Dioxin emission reduction optimization method based on dual-mode evaluation mechanism swarm intelligence algorithm

Abstract

The invention provides a dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligence algorithm, and relates to the technical field of solid waste incineration. Comprising the steps of obtaining an incineration process data set; performing space division on the incineration process data set according to an IT2FDT algorithm to obtain a subspace set; based on the IT2FNN, constructing a DXN concentration prediction model according to the subspace set; constructing a DXN emission parameter optimization model according to the DXN concentration prediction model and the DXN emission process parameters; solving the DXN emission parameter optimization model by using an improved PSO (Particle Swarm Optimization) algorithm to obtain an optimized process variable value; and substituting the optimized process variable value into the DXN concentration prediction model to obtain the DXN optimized emission concentration. According to the invention, the problem that the low DXN emission concentration process variable value is difficult to obtain in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of solid waste incineration, and in particular to a dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligent algorithm. Background Art

[0002] The annual increase in the total amount of global municipal solid waste (MSW) has aggravated the degree of environmental pollution and also affected the sustainable development of ecological civilization. Traditional landfill technology has the disadvantages of large land resource consumption and long treatment cycle, which makes MSW incineration (MSWI) technology, which has the advantages of harmlessness, reduction and resource utilization, gradually become one of the mainstream ways to treat MSW, and is also an important part of my country's green environmental protection industry and ecological civilization construction.

[0003] Due to the differences in MSW components, equipment types, operation and maintenance levels, and management models at home and abroad, my country's MSWI power plants generally adopt manual control modes, resulting in fluctuations in the concentration of tail gas pollutants, causing incineration enterprises to be listed in the list of polluting enterprises by the state. Among them, the emission control of dioxins (DXN), which are difficult to measure trace pollutants with strong carcinogenicity and mutagenicity, has become the focus of attention of environmental protection departments and the public. The state conducts unannounced flight inspections on DXN emission concentrations, and if the concentration exceeds the standard, the enterprise will be fined a huge amount. Therefore, my country's MSWI power plants generally adopt the method of excessive injection of purification materials such as activated carbon to avoid exceeding the DXN emission standard. Obviously, this method has a serious problem of waste of resources.

[0004] Research shows that DXN emission concentration is closely related to many process parameters, including manipulated variables (MV) such as air distribution and material distribution, controlled variables (CV) such as furnace temperature and flue gas oxygen content, auxiliary variables (AV) such as combustion grate temperature and reactor inlet temperature, and environmental indicators (EI) such as CO and HCl. In actual industrial sites, experienced domain experts perform CV control and MV operation based on CO emission concentration to achieve rough control of DXN emission concentration. Obviously, this method has serious randomness and it is difficult to maintain stable DXN emission concentration in the long term. Therefore, how to provide domain experts with the optimal key process parameter values ​​and reduce DXN emission concentration is an important problem that needs to be solved urgently in my country's MSWI power plants, and it is also the key to improving the economic and social benefits of enterprises.

[0005] Obtaining soft-measured values ​​of DXN emission concentration can provide valuable references for the operation of domain experts, but in actual industrial processes, domain experts still need to refer to easily measured EIs, CVs, and AVs based on their own experience to adjust the output values ​​of MVs. Obviously, this model that relies on expert experience has a large degree of randomness and subjectivity, which can easily cause fluctuations in DXN emissions. At this time, it is more effective to use a swarm intelligence optimization algorithm with strong search capabilities and high adaptability to complex problems to obtain the optimal values ​​of key process parameters with the goal of minimizing DXN emission concentration. There is no report on how to combine the static and dynamic information of the population for evaluation to achieve optimization for DXN emission concentration reduction.

[0006] Dioxins (DXN) emitted during municipal solid waste incineration (MSWI) are trace and highly toxic pollutants that are difficult to measure and affect the sustainable development of the ecological environment and the survival of human beings. Effective control of DXN has always been the focus of the incineration industry and environmental protection departments. Limited by the hysteresis of detection values, expensive detection costs, the complexity of the whole process incineration mechanism, and the uncertainty of the "memory effect", the DXN operation optimization for the MSWI process has problems such as difficulty in building an efficient operation index model, difficulty in determining optimization decision variables, and difficulty in solving optimization algorithms. Summary of the invention

[0007] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligence algorithm. The present invention solves the problems in the prior art that the traditional landfill technology and the adjustment method relying on the experience of experts in the field are insufficient in dealing with environmental pollution and ecological sustainable development. This model relying on expert experience has a large degree of randomness and subjectivity, which easily leads to fluctuations in DXN emissions and low accuracy in obtaining DXN emission concentration measurements.

[0008] To achieve the above object, the present invention provides the following solutions:

[0009] A dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligence algorithm, comprising:

[0010] Obtain incineration process dataset;

[0011] According to the IT2FDT algorithm, the incineration process data set is spatially divided to obtain a subspace set;

[0012] Based on IT2FNN, a DXN concentration prediction model is constructed according to the subspace set;

[0013] Constructing a DXN emission parameter optimization model based on the DXN concentration prediction model and the process parameters of DXN emission;

[0014] The DXN emission parameter optimization model is solved by using an improved PSO algorithm to obtain optimized process variable values;

[0015] The optimized process variable values ​​are brought into the DXN concentration prediction model to obtain the optimized DXN emission concentration.

