Energy-saving operation control method for desulfurization system

Abstract

The invention discloses an energy-saving operation control method for a desulfurization system, and relates to the field of wet flue gas desulfurization of a power plant. The method comprises the following steps: (1) establishing a clean flue gas SO2 concentration prediction model; (2) optimization control of energy-saving operation of the desulfurization system: selecting an optimal feature subset from the selected feature subset, and searching an optimal control value of the optimal feature subset under each working condition; and establishing an energy-saving operation optimization target and constraint conditions of the desulfurization system, and then adjusting operation parameters according to outlet data to ensure energy-saving operation. The method comprises the following steps: carrying out feature selection on desulfurization online operation parameters, and selecting an optimal feature subset to establish a clean flue gas SO2 concentration prediction model; by expanding the data acquisition range, introducing the deep learning algorithm and dynamically updating the data, the SO2 concentration of the purified flue gas can be predicted more accurately, and a reliable basis is provided for optimization control.

Description

Technical Field

[0001] The present invention relates to the field of wet flue gas desulfurization in power plants, and more specifically, it is a method for controlling the energy-saving operation of a desulfurization system. Background Art

[0002] In the current field of ultra-low emission transformation of SO2, the wet flue gas desulfurization technology occupies a relatively high market share due to its many remarkable advantages. On the one hand, it has greatly helped coal-fired power enterprises achieve remarkable results in the reduction of ultra-low emission pollutants, contributing a great deal to environmental protection; on the other hand, however, it has also significantly increased the operating energy consumption of the unit, resulting in a substantial increase in the operating cost of the coal-fired power unit. With the increasingly strict requirements of society for environmental protection, the coal-fired power industry is facing huge emission reduction pressures. If the energy consumption of the desulfurization system cannot be effectively reduced, it will not only increase the operating cost of the enterprise, but may also affect the sustainable development of the enterprise. Therefore, deeply analyzing the energy consumption characteristics of the desulfurization system after ultra-low emission of the unit and actively carrying out research on the operation optimization of the unit desulfurization system are of extremely important and practical significance for reducing the energy consumption of the unit during the pollutant treatment process, improving the economy of the technology, and achieving the development goal of energy conservation and emission reduction in the coal-fired power industry.

[0003] The core of the optimization work of desulfurization operation lies in deeply exploring how to reduce the operating cost on the premise of ensuring that SO2 can meet the emission standards, that is, not only to achieve the emission reduction goal, but also to effectively reduce the operating energy consumption. At present, the tense situation of energy resources is becoming increasingly prominent. Reducing energy consumption is not only the need for enterprises to reduce costs, but also an important measure to ensure energy security. This is undoubtedly a key and important issue faced by current coal-fired power enterprises. At the present stage, when most power plant units carry out the ultra-low emission transformation of SO2, they are often designed based on relatively unfavorable sulfur content. However, during the actual operation of the unit, the sulfur content is usually lower than the design value, and there is a phenomenon of excessive margin, which leads to relatively poor operating economy.

[0004] In the context of global climate change, reducing greenhouse gas emissions and energy consumption has become an urgent task. As an important field of energy consumption and greenhouse gas emissions, the coal-fired power industry must take effective measures to improve energy utilization efficiency.

[0005] Under the situation of ultra-low emissions, the SO2 concentration in the clean flue gas is the most important indicator for evaluating the performance of the desulfurization system. If the hourly average value of the SO2 concentration in the clean flue gas is less than the ultra-low emission limit, the ultra-low emission electricity price subsidy can be obtained. If it is greater than the emission limit, not only the ultra-low emission electricity price subsidy cannot be obtained, but also penalties will be imposed. The SO2 concentration in the clean flue gas is also an important factor affecting the operating cost of the desulfurization system. Low SO2 emission concentration and low desulfurization operating cost are two opposing goals. In actual operation, one-sided pursuit of reducing the SO2 concentration in the clean flue gas will inevitably lead to an increase in the desulfurization operating cost and a decrease in the operating economy of the unit. While simply aiming at reducing the desulfurization operating cost, it may lead to the SO2 emission concentration in the clean flue gas not meeting the ultra-low emission limit requirements.

[0006] Therefore, under this new normal situation of ultra-low emissions, vigorously carrying out research on the optimized operation of the desulfurization system and deeply exploring the key points for energy conservation and consumption reduction of the desulfurization system under different sulfur contents and load conditions have extremely important and practical significance for ensuring the safe and stable operation of the desulfurization system and reducing the desulfurization cost. Summary of the Invention

[0007] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art, and provide an energy-saving operation control method for a desulfurization system. Aiming at the core problem of the operation optimization of wet desulfurization technology, based on the composition of the operation cost of the desulfurization system and its energy consumption characteristics, a prediction model for the SO2 concentration in the clean flue gas is established. Combining the analysis of the operation energy consumption of the desulfurization system, the adjustable parameters in the operation are optimized through an intelligent optimization algorithm to achieve low-cost compliance emissions of SO2 in the clean flue gas and reduce the operation cost of the desulfurization system.

[0008] In order to achieve the above purpose, an energy-saving operation control method for a desulfurization system of the present invention includes the following steps:

[0009] (1) Establish a prediction model for the SO2 concentration in the clean flue gas:

[0010] a. Collect the online operation parameters of a plurality of desulfurization systems for several days to obtain a data matrix of the online operation parameters;

[0011] b. Based on the random forest method, conduct a rough selection of the online operation characteristic parameters for the prediction model of the SO2 concentration in the clean flue gas; screen out the characteristics that have a great influence on the prediction model of the SO2 concentration in the clean flue gas, and rank the importance of each online operation characteristic parameter for the prediction model of the SO2 concentration in the clean flue gas;

[0012] c. Apply the particle swarm optimization algorithm to conduct a refined selection of the online operation characteristic parameters obtained from the rough selection to obtain a refined feature subset;

[0013] d. Use the Gaussian process verification for the refined feature subset to establish a prediction model for the SO2 concentration in the clean flue gas;

[0014] (2) Energy-saving operation optimization control of desulfurization system:

[0015] Select the optimal feature subset from the selected feature subsets, and find the best control values of the optimal feature subsets under various working conditions; establish the energy-saving operation optimization objectives and constraints of the desulfurization system, and then adjust the operating parameters according to the outlet data to ensure energy-saving operation.

