A pollution reduction and efficiency increase intelligent optimization method for urban solid waste incineration process

Abstract

The present application relates to a kind of urban solid waste incineration process pollution reduction and efficiency increasing intelligent optimization method, through the collaborative operation optimization of multiple process links, improve the main steam flow and its stability, reduce the concentration of NOx, SO2, HCl and other pollutant emissions, while reducing the amount of urea solution and limestone slurry.It includes the following steps: first, obtain data;Determine input and output, and use fuzzy neural network to establish operation index model;Second, according to the response time of solid waste incineration link, waste heat utilization link and flue gas treatment link, a multi-time scale operation optimization objective function is established;Design double-layer multi-objective competitive group optimization algorithm to solve the dynamic collaborative operation optimization problem of urban solid waste incineration process;Finally, according to the optimization characteristics of different time scales, determine the optimal decision, and use the actual urban solid waste incineration process data to verify the optimization performance.The present application realizes the pollution reduction and efficiency increasing of urban solid waste incineration process, and has important theoretical significance and application value.

Description

Technical Field

[0001] This invention relates to an intelligent optimization method for pollution reduction and efficiency improvement in urban solid waste incineration processes. A model was established for the pollutant generation concentration, main steam flow rate, and pollutant emission concentration during the urban solid waste incineration process. A dynamic collaborative operation optimization method involving multiple stages and time scales was designed, and an improved competitive group optimization algorithm and dynamic response strategy were proposed. This method achieves operational optimization of solid waste incineration, waste heat utilization, and flue gas purification processes. It belongs to both the field of urban solid waste management and the field of intelligent optimization. Background Technology

[0002] Urban solid waste incineration technology, characterized by significant volume and weight reduction, resource reuse, and thorough harmless treatment, has become a major method for urban solid waste management. However, the calorific value of solid waste fluctuates greatly, leading to instability in the incineration process, low waste heat utilization efficiency, and high pollutant emission concentrations. Therefore, achieving dynamic and coordinated optimization of the urban solid waste incineration process to improve waste heat utilization efficiency, reduce pollutant emission concentrations, and lower operating costs has significant theoretical and practical value. Summary of the Invention

[0003] This invention provides a dynamic and coordinated operation optimization method for the solid waste incineration, waste heat utilization, and flue gas purification processes in urban solid waste incineration, so as to achieve compliance with nitrogen oxide emission concentration standards in the urban solid waste incineration process while improving combustion efficiency.

[0004] The present invention adopts the following technical solution and implementation steps:

[0005] 1. Data collection;

[0006] 2. Determine the input variables and corresponding output variables of the operation index models for the urban solid waste incineration process, including NOx generation concentration, SO2 generation concentration, HCl generation concentration, main steam flow rate, NOx emission concentration, SO2 emission concentration, and HCl emission concentration. Establish operation index models using fuzzy neural networks (FNN).

[0007] (1) Determine the input variables for each operational indicator model:

[0008] The established urban solid waste incineration process operation index model includes the solid waste incineration process, the waste heat utilization process, and the flue gas treatment process, and contains the key process variables of each process.

[0009] The model for NOx generation concentration, SO2 generation concentration, HCl generation concentration, and main steam flow rate includes key process variables for both solid waste incineration and waste heat utilization.

[0010] Model 1 is a NOx generation concentration model, with input variables X1 = [x FT ,xFO ,x FP ,x GT ,x NOx-p ] T The input dimension n1 = 5, where x FT The average temperature of the furnace, x FO For the oxygen content of flue gas, x FP For furnace negative pressure, x GT For the flue gas temperature at the furnace outlet, x NOx-p The NOx concentration at the previous time step is represented by the output variable Y1, which represents the NOx concentration at the previous time step.

[0011] Model 2 is a model for SO2 production concentration, with input variables X2 = [x FT ,x FO ,x FP ,x GT ,x SO2-p ] T The input dimension is n² = 5, where x SO2-p The output variable Y2 represents the SO2 concentration generated at the previous time step.

[0012] Model 3 is the HCl production concentration model, and its input variable is X3 = [x FT ,x FO ,x FP ,x GT ,x HCl-p ] T The input dimension is n3 = 5, where x HCl-p The output variable Y3 represents the concentration of HCl produced at the previous moment;

[0013] Model 4 is the main steam flow model, and its input variable is X4 = [x FT ,x FO ,x FP ,x GW ,x MSF ] T The input dimension is n4 = 5, where x GW Main water supply flow rate, x MSF Y4 represents the main steam flow rate at the previous moment; the output variable Y4 represents the main steam flow rate.

[0014] The operational indicator models for NOx emission concentration, SO2 emission concentration, and HCl emission concentration include key process variables from the solid waste incineration and flue gas treatment processes:

[0015] Model 5 is a NOx emission concentration model, and its input variable is X5 = [x FT ,x FO ,x S ,x NOx-m ]T The input dimension n5 = 4, where x S x represents the amount of urea solution used. NOx-m The NOx emission concentration is the value at the previous time step; the output variable Y5 represents the NOx emission concentration.