[0016] Preferably, the incineration process data set is spatially divided according to the IT2FDT algorithm to obtain a subspace set, including:

[0017] Extracting features from the incineration process data set and performing spatial segmentation according to the extracted features to obtain a left subspace and a right subspace;

[0018] Calculating a total mean square error (TMSE) for the left subspace and the right subspace;

[0019] Determine the segmentation variable and segmentation point with the smallest total mean square error to obtain the current divided subspace;

[0020] The IT2FDT algorithm is used to recursively partition the currently divided subspace to obtain a subspace set.

[0021] Preferably, the IT2FNN-based DXN concentration prediction model is constructed according to the subspace set, including:

[0022] Acquire the subspace set and perform fuzzification processing to obtain a fuzzy set corresponding to the subspace set;

[0023] The membership degree of fuzzy sets is calculated using Gaussian membership function;

[0024] Calculating a corresponding lower bound and an upper bound of the membership value according to the membership;

[0025] Determine the activation strength of the corresponding combination rule according to the lower bound of the membership value and the upper bound of the membership value;

[0026] Obtaining a lower bound value and an upper bound value of the subspace according to the activation strength of the combination rule and the corresponding consequent parameter;

[0027] Defuzzifying the lower bound value and the upper bound value of the subspace to obtain a feature mapping result;

[0028] According to the feature mapping results, the subspace set is updated by using the gradient descent method to obtain a DXN concentration prediction model.

[0029] Preferably, the expressions of the lower bound of the membership value and the upper bound of the membership value are respectively:

[0030]

[0031]

[0032] in, and The subspace R k The mth k The input variable corresponds to The membership function does not determine the lower bound, upper bound and width of the center. is the lower bound of the membership value, is the upper bound of the membership value, x kj mk is the mth value of the jth sample in the kth subspace k feature values.

[0033] Preferably, the expressions of the lower bound and upper bound of the subspace are respectively:

[0034]

[0035] in, is the weight of the consequent interval, and Respectively The lower and upper bounds of the activation strength of the rules, z k,j is the lower bound of the subspace, is the upper bound of the subspace.

[0036] Preferably, the expression of the DXN concentration prediction model is:

[0037]

[0038] Among them, Θ k is the output of the IT2FNN network trained in the kth subspace, I(·) is the indicator function, and f(·) is the operation indicator model based on IT2FDT.

[0039] Preferably, the expression of the DXN emission parameter optimization model is:

[0040]

[0041] Among them, D is the dimension of the decision variable after screening.

[0042] Preferably, the improved PSO algorithm is used to solve the DXN emission parameter optimization model to obtain the optimized process variable value, including:

[0043] Determine dynamic performance indicators and static performance indicators;

[0044] Performing trend judgment on the dynamic performance index and the static performance index to obtain a judgment result, if the judgment result is that both the dynamic performance index and the static performance index are in a downward trend, adjusting the parameter in a first manner, if the judgment result is that the dynamic performance index is in a downward trend, while the static performance index is increasing, adjusting the parameter in a second manner, if the judgment result is that the dynamic performance index is in an upward trend, while the static performance index is decreasing, adjusting the parameter in a third manner, if the judgment result is that both the dynamic performance index and the static performance index are in an upward trend, adjusting the parameter in a fourth manner, if the judgment result is that the dynamic performance index or the static performance index remains unchanged, adjusting the parameter in a fifth manner;

[0045] The optimized process variable value is obtained according to the adjusted corresponding parameters.

[0046] The present invention discloses the following technical effects:

[0047] The present invention provides a dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligent algorithm, comprising: obtaining an incineration process data set; spatially dividing the incineration process data set according to an IT2FDT algorithm to obtain a subspace set; constructing a DXN concentration prediction model according to the subspace set based on an IT2FNN; constructing a DXN emission parameter optimization model according to the DXN concentration prediction model and process parameters of DXN emissions; solving the DXN emission parameter optimization model using an improved PSO algorithm to obtain optimized process variable values; and bringing the optimized process variable values ​​into the DXN concentration prediction model to obtain the DXN optimized emission concentration. The present invention proposes an emission reduction optimization strategy that uses an intelligent optimization algorithm to obtain the optimal values ​​of multiple process parameters with the goal of minimizing the DXN pollutant emission concentration in the MSWI process, providing decision support for field experts; an IT2FDT algorithm is proposed, which enhances the ability of the leaf nodes of the tree structure algorithm in mining the mapping relationship between input features and output target values ​​by replacing the sample mean calculation method in the traditional DT algorithm, simplifies the indicator model structure and enhances the uncertainty processing capability; a PSO (DME-PSO) algorithm based on the DME mechanism is proposed, which realizes adaptive adjustment of control parameters while fully mining population status information, significantly improving the algorithm's solution capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. 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 creative labor.

