviation (DEV max ) to evaluate the controller performance, and the calculation formula is as follows: DEV max =max{|e(t)|} (32)

[0149] Where t0 and t f Indicates the control start and end time; the indicator change curves are shown in Figures 10a-10c and Figures 11a-11c, and the statistical results are shown in Table 7;

[0150] Table 7 Statistical results of controller indicators

[0151] From the index change curve and statistical results, we can see that although the SNA-PID controller is max The performance of the SNA-PID controller is poor in terms of the indicators, but it has higher accuracy in terms of the IAE and ISE indicators. The reason is that the SNA-PID controller continuously adjusts its output through the error and controller parameters, thereby achieving a faster adjustment speed; when facing the situation of set value changes, due to the large error at this time, the SNA-PID controller makes the controller output amplitude larger without changing the parameters, resulting in DEV max The indicators perform worse than those of the PID controller;

[0152] The results of multi-objective optimization are as follows: the present invention first normalizes the pollutant emission concentration and then sums it as the comprehensive pollutant emission concentration, and then uses the IMOPSO algorithm to solve the optimized set value of the controlled variable. The relevant parameters are as follows: the population size is 200, the upper limit of the external archive particles is 200, the number of iterations is 500, the inertia factor is 0.9, the learning factor c1=c2=2, the number of grids is 20, the mutation rate is 0.5, the maximum particle speed is [1, 0.1, 0.1], and the iteration termination condition e stop The output Pareto front is shown in Figures 12a and 12b. It can be seen that: 1) the two algorithms provide similar results for the Pareto front, demonstrating that the proposed IMOPSO algorithm is capable of solving the multi-objective optimization problem of the MSWI process. 2) Compared with the IMOPSO algorithm, the MOPSO algorithm has a more concentrated population near the Pareto front after iteration. This is because the population gradually approaches the global optimum as the number of iterations increases, further demonstrating the effectiveness of both algorithms in obtaining the Pareto front. However, for optimization problems, researchers are more interested in the Pareto front. Therefore, the proposed IMOPSO algorithm can accurately solve the Pareto front while reducing iteration time, which has greater engineering application significance.

[0153] The data-driven multi-objective optimization control method for the municipal solid waste incineration process provided by the present invention establishes a multi-objective optimization model based on the MSWI process analysis for multi-objective optimization, with key controlled variables as decision variables, minimizing the comprehensive pollutant emission concentration and maximizing the MSW combustion efficiency, and proposes a corresponding multi-objective optimization control strategy. Specifically, first, a full-process model of the MSWI process for pollutant emissions is established, including the controlled object models of furnace temperature, boiler steam flow and flue gas oxygen content, and the CO, CO2 and NOx pollutant emission index models; then, a single neuron adaptive PID is used to establish a multi-input and multi-output loop controller for key controlled variables to achieve stable control of the combustion process; finally, combined with the knowledge of domain experts, an adaptive solution of the optimal set values ​​of key controlled variables based on the multi-objective particle swarm optimization algorithm is realized. Based on the field data of a certain MSWI power plant, it is verified that the proposed method can obtain the optimal set values ​​of three key controlled variables, reduce the pollutant emission concentration, and improve the combustion efficiency.

[0154] The present invention 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 data-driven multi-objective optimization control method for the urban solid waste incineration process, characterized in that It includes the following steps: Step 1: Analyze the influencing factors for multi-objective optimization; Step 2: Detect the multi-objective optimization control model according to the influencing factors.

2. The data-driven multi-objective optimization control method for the urban solid waste incineration process according to claim 1, wherein, In Step 1, when analyzing the influencing factors for multi-objective optimization, specifically: The constraints of MSWI are as follows: In the formula, t is the environmental or time variable, and x is the set value of the process operation variable; To ensure compliance with environmental protection indicators, there are: To improve product indicators, there are: To improve economic indicators, there are: Align the optimization objectives in the same direction, and express the optimization objective of MSWI as: By solving for x, the optimized operation of the MSWI process is achieved, where x includes the furnace temperature, boiler steam flow rate, and flue gas oxygen content.

3. The data-driven multi-objective optimization control method for the urban solid waste incineration process according to claim 2, wherein In Step 1, when detecting the multi-objective optimization control model according to the influencing factors, it specifically includes the following steps: Step 301: Establish a full-process model for pollutant reduction as the controlled object model; Step 302: Establish a multi-input multi-output loop controller of single neuron adaptive PID to stably control the controlled object model; Step 303: Establish an optimized setpoint solving model for the controlled variables based on multi-objective PSO, which is used to adaptively solve the optimized set values of the key controlled variables, and send the solution results to the multi-input multi-output loop controller of single neuron adaptive PID to stably control the controlled object model.

