Segmented planned load and switching mill vector coupling-oriented coal blending combustion optimization method

Abstract

The invention provides a coal blending combustion optimization method for segmented planned load and switching mill vector coupling, and relates to the technical field of power plant energy conservation and environmental protection. The method comprises the following steps: firstly, determining an objective function of a coupling distribution combustion optimization model between a segmented plan load and a switching mill vector according to the optimal coal blending cost, and introducing a constraint condition; converting the established coupling combustion optimization model into a standard type and a model type; aiming at the model, adopting two-stage simplex solution to obtain a coal blending scheme; and finally, calculating a sulfur constraint upper bound by adopting an inversion method, and establishing an improved random configuration network to carry out feed-forward compensation on the sulfur constraint bound based on a historical burning case database. And based on the real-time coal blending data each time, updating the historical blending combustion case database. According to the method, the coupling relation between the segmented planned load and the distribution combustion optimization model is established through the switching mill vector of the coal mill unit, and digital distribution combustion based on the computer optimization model is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power plant energy saving and environmental protection technology, and particularly relates to a coal blending and burning optimization method for segmented planning load and switching mill vector coupling. BACKGROUND

[0003] Under the premise of ensuring safe operation of the unit, environmental protection and load scheduling, the coal blending and burning (referred to as blending) technology is the key means to realize the optimization of coal-fired cost. The international advanced thermal power plant has formed an intelligent blending system with computer algorithm as the core, through warehouse mixing, belt mixing and in-furnace direct mixing processes, to realize the collaborative optimization of power generation efficiency and environmental performance. This automatic solution based on the theory of coal blending and burning marks that the blending technology has entered the stage of digitization.

[0004] The development of coal blending and burning technology in China has experienced two important stages: 1) In the 1990s, Chinese thermal power enterprises could not obtain sufficient supply of single design coal, in addition, the power industry also needed to blend some low calorific value coal to improve enterprise efficiency, and the demand for coal blending and burning by enterprises prompted teams mainly composed of research institutes and universities to study the combustion characteristics of mixed coal; 2) At present, the coal blending and burning technology of most domestic power plants still remains in the non-intelligent stage of "manual experience + Excel auxiliary", lacking advanced methods and professional software, and urgently needs to be transformed to digitized blending. Digitized blending requires thermal power enterprises to fundamentally change the way of work and business model, and needs to use optimization technology and intelligent technology to solve the non-optimal operation problem of the whole cycle of fuel blending, and fully tap the potential and benefits of the whole process and cycle of blending. SUMMARY

[0005] To solve the above technical problems, the technical solution adopted by the present application is: the present application provides a coal blending and burning optimization method for segmented planning load and switching mill vector coupling, comprising: Establishing a logical business coupling relationship between segmented planning load, switching mill vector and coal blending, and determining the objective function of the coupling blending optimization model between segmented planning load and switching mill vector according to the optimal coal blending cost; Introducing calorific value constraint conditions, environmental protection constraint conditions, safety constraint conditions and coal type proportion constraint conditions for the coupling blending optimization model; Converting the established coupling blending optimization model into a standard type and a canonical type; For the canonical type of the coupling blending optimization model, a two-stage simplex method is used to obtain a coal blending scheme; An inversion method is used to calculate the upper bound of sulfur constraint, and based on the historical blending case database, an improved random configuration network is established to feed forward compensate the sulfur constraint boundary; Based on each real-time coal blending data, iterative learning technology is used to dynamically and continuously update the historical coal blending case database to ensure the accuracy of the modeling case data.

[0006] Preferably, the logical business coupling relationship between establishing segmented planned load, switching mill vectors, and coal blending is established, and the objective function of the coupled blending optimization model is determined according to the optimal blending cost, as shown in the following formula: (1); In the formula, To minimize the cost of blended coal, For the cutting and grinding vector coefficients, and , This is a segmented load coal distribution function. For the first The number of coal mills added to each planned load segment, For the first The planned load segment The price of each type of coal For the first The first load segment The proportion of blended coal types for each coal type For the first A vector representing the proportion of coal types blended in each planned load segment. To adjust the proportion of coal types in three stages: deep load, medium-high load, and high load. The vector formed by these vectors.