[0049] Figure 1 A flow chart of a dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligent algorithm provided by an embodiment of the present invention;

[0050] Figure 2 A detailed schematic diagram of a dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligent algorithm provided by an embodiment of the present invention;

[0051] Figure 3 Schematic diagram of model fitting curve provided in an embodiment of the present invention, wherein (a) is a schematic diagram of training set fitting curve, and (b) is a schematic diagram of test set fitting curve;

[0052] Figure 4 A schematic diagram of the DXN emission concentration operation optimization results provided by an embodiment of the present invention;

[0053] Figure 5 A schematic diagram of furnace temperature comparison results provided by an embodiment of the present invention;

[0054] Figure 6 A schematic diagram of the comparison results of the secondary air volume provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] 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.

[0057] like Figure 1 As shown, the present invention provides a dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligence algorithm, comprising:

[0058] Step 100: Obtaining an incineration process data set;

[0059] Step 200: spatially partitioning the incineration process data set according to the IT2FDT algorithm to obtain a subspace set;

[0060] Step 300: Based on IT2FNN, a DXN concentration prediction model is constructed according to the subspace set;

[0061] Step 400: constructing a DXN emission parameter optimization model according to the DXN concentration prediction model and the process parameters of DXN emission;

[0062] Step 500: using an improved PSO algorithm to solve the DXN emission parameter optimization model to obtain optimized process variable values;

[0063] Step 600: Substitute the optimized process variable value into the DXN concentration prediction model to obtain the optimized DXN emission concentration.

[0064] Specifically, firstly, the mechanism characteristics of DXN in the MSWI process are analyzed, and it is determined to use multiple types of process parameters as the input of the operation index model; secondly, the Interval Type-2 Fuzzy decision tree (IT2FDT) algorithm is proposed to construct the DXN operation index model to enhance the interpretability and improve the uncertainty handling ability; finally, the dual-mode evaluation (DME) mechanism that combines static and dynamic is introduced into the particle swarm optimization (PSO) algorithm to fully mine the evolutionary information of the population and realize parameter adaptive adjustment, and the global convergence of the DME-PSO algorithm is proved.

[0065] MSWI process analysis for DXN emission concentration optimization:

[0066] From the perspective of the whole life cycle of DXN generation, combustion, regeneration, adsorption and emission concentration, taking a certain MSWI power plant in Beijing as an example, it can be divided into solid phase combustion area, gas phase combustion area, high temperature heat exchange area, low temperature heat exchange area, flue gas purification area and flue gas emission area. The specific description is as follows.

[0067] (1) MSW contains chlorine-containing compounds required for the production of DXN and may contain trace amounts of DXN (about 0.8 ngTEQ / kg). In the furnace, after the combustible components of MSW are fully burned, in addition to releasing the DXN it contains, key DXN precursors, which mainly include chlorobenzene, polychlorobenzenes, chlorophenols, polycyclic aromatic hydrocarbons and other incomplete combustion products, will be generated under local hypoxic conditions;

[0068] (2) The volatiles and products in the solid phase combustion zone are decomposed into CO during the high temperature combustion process. 2 To ensure that DXN can be completely decomposed in the furnace, the gas phase combustion process needs to meet strict "3T+E" process requirements, that is, the flue gas temperature is controlled above 850℃, the high temperature flue gas residence time exceeds 2 seconds, and sufficient flue gas turbulence intensity is ensured;

[0069] (3) The high-temperature gas phase synthesis of DXN mainly occurs in the high-temperature heat exchange zone, among which the oxidative coupling of chlorophenol is the main pathway for the formation of DXN. The reaction mechanism can be divided into three steps: the formation, coupling and cyclization of chlorophenoxyl radicals;

[0070] (4) The low-temperature heat exchange zone is dominated by low-temperature heterogeneous catalytic reactions, including precursor surface catalytic reactions and de novo synthesis reactions. Precursors such as chlorophenol and chlorobenzene produced by flue gas entrainment and residual carbon decomposition will use metals such as CuO as catalysts to generate DXN based on the Eley-Rideal and Langmuir-Hinshelwood mechanisms during the cooling process. In addition, the de novo synthesis reaction is based on carbon and generates DXN through elementary reactions such as chlorination and oxidation. Part of the DXN generated by the above reactions will diffuse into the flue gas, and most of it will remain in the fly ash. The generation of DXN at this stage is affected by many factors such as temperature, reaction time, residual carbon content in fly ash, chlorine form and content, catalyst, pH value and reaction environment (O 2 , H 2 O、Cl 2 and HCl), etc.

[0071] (5) The cooled flue gas passes through the deacidification reactor and bag filter in turn, and is physically adsorbed by activated carbon, Ca(OH) 2 Chemical removal and other methods are used to achieve emission concentrations of harmful substances such as acid gases, heavy metals and DXN that meet standards.

[0072] (6) Containing HCl and SO 2 、NO x The flue gas that meets the standards for substances such as DXN enters the chimney through the induced draft fan and is discharged into the atmosphere.

[0073] In the above-mentioned stages, there are problems such as unclear chemical reaction mechanism caused by factors such as uncertain MSW composition and unclear release mechanism of DXN "memory effect" attached to the inner wall of the equipment, which seriously increases the difficulty of building a high-efficiency and high-precision DXN soft measurement model.