4. The data-driven multi-objective optimization control method for the urban solid waste incineration process according to claim 3, wherein In Step 301, when establishing a full-process model for pollutant reduction, specifically: Establish a full-process model for pollutant reduction, including a series-connected controlled object model and a parallel pollutant index model. The multi-input multi-output loop controller of single neuron adaptive PID is connected to the series-connected controlled object model, the series-connected controlled object model is connected to the parallel pollutant index model, and the parallel pollutant index model is connected to the optimized setpoint solving model for the controlled variables based on multi-objective PSO; The series-connected controlled object model includes a furnace temperature model, a boiler steam flow model, and a flue gas oxygen content model. Among them, the furnace temperature model, the boiler steam flow model, and the flue gas oxygen content model are all constructed by a linear regression decision tree algorithm based on Tikhonov regularization. The multi-input multi-output loop controller of the single-neuron adaptive PID is connected to the furnace temperature model, the boiler steam flow model, and the flue gas oxygen content model, and inputs u PriAir , u Feeder , and u Dry into the furnace temperature model, inputs uFeederWater into the boiler steam flow model, and inputs u SecAir into the flue gas oxygen content model. The furnace temperature model is connected to the boiler steam flow model and the flue gas oxygen content model. The furnace temperature model, the boiler steam flow model, and the flue gas oxygen content model are connected to the parallel pollutant index model. Among them, uFeederWater, u PriAir , u Feeder , u Dry , and u SecAir respectively represent the feed water volume, the primary air volume, the feeder speed, the drying grate speed, and the secondary air volume output; The parallel pollutant index model includes a CO model, a CO2 model, and a NOx model. Among them, the CO model, the CO2 model, and the NOx model are all constructed by the linear regression decision tree algorithm based on Tikhonov regularization. The furnace temperature model, the boiler steam flow rate model, and the flue gas oxygen content model are respectively connected to the CO model, the CO2 model, and the NOx model, and the CO model, the CO2 model, and the NOx model are connected to the optimized setpoint solving model for the controlled variables based on multi-objective PSO.

5. The data-driven multi-objective optimization control method for the urban solid waste incineration process according to claim 4, wherein In Step 302, when establishing a multi-input multi-output loop controller of single neuron adaptive PID, specifically: The multi-input multi-output loop controller of single neuron adaptive PID includes a furnace temperature controller with the feeder speed, the drying grate speed, and the primary air volume as the operation variables, a boiler steam flow rate controller with the water supply volume as the operation variable, and a flue gas oxygen content controller with the secondary air volume as the operation variable. Among them, the furnace temperature controller, the boiler steam flow rate controller, and the flue gas oxygen content controller are all implemented by the single neuron adaptive PID algorithm. The optimized setpoint solving model for the controlled variables based on multi-objective PSO is respectively connected to the furnace temperature controller, the boiler steam flow rate controller, and the flue gas oxygen content controller. The furnace temperature controller is connected to the furnace temperature model, the boiler steam flow rate controller is connected to the boiler steam flow rate model, and the flue gas oxygen content controller is connected to the flue gas oxygen content model.

6. The data-driven multi-objective optimization control method for the urban solid waste incineration process according to claim 5, wherein In Step 303, when establishing an optimized setpoint solving model for the controlled variables based on multi-objective PSO, specifically: Taking the sum of the output values of the CO model, CO2 model, and NOx model, that is, the sum of the emission values of CO, CO2, and NOx, as the comprehensive pollutant emission concentration to be minimized, and maximizing the MSW combustion efficiency, a solution model for optimizing the setting of controlled variables based on multi-objective PSO is established as follows: In the formula, And respectively represent the output of the CO, CO2, and NOx index models, r FT , r BSF and r OX respectively represent the set values of the furnace temperature, boiler steam flow rate, and oxygen content in flue gas in the multi-loop control system. The controlled variables, i.e., the optimized set values of the furnace temperature, boiler steam flow rate, and oxygen content in flue gas, are optimized and solved based on the improved multi-objective particle swarm optimization algorithm. Among them, the improved multi-objective particle swarm optimization algorithm includes population initialization, updating the individual best, global best, and archive of particles, updating the velocity and position of particles, mutating the population, detecting boundary limitations, calculating fitness values, judging termination conditions, and judging and outputting the optimal set value according to expert experience.

Images