[0007] Preferably, the calorific value constraints, environmental protection constraints, safety constraints, and coal type ratio constraints of the coupled combustion optimization model are specifically as follows: The aforementioned calorific value constraint means that the calorific value of the coal supplied to each load segment of the unit must be greater than or equal to the lower limit of the calorific value determined by the boiler load of each load segment, as shown in the following formula: (2); In the formula, This is the weighted average of the calorific value of the blended coal in the unit. This is the coal blending function based on the calorific value of the load. For the first The planned load segment The calorific value of each type of coal For the first The number of coal mills operating in each planned load segment. This is the lower limit function vector of calorific value corresponding to segmented loads. For the first The lower limit of calorific value for each planned load segment is shown in the following formula: (3); In the formula, For the first The planned load of each planned load segment For the first The first planned load segment corresponding to the planned load. Maximum coal feed rate of the coal mill. For the first Unit coal consumption corresponding to the planned load of each planned load segment, unit coal consumption With the efficiency and load of the unit related, This represents the total number of coal mills operating at deep load, medium-high load, or high load. The environmental constraints, namely sulfur content constraints, are shown in the following formula: (4); In the formula, This represents the weighted average sulfur content of the coal blends under different planned load ranges for the coal mill unit, where the sum of the proportions of each blend is 1. This represents the increased sulfur-coal blending function vector under segmented load conditions. For the first The planned load segment The sulfur content of each type of coal, This is the upper limit function vector for sulfur corresponding to the segmented load. For the first The upper limit of desulfurization under each planned load segment is obtained by inverting the capacity of the desulfurization equipment using the following formula: (5); In the formula, sulfur Converted to sulfur dioxide Conversion efficiency For the desulfurization efficiency of desulfurization equipment, Removes sulfur dioxide from the desulfurization equipment per hour Ability; The sulfur constraint boundary in equation (5) is derived by inversion from the maximum desulfurization capacity of the desulfurization equipment. In order to ensure that environmental protection standards are not exceeded, in practical applications, equation (5) is simplified to the following equation: (6); The safety constraints are shown in the following formula: (7); In the formula, For volatiles under segmented load , moisture or ash The weighted average of the values, where the sum of the coal blending ratios is 1. and Increased volatile matter under segmented load , moisture or ash The coal blending function vector, For the first The planned load segment The volatile matter, moisture, or ash content of each type of coal. and These are the upper and lower bound vectors for volatile matter, moisture, or ash content corresponding to the segmented loads; The coal type ratio constraint is shown in the following formula: (8); (9); In the formula, The vector combining the maximum coal feed rate of the coal mill increased by the three load sections. and Indicates the first The maximum coal feed rate of one or two additional coal mills in each load section; The proportion coefficient of the selected coal type in the coal bunker in the coupled combustion optimization model is constrained by the following formula: (10); In the formula, and It is the coal feeding coefficient of the reclaimer. For the first The first load segment The maximum coal feeding rate of the coal mill; the coal blending optimization model requires that the coal blending amount of each coal type be practically meaningful and conducive to the coal feeding of the reclaimer, therefore the coal feeding coefficient of the reclaimer is often taken as... , ; The constraints of the coupled combustion optimization model also include coal quality characteristic index constraints, coal type priority constraints at the coal yard, and coal type constraints based on the structure and parameter characteristics of the coal mill unit.

[0008] Preferably, the established coupled combustion optimization model is transformed into a standard form and a canonical form. For the canonical form of the coupled combustion optimization model, the specific method for obtaining the coal blending scheme using a two-stage simplex solution is as follows: For the inequality constraints in the coupled firing optimization model, after adding relaxation vectors and residual vectors, it is transformed into the following standard form, namely the standard firing optimization model: (11); In the formula, For generalized vectors, For the proportion vector of coal types in coal blending, For the remaining vector, the slack vector, , the constraint vector, the coefficient matrix; , , , the identity matrix with 1 on the main diagonal ) the identity matrix ) the identity matrix with -1 on the main diagonal ) the identity matrix ) Since the coefficient matrix does not have a standard admissible basis, artificial variables are introduced to construct the following canonical form: (12); where, , , , is an artificial variable added to form the initial basis vector in linear programming, , is the initial basis vector, is the identity matrix with 1 on the main diagonal ) the identity matrix ) The first stage of the simplex method is used to solve equation (12). The optimal solution of equation (12) is obtained by iterative calculation of the artificial variable. If the optimal solution does not contain artificial variables, the optimal solution of equation (12) is a basic feasible solution of equation (11). At this time, the coefficient vector of the artificial variable is transferred to , and is used to replace in equation (11), obtaining a canonical form of linear programming equivalent to equation (11): (13); where, is the coefficient matrix containing the identity matrix after the first stage transformation of the simplex method, is the constraint vector after the first stage transformation of the simplex method; The second stage of the simplex method is used to solve the canonical form of linear programming in equation (13) to obtain the optimal solution of the blending optimization model; For the coupled coal blending optimization model of the typical segmented planned load and the switching mill vector coupling, the classical two-stage simplex method is used to solve it; if there is a solution, the coal blending scheme is output and the coal blending is completed; if there is no solution, the coal feed rate of the pulverizer corresponding to the deep adjustment load is adjusted. The coal feed rate of the pulverizer corresponding to the deep adjustment load is gradually reduced by setting a step size, and the rate adjustment is as follows: (14); In the formula, To adjust the step size of the coal feed rate, For the number of iterations, For the first phase of load The first coal mill The coal feed rate in the next iteration; At each iteration's coal feed rate, let... Then solve equation (11) again until an optimal solution is found, and stop iterating.