[0074] From the above description, it can be seen that the DXN emission concentration is closely related to many variables in the entire MSWI process. Based on the actual data of a certain MSWI power plant in Beijing, the correlation analysis of all process variables and DXN emission concentration was carried out using mutual information, and 98 key process parameters including MVs, AVs, CVs and conventional EIs were selected based on expert experience as shown in Table 1 below:

[0075] Table 1 Summary of 98 key process parameters selected based on expert experience

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082] Further, such as Figure 2 As shown, the incineration process data set is spatially divided according to the IT2FDT algorithm to obtain a subspace set, including:

[0083] Extracting features from the incineration process data set and performing spatial segmentation according to the extracted features to obtain a left subspace and a right subspace;

[0084] Calculating a total mean square error (TMSE) for the left subspace and the right subspace;

[0085] Determine the segmentation variable and segmentation point with the smallest total mean square error to obtain the current divided subspace;

[0086] The IT2FDT algorithm is used to recursively partition the currently divided subspace to obtain a subspace set.

[0087] Specifically, the IT2FDT algorithm proposed in this paper adopts the same subspace partitioning method as the CART algorithm, which is described as follows.

[0088] Given the DXN training dataset Where N and M are the number and dimension of training data samples respectively, and the input feature space is defined as

[0089] First, in order to obtain the optimal segmentation subspace, the data set is divided into two subspaces, left and right, by traversing all features and trying multiple segmentation points for each feature.

[0090]

[0091] Among them, the subscript i represents the i-th sample, and the superscript m represents the m-th feature (variable). represents the value of the mth feature of the ith sample, and s represents the set threshold.

[0092] Next, the total mean square error of the two subspaces after division is calculated.

[0093]

[0094] Among them, x m is the mth segmentation variable, y i is the true value of the i-th sample; c L and c R are the sample means of the left and right subspaces, respectively, N L and N RThey represent the number of samples in the left and right subspaces respectively, and the superscripts L and R represent the left subspace and the right subspace respectively.

[0095]

[0096] Among them, N L and N R is the number of samples in the left and right subspaces.

[0097] Then, the segmentation variable and segmentation point with the smallest total mean square error are selected to end the subspace division, so that,

[0098]

[0099] Repeat the above process until the number of samples in all subspaces is less than the minimum number of samples in leaf nodes set based on experience.

[0100] Finally, the IT2FDT algorithm transforms the feature space Recursive partitioning of to obtain K disjoint subspaces R 1 ,R 2 ,...,R K ,

[0101]

[0102] Furthermore, the IT2FNN-based DXN concentration prediction model is constructed according to the subspace set, including:

[0103] Acquire the subspace set and perform fuzzification processing to obtain a fuzzy set corresponding to the subspace set;

[0104] The membership degree of fuzzy sets is calculated using Gaussian membership function;

[0105] Calculating a corresponding lower bound and an upper bound of the membership value according to the membership;

[0106] Determine the activation strength of the corresponding combination rule according to the lower bound of the membership value and the upper bound of the membership value;

[0107] Obtaining a lower bound value and an upper bound value of the subspace according to the activation strength of the combination rule and the corresponding consequent parameter;

[0108] Defuzzifying the lower bound value and the upper bound value of the subspace to obtain a feature mapping result;

[0109] According to the feature mapping results, the subspace set is updated by using the gradient descent method to obtain a DXN concentration prediction model.

[0110] Specifically, for each subspace R k,The CART algorithm uses the mean of its internal sample labels as the predicted value of the subspace. Obviously, this method is difficult to accurately mine the mapping relationship between input features and output features, and makes the model output lose smoothness. In this regard, this paper uses IT2FNN to fit the subspace samples to enhance the robustness of the model and improve its modeling ability for DXN.

[0111] Define subspace R k The training data samples in are Where N k and M k The subspace R k The number and dimension of training data samples in M k <<M. Taking subspace R k Taking the jth sample in as an example, a DXN model based on IT2FNN is constructed, which includes the antecedent network and the consequent network. It can be divided into input layer, membership function layer, fuzzy rule layer, reduction layer and output layer. The specific model construction steps are described as follows.

[0112] (1) Precondition Network

[0113] a) Input layer, the IT2FNN network input is passed to the membership function layer. The weight of this layer is 1 and the number of nodes is M k .

[0114] b) Membership function layer, which fuzzifies the input variables and expresses them in interval form, is used to calculate the membership of the input variables. Each node represents an interval type II membership function. The weight of this layer is 1, and the number of nodes is (in, is the subspace R k This paper uses the Gaussian membership function with uncertain mean to calculate the membership of fuzzy sets, where the subspace R k The mth k The input feature corresponds to the The lower bound of the membership value of the rule and upper bound The calculation is as follows:

[0115] The expressions of the lower bound of the membership value and the upper bound of the membership value are respectively:

[0116]

[0117] in, and The subspace R k The mth k The input variable corresponds to The membership function does not determine the lower bound, upper bound and width of the center. is the lower bound of the membership value, is the upper bound of the membership value, x kj mk is the mth value of the jth sample in the kth subspace k feature values.