[0016] In the above technical solution, in step (2), the optimal feature subsets are the current of the slurry circulation pump and the pH value of the slurry;

[0017] Establish the energy-saving operation optimization objectives and constraints of the desulfurization system as follows:

[0018] minC = C1 + C2

[0019]

[0020] Among them, minC is the minimum cost, C1 is the absorbent cost; C2 is the operating power consumption cost; I0 is the current of the slurry circulation pump; pH1 is the pH value of the slurry; SO2,out is the SO2 concentration in the clean flue gas; ffitrgp (optimal feature subset) is the Gaussian process regression model;

[0021] Among them, the SO2 concentration in the clean flue gas is not higher than 30 mg / m 3 , the pH1 value of the slurry is limited to 4.5 - 6.5, and the total current of the slurry circulation pump is limited to 200 - 600 A.

[0022] In the above technical solution, in step a, collect the online operation parameters of a desulfurization system for one month, export the online operation characteristic parameters by DCS, and obtain the data matrix of the online operation characteristic parameters; according to the DCS physical parameter coding rule, complete the identification of the online operation characteristic parameters.

[0023] In the above technical solution, in step b, the random forest method measures the importance of the online operation characteristic parameters, which is achieved by calculating the out-of-bag data accuracy or Gini index of the characteristic parameters;

[0024] The Gini index measures the importance of a characteristic parameter by the average change in the node splitting impurity of all decision trees in the random forest method for a certain online operation characteristic parameter; the out-of-bag data accuracy measures the importance of a characteristic parameter by the average error change of the prediction results before and after randomly permuting a certain online operation characteristic parameter in the out-of-bag samples of each decision tree.

[0025] In the above technical solution, the importance of the online operation characteristic parameter X j is calculated through the following steps:

[0026] Step 1, in the random forest method, A training samples are obtained by the sampling method with replacement, denoted as β1, β2, …, βA, and at the same time, there are A out-of-bag samples, denoted as oob1, oob2, …, oobA;

[0027] Step 2, for a certain training set βa, a = 1, 2, …, A, a decision tree ka is created on the training samples, and the out-of-bag sample ooba is predicted using the decision tree ka, and its prediction result is denoted as ERRa;

[0028] Step 3, the values of the feature parameter X j in ooba are randomly permuted, and the permuted sample is denoted as oobaj, and the decision tree ka is used to predict the oobaj sample data, and its prediction result is denoted as ERR aj ;

[0029] Step 4, for a = 2, 3, …, Z, repeat the above steps;

[0030] Step 5, the importance measure I j of the feature parameter X x is calculated by the following formula:

[0031]

[0032] The larger the Ix value before and after randomly permuting a certain feature parameter, the greater the influence of the feature parameter on the model precision rate, and the relatively higher the importance of the feature parameter. On the contrary, the importance of the feature parameter is relatively low.

[0033] In the above technical solution, in step d, a non-parametric model for regression analysis of data is carried out under the prior hypothesis of Gaussian process verification, and a complete posterior distribution of the prediction samples is given, providing the uncertainty of the prediction results, making the model have good interpretability; when the likelihood is a normal distribution, the posterior distribution of the prediction samples has a closed-form solution in Gaussian form;

[0034] A prediction model for the net flue gas SO2 concentration of each online operating parameter feature is established; then, according to the net flue gas SO2 concentration data, the online operating parameters are adjusted and optimized.

[0035] In the above technical solution, according to the random forest method, the importance ranking of each influencing factor in the net flue gas sulfur dioxide prediction model is as follows: the concentration of SO2 in the raw flue gas, the current of the first slurry circulation pump, the current of the second slurry circulation pump, the flow rate of the raw flue gas, the slurry replenishment flow rate of the absorption tower, the dust concentration of the raw flue gas, the current of the first oxidation blower, the current of the first gypsum discharge pump, the current of the second gypsum discharge pump, the current of the third gypsum discharge pump, the current of the fourth slurry circulation pump, the pH1 at the measuring point below the slurry pool, the liquid level of the absorption tower, the unit load, the current of the second oxidation blower, the current of the third slurry circulation pump, the temperature of the raw flue gas, the pressure of the raw flue gas, the current of the fifth slurry circulation pump, the current of the third oxidation blower, the O2 content of the raw flue gas, the HO2 content of the net flue gas, the temperature of the net flue gas, the flushing water flow rate of the demister, the pH2 at the measuring point above the slurry pool, the O2 content of the net flue gas, the dust concentration of the net flue gas, the pressure of the net flue gas, and the slurry replenishment density of the absorption tower.

[0036] In the above technical solution, according to the importance score ranking of the on-line operation characteristic parameters of desulfurization, combined with the mechanism analysis and experimental analysis of SO2 removal in the desulfurization tower, the characteristic parameters that are relatively unimportant for the net flue gas SO2 concentration prediction model are excluded, including at least the slurry replenishment density of the absorption tower, the pressure of the net flue gas, the dust concentration of the net flue gas, the pH2 at the measuring point above the slurry pool, and the flushing water flow rate of the demister.

[0037] In the above technical solution, according to the importance score ranking of the on-line operation characteristic parameters of desulfurization, combined with the mechanism analysis and experimental analysis of SO2 removal in the desulfurization tower, according to the importance ranking and influencing factors, the characteristic parameters that have a greater impact on the accuracy of the net flue gas SO2 concentration prediction model are selected, including at least the concentration of SO2 in the raw flue gas, the current of the absorption tower circulation pump, the flow rate of the raw flue gas, the dust concentration of the raw flue gas, the current of the gypsum discharge pump, the current of the oxidation blower, the pH1 at the measuring point below the slurry pool, and the liquid level of the absorption tower;

[0038] The current of the slurry circulation pump includes the sum of the current of the first slurry circulation pump, the current of the second slurry circulation pump, the current of the third slurry circulation pump, the current of the fourth slurry circulation pump, and the current of the fifth slurry circulation pump; the current of the gypsum discharge pump includes the sum of the current of the first gypsum discharge pump, the current of the second gypsum discharge pump, and the current of the third gypsum discharge pump; the current of the oxidation blower includes the sum of the current of the first oxidation blower, the current of the second oxidation blower, and the current of the third oxidation blower.