[0009] Preferably, based on the calculation of the upper limit of the sulfur constraint using the inversion method, and based on a historical calcination case database, the specific method for establishing an improved random configuration network to perform feedforward compensation of the sulfur constraint boundary is as follows: The sulfur constraint adjustment is related to the sulfur constraint setpoint in the load and combustion model. An improved stochastic configuration network is used to establish a two-input-single-output nonlinear model for sulfur constraint compensation: (15); In the formula, u represents the number in the real number field. The two input variables of the compensated nonlinear improved stochastic allocation network model are y in the real number domain. Compensation for a single output variable in a nonlinear model The input-output function relationship established for the improved stochastic configuration network. For load, Sulfur confinement boundary, This is the feedforward compensation amount for the sulfur constraint boundary; Given a historical cooking case database for training an improved stochastic configuration network Group data , where input The input of the kth training data And the corresponding output , During network training, assuming the improved randomly configured network has constructed L-1 hidden layer nodes, then the output of the L-1th hidden layer node is... Represented as: (16); In the formula, is the output of the L-1th hidden layer node at the kth data input, is the activation function of the hidden layer, represents the output of the L-1th hidden layer node corresponding to the kth data input, and are the input weight vector and bias of the L-1th hidden node, respectively, and the output of the improved randomly configured network based on the training data input is: (17) ; wherein, is the output predicted by the improved randomly configured network with L-1 hidden nodes according to the training data set input, and are the input weight vector and bias of the lth hidden node, respectively, is the output weight of the lth hidden node, calculated by: (18) ; wherein, is the output of the lth hidden layer node of the ISCN network, and the network residual error at this time is is: (19) ; the initial residual error of the network , the network tolerance limit is set to , if the residual error of the improved randomly configured network , the Lth hidden node is added to the hidden layer, and the new node parameters are configured using the IPOA algorithm.

[0010] Preferably, the IPOA algorithm is used to configure the new node parameters The process is as follows: Step S1. Set the IPOA optimization target function and constraints for the new Lth hidden node parameters of the improved randomly configured network as follows: (20) ; wherein, is the maximum of the target function, 0 <1, non-negative sequence , , L is the number of hidden nodes of the improved randomly configured network, initially 1, is the error of the improved randomly configured network with L-1 hidden nodes in the hidden layer, is the output of the Lth hidden node; Step S2. Initialize the population: (21) ; (twenty two); set up For the first The pelican seeks the position corresponding to the optimal parameter vector. For the first The Pelicans in The position of the dimension , Let be the population size of the pelicans, and the location matrix of the pelican population is as follows: The scalar function matrix of the pelican population is , For the first The objective function value corresponding to the position of each individual pelican. It is a chaotic sequence. A random number in the interval [0,1]. and These represent the upper and lower boundaries of the domain of the newly added hidden node parameters. A constant less than 1; Step S3. Randomly generate the prey target locations for the pelican population to search within the defined space. The pelican population moves towards its prey location, and the location update is shown in the following formula: (twenty three); In the formula, For the prey in the The position of the dimension For the first After the pelican moved toward its prey, it was in the first... Wei's new position The objective function value corresponding to the prey's location. A random number within the range [0,1] For a random number equal to 1 or 2, update the position of the parameter vector represented by each pelican in the population using the following formula: (twenty four); In the formula, For the Pelicans' new position The objective function value; Step S4. Local search using IPOA to find the new position of the parameter vector: (25); In the formula, For the first local search Only the Pelicans in The new position of the dimension parameter, The average value of the population objective function. The optimal position for the current population is at the th position. The coordinates of the Cauchy (0, 1) are standard Cauchy distribution, is the current iteration number, is the maximum iteration number, is the search space radius; after the local search is completed, the position corresponding to the individual optimization parameter is updated based on equation (24); Step S5. Set the target function change threshold If the change in the target function value before updating the population optimal position of IPOA is less than , then the population optimal position is sought The parameter configuration of the newly added Lth hidden node of the improved randomly configured network is completed; Step S6. Calculate the output of the improved randomly configured network after adding the node and the network residual error: (26) ; (27) ; If the residual error of the improved randomly configured network is , then the training of the improved randomly configured network is completed, otherwise, the (L+1)th hidden node is added to the hidden layer, and steps S1-S6 are repeated until the residual error of the improved randomly configured network is ; After the training of the improved randomly configured network is completed, the test data is input At this time, the sulfur constraint compensation predicted by the improved randomly configured network is: (28) ; In the formula, is the compensation amount of the upper bound of the sulfur constraint.