[0118] Fuzzy rule layer, constructs fuzzy rules and implements fuzzy reasoning calculations. Each node represents a fuzzy rule. The weight of this layer is 1, and the number of nodes is Its input is the output of the membership function layer, and its output is The activation strength of the rule The formula is as follows:

[0119]

[0120] in, and Respectively The lower and upper bounds of the activation strength of the rule.

[0121] The reduction layer combines the activation strength of the rule with the corresponding rule consequent parameter to complete the reduction operation. The weight of this layer is the weight of the consequent interval, and the number of nodes is 2; the input is the rule activation strength output by the fuzzy calculation layer, and the output is the subspace R k The lower bound z k,j Value and upper bound

[0122]

[0123] in, is the weight of the consequent interval, and Respectively The lower and upper bounds of the activation strength of the rule, z k,j is the lower bound of the subspace, is the upper bound of the subspace.

[0124] Output layer, defuzzification to get the network output. The weight of this layer is the lower bound ratio value qk and the upper bound ratio (1-q k ). Its input is the output of the degradation layer, and its output is the feature mapping result. The formula is as follows:

[0125]

[0126] Post-processing network:

[0127] The input layer passes the network output to the subsequent network without performing any operations, as follows:

[0128]

[0129] Hidden layer, calculates the weight of fuzzy rules in the antecedent network, and the number of nodes in this layer is as follows:

[0130]

[0131] in, and They are the first Each rule outputs weights and biases; For the mk Input to the The weight of a rule.

[0132] Finally, the gradient descent method is used to update the network of each subspace in the IT2FDT algorithm to obtain the trained IT2FDT model (the DXN concentration prediction model), as follows:

[0133]

[0134] I(·) is the indicator function, when x∈R k 1 if exists, 0 otherwise.

[0135] Among them, Θ k is the network output of IT2FNN trained in the kth subspace, and f(·) is the operation indicator model based on IT2FDT.

[0136] Furthermore, the goal of the DXN emission reduction optimization proposed in this paper is to reduce the DXN emission concentration as much as possible within the allowable range of many MVs, CVs, AVs and easily detectable EIs. Based on the experience of experts in the field and correlation analysis, 98 process parameters in the attached table are used as decision variables; further, the DXN emission reduction optimization model (the DXN emission parameter optimization model) can be expressed as:

[0137]

[0138] Among them, D is the dimension of the decision variable after screening (98 in this article).

[0139] Furthermore, the improved PSO algorithm is used to solve the DXN emission parameter optimization model to obtain the optimized process variable values, including:

[0140] Determine dynamic performance indicators and static performance indicators;

[0141] Performing trend judgment on the dynamic performance index and the static performance index to obtain a judgment result, if the judgment result is that both the dynamic performance index and the static performance index are in a downward trend, adjusting the parameter in a first manner, if the judgment result is that the dynamic performance index is in a downward trend, while the static performance index is increasing, adjusting the parameter in a second manner, if the judgment result is that the dynamic performance index is in an upward trend, while the static performance index is decreasing, adjusting the parameter in a third manner, if the judgment result is that both the dynamic performance index and the static performance index are in an upward trend, adjusting the parameter in a fourth manner, if the judgment result is that the dynamic performance index or the static performance index remains unchanged, adjusting the parameter in a fifth manner;

[0142] The optimized process variable value is obtained according to the adjusted corresponding parameters.

[0143] Specifically, as mentioned above, the algorithm for achieving DXN emission reduction optimization needs to have strong robustness and the ability to handle uncertainty. In this regard, this paper designs a PSO (DME-PSO) algorithm based on a dual-mode evaluation mechanism. The core is: using dynamic performance indicators and static performance indicators to achieve dual evaluation of the population to obtain the current population evolution state, and to achieve adaptive transformation of control parameters to improve the global exploration ability of the algorithm. At the same time, the original characteristics of some particles in the population are maintained to ensure that the local exploration ability of the algorithm is not weakened. The description is as follows.

[0144] Typically, the speed and position update formulas of the population particles in the standard PSO algorithm are as follows:

[0145]

[0146] in, and They represent the velocity and position of the d-th dimension of the ith particle in the population at the t-th iteration respectively; and They represent the individual optimum and global optimum of the ith particle in the dth dimension at the tth iteration respectively; r 1 and r 2 represents a random number generated in [0,1], which is used to promote population diversity changes; w represents velocity momentum; c 1 and c 2 Represent individual learning factor and social learning factor respectively.

[0147] From formula (18), we can see that the particles in the population change their flight direction and speed under the influence of their own historical information and population information, thereby realizing the exploration of the decision space. Obviously, speed is dynamic information that cannot be ignored in the PSO algorithm, which affects the particle's ability to explore the decision space in the next iteration; at the same time, position information cannot be ignored either, which is the key to realizing the solution of the PSO algorithm. To this end, this paper defines particle speed and position as population dynamic information and static information, respectively, so as to realize a comprehensive evaluation of the population state.