[0039] In the above technical solution, in step c, the method of applying the particle swarm optimization algorithm to select the on-line operation characteristic parameters obtained by rough selection to obtain the selected characteristic subset is as follows:

[0040] The random forest method is used to roughly select the characteristic subset S1 of the desulfurization influencing factors. Then, the particle swarm optimization algorithm is applied to randomly initialize the particle swarm S2 of the on-line operation characteristic parameters of desulfurization obtained by rough selection. Next, the particle velocity and position are calculated S3. If the fitness of the current position is greater than the fitness value of the individual optimal position S4, the individual position is updated S5 to make the fitness of the individual optimal position greater than the fitness value of the global optimal position S6. If the fitness of the current position is less than the fitness value of the individual optimal position S4, the fitness of the individual optimal position is made greater than the fitness value of the global optimal position S6. The global optimal position is updated S7 until the maximum number of iterations S8 is reached, and the selected characteristic subset S9 in the desulfurization influencing factors is obtained from the global optimal position. If the maximum number of iterations S8 is not reached, the calculation of the particle velocity and position S3 is returned. If the fitness of the individual optimal position is less than the fitness value of the global optimal position S6, the selected characteristic subset S9 in the desulfurization influencing factors is obtained from the global optimal position.

[0041] A desulfurization system energy-saving operation control method of the present invention has the following advantages:

[0042] 1) Feature selection is performed on the on-line operation parameters of desulfurization, and the optimal feature subset is selected to establish a net flue gas SO2 concentration prediction model. By expanding the data acquisition range, introducing deep learning algorithms, and dynamically updating data, the net flue gas SO2 concentration can be predicted more accurately, providing a reliable basis for optimization control. Different types of power plant data can cover more working conditions and equipment characteristics. Deep learning algorithms can mine complex relationships in the data, and real-time updated data can reflect the latest system state. An accurate prediction model can anticipate the change trend of the net flue gas SO2 concentration in advance, so as to adjust the operation parameters in time to ensure that the desulfurization system always operates in the high-efficiency range. For example, when it is predicted that the SO2 concentration in the raw flue gas is about to increase, increasing the current of the slurry circulation pump in advance or raising the slurry pH1 value can effectively avoid the decline of desulfurization efficiency and ensure compliance with emissions.

[0043] 2) Expanding the optimized parameters and introducing intelligent control strategies can more effectively reduce the operation cost of the desulfurization system. The particle swarm optimization algorithm is used to optimize the adjustable operation parameters of the desulfurization system, so as to realize the economic operation of the desulfurization system. Taking the introduction of slurry density as an example, reasonably controlling the slurry density can reduce the usage amount of absorbent and lower the absorbent cost. It can quickly adjust parameter control according to the real-time changing working conditions to avoid excessive consumption of energy and materials. Intelligent control of the combination of slurry circulation pumps can cope with different working conditions and save electricity consumption to the greatest extent.

[0044] 3) More accurate prediction models and comprehensive optimization control strategies help to discover potential equipment failures and operation anomalies in advance, perform maintenance and adjustment in time, reduce the downtime and maintenance costs caused by failures, and improve the overall reliability and stability of the system.

[0045] 4) By optimizing parameters such as the current of the slurry circulation pump and the pH1 value of the slurry, the system can minimize energy consumption while meeting the emission standards, achieve efficient utilization of energy, and help alleviate the energy shortage situation. For example, reasonably adjusting the operating current of the slurry circulation pump can avoid energy waste and ensure the desulfurization effect at the same time.

[0046] 5) The optimized operation control method can reduce the emissions of secondary pollutants such as waste residue and waste water generated during the desulfurization process and reduce the potential harm to the surrounding environment.

[0047] 6) The expansion of optimized parameters and the introduction of intelligent control strategies can more precisely control the chemical reaction conditions during the desulfurization process, improve the utilization rate of desulfurization agents, and thus enhance the efficiency of the desulfurization system. For example, reasonably controlling the slurry density can enhance the contact effect between the desulfurization agent and SO2, and optimizing the oxidation air volume can promote the oxidation of sulfite, which all contribute to improving the desulfurization efficiency. At the same time, comprehensively considering the environmental benefit assessment can prompt enterprises to pursue energy conservation without sacrificing the efficiency of the desulfurization system and achieve a balance between economic and environmental benefits. Description of the Drawings

[0048] Figure 1 Flow chart of the particle swarm optimization algorithm for calculating the selected feature subset in the desulfurization influencing factors.

[0049] Figure 2 Schematic diagram of the energy-saving operation control method for the desulfurization system of the present invention. Detailed Embodiments

[0050] The following details the implementation of the present invention in conjunction with the drawings, but it does not limit the present invention and is only for illustration purposes. At the same time, the advantages of the present invention will become clearer and easier to understand.

[0051] Refer to Figure 1 、 Figure 2 It can be known that an energy-saving operation control method for a desulfurization system of the present invention includes the following steps:

[0052] (1) Establish a prediction model for the SO2 concentration in the clean flue gas:

[0053] a. Collect the online operation parameters of multiple desulfurization systems for several days to obtain a data matrix of the online operation parameters;

[0054] In step a, the online operation parameters of multiple desulfurization systems for one month are collected, the online operation characteristic parameters are exported by DCS, and a data matrix of the online operation characteristic parameters is obtained; according to the DCS physical parameter coding rule, the identification of the online operation characteristic parameters is completed.

[0055] b. Based on the random forest method, conduct a rough selection of the online operation characteristic parameters for the net flue gas SO2 concentration prediction model; screen out the characteristics that have a great impact on the net flue gas SO2 concentration prediction model, and rank the importance of each online operation characteristic parameter for the net flue gas SO2 concentration prediction model;

[0056] In step b, the random forest method measures the importance of the online operation characteristic parameters by calculating the out-of-bag data accuracy rate or the Gini index of the characteristic parameters;

[0057] The Gini index measures the importance of the characteristic parameters by the average change in the node splitting impurity of a certain online operation characteristic parameter in all decision trees of the random forest method; the out-of-bag data accuracy rate measures the importance of the characteristic parameters by the average error change in the prediction results of each decision tree before and after randomly permuting a certain online operation characteristic parameter in the out-of-bag samples.

[0058] The importance of the online operation characteristic parameter X j is calculated through the following steps:

[0059] Step 1, in the random forest method, obtain A training samples by the method of sampling with replacement, denoted as β1, β2,..., βa, and at the same time there are A out-of-bag samples, denoted as oob1, oob2,..., ooba;

[0060] Step 2, for a certain training set βa, a = 1, 2,..., A, create a decision tree ka on the training samples, and use the decision tree ka to predict the out-of-bag sample ooba, and its prediction result is denoted as ERRa;

[0061] Step 3, randomly permute the value of the characteristic parameter X j in ooba, and the permuted sample is denoted as oobaj, and use the decision tree ka to predict the oobaj sample data, and its prediction result is denoted as ERR aj ;

[0062] Step 4, for a = 2, 3,..., Z, repeat the above steps;

[0063] Step 5, the importance measure I j of the characteristic parameter X x is calculated through the following formula:

[0064]

[0065] The larger the Ix value before and after randomly permuting a certain characteristic parameter, the greater the impact of the characteristic parameter on the model precision rate, and the relatively higher the importance of the characteristic parameter. Conversely, the importance of the characteristic parameter is relatively lower.