[0016] Model 6 is an SO2 emission concentration model, and its input variable is X6 = [x FT ,x FO ,x C ,x SO2-m ] T The input dimension is n6 = 4, where x C x represents the amount of limestone slurry used. SO2-m The SO2 emission concentration at the previous moment is given; the output variable Y6 represents the SO2 emission concentration.

[0017] Model 7 is an HCl emission concentration model, and its input variable is X7 = [x FT ,x FO ,x C ,x HCl-m ] T The input dimension is n7 = 4, where x HCl-m Y7 represents the HCl emission concentration at the previous moment; the output variable Y7 represents the HCl emission concentration.

[0018] (2) The above-mentioned operational indicator models are established using FNN, as follows:

[0019] The performance indicator model based on FNN is represented as follows:

[0020]

[0021] Where k∈[1,Z], Z=7 is the number of operational indicator models, Y k (X k (t) represents the output of the k-th model, X k (t) represents the input of the k-th model at time t, and n k Let r = 15 be the number of input variables for the k-th model, and c be the number of fuzzy rules in the FNN. k,ij (t) and σ k,ij (t) represents the center and width of the i-th input and j-th membership function of the k-th model at time t, respectively; t = 1, 2, 3, ... represents the sampling time of the urban solid waste incineration process;

[0022] A k,j (X k (t) is a first-order Takagi-Sugeno-Kang, i.e., TSK fuzzy rule, defined as:

[0023]

[0024] Where a k,j0 (t) and a k,ji (t) represents the first-order TSK fuzzy rule coefficient at time t;

[0025] For the center c of the kth model k,ij (t), width σ k,ij (t) and fuzzy rule coefficient a k,ji (t), whose value is to be determined, is solved by the online learning algorithm given by equations (4)-(7);

[0026] The absolute percentage error (MAPE) of the k-th model is:

[0027]

[0028] Where Y k,p (X k (t) represents the actual output of the k-th model on the p-th sample at time t. Let N be the expected output of the k-th model on the p-th sample at time t. m The amount of data used to train the model;

[0029] Set the model fitness threshold δ = 0.05, if MAPE k If (t) > δ, then the model is mismatched and will be updated online; otherwise, the current model will be maintained.

[0030] The loss function for online model learning is defined as:

[0031]

[0032] in Let e ​​be the loss function of the k-th model at time t. k,p (t) represents the output error of the k-th model on the p-th sample at time t;

[0033] An adaptive Levenberg-Marquardt (LM) algorithm is proposed to update model parameters. The online learning method is as follows:

[0034]

[0035] in The number of iterations for the adaptive LM algorithm. Let E be the maximum number of iterations for the adaptive LM algorithm, and E be the identity matrix. Let be the parameter matrix of the k-th model during the τ-th iteration of the algorithm; and Let the Hessian matrix and gradient vector at the τ-th iteration be respectively, and defined as:

[0036]

[0037] Adaptive learning rate at the τth iteration of the algorithm for:

[0038]

[0039] Where ||·|| is the L2 norm. Let ω be the output error vector of the k-th model at time t in the τ-th iteration, where ω = 0.5 is a constant.

[0040] 3. Based on the response times of the solid waste incineration, waste heat utilization, and flue gas treatment stages, establish an objective function for optimizing the pollution reduction and efficiency improvement of the urban solid waste incineration process;

[0041] Based on the different response times of the solid waste incineration, waste heat utilization, and flue gas treatment stages involved in urban solid waste incineration, the optimization of pollution reduction and efficiency improvement in urban solid waste incineration is divided into upper-level optimization and lower-level optimization at different time scales; multiple time scales are defined as:

[0042]

[0043] Where t1 is the upper-level optimization time, T1 is the upper-level optimization time resolution, set to 1 hour, t2 is the lower-level optimization time, T2 is the lower-level optimization time resolution, set to 10 minutes, i = 1, 2, ..., j = 1, 2, ..., T1 / T2;

[0044] When t = t1, upper-level optimization is carried out. Upper-level optimization targets the solid waste incineration and waste heat utilization stages of the urban solid waste incineration process. The optimization objectives are to reduce pollutant concentration and increase the main steam flow rate and its stability to ensure a relatively stable incineration process. The upper-level optimization problem is defined as follows:

[0045]

[0046] The decision variable D(t1) of the upper-level optimization problem is [x FT (t1),x FO (t1),x GW (t1)] T Std.(·) is the standard deviation calculation function, f1(D(t1)) is the nonlinear functional relationship between D(t1) and NOx concentration, f2(D(t1)) is the nonlinear functional relationship between D(t1) and SO2 and HCl concentration, and f3(D(t1)) is the nonlinear functional relationship between D(t1) and main steam flow rate.

[0047] The constraints for upper-level optimization are:

[0048]

[0049] Where x FT,max and x FT,min x represents the maximum and minimum furnace temperatures, respectively. FO,max and x FO,min These represent the maximum and minimum oxygen content in the flue gas, respectively. GW,max and x GW,min These are the maximum and minimum values ​​of the main water supply, respectively.