[0148] Dynamic performance indicators:

[0149] The particle velocity carries both direction information and step information. Here, cosine similarity is used to measure the consistency of the particle velocity direction:

[0150]

[0151] in, represents the similarity of the velocity direction of the population at the tth iteration; N represents the number of particles in the population; It represents the velocity direction similarity index of the i-th particle in the population at the t-th iteration, as follows:

[0152]

[0153] in, It represents the mean velocity of the d-th dimension of particles in the population at the t-th iteration.

[0154] To further evaluate the speed step information, the population speed step index is defined as follows:

[0155]

[0156] Among them, (t) represents the population speed step size at the tth iteration; It represents the speed step index of the i-th particle in the population at the t-th iteration, as follows:

[0157]

[0158] Then, the direction information and step length information are combined to obtain the population dynamic performance index v at the tth iteration. (t) :

[0159]

[0160] When v (t) The larger the value is, the more similar the movement directions of particles in the population are and the longer the moving step is. Otherwise, it means the directions are divergent but the step length is shorter. In this case, particles fly slowly in a small decision space and have stronger local exploration capabilities.

[0161] Static performance indicators:

[0162] The position of particles in the population is the key to solving the PSO algorithm, and it is also an important static information for evaluating the state of the population. This paper uses the Euclidean distance to evaluate the static performance index of the population, which is defined as follows:

[0163]

[0164] Among them, O (t) represents the static index of the population at the tth iteration, represents the Euclidean distance between the i-th particle in the population and the average position at the t-th iteration, as follows:

[0165]

[0166] in, represents the mean position of particles in the population at the tth iteration.

[0167] when The larger the value is, the more dispersed the position distribution of the population particles is; otherwise, it means that the position of the population particles is highly concentrated.

[0168] Population adaptive adjustment:

[0169] Since a single indicator is difficult to fully and accurately measure the state of the population, this paper conducts a comprehensive evaluation of the dynamic performance indicators and static performance indicators of the population to more balancedly reflect the multidimensional characteristics of the population state. On the basis of comprehensive evaluation, in order to ensure that the population can maintain diversity in each iteration and enhance the global exploration ability of the algorithm, this paper first designs a nonlinear function Φ(·) based on the dual-mode evaluation results to measure the change amplitude of the control parameters:

[0170]

[0171] Furthermore, in view of the possibility of the existence of dual-mode evaluation results, this paper divides the population state changes into the following five categories and describes them in detail:

[0172] a) when and , indicating that both the current dynamic performance index and the static performance index are on a downward trend. This reflects that the particles in the population are exploring in a local range in a small step and multi-directional manner. At this time, the local exploration ability is strong, but the population is at risk of falling into the local optimum. At this time, it is necessary to enhance the global exploration ability of the population. In response to this phenomenon, this paper designs the following control parameter dynamic adjustment method:

[0173]

[0174] Assuming that the population is in a local optimal area in the decision space, the individual optimality of the particle is usually replaced by the global optimality. In order to avoid the redundant and repeated effects of the two on the particle speed and position update, this paper appropriately reduces the individual learning parameters and social learning parameters, and at the same time increases the speed inertia parameters to enhance the ability and probability of particles jumping out of this area.

[0175] b) When and When the dynamic performance index shows a downward trend, the static performance index increases. This shows that the particles are diverging in a small step and multi-directional manner, and the global exploration ability of the population has been enhanced, but its efficiency is still low. In order to further improve the computational efficiency of DME-PSO, this paper designs the following control parameter dynamic adjustment method:

[0176]

[0177] At this time, the population gradually diverges from a certain area, and its individual optimum and global optimum may be in this area. In order to avoid falling into the problem of excessive divergence or local optimum, this paper appropriately increases the speed inertia parameter to enhance the ability of population particles to jump out of the current area, while keeping the individual learning parameters and social learning parameters unchanged, thereby stabilizing the population learning characteristics while accelerating its global search efficiency.

[0178] c) When and When , the dynamic performance index shows an upward trend, while the static performance index decreases. This shows that the particles in the population are gathering in a large step and in the same direction, and the local search ability is gradually enhanced, but the global exploration ability is gradually weakened. At this time, the population is very likely to fall into the local optimal area. To avoid this problem, we need to enhance the global exploration ability of the population. This paper designs the following control parameter dynamic adjustment method:

[0179]

[0180] Under this condition, particles in the population are generally affected by the global optimum and gather in a specific area, thus falling into the local optimum. In order to suppress the premature convergence problem of the population, this paper appropriately reduces the speed inertia parameter and the social learning parameter, and appropriately increases the individual learning parameter. Through this strategy, the particles in the population are able to explore the individual optimum, thereby improving the diversity of the population and enhancing its ability to escape the local optimum.

[0181] d) When and When , both the dynamic performance index and the static performance index show an upward trend. This shows that the particles in the population are dispersing in a small step and multi-directional manner, and the global exploration ability is gradually enhanced. In this case, in order to maintain the current state and ensure the exploration performance of the population, its control parameters remain unchanged at this time:

[0182]

[0183] Under this condition, the population has gradually entered a stable exploration phase, and the control parameters do not need to be adjusted to avoid destroying the current exploration ability of the population. At the same time, keeping the parameters unchanged can balance local development and global exploration, providing a good foundation for further optimization of the population.