[0066] According to the random forest method, the importance ranking of each influencing factor in the net flue gas sulfur dioxide prediction model is as follows: raw flue gas SO2 concentration, current of the first slurry circulation pump, current of the second slurry circulation pump, raw flue gas flow rate, absorption tower make-up slurry flow rate, raw flue gas dust concentration, current of the first oxidation fan, current of the first gypsum discharge pump, current of the second gypsum discharge pump, current of the third gypsum discharge pump, current of the fourth slurry circulation pump, pH1 at the measuring point below the slurry pool, absorption tower liquid level, unit load, current of the second oxidation fan, current of the third slurry circulation pump, raw flue gas temperature, raw flue gas pressure, current of the fifth slurry circulation pump, current of the third oxidation fan, O2 content in the raw flue gas, HO2 content in the net flue gas, net flue gas temperature, demister flushing water flow rate, pH2 at the measuring point above the slurry pool, O2 content in the net flue gas, net flue gas dust concentration, net flue gas pressure, absorption tower make-up slurry density.

[0067] Based on the importance score ranking of the on-line operation characteristic parameters of desulfurization, combined with the mechanism analysis and experimental analysis of SO2 removal in the desulfurization tower, exclude the characteristic parameters that are relatively unimportant for the net flue gas SO2 concentration prediction model, including at least absorption tower make-up slurry density, net flue gas pressure, net flue gas dust concentration, pH2 at the measuring point above the slurry pool, and demister flushing water flow rate.

[0068] Based on the importance score ranking of the on-line operation characteristic parameters of desulfurization, combined with the mechanism analysis and experimental analysis of SO2 removal in the desulfurization tower, according to the importance ranking and influencing factors, select the characteristic parameters that have a greater impact on the accuracy of the net flue gas SO2 concentration prediction model, including at least raw flue gas SO2 concentration, absorption tower circulation pump current, raw flue gas flow rate, raw flue gas dust concentration, gypsum discharge pump current, oxidation fan current, pH1 at the measuring point below the slurry pool, and absorption tower liquid level;

[0069] The current of the slurry circulation pump includes the sum of the current of the first slurry circulation pump, the current of the second slurry circulation pump, the current of the third slurry circulation pump, the current of the fourth slurry circulation pump, and the current of the fifth slurry circulation pump; the current of the gypsum discharge pump includes the sum of the current of the first gypsum discharge pump, the current of the second gypsum discharge pump, and the current of the third gypsum discharge pump; the current of the oxidation fan includes the sum of the current of the first oxidation fan, the current of the second oxidation fan, and the current of the third oxidation fan.

[0070] c. Apply the particle swarm optimization algorithm to refine the on-line operation characteristic parameters obtained by rough selection to obtain a refined feature subset;

[0071] In step c, the method of using the particle swarm optimization algorithm to select the online operation characteristic parameters obtained by rough selection to obtain a selected feature subset is as follows: The random forest method is used to roughly select the desulfurization influence factor feature subset S1, and then the particle swarm optimization algorithm is applied to randomly initialize the particle swarm S2 of the desulfurization online operation characteristic parameters obtained by rough selection, and then the particle velocity position S3 is calculated. If the fitness of the current position is greater than the fitness value of the individual optimal position S4, then the individual position S5 is updated to make the fitness of the individual optimal position greater than the fitness value of the global optimal position S6; if the fitness of the current position is less than the fitness value of the individual optimal position S4, then the fitness of the individual optimal position is made greater than the fitness value of the global optimal position S6; the global optimal position S7 is updated, the maximum number of iterations S8 is reached, and the selected feature subset S9 in the desulfurization influence factors is obtained from the global optimal position. If the maximum number of iterations S8 is not reached, then return to calculate the particle velocity position S3; if the fitness of the individual optimal position is less than the fitness value of the global optimal position S6, then enter the process of obtaining the selected feature subset S9 in the desulfurization influence factors from the global optimal position.

[0072] d. Use the Gaussian process verification to establish a net flue gas SO2 concentration prediction model for the selected feature subset;

[0073] In step d, a non-parametric model for regression analysis of data is carried out under the prior assumption of Gaussian process verification, which gives the complete posterior distribution of the prediction samples, provides the uncertainty of the prediction results, and makes the model have good interpretability; when the likelihood is a normal distribution, the posterior distribution of the prediction samples has a closed-form solution in Gaussian form;

[0074] Establish a net flue gas SO2 concentration prediction model for each online operation parameter feature; then, according to the net flue gas SO2 concentration data, adjust and optimize the online operation parameters.

[0075] The net flue gas SO2 concentration prediction model uses Gaussian process for regression analysis. Gaussian Process is a model framework with probabilistic significance that combines Bayesian theory and statistical learning theory. It is a non-parametric model for regression analysis of data under the prior assumption of Gaussian process, which can give the complete posterior distribution of the prediction samples, provide the uncertainty of the prediction results, and make the model have good interpretability.

[0076] f(x)~GP(μ(x),k(x,x'))

[0077] In the above formula, x and x' are any vectors in X. After preprocessing the data and taking the mean value as 0, the linear regression model with Gaussian white noise is as follows:

[0078] y=f(x)+ε,f(x)=xTw

[0079] In the above formula, x is the input vector of the selected feature subset that affects the net flue gas sulfur dioxide concentration, y is the output vector of the net flue gas sulfur dioxide concentration, f(x) is the objective function, ε is the noise, w is the weight vector, and it is assumed that all follow a Gaussian distribution, w~N(0,Σp), where Σp is the covariance; ε~N(0,σn2), where σn2 is the variance. The prior distribution of the output vector y can be obtained as follows:

[0080] y~N(0,K(X,X)+σn 2 In);

[0081] The joint prior distribution of the predicted value f* of the net flue gas sulfur dioxide concentration and the output value y is as follows:

[0082]

[0083] In the above formula, K(X,x*)=K(x*,X)T is the n×1 covariance matrix of x* and X, K(x*,x*) is the self-covariance of x*; In is the identity matrix; K(X,X)=Kn=(kij)n×n is the n×n symmetric positive definite covariance matrix. Furthermore, the posterior distribution of the predicted value f* of x* can be calculated as the following formula:

[0084]

[0085] cov(f*)=kx*,x*-K(x*,X)[K(X,X+σn 2 In -1 K(X,x*)

[0086] In the above formula, is the mean of the sample predicted value, and cov(f*) is the variance;

[0087] Input all the operation data and perform regression through Gaussian to obtain the predicted value of the sulfur dioxide concentration in the desulfurized net flue gas.