[0011] Preferably, the specific method of dynamically rolling updating the historical blending case database based on real-time blending data each time using iterative learning technology is: Real-time monitoring of the actual sulfur dioxide emission concentration after desulfurization of boiler mixed coal combustion , and the deviation of the expected emission concentration of the environmental protection standard is quantified, so as to correct the upper bound of the sulfur constraint, and the iterative learning compensation of the upper bound of the sulfur constraint: (29) ; In the formula, and are the upper bounds of the sulfur constraint of the coupled blending optimization model in the dth and (d+1)th iterations, respectively, and the initial value is set according to equation (5), is the feedforward compensation amount of the upper bound of the sulfur constraint predicted by the improved randomly configured network in the (d+1)th iteration, is the deviation of the actual sulfur dioxide emission concentration after desulfurization from the environmental protection standard emission concentration in the dth iteration, The sulfur dioxide emission deviation in the d-th iteration The deviation of the sulfur constraint setpoint obtained from the inversion calculation This represents the actual sulfur dioxide emission concentration after desulfurization during the d-th iteration of coal-fired combustion. Let d be the flue gas emission flow rate in the d-th iteration. Let be the flue gas density in the d-th iteration. For coal sulfur conversion efficiency, The maximum coal feeding rate of the k-th coal mill in the i-th planned load segment; The historical calcination case database of the improved stochastic configuration network is updated using the results of the sulfur constraint upper bound iteration, and the parameters of the improved stochastic configuration network are retrained to improve the model prediction accuracy.

[0012] Preferably, the historical food preparation case database update steps are as follows: Step C1. Convert the training data in the historical firing case database into triplet case format. Suppose that the case database already stores W source cases, where the w-th source case... The triplet form is as follows: (30); In the formula, Input for the w-th case, These represent the load and sulfur constraint upper bound for the w-th case, respectively. Output for the w-th case. The evaluation metric for the w-th case is: (31); In the formula, This is the upper bound of the sulfur constraint in the w-th case. The deviation of the upper limit setting value for the sulfur constraint in the w-th case; Step C2. Case retrieval, for a new case The similarity between a case and a source case in the case library is retrieved using the Euclidean distance shown in the following formula: (32); In the formula, For new cases To the source case European distance, For new cases With source case Based on the overall similarity, the source cases in the case library are traversed according to the above formula to retrieve cases similar to the new cases. Maximum similarity Source Case ,in, Each is a new case input, output and evaluation index of the source case with the maximum similarity; Step C3. Evaluate the overall similarity of the new case; set the overall similarity evaluation threshold If , go to step C4, otherwise go to step C6; Step C4. Evaluate the single-dimensional similarity of the new case; set the load evaluation threshold and sulfur constraint boundary threshold If the load similarity and sulfur constraint boundary similarity are both within the threshold range, go to step C5, otherwise go to step C6; Step C5. Case evaluation and update, if the new case evaluation index , discard the case; otherwise, the new case is assigned to replace the source case , ; Step C6. If the evaluation index of the new case , is the evaluation threshold, the new case is stored in the historical blending case database.