[0184] e) When or When the dynamic performance index or the static performance index remains unchanged, it indicates that the population is in a stagnant state. In this case, the adjustment of the control parameters has a relatively weak effect on the population. Thanks to the memory ability of the PSO algorithm, this paper reinitializes some particles in the population on the premise of preserving the current global optimality to enhance its global exploration ability. The position update formula is as follows:

[0185]

[0186] In this way, the distribution of some particles in the population can be re-randomized to avoid falling into the local optimal solution, thereby further enhancing the global exploration ability of the population and providing the algorithm with more optimization possibilities.

[0187] The core goal of the dual-mode evaluation mechanism proposed in this paper is to improve the global exploration ability of the population. It is well known that the PSO algorithm is prone to fall into the local optimal area, but this defect actually reflects the significant advantage of its local exploration ability. Therefore, this paper does not adjust its local exploration ability, thereby effectively reducing the computational complexity of DME-PSO. At the same time, this paper only adopts the adaptive control parameter adjustment strategy based on dual-mode evaluation for some particles in the population, while the control parameters of the remaining particles remain unchanged, so as to maintain the original local exploration ability of the PSO algorithm while reducing the computational overhead of the algorithm solution.

[0188] Different from the improvement methods of other PSO variants, the focus of this paper is to improve the coverage of the population in the decision space during the iteration process, that is, to improve the diversity of the population. Thanks to the memory characteristics of the PSO algorithm, when the population has a high enough coverage of the decision space, it will inevitably be able to search for the global optimal area.

[0189] Furthermore, this embodiment provides experimental verification:

[0190] This paper conducts engineering verification based on the actual DXN data of a certain MSWI power plant in Beijing. In the engineering verification, a total of 136 sets of actual DXN data were obtained, of which 2 / 3 were used as training sets and 1 / 3 were used as test sets.

[0191] Running the indicator model fitting results:

[0192] In order to verify the effectiveness of the IT2FDT algorithm proposed in this paper, this section uses the CART and IT2FNN algorithms to establish operation index models for comparative experiments. Among them, the IT2FDT parameters are set as follows: the minimum number of samples is 10, the number of rules is 10, the number of iterations is 1000, and the gradient descent learning rate is 0.1; the CART algorithm parameters are set as follows: the minimum number of samples is 10; the IT2FNN parameters are set as follows: the number of rules is 10, the number of iterations is 1000, and the gradient descent learning rate is 0.1.

[0193] The root mean square error (RMSE) and mean absolute error (MAE) indicators are used to evaluate the performance of the above model. The calculation formula is as follows:

[0194]

[0195] in, and yi They represent the model prediction value and the true value respectively; N represents the number of samples.

[0196] The model fitting curve and statistical results are shown in Figure 3 As shown in Table 2. Table 2 is the statistical result table of the running indicator model, as shown in Table 2:

[0197] Table 2 Statistical results of the operating indicator model

[0198]

[0199]

[0200] Depend on Figure 3 As shown in Table 2, under the same parameter setting conditions, the fitting curves and statistical results of the IT2FDT method in both the training set and the test set are better than those of the comparison method, especially in the training set. In addition, during the experimental implementation, we found that the performance of IT2FNN is highly random, which indicates that it is difficult to accurately capture the mapping relationship between input features and output features.

[0201] In summary, the operation index model constructed based on the IT2FDT algorithm has high accuracy and good robustness, laying a solid foundation for subsequent DXN emission reduction optimization research.

[0202] Based on the operation index model constructed by IT2FDT, the DME-PSO algorithm is used to solve the optimal value of the process variable under the condition of minimum DXN emission concentration. The number of population particles is set to 30 and the number of iterations is set to 100. The operation optimization results of DXN emission concentration are shown in Figure 2. Figure 4 shown.

[0203] Depend on Figure 4 It can be seen that the DXN emission concentration level after operation optimization is relatively stable, basically maintaining at 0.001ng-TEQ / Nm 3 Compared with the actual results, the DXN emission concentration after running the optimization was reduced by an average of 77.88%. The results show that the proposed algorithm can significantly reduce the DXN emission concentration. Figure 5-6 The comparison results of furnace temperature and secondary air volume after operation optimization solved by DME-PSO algorithm are shown.

[0204] Depend on Figure 5 It can be seen that the furnace temperature solved by the DME-PSO algorithm can meet the incineration requirements (greater than 850°C). It is worth noting that due to the limitations of DXN emission concentration detection technology, the actual DXN data set used in this article is not obtained by short-term continuous sampling, and there are differences in the operating conditions between each sample. Therefore, the optimal furnace temperature solved by the DME-PSO algorithm only represents the optimal furnace temperature under the current sample, and has no direct correlation with the actual continuous operating conditions of the MSWI power plant. At the same time, it does not mean that the proposed method sacrifices the stability of the internal process variables of the furnace while obtaining the minimum DXN emission concentration. In addition, by Figure 6 It can be seen that the secondary air volume after operation optimization is generally higher than the actual value. Obviously, this result is similar to the basic cognition of experts in the field: by increasing the secondary air volume, the turbulence intensity of the flue gas in the furnace can be enhanced, thereby promoting the effective decomposition of DXN and ultimately reducing its emission concentration.