[0088] (2) Energy-saving operation optimization control of the desulfurization system:

[0089] Select the optimal feature subset from the selected feature subsets, and find the best control values of the optimal feature subsets under various working conditions; establish the energy-saving operation optimization objectives and constraints of the desulfurization system, and then adjust the operation parameters according to the outlet data to ensure energy-saving operation.

[0090] The optimal feature subsets are the current of the slurry circulation pump and the slurry pH1 value;

[0091] Establish the energy-saving operation optimization objectives and constraints of the desulfurization system as follows:

[0092] minC=C1+C2;

[0093]

[0094] Among them, minC is the minimum cost, C1 is the absorbent cost; C2 is the operating power consumption cost; I0 is the current of the slurry circulation pump; pH1 is the pH value of the measuring point below the slurry tank; SO2,out is the SO2 concentration in the clean flue gas; ffitrgp (optimal feature subset) is a Gaussian process regression model;

[0095] where the SO2 concentration in the clean flue gas is not higher than 30 mg / m 3 , the pH1 value of the slurry is limited to 4.5 - 6.5, and the total current of the slurry circulation pump is limited to 200 - 600 A.

[0096] Example 1

[0097] A large-scale coal-fired power plant adopts the energy-saving operation control method of the desulfurization system of the present invention. First, in the stage of establishing an accurate prediction model for the SO2 concentration in the clean flue gas, the power plant collected a large amount of operation data from different units in the past six months, including the SO2 concentration in the raw flue gas, the raw flue gas temperature, the oxygen content in the raw flue gas, the dust concentration in the raw flue gas, the raw flue gas flow rate, the current of the slurry circulation pump, the pH value of the slurry, the slurry density, the oxidation air volume, etc., a total of 32 parameters.

[0098] Through the random forest method, the characteristic parameters are roughly selected to screen out the characteristics that have a greater impact on the prediction model of the SO2 concentration in the clean flue gas. When running the random forest method, the decision tree is set to 300, and the number of characteristic parameters for node splitting is one-third of the total number of characteristics. At the same time, the collected data is preprocessed to remove outliers and noise data, and a data matrix of 10000×32 is obtained. After five operations of the random forest method to calculate its average value, it is used as the evaluation basis for the importance of each characteristic to the prediction model of the SO2 concentration in the clean flue gas.

[0099] On this basis, combined with the particle swarm optimization algorithm, the feature selection is carried out. During the selection process, fully considering the situation that there are spares in the slurry circulation pump, oxidation blower, etc. in the actual operation of the power plant unit, the current characteristic parameters of a single pump or blower are not included, and only the total current of the slurry circulation pump, the total current of the oxidation blower, and the total current of the gypsum discharge pump are considered as characteristic parameters. Finally, 14 key characteristic parameters are determined from numerous parameters, such as the SO2 concentration in the raw flue gas, the total current of the slurry circulation pump, the total current of the oxidation blower, the liquid level of the absorption tower, etc.

[0100] Use Gaussian process to establish a prediction model for the SO2 concentration in the clean flue gas. In order to improve the accuracy of the model, at least two months of data during stable operation are extracted as the benchmark, and the data is normalized. Under the prior assumption of the Gaussian process, regression analysis is carried out on the data, which can give the complete posterior distribution of the prediction samples, provide the uncertainty of the prediction results, and make the model have good interpretability.

[0101] During actual operation, when it is real-time monitored that the SO2 concentration in the raw flue gas suddenly increases from 1500 mg / m 3 to 2500 mg / m 3 , the prediction model quickly responds. The original total current of the slurry circulation pumps is 400 A. The system increases it to 550 A according to the model's suggestion. At the same time, the slurry pH1 value is increased from 5.0 to 5.8, ensuring effective response to the substantial increase in SO2 concentration in a short time and ensuring that the SO2 emission concentration in the clean flue gas is not higher than 30 mg / m 3 , maintaining a high desulfurization efficiency.

[0102] In addition, under the long-term low-load operation condition of the unit, the SO2 concentration in the raw flue gas remains stable at about 800 mg / m 3 . The model suggests gradually reducing the total current of the slurry circulation pumps to 220 A and controlling the slurry pH1 value at about 4.6. At the same time, according to the intelligent control strategy, the amount of oxidation air is appropriately reduced, optimizing the oxidation effect of the slurry and further reducing the operation cost.

[0103] After three months of operation monitoring, compared with the same period using traditional control methods, the operation power consumption cost of the desulfurization system of this power plant is reduced by 20%, and the absorbent cost is reduced by 12%. At the same time, the SO2 emission concentration in the clean flue gas always meets the national standards, and the desulfurization system efficiency is stable above 99%.

[0104] It can be clearly seen from this embodiment that the energy-saving operation control method of this desulfurization system can significantly improve the energy-saving effect and desulfurization efficiency of the power plant in practical applications, achieving a win-win situation of economic and environmental benefits.

[0105] Embodiment 2

[0106] 1. Establishing an accurate prediction model for the SO2 concentration in the clean flue gas is the basis for the operation optimization of the desulfurization system.

[0107] Through on-site tests, it is analyzed that flue gas parameters and operation parameters have an impact on the desulfurization performance indicators. Among them, the flue gas parameters at least include the SO2 concentration in the raw flue gas, the oxygen content in the raw flue gas (O2 content in the raw flue gas), the dust concentration in the raw flue gas, the raw flue gas temperature, the raw flue gas flow rate; and the operation parameters at least include the liquid-gas ratio, the slurry pH value, the slurry liquid level, the slurry density, the combination mode of slurry circulation pumps (referring to the combination form of multiple slurry circulation pumps), the oxygen-sulfur ratio, and the oxygen supply mode of oxidation air have an important impact on the prediction model of the SO2 concentration in the clean flue gas.