[0050] When t = t2, lower-level optimization is carried out; the lower-level optimization targets the flue gas treatment process, and the optimization objectives are to reduce the concentration of pollutant emissions and reduce the usage of urea solution and limestone slurry. The lower-level optimization problem is defined as follows:

[0051]

[0052] The decision variable D(t2) for the lower-level optimization problem is: [x S (t2),x C (t2)] T x S (t2) represents the amount of urea solution used, x C (t2) represents the amount of limestone slurry used, f4(D(t2)) represents the nonlinear functional relationship between D(t2) and the emission concentrations of NOx, SO2 and HCl, and f5(D(t2)) represents the amount of flue gas treatment agent used. and The optimal setting value for the upper-level optimization is the solution result of equation (9), where Optimal furnace temperature setting. The optimal setpoint for the oxygen content in the flue gas;

[0053] The constraints for lower-level optimization are:

[0054]

[0055] Where x S,max and x S,min These represent the maximum and minimum values ​​of urea solution usage, x C,max and x C,min These represent the maximum and minimum values ​​for the amount of limestone slurry used, respectively.

[0056] 4. Design a two-level multi-objective competitive swarm optimization (DMCSO) algorithm to solve the optimization problem of pollution reduction and efficiency improvement in urban solid waste incineration process;

[0057] (1) Competitive strategy

[0058] Pareto dominance definition: For decision variables D1(t) and D2(t), when There is fi (D1(t))≤f j (D2(t)), and j∈[1,M], f i (D1(t))<f j If (D2(t)), then D1(t) is said to dominate D2(t); where M is the number of optimization objectives;

[0059] Define the DMCSO algorithm population, consisting of N = 100 individuals, where the i-th individual is defined as I. i =[V i ,P i ] T V i =[v i,1 ,v i,2 ,...,v i,d ] T Let P be the velocity of the i-th individual. i =[p i,1 ,p i,2 ,...,p i,d ] T Let d be the position of the i-th individual, and d be the dimension of the decision variable in the optimization problem; a pair of individuals are randomly selected from the population to compete, and the competition strategy is expressed as:

[0060] 1) If a Pareto dominance relationship exists between a pair of individuals, the dominated individual is the loser and the dominating individual is the winner.

[0061] 2) If there is no Pareto dominance relationship between individuals, then the index is used. Calculating and obtaining an individual's comprehensive performance evaluation has significant advantages. The individual who results is a loser, while those with a smaller... The individual who achieves the result is the winner;

[0062] Taking the i-th individual in the population as an example, its index-based... The overall performance evaluation of the calculation is as follows:

[0063]

[0064] Where t is the sampling time; when t = t1, it represents the upper-level optimization; when t = t2, it represents the lower-level optimization. This represents the number of iterations in the DMCSO algorithm. Let P be the maximum number of iterations for the algorithm, M be the target number, ρ(·) be the sorting function in ascending order, and P be the target number. i τ (t) represents the position of the i-th individual at time t during the τ-th iteration of the algorithm, f m (P i τ(t) represents the target value of the i-th individual on the m-th target at time t during the τ-th iteration of the algorithm. and These are the maximum and minimum values ​​on the m-th target at time t during the τ-th iteration of the algorithm;

[0065] The competition is repeated until all individuals in the population participate in the competition, and the population is divided into two camps: winners and losers.

[0066] (2) Upper-level optimization strategy

[0067] When t = t1, perform upper-level optimization, that is, solve the optimization problem given by equation (9) to obtain the optimal setpoint of the upper-level optimization. The upper-level optimization problem is a complex multimodal problem. The algorithm should have a good ability to explore the global optima and get rid of local optima. Based on this, optimization strategies are designed for the winners and losers respectively.

[0068] 1) The learning strategies of losers are:

[0069]

[0070] Where V l τ (t1) represents the velocity of the loser in the τth iteration of the DMCSO algorithm at time t1, P l τ (t1) represents the position of the loser at time t1 during the τth iteration of the algorithm. Let t1 be the position of the winner in the τth iteration of the algorithm. and These are vectors composed of random numbers generated during the τth iteration of the algorithm, where ψ = 1.2 is the learning rate of the algorithm. P represents the position of the individual with the best overall performance in the population at time t1 during the τth iteration of the algorithm. τ (t1) is the matrix consisting of the positions of all individuals in the population at time t1 during the τth iteration of the algorithm;

[0071] 2) The winner's learning strategy is:

[0072]

[0073] in Let be the velocity of the winner at time t1 during the τth iteration of the algorithm, and let α and β be vectors following a Gaussian distribution. The variance σ of the Gaussian distribution α and σ β They are respectively:

[0074]

[0075] Where Γ(·) is the gamma function;

[0076] (3) Lower-level optimization strategy

[0077] When t = t2, perform lower-level optimization, that is, solve the optimization problem given by equation (11) to obtain the optimal setpoint of the lower-level optimization. The lower-level optimization problem is a typical unimodal problem; based on this, an optimization strategy is designed.