[0205] Due to the complex characteristics of DXN, we selected some controlled variables in the process of decision variable selection. In the actual operation process, it is necessary to use advanced intelligent controllers to achieve stable tracking based on the optimal values ​​of the controlled variables obtained. In this regard, this article does not discuss the relevant content in depth.

[0206] To further prove the superiority of the DME-PSO algorithm proposed in this paper, the statistical results of running PSO, LDW-PSO and NLDW-PSO 30 times and taking the average are shown in Table 3. Table 3 is the running optimization statistical results table, as shown below:

[0207] Table 3 Optimization statistics results

[0208]

[0209] As shown in Table 3, after the optimization algorithm is adopted, the DXN emission concentration is greatly reduced. In particular, the operation optimization results of the DME-PSO algorithm are significantly better than the three comparison methods, and its variance is the smallest. It is worth noting that compared with the statistical results of the benchmark test function, the NLDW-PSO algorithm performs poorly in the actual engineering test.

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

[0211] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligence algorithm, characterized in that: include: Obtain incineration process dataset; According to the IT2FDT algorithm, the incineration process data set is spatially divided to obtain a subspace set; Based on IT2FNN, a DXN concentration prediction model is constructed according to the subspace set; Constructing a DXN emission parameter optimization model based on the DXN concentration prediction model and the process parameters of DXN emission; The DXN emission parameter optimization model is solved by using an improved PSO algorithm to obtain optimized process variable values; The optimized process variable values ​​are brought into the DXN concentration prediction model to obtain the optimized DXN emission concentration.

2. The dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligent algorithm according to claim 1 is characterized in that: The incineration process data set is spatially divided according to the IT2FDT algorithm to obtain a subspace set, including: Extracting features from the incineration process data set and performing spatial segmentation according to the extracted features to obtain a left subspace and a right subspace; Calculating a total mean square error (TMSE) for the left subspace and the right subspace; Determine the segmentation variable and segmentation point with the smallest total mean square error to obtain the current divided subspace; The IT2FDT algorithm is used to recursively partition the currently divided subspace to obtain a subspace set.

3. The dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligent algorithm according to claim 1, characterized in that: The IT2FNN-based DXN concentration prediction model is constructed according to the subspace set, including: Acquire the subspace set and perform fuzzification processing to obtain a fuzzy set corresponding to the subspace set; The membership degree of fuzzy sets is calculated using Gaussian membership function; Calculating a corresponding lower bound and an upper bound of the membership value according to the membership; Determine the activation strength of the corresponding combination rule according to the lower bound of the membership value and the upper bound of the membership value; Obtaining a lower bound value and an upper bound value of the subspace according to the activation strength of the combination rule and the corresponding consequent parameter; Defuzzifying the lower bound value and the upper bound value of the subspace to obtain a feature mapping result; According to the feature mapping results, the subspace set is updated by using the gradient descent method to obtain a DXN concentration prediction model.

4. The dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligence algorithm according to claim 3 is characterized in that: The expressions of the lower bound of the membership value and the upper bound of the membership value are respectively: in, and The subspace R k The mth k The input variable corresponds to The membership function does not determine the lower bound, upper bound and width of the center. is the lower bound of the membership value, is the upper bound of the membership value, x kj mk is the mth value of the jth sample in the kth subspace k characteristic values.

5. The dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligent algorithm according to claim 4, characterized in that: The expressions of the lower and upper bounds of the subspace are: in, is the weight of the consequent interval, and Respectively The lower and upper bounds of the activation strength of the rule, z k,j is the lower bound of the subspace, is the upper bound of the subspace.

6. The dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligent algorithm according to claim 5, characterized in that: The expression of the DXN concentration prediction model is: in,, Θk is the IT2FNN network output trained in the kth subspace, I (·) is the indicator function, f ( g ) is the operation indicator model based on IT2FDT.

7. The dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligence algorithm according to claim 6, characterized in that: The expression of the DXN emission parameter optimization model is: Among them, D is the dimension of the decision variable after screening.

8. The dioxin emission reduction optimization method based on a dual-mode evaluation mechanism swarm intelligence algorithm according to claim 6, characterized in that: The improved PSO algorithm is used to solve the DXN emission parameter optimization model to obtain the optimized process variable value, including: Determine dynamic performance indicators and static performance indicators; Performing trend judgment on the dynamic performance index and the static performance index to obtain a judgment result, if the judgment result is that both the dynamic performance index and the static performance index are in a downward trend, adjusting the parameter in a first manner, if the judgment result is that the dynamic performance index is in a downward trend, while the static performance index is increasing, adjusting the parameter in a second manner, if the judgment result is that the dynamic performance index is in an upward trend, while the static performance index is decreasing, adjusting the parameter in a third manner, if the judgment result is that both the dynamic performance index and the static performance index are in an upward trend, adjusting the parameter in a fourth manner, if the judgment result is that the dynamic performance index or the static performance index remains unchanged, adjusting the parameter in a fifth manner; The optimized process variable value is obtained according to the adjusted corresponding parameters.

Images