[0108] The state parameters of the desulfurization system in coal-fired power plants can be divided into static parameters and dynamic parameters. Among them, the static parameters basically do not change after the completion of the infrastructure construction or transformation of the desulfurization system. For example, the size of the desulfurization tower, the design parameters of each equipment such as the slurry circulation pump and the oxidation fan, etc.; the dynamic parameters mainly include flue gas parameters and parameters controlled during operation, etc. In the test unit, 32 online operation parameters of the desulfurization system were collected for several days. The data export method of DCS (Distributed Control System) is relatively fixed, and a data matrix of 8640×32 is obtained (calculated based on one month's data). In addition, according to the DCS physical parameter coding rule, parameter identification can be completed.

[0109] Based on the random forest method, a rough selection of the online operation characteristic parameters of the net flue gas SO2 concentration prediction model is carried out. Measuring the importance of characteristic parameters by the random forest method is achieved by calculating the out-of-bag (OOB) accuracy rate or Gini index (gini) of the characteristic parameters. The Gini index measures the importance of characteristic parameters by the average change in the node splitting impurity of a certain characteristic in all decision trees of the random forest. The out-of-bag (OOB) accuracy rate measures the importance of characteristic parameters by the average error change of the prediction results before and after randomly permuting a certain characteristic parameter in the out-of-bag samples for each decision tree. Specifically, for the characteristic parameter X j The importance is calculated according to the following steps:

[0110] (1) In the random forest method, A training samples are obtained by the sampling method with replacement, denoted as β1, β2, …, βa, and at the same time, there are A out-of-bag samples, denoted as oob1, oob2, …, ooba.

[0111] (2) For a certain training set βa, a = 1, 2, …, A, a decision tree ka is created on the training samples, and the out-of-bag sample ooba is predicted using the decision tree ka, and its prediction result is denoted as ERRa.

[0112] (3) For the characteristic parameter X j in ooba, the value is randomly permuted, and the permuted sample is denoted as oobaj. The out-of-bag aj sample data is predicted using the decision tree ka, and its prediction result is denoted as ERR aj .

[0113] (4) For a = 2, 3, …, Z, repeat the above steps.

[0114] (5) The importance measure I j of the characteristic X x is calculated by the following formula:

[0115]

[0116] The larger the Ix value before and after randomly permuting a certain characteristic parameter, the greater the impact of the characteristic parameter on the model accuracy rate, indicating that the importance of the characteristic parameter is relatively high. Conversely, the importance of the characteristic parameter is relatively low. The random forest method is used to conduct a rough selection of characteristics for the desulfurization on-line operation parameters, and the characteristics that have a greater impact on the net flue gas SO2 concentration prediction model are screened out. When running the random forest method, the number of decision trees is 200, the number of characteristics for node splitting is one-third of the total number of characteristics, and other parameters use the default settings. For the 7300×32 new desulfurization on-line operation parameter matrix obtained after preprocessing the above text, the random forest method is run five times and its average value is used as the basis for evaluating the importance of each characteristic to the net flue gas SO2 concentration prediction model.

[0117] According to the random forest method, the importance rankings of the influencing factors in the net flue gas sulfur dioxide prediction model are as follows: concentration of SO2 in the original flue gas, current of the first slurry circulation pump, current of the second slurry circulation pump, flow rate of the original flue gas, make-up slurry flow rate of the absorption tower, dust concentration of the original flue gas, current of the first oxidation fan, current of the first gypsum discharge pump, current of the second gypsum discharge pump, current of the third gypsum discharge pump, current of the fourth slurry circulation pump, pH1 at the measuring point below the slurry pool, liquid level of the absorption tower, unit load, current of the second oxidation fan, current of the third slurry circulation pump, temperature of the original flue gas, pressure of the original flue gas, current of the fifth slurry circulation pump, current of the third oxidation fan, O2 content of the original flue gas, H2O content of the net flue gas, temperature of the net flue gas, flushing water flow rate of the demister, pH2 at the measuring point above the slurry pool, O2 content of the net flue gas, dust concentration of the net flue gas, pressure of the net flue gas, make-up slurry density of the absorption tower.

[0118] The importance score of the characteristic parameter obtained by the random forest method is a measure of the accuracy of the prediction model. Based on the ranking of the importance scores of the desulfurization on-line operation characteristic parameters, combined with the mechanism analysis and experimental analysis of SO2 removal in the desulfurization tower, the characteristic parameters that are relatively unimportant to the net flue gas SO2 concentration prediction model are excluded, such as make-up slurry density of the absorption tower, pressure of the net flue gas, dust concentration of the net flue gas, pH2 at the measuring point above the slurry pool, flushing water flow rate of the demister, etc.; According to the importance ranking and influencing factors, the characteristic parameters that have a greater impact on the accuracy of the net flue gas SO2 concentration prediction model are: concentration of SO2 in the original flue gas, current of the absorption tower circulation pump, flow rate of the original flue gas, dust concentration of the original flue gas, current of the gypsum discharge pump, current of the oxidation fan, pH1 at the measuring point below the slurry pool, and liquid level of the absorption tower, etc.

[0119] Among them, when the characteristic parameters of the random forest method are roughly selected, the currents of each slurry circulation pump, the currents of each oxidation fan, and the currents of each gypsum discharge pump are considered as an independent characteristic parameter. After screening, 21 characteristic parameters are retained (the concentration of SO2 in the original flue gas, the current of the first slurry circulation pump, the current of the second slurry circulation pump, the flow rate of the original flue gas, the slurry replenishment flow rate of the absorption tower, the dust concentration of the original flue gas, the current of the first oxidation fan, the current of the first gypsum discharge pump, the current of the second gypsum discharge pump, the current of the third gypsum discharge pump, the current of the fourth slurry circulation pump, the pH1 measured point below the slurry pool, the liquid level of the absorption tower, the unit load, the current of the second oxidation fan, the current of the third slurry circulation pump, the temperature of the original flue gas, the pressure of the original flue gas, the current of the fifth slurry circulation pump, the current of the third oxidation fan, the O2 content of the original flue gas). During the actual operation of the unit, there are standby situations for slurry circulation pumps, oxidation fans, gypsum discharge pumps, etc. Therefore, when applying the particle swarm optimization algorithm for the fine selection of characteristic parameters, the characteristic parameters of the current of a single pump or fan are not considered, and only the total current of the slurry circulation pumps, the total current of the oxidation fans, and the total current of the gypsum discharge pumps are considered as characteristic parameters, thus simplifying to 13 characteristic parameters (the concentration of SO2 in the original flue gas, the total current of the slurry circulation pumps, the flow rate of the original flue gas, the slurry replenishment flow rate of the absorption tower, the dust concentration of the original flue gas, the total current of the oxidation fans, the total current of the gypsum discharge pumps, the pH1 measured point below the slurry pool, the liquid level of the absorption tower, the unit load, the temperature of the original flue gas, the pressure of the original flue gas, the O2 content of the original flue gas). The data of these 13 characteristic parameters are used as input parameters, and the concentration of SO2 in the clean flue gas is used as the output characteristic parameter data.