[0078] The feasible subgroup of the algorithm in each iteration is defined as:

[0079]

[0080] in Let t2 be the i-th individual in the population during the τ-th iteration of the DMCSO algorithm, and Ω be the set that satisfies the constraint of equation (12). Let i be the i-th individual in the feasible subgroup at time t2 during the τ-th iteration of the algorithm.

[0081] Therefore, the dynamic feasible solution space is:

[0082]

[0083] in and The decision variable x is the decision variable x at the (τ+1)th iteration of the algorithm at time t2. S (t2) and x C The feasible upper bound of (t2) and The decision variable x is the decision variable x at the (τ+1)th iteration of the algorithm at time t2. S (t2) and x C A feasible lower bound for (t2);

[0084] The learning strategies of winners are represented as follows:

[0085]

[0086] Where N S =0.2N is the number of sparks produced by the explosion. Let t2 be the position of the winner in the τth iteration of the algorithm. P represents the position of the individual with the best overall performance in the population at time t2 during the τth iteration of the algorithm. τ (t2) is the matrix consisting of the positions of all individuals in the population at time t2 during the τth iteration of the algorithm. Let t2 be the i-th random vector following a Gaussian distribution during the τ-th iteration of the algorithm.

[0087] The loser also adopts the optimization strategy given in equation (14);

[0088] (4) External archive maintenance

[0089] During the algorithm's optimization process, the discovered non-dominated solutions will be stored in an external archive, which is as follows:

[0090]

[0091] Where t is the sampling time, when t = t1, it is the upper-level optimization, when t = t2, it is the lower-level optimization, rep τ For the external archive of the τth iteration, Let i be the i-th non-dominated solution in the external archive at time t during the τ-th iteration of the algorithm. Let be the number of non-dominated solutions in the external archive during the τth iteration of the algorithm;

[0092] External archive maintenance is represented as:

[0093]

[0094] in Let be the solution to be deleted from the external archive during the τ-th iteration at time t. The performance evaluation index is given by equation (13).

[0095] 5. Based on the characteristics of upper-level and lower-level optimization, design a dual decision-making strategy to obtain the optimal setpoint;

[0096] For upper-level optimization, the optimal setpoint is selected as the non-dominated solution that achieves the best overall improvement in NOx, SO2, and HCl concentrations, as well as the stability of the main steam flow rate. The optimal decision for upper-level optimization is expressed as:

[0097]

[0098] Where S * (t1) is the optimal setpoint selected at time t1, S i,k (t1) represents the result of the i-th non-dominated solution in the external archive at time t1 on the performance index corresponding to the k-th model, Y. k (X k (t1)) represents the output of the k-th model at time t1, X k (t1) is the input of the k-th model at time t1;

[0099] For the lower-level optimization, the non-dominated solution that achieves the best overall improvement in NOx, SO2, and HCl emission concentrations as well as the amount of reagent used is selected as the optimal setpoint. The optimal decision for the lower-level optimization is expressed as:

[0100]

[0101] Where S * (t2) represents the optimal setpoint selected at time t2, S i,k (t2) represents the result of the i-th non-dominated solution in the external archive at time t2 on the performance index corresponding to the k-th model. i,j (t2) is the j-th dimension decision variable of the i-th non-dominated solution in the external archive at time t2, Y k (X k (t2)) represents the output of the k-th model at time t2. Let x be the input of the k-th model at time t2. k,j (t2) represents the j-th dimension input of the k-th model at time t2, where n k Let be the input dimension of the k-th model.

[0102] This invention has the following obvious advantages and beneficial effects:

[0103] This invention achieves dynamic collaborative operation optimization of multiple stages in the urban solid waste incineration process based on neural networks and competitive group optimization algorithms. It establishes an accurate and effective operational index model for the solid waste incineration process and realizes efficient optimization solutions. This can improve the main steam flow rate and stability, reduce NOx, SO2, and HCl emission concentrations, and reduce flue gas treatment costs, which is of great significance for the stable operation and sustainable development of the urban solid waste incineration process. Attached Figure Description

[0104] Figure 1 This is a flowchart of the present invention;

[0105] Figure 2 An optimization diagram for the dynamic and coordinated operation of urban solid waste incineration process across multiple time scales;

[0106] Figure 3 Results of the operational index model for urban solid waste incineration process;

[0107] Figure 4 The results of dynamic and coordinated operation optimization for urban solid waste incineration processes;

[0108] Figure 5 The optimal setpoints for furnace temperature, flue gas oxygen content, and main feedwater are used to optimize the decision variables for the upper level.

[0109] Figure 6 The optimal settings for the usage of urea solution and limestone slurry are used as decision variables for lower-level optimization. Detailed Implementation

[0110] This invention utilizes collected historical data to establish operational index models for the NOx, SO2, and HCl generation concentrations, main steam flow rate, and NOx, SO2, and HCl emission concentrations during urban solid waste incineration. It also utilizes the sampling of the dataset to conduct dynamic collaborative operation optimization of the urban solid waste incineration process according to the corresponding time scales.

[0111] As an example, the effectiveness of the proposed method was verified using actual data from a solid waste incineration plant in Beijing. The original data sampling interval was 1 minute, resulting in 16.7 hours of experimental data. Variables required for obtaining the operational index model were extracted as input variables for the model, as detailed in step 2. The first 2 hours of data were used for pre-training of the index model, and the remaining 14.7 hours of data were read at sampling intervals and fed into the algorithm to test the online dynamic collaborative operation optimization effect of multiple stages in the urban solid waste incineration process.

[0112] (1) An operational index model based on a fuzzy neural network was established using the method in step 2. Based on 100 sets of historical data, the adaptive LM algorithm in step 2 was used to train the operational index model for NOx, SO2, and HCl generation concentrations, main steam flow rate, and NOx, SO2, and HCl emission concentrations. The model pre-training results are as follows: Figure 3 As shown, the X-axis represents time in hours, and the Y-axis represents NOx concentration in mg / m³. 3 SO2 concentration, unit is mg / m³ 3 The concentration of HCl produced is expressed in mg / m³. 3 Main steam flow rate, in t / h; NOx emission concentration, in mg / m³. 3 SO2 emission concentration, unit is mg / m³ 3 HCl emission concentration, unit is mg / m³ 3 .

[0113] (2) Based on the obtained operational index model of the urban solid waste incineration process, the multi-timescale dynamic collaborative operation optimization scheme in step 3 was adopted to establish the upper and lower level optimization objective functions and constraints. Based on 14.7 hours of data, the DMCSO algorithm in step 4 was used to solve the upper and lower level optimization problems according to the time scale. The dynamic collaborative operation optimization results of the urban solid waste incineration process are as follows: Figure 4 As shown, the X-axis represents time in hours, and the Y-axis represents NOx emission concentration in mg / m³. 3 SO2 emission concentration, unit is mg / m³ 3 HCl emission concentration, unit is mg / m³ 3 The main steam flow rate is expressed in tons per hour (t / h). The optimal setpoints for the upper-level optimization decision variables are as follows: Figure 5As shown, the X-axis represents time in hours, the Y-axis represents furnace temperature in °C, flue gas oxygen content in %, and main feedwater in t / h. The optimal setpoints for the lower-level optimization decision variables are as follows: Figure 6 As shown, the X-axis represents time in hours, and the Y-axis represents urea solution in L / h and limestone slurry in t / h.

[0114] (3) The optimization results and improvement rate were used to quantitatively evaluate the operational optimization effect. The calculation results were as follows: the original actual average NOx emission concentration was 150.60 mg / m³. 3 The optimized average NOx emission concentration was 104.09 mg / m³. 3 NOx emission concentration decreased by an average of 30.16%, while the original average SO2 emission concentration was 0.93 mg / m³. 3 The optimized average SO2 emission concentration is 0.50 mg / m³. 3 SO2 emission concentration decreased by an average of 42.75%, while the original average HCl emission concentration was 4.42 mg / m³. 3 The optimized average HCl emission concentration was 3.36 mg / m³. 3 The average NOx emission concentration was reduced by 23.65%, and the original actual main steam flow rate was 77.41 t / h, while the optimized average main steam flow rate was 82.25 mg / m³. 3 The main steam flow rate increased by 6.84%, and the original standard deviation of the actual main steam flow rate was 6.01, while after optimization, the standard deviation of the main steam flow rate was 2.94, resulting in a 51.08% improvement in steam flow stability. The original actual urea solution usage was 26.08 L / h, while after optimization, the urea solution usage was 12.29 L / h, a reduction of 52.86%. The original limestone slurry usage was 26.08 L / h, while after optimization, the urea solution usage was 12.29 L / h, a reduction of 52.86%. The original limestone slurry usage was 10.51 t / h, while after optimization, the limestone slurry usage was 5.12 t / h, a reduction of 51.31%.

[0115] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. It should be noted that the above descriptions are merely specific embodiments of this invention and do not limit the invention. All modifications and optimizations made within the spirit and principles of this invention should be covered by the claims of this invention.