[0120] Introduce the deep learning algorithm: Combine deep learning technologies such as neural networks to mutually verify and supplement with the random forest method to further improve the accuracy of the prediction model. Use the Gaussian process to verify the prediction model of the SO2 concentration in the clean flue gas. The non-parametric model for regression analysis of data under the prior assumption of the Gaussian process can give the complete posterior distribution of the prediction samples, provide the uncertainty of the prediction results, and make the model have good interpretability. When the likelihood is a normal distribution, the posterior distribution of the prediction samples has a closed-form solution in the form of a Gaussian. For different projects, the selection of DCS data can be appropriately controlled. Usually, at least one month of data during stable operation is selected as the benchmark. For each project, the model construction process of the present invention can be adopted to establish the prediction model of the SO2 concentration in the clean flue gas for each project. Then, according to the outlet data, the operating parameters are appropriately adjusted to ensure energy-saving operation.

[0121] An accurate prediction model can anticipate the change trend of the SO2 concentration in the clean flue gas in advance, so as to timely adjust the operating parameters to ensure that the desulfurization system always operates in the high-efficiency range. For example, when it is predicted that the concentration of SO2 in the original flue gas is about to increase, increasing the current of the slurry circulation pump in advance or raising the pH1 value of the slurry can effectively avoid the decline of the desulfurization efficiency and ensure that the emissions meet the standards.

[0122] II. Energy-saving Operation Optimization Control Method for Desulfurization System

[0123] 1) The optimization of desulfurization operation includes two elements. One is whether the SO2 emission concentration of the clean flue gas meets the emission standard, and the other is whether the total operation cost can be minimized. Establishing a prediction model for the SO2 concentration in the clean flue gas is the basis of the entire cost model. Among them, parameters such as the SO2 concentration in the raw flue gas, the temperature of the raw flue gas, the O2 content in the raw flue gas, the flow rate of the raw flue gas, the pressure of the raw flue gas, and the dust concentration in the raw flue gas are not adjustable during operation; the liquid level in the absorption tower, the pH1 value of the slurry, the current of the gypsum discharge pump, the current of the oxidation fan, and the current of the slurry circulation pump reflect the power consumption and water consumption costs in the desulfurization operation cost. The absorbent concentration and the absorbent supplement amount, etc., reflect the absorbent cost. The absorbent cost, power consumption, and water consumption costs constitute the total cost; Considering comprehensively, the current of the slurry circulation pump and the pH1 value of the slurry are selected as the optimization parameters to find the optimal control values of the optimization parameters under various working conditions, so as to achieve the purpose of reducing the operation cost of the desulfurization system; As Figure 2 shown.

[0124] The operation optimization objectives and constraints of the desulfurization system are described as follows:

[0125] minC = C1 + C2

[0126]

[0127] minC is the low-cost electricity; C1 - absorbent cost; C2 - operation power consumption cost; I0 is the current of the slurry circulation pump; During the operation optimization process of the desulfurization system, it is specified that the SO2 concentration in the clean flue gas is not higher than 30mg / m 3 ; The pH1 value of the slurry is limited to 4.5 - 6.5; The total current of the slurry circulation pump is limited to 200 - 600A; SO2,out is the SO2 concentration in the clean flue gas; In the formula, ffitrgp (optimal feature subset) is the Gaussian process regression model (i.e., the Gaussian model formula).

[0128] Other parts not described in detail belong to the prior art.

Claims

1. An energy-saving operation control method for a desulfurization system, characterized in that, It includes the following steps: (1) Establish a prediction model for the SO2 concentration in the clean flue gas: a. Collect the online operation parameters of multiple desulfurization systems for several days to obtain the data matrix of the online operation parameters; b. Based on the random forest method, conduct a rough selection of the online operation characteristic parameters for the prediction model of the SO2 concentration in the clean flue gas; Screen out the characteristics that have a great impact on the prediction model of the SO2 concentration in the clean flue gas, and rank the importance of each online operation characteristic parameter for the prediction model of the SO2 concentration in the clean flue gas; c. Apply the particle swarm optimization algorithm to finely select the online operation characteristic parameters obtained from the rough selection to obtain a finely selected feature subset; d. Use the Gaussian process verification on the finely selected feature subset to establish a prediction model for the SO2 concentration in the clean flue gas; (2) Optimize the energy-saving operation control of the desulfurization system: Select the optimal feature subset from the finely selected feature subsets, and find the best control values of the optimal feature subsets under various working conditions; establish the energy-saving operation optimization objectives and constraints of the desulfurization system, and then adjust the operation parameters according to the outlet data to ensure energy-saving operation.

2. The energy-saving operation control method of a desulfurization system according to claim 1, wherein In step (2), the optimal feature subsets are the current of the slurry circulation pump and the pH value of the slurry; Establish the energy-saving operation optimization objectives and constraints of the desulfurization system as follows: minC = C1 + C2 where minC is the minimum cost, C1 is the absorbent cost; C2 is the operating power consumption cost; I0 is the current of the slurry circulation pump; pH1 is the pH value of the slurry; SO2,out is the SO2 concentration in the clean flue gas; ffitrgp (optimal feature subset) is the Gaussian process regression model; Among them, the concentration of SO2 in the clean flue gas is not higher than 30 mg / m 3 , the pH1 value of the slurry is limited to 4.5 - 6.5, and the total current of the slurry circulation pumps is limited to 200 - 600 A.

3. The energy-saving operation control method of a desulfurization system according to claim 1, characterized in that In step a, collect the online operation parameters of multiple desulfurization systems for one month, export the online operation characteristic parameters by DCS, and obtain the data matrix of the online operation characteristic parameters; complete the identification of the online operation characteristic parameters according to the DCS physical parameter coding rules.

4. A method for controlling the energy-saving operation of a desulfurization system according to claim 1, characterized in that In step b, the random forest method measures the importance of the online operation characteristic parameters, which is achieved by calculating the out-of-bag data accuracy or Gini index of the characteristic parameters; The Gini index measures the importance of the characteristic parameters by the average change in the node splitting impurity of a certain online operation characteristic parameter in all decision trees of the random forest method; the out-of-bag data accuracy measures the importance of the characteristic parameters by the average error change in the prediction results of each decision tree for a certain online operation characteristic parameter before and after random permutation of the out-of-bag samples.

5. A method for controlling the energy-saving operation of a desulfurization system according to claim 4, characterized in that The online operating characteristic parameter X j is calculated through the following steps: Step 1, in the random forest method, obtain A training samples through the sampling method with replacement, denoted as β1, β2,..., βa, and at the same time there are A out-of-bag samples, denoted as oob1, oob2,..., ooba; Step 2, for a certain training set βa, a = 1, 2,..., A, create a decision tree ka on the training samples, and use the decision tree ka to predict the out-of-bag sample ooba, and its prediction result is denoted as ERRa; Step 3, randomly permute the values of the feature parameter X in ooba j The permuted sample is denoted as oobaj, and use the decision tree ka to predict the oobaj sample data, and its prediction result is denoted as ERR aj ; Step 4, for a = 2, 3,..., Z, repeat the above steps; Step 5, Feature Parameter X j Importance Measure I x Calculate by the following formula: The larger the Ix value before and after randomly permuting a certain characteristic parameter, the greater the impact of the characteristic parameter on the model accuracy, and the relatively higher the importance of the characteristic parameter. On the contrary, the importance of the characteristic parameter is relatively lower.

6. The energy-saving operation control method of a desulfurization system according to claim 5, characterized in that In step d, a non-parametric model for regression analysis of data is performed under the prior assumption verified by the Gaussian process, giving the complete posterior distribution of the predicted samples, providing the uncertainty of the prediction results, and making the model have good interpretability; when the likelihood is a normal distribution, the posterior distribution of the predicted samples has a closed-form solution in Gaussian form. Establish a prediction model for the net flue gas SO2 concentration of each on-line operating parameter characteristic; then, according to the net flue gas SO2 concentration data, adjust and optimize the on-line operating parameters.

7. A method for controlling the energy-saving operation of a desulfurization system according to claim 5, characterized in that, According to the random forest method, the importance rankings of the influencing factors in the net flue gas sulfur dioxide prediction model are as follows: the concentration of SO2 in the raw flue gas, the current of the first slurry circulation pump, the current of the second slurry circulation pump, the flow rate of the raw flue gas, the make-up slurry flow rate of the absorption tower, the dust concentration of the raw flue gas, the current of the first oxidation blower, the current of the first gypsum discharge pump, the current of the second gypsum discharge pump, the current of the third gypsum discharge pump, the current of the fourth slurry circulation pump, the pH1 at the measuring point below the slurry pool, the liquid level of the absorption tower, the unit load, the current of the second oxidation blower, the current of the third slurry circulation pump, the temperature of the raw flue gas, the pressure of the raw flue gas, the current of the fifth slurry circulation pump, the current of the third oxidation blower, the O2 content of the raw flue gas, the HO2 content of the net flue gas, the temperature of the net flue gas, the flushing water flow rate of the demister, the pH2 at the measuring point above the slurry pool, the O2 content of the net flue gas, the dust concentration of the net flue gas, the pressure of the net flue gas, and the make-up slurry density of the absorption tower.

8. A method for controlling the energy-saving operation of a desulfurization system according to claim 7, characterized in that, Based on the importance score ranking of the on-line operating characteristic parameters of desulfurization, combined with the mechanism analysis and experimental analysis of SO2 removal in the desulfurization tower, exclude the characteristic parameters that are relatively unimportant for the net flue gas SO2 concentration prediction model, including at least the make-up slurry density of the absorption tower, the pressure of the net flue gas, the dust concentration of the net flue gas, the pH2 at the measuring point above the slurry pool, and the flushing water flow rate of the demister.

9. The energy-saving operation control method of a desulfurization system according to claim 7, characterized in that Based on the importance score ranking of the on-line operating characteristic parameters of desulfurization, combined with the mechanism analysis and experimental analysis of SO2 removal in the desulfurization tower, according to the importance ranking and influencing factors, select the characteristic parameters that have a greater impact on the accuracy of the net flue gas SO2 concentration prediction model, including at least the concentration of SO2 in the raw flue gas, the current of the absorption tower circulation pump, the flow rate of the raw flue gas, the dust concentration of the raw flue gas, the current of the gypsum discharge pump, the current of the oxidation blower, the pH1 at the measuring point below the slurry pool, and the liquid level of the absorption tower. The current of the slurry circulation pump includes the sum of the current of the first slurry circulation pump, the current of the second slurry circulation pump, the current of the third slurry circulation pump, the current of the fourth slurry circulation pump, and the current of the fifth slurry circulation pump; the current of the gypsum discharge pump includes the sum of the current of the first gypsum discharge pump, the current of the second gypsum discharge pump, and the current of the third gypsum discharge pump; the current of the oxidation blower includes the sum of the current of the first oxidation blower, the current of the second oxidation blower, and the current of the third oxidation blower.

10. A method for controlling the energy-saving operation of a desulfurization system according to claim 1, characterized in that, In step c, the method for applying the particle swarm optimization algorithm to carefully select the on-line operating characteristic parameters obtained by rough selection to obtain a carefully selected feature subset is as follows: Coarsely select the characteristic subset of desulfurization influencing factors (S1) using the random forest method. Then, apply the particle swarm optimization algorithm to randomly initialize the particle swarm of the on-line operation characteristic parameters of desulfurization obtained by the coarse selection (S2). Next, calculate the particle velocity and position (S3). If the fitness of the current position is greater than the fitness value of the individual optimal position (S4), then update the individual position (S5) to make the fitness of the individual optimal position greater than the fitness value of the global optimal position (S6); if the fitness of the current position is less than the fitness value of the individual optimal position (S4), then make the fitness of the individual optimal position greater than the fitness value of the global optimal position (S6); update the global optimal position (S7), reach the maximum number of iterations (S8), and obtain the selected characteristic subset in the desulfurization influencing factors from the global optimal position (S9). If the maximum number of iterations is not reached (S8), then return to calculate the particle velocity and position (S3); if the fitness of the individual optimal position is less than the fitness value of the global optimal position (S6), then enter to obtain the selected characteristic subset in the desulfurization influencing factors from the global optimal position (S9).

Images