Claims

1. A smart optimization method for pollution reduction and efficiency improvement in urban solid waste incineration processes, characterized in that, Includes the following steps: Step 1: Data Collection; Step 2: Determine the input and output variables of the operating index model, and use a fuzzy neural network (FNN) to establish operating index models for NOx generation concentration, SO2 generation concentration, HCl generation concentration, main steam flow rate, NOx emission concentration, SO2 emission concentration, and HCl emission concentration. Step 3: Based on the response times of the solid waste incineration, waste heat utilization, and flue gas treatment stages, establish the objective function for optimizing the pollution reduction and efficiency improvement of the urban solid waste incineration process. Step 4: Design a two-layer multi-objective competitive group optimization (DMCSO) algorithm to solve the optimization problem of pollution reduction and efficiency improvement in the urban solid waste incineration process; Step 5: Obtain the optimal decision by determining the optimized main steam flow rate, NOx, SO2, HCl emission concentrations, and flue gas treatment agent input, namely the amount of urea solution and limestone slurry used. In step 3, an objective function is established for optimizing the pollution reduction and efficiency improvement problem of the urban solid waste incineration process based on the response times of the solid waste incineration, waste heat utilization, and flue gas treatment processes. Based on the different response times of the solid waste incineration, waste heat utilization, and flue gas treatment stages involved in urban solid waste incineration, the optimization of pollution reduction and efficiency improvement in urban solid waste incineration is divided into upper-level optimization and lower-level optimization at different time scales; multiple time scales are defined as: Where t1 is the upper-level optimization time, T1 is the upper-level optimization time resolution, set to 1 hour, t2 is the lower-level optimization time, T2 is the lower-level optimization time resolution, set to 10 minutes, i = 1, 2, ..., j = 1, 2, ..., T1 / T2; When t = t1, upper-level optimization is carried out. Upper-level optimization targets the solid waste incineration and waste heat utilization stages of the urban solid waste incineration process. The optimization objectives are to reduce pollutant concentration and increase the main steam flow rate and its stability to ensure a relatively stable incineration process. The upper-level optimization problem is defined as follows: The decision variable D(t1) of the upper-level optimization problem is [x FT (t1),x FO (t1),x GW (t1)] T Std.(·) is the standard deviation calculation function, f1(D(t1)) is the nonlinear functional relationship between D(t1) and NOx concentration, f2(D(t1)) is the nonlinear functional relationship between D(t1) and SO2 and HCl concentration, and f3(D(t1)) is the nonlinear functional relationship between D(t1) and main steam flow rate. The constraints for upper-level optimization are: Where x FT,max and x FT,min x represents the maximum and minimum furnace temperatures, respectively. FO,max and x FO,min These represent the maximum and minimum oxygen content in the flue gas, respectively. GW,max and x GW,min These are the maximum and minimum values ​​of the main water supply, respectively; When t = t2, lower-level optimization is carried out; the lower-level optimization targets the flue gas treatment process, and the optimization objectives are to reduce the concentration of pollutant emissions and reduce the usage of urea solution and limestone slurry. The lower-level optimization problem is defined as follows: The decision variable D(t2) for the lower-level optimization problem is: [x S (t2),x C (t2)] T x S (t2) represents the amount of urea solution used, x C (t2) represents the amount of limestone slurry used, f4(D(t2)) represents the nonlinear functional relationship between D(t2) and the emission concentrations of NOx, SO2 and HCl, and f5(D(t2)) represents the amount of flue gas treatment agent used. and The optimal setting value for the upper-level optimization is the solution result of equation (9), where Optimal furnace temperature setting. The optimal setpoint for the oxygen content in the flue gas; The constraints for lower-level optimization are: Where x S,max and x S,min These represent the maximum and minimum values ​​of urea solution usage, x C,max and x C,min These represent the maximum and minimum values ​​for the amount of limestone slurry used, respectively. In step 4, a two-layer multi-objective competitive swarm optimization (DMCSO) algorithm is designed to solve the optimization problem of pollution reduction and efficiency improvement in the urban solid waste incineration process: (1) Competitive strategy Pareto dominance definition: For decision variables D1(t) and D2(t), when There is f i (D1(t))≤f j (D2(t)), and f i (D1(t)) <f j If (D2(t)), then D1(t) is said to dominate D2(t); where M is the number of optimization objectives; Define the DMCSO algorithm population, consisting of N = 100 individuals, where the i-th individual is defined as I. i =[V i ,P i ] T V i =[v i,1 ,v i,2 ,...,v i,d ] T Let P be the velocity of the i-th individual. i =[p i,1 ,p i,2 ,...,p i,d ] T Let d be the position of the i-th individual, and d be the dimension of the decision variable in the optimization problem; a pair of individuals are randomly selected from the population to compete, and the competition strategy is expressed as: 1) If a Pareto dominance relationship exists between a pair of individuals, the dominated individual is the loser and the dominating individual is the winner. 2) If there is no Pareto dominance relationship between individuals, the comprehensive performance evaluation of individuals is obtained by using the index J. Individuals with larger J results are considered losers, while individuals with smaller J results are considered winners. Taking the i-th individual in the population as an example, its comprehensive performance evaluation based on index J is as follows: Where t is the sampling time, when t = t1, it is the upper-level optimization, when t = t2, it is the lower-level optimization, and τ∈[1,T]. q [T] represents the number of iterations in the DMCSO algorithm. q =50 is the maximum number of iterations for the algorithm, M is the target number for optimization, ρ(·) is the sorting function from smallest to largest, and P i τ (t) represents the position of the i-th individual at time t during the τ-th iteration of the algorithm, f m (P i τ (t) represents the target value of the i-th individual on the m-th target at time t during the τ-th iteration of the algorithm. and These are the maximum and minimum values ​​on the m-th target at time t during the τ-th iteration of the algorithm; The competition is repeated until all individuals in the population participate in the competition, and the population is divided into two camps: winners and losers. (2) Upper-level optimization strategy When t = t1, perform upper-level optimization, that is, solve the optimization problem given by equation (9) to obtain the optimal setpoint of the upper-level optimization. The upper-level optimization problem is a complex multimodal problem. The algorithm should have a good ability to explore the global optima and get rid of local optima. Based on this, optimization strategies are designed for the winners and losers respectively. 1) The learning strategies of losers are: Where V l τ (t1) represents the velocity of the loser in the τth iteration of the DMCSO algorithm at time t1, P l τ (t1) represents the position of the loser at time t1 during the τth iteration of the algorithm. Let t1 be the position of the winner in the τth iteration of the algorithm. and These are vectors composed of random numbers generated during the τth iteration of the algorithm, where ψ = 1.2 is the learning rate of the algorithm. P represents the position of the individual with the best overall performance in the population at time t1 during the τth iteration of the algorithm. τ (t1) is the matrix consisting of the positions of all individuals in the population at time t1 during the τth iteration of the algorithm; 2) The winner's learning strategy is: in Let be the velocity of the winner at time t1 during the τth iteration of the algorithm, and let α and β be vectors following a Gaussian distribution. The variance σ of the Gaussian distribution α and σ β They are respectively: Where Γ(·) is the gamma function; (3) Lower-level optimization strategy When t = t2, perform lower-level optimization, that is, solve the optimization problem given by equation (11) to obtain the optimal setpoint of the lower-level optimization. The lower-level optimization problem is a typical unimodal problem; based on this, an optimization strategy is designed. The feasible subgroup of the algorithm in each iteration is defined as: in Let t2 be the i-th individual in the population during the τ-th iteration of the DMCSO algorithm, and Ω be the set that satisfies the constraint of equation (12). Let i be the i-th individual in the feasible subgroup at time t2 during the τ-th iteration of the algorithm. Therefore, the dynamic feasible solution space is: in and The decision variable x is the decision variable x at the (τ+1)th iteration of the algorithm at time t2. S (t2) and x C The feasible upper bound of (t2) and The decision variable x is the decision variable x at the (τ+1)th iteration of the algorithm at time t2. S (t2) and x C A feasible lower bound for (t2); The learning strategies of winners are represented as follows: Where N S =0.2N is the number of sparks produced by the explosion. Let t2 be the position of the winner during the τth iteration of the algorithm. P represents the position of the individual with the best overall performance in the population at time t2 during the τth iteration of the algorithm. τ (t2) is the matrix consisting of the positions of all individuals in the population at time t2 during the τth iteration of the algorithm. Let t2 be the i-th random vector that follows a Gaussian distribution during the τ-th iteration of the algorithm. The loser also adopts the optimization strategy given in equation (14); (4) External archive maintenance During the algorithm's optimization process, the discovered non-dominated solutions will be stored in an external archive, which is as follows: Where t is the sampling time, when t = t1, it is the upper-level optimization, when t = t2, it is the lower-level optimization, rep τ For the external archive of the τth iteration, Let i be the i-th non-dominated solution in the external archive at time t during the τ-th iteration of the algorithm. Let be the number of non-dominated solutions in the external archive during the τth iteration of the algorithm; External archive maintenance is represented as: in Let J(·) be the solution to be deleted from the external archive in the τth iteration at time t, and let J(·) be the comprehensive performance evaluation given by equation (6).

2. The intelligent optimization method for pollution reduction and efficiency improvement in urban solid waste incineration process according to claim 1, characterized in that, In step 2, the method for selecting input and output variables and establishing an FNN-based performance indicator model is as follows: (1) Determine the input variables for each operational indicator model: The established urban solid waste incineration process operation index model includes the solid waste incineration process, the waste heat utilization process, and the flue gas treatment process, and contains the key process variables of each process. The model for NOx generation concentration, SO2 generation concentration, HCl generation concentration, and main steam flow rate includes key process variables for both solid waste incineration and waste heat utilization. Model 1 is a NOx generation concentration model, with input variables X1 = [x FT ,x FO ,x FP ,x GT ,x NOx-p ] T The input dimension n1 = 5, where x FT The average temperature of the furnace, x FO For the oxygen content of flue gas, x FP For furnace negative pressure, x GT For the flue gas temperature at the furnace outlet, x NOx-p The NOx concentration at the previous time step is given; the output variable Y1 represents the NOx concentration. Model 2 is a model for SO2 production concentration, and its input variables are: The input dimension n² = 5, where The output variable Y2 represents the SO2 concentration generated at the previous time step. Model 3 is the HCl production concentration model, and its input variable is X3 = [x FT ,x FO ,x FP ,x GT ,x HCl-p ] T The input dimension is n3 = 5, where x HCl-p The output variable Y3 represents the concentration of HCl produced at the previous moment; Model 4 is the main steam flow model, and its input variable is X4 = [x FT ,x FO ,x FP ,x GW ,x MSF ] T The input dimension is n4 = 5, where x GW Main water supply flow rate, x MSF Y4 represents the main steam flow rate at the previous moment; the output variable Y4 represents the main steam flow rate. The operational indicator models for NOx emission concentration, SO2 emission concentration, and HCl emission concentration include key process variables from the solid waste incineration and flue gas treatment processes: Model 5 is a NOx emission concentration model, and its input variable is X5 = [x FT ,x FO ,x S ,x NOx-m ] T The input dimension n5 = 4, where x S x represents the amount of urea solution used. NOx-m This represents the NOx emission concentration at the previous moment. The output variable Y5 represents the NOx emission concentration; Model 6 is an SO2 emission concentration model, and its input variables are: The input dimension is n6 = 4, where x C This refers to the amount of limestone slurry used. The SO2 emission concentration at the previous moment is given; the output variable Y6 represents the SO2 emission concentration. Model 7 is an HCl emission concentration model, and its input variable is X7 = [x FT ,x FO ,x C ,x HCl-m ] T The input dimension is n7 = 4, where x HCl-m This represents the HCl emission concentration at the previous moment; The output variable Y7 represents the HCl emission concentration; (2) The above-mentioned operational indicator models are established using FNN, as follows: The performance indicator model based on FNN is represented as follows: Where k∈[1,Z], Z=7 is the number of operational indicator models, Y k (X k (t) represents the output of the k-th model, X k (t) represents the input of the k-th model at time t, and n k Let r = 15 be the number of input variables for the k-th model, and c be the number of fuzzy rules in the FNN. k,ij (t) and σ k,ij (t) represents the center and width of the i-th input and j-th membership function of the k-th model at time t, respectively; t = 1, 2, 3, ... represents the sampling time of the urban solid waste incineration process; A k,j (X k (t) is a first-order Takagi-Sugeno-Kang, i.e., TSK fuzzy rule, defined as: Where a k,j0 (t) and a k,ji (t) represents the first-order TSK fuzzy rule coefficient at time t; For the center c of the kth model k,ij (t), width σ k,ij (t) and fuzzy rule coefficient a k,ji (t), whose value is to be determined, is solved by the online learning algorithm given by equations (4)-(7); The absolute percentage error (MAPE) of the k-th model is: Where Y k,p (X k (t) represents the actual output of the k-th model on the p-th sample at time t. Let N be the expected output of the k-th model on the p-th sample at time t. m The amount of data used to train the model; Set the model fitness threshold δ = 0.05, if MAPE k If (t)>δ, the model is mismatched and will be updated online; otherwise, the current model will be maintained. The loss function for online model learning is defined as: Where L k (t) is the loss function of the k-th model at time t, e k,p (t) represents the output error of the k-th model on the p-th sample at time t; An adaptive Levenberg-Marquardt (LM) algorithm is proposed to update model parameters. The online learning method is as follows: Where τ∈[1,T] m [T] represents the number of iterations in the adaptive LM algorithm. m =50 represents the maximum number of iterations in the adaptive LM algorithm, and E is the identity matrix. Let be the parameter matrix of the k-th model during the τ-th iteration of the algorithm; and Let the Hessian matrix and gradient vector at the τ-th iteration be respectively, and defined as: Adaptive learning rate at the τth iteration of the algorithm for: Where ||·|| is the L2 norm. Let ω be the output error vector of the k-th model at time t in the τ-th iteration, where ω = 0.5 is a constant.

3. The intelligent optimization method for pollution reduction and efficiency improvement in urban solid waste incineration process according to claim 1, characterized in that, In step 5, a dual-decision strategy is designed to obtain the optimal setpoint: For upper-level optimization, the non-dominated solution that achieves the best overall improvement in NOx, SO2, and HCl concentrations, as well as the stability of the main steam flow rate, is selected as the optimal setpoint. The optimal decision for upper-level optimization is expressed as: Where S * (t1) is the optimal setpoint selected at time t1, S i,k (t1) represents the result of the i-th non-dominated solution in the external archive at time t1 on the performance index corresponding to the k-th model, Y. k (X k (t1)) represents the output of the k-th model at time t1, X k (t1) is the input of the k-th model at time t1; For the lower-level optimization, the non-dominated solution that achieves the best overall improvement in NOx, SO2, and HCl emission concentrations as well as the amount of reagent used is selected as the optimal setpoint. The optimal decision for the lower-level optimization is expressed as: Where S * (t2) represents the optimal setpoint selected at time t2, S i,k (t2) represents the result of the i-th non-dominated solution in the external archive at time t2 on the performance index corresponding to the k-th model. i,j (t2) is the j-th dimension decision variable of the i-th non-dominated solution in the external archive at time t2, Y k (X k (t2)) represents the output of the k-th model at time t2. Let x be the input of the k-th model at time t2. k,j (t2) represents the j-th dimension input of the k-th model at time t2, where n k Let be the input dimension of the k-th model.

Images