[0013] Preferably, the method also develops a blending optimization software system using B / S architecture, and automatically blends based on the coupled coal blending optimization model. The specific method is as follows: Develop a relational database interface based on ADO.NET and establish a SQL Server relational database table of coal yard coal types, including coal type name, heat value, sulfur content, moisture content, volatile matter, ash content, price, and coal yard number information; based on the coupling relationship of segmented load and mill vector, develop a coupled coal blending optimization model running program using C#, including a static blending optimization model and a dynamic blending optimization model; the static blending optimization model only uses the inversion calculation method to calculate the upper bound of the sulfur constraint for each coal blending of the blending optimization model, and the dynamic blending optimization model uses the inversion calculation method, improved random configuration network, and iterative learning comprehensive technology to dynamically calculate the sulfur constraint boundary throughout the cycle for each coal blending of the blending optimization model; develop a software module using C# to solve the static blending optimization model and the dynamic blending optimization model to calculate the coal blending scheme; use ASP.NET to develop a front-end interface of the system to display multiple coal blending schemes, and the coal blending personnel selects the scheme according to actual needs; based on the selected coal blending scheme, transmit the coal types and proportions of each coal bunker to the reclaimer control system to coordinate the two reclaimers to feed coal to the coal bunker; read the real-time load data of the SIS system through the real-time database interface, and automatically switch the coal mill according to the coal blending optimization model; within each load segment, adjust the mill output by the mill control system to adapt to the load change requirements.

[0014] The beneficial effects produced by the above technical solution are as follows: (1) the present application digitizes the logical business coupling and interface design between the segmented planned load, the coal mill switching mill and the coal blending, and for the first time, systematically establishes a coupled coal blending optimization model facing the segmented planned load and the switching mill vector coupling, thereby ensuring the optimality and rapidity of the coal blending cost.

[0015] (2) for the problem that the key sulfur constraint boundary in the coupled coal blending optimization model is difficult to accurately determine, the present application proposes an inversion method based on the desulfurization device to calculate the sulfur constraint boundary, thereby ensuring that the coal blending and blending of the unit reaches the hard requirement of the environmental protection index.

[0016] (3) further considering that the calculation of the sulfur constraint boundary by the inversion method is affected by the uncertainties such as the seasonal changes of the sulfur conversion efficiency of the coal, the desulfurization efficiency of the desulfurization equipment and the seawater desulfurization capacity of the equipment, which often leads to the artificial further restriction of the sulfur constraint boundary, and the excessive guarantee of the environmental protection index requirement specified by the international, and the significant reduction of the high-sulfur coal blending, and the loss of the economy. For this problem, the present application first proposes to improve the stochastic configuration network and the iterative learning technology to feed forward the dynamic compensation of the sulfur constraint boundary, thereby maximizing the saving of the coal blending cost under the condition of meeting the environmental protection index.

[0017] (4) based on the coupled coal blending optimization model and the solving method, a coal blending optimization software system is developed, which realizes the automatic power coal blending, enables the general operator to rapidly possess the expert-level knowledge, and achieves good economic benefits.

[0018] The coal blending and blending optimization method facing the segmented planned load and the switching mill vector coupling provided by the present application digitizes the coupling relationship between the segmented planned load, the switching mill vector and the coal blending model, determines the objective function and the constraint condition of the coupled coal blending optimization model according to the optimal coal blending cost; for the problem that the key environmental protection index sulfur constraint upper limit in the model is difficult to accurately determine, an inversion calculation constraint upper limit based on the desulfurization equipment capacity, a dynamic feed forward compensation constraint boundary based on the improved stochastic configuration network and the historical data, and a comprehensive strategy based on the real-time data full-cycle iterative optimization case sample database are established; for the model, it is further converted into a canonical type and solved by a two-stage simplex method to obtain an optimal coal blending scheme; on the basis of the optimization model and the solving method, a practical application system is developed by using the B / S architecture, which significantly reduces the coal blending and blending cost under the premise of ensuring the safe and environmentally-friendly operation of the power plant.

[0019] In summary, the present application has great economic value, social value and scientific value. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A physical process flow chart of coal blending and burning provided for the embodiment of the present application for a thermal power plant; Figure 2 A technical route implementation chart of the coal blending and burning optimization method provided for the embodiment of the present application for the segmented planned load and switching mill vector coupling; Figure 3 A load-sulfur conversion efficiency and load-desulfurization efficiency curve provided for the embodiment of the present application; Figure 4 A coupling coal blending optimization model solving process chart under different load modes provided for the embodiment of the present application; Figure 5 A flow chart of the improved random configuration network algorithm provided for the embodiment of the present application; Figure 6 A sulfur constraint boundary iteration compensation flow chart provided for the embodiment of the present application; Figure 7 A software interface chart of the coal blending and burning optimization software system provided for the embodiment of the present application. DETAILED DESCRIPTION

[0021] The specific embodiments of the present application will be further described in detail below in combination with the drawings and embodiments. The following embodiments are used to illustrate the present application, but not to limit the scope of the present application.

[0022] This embodiment takes a unit of a Dalian power plant as an example, and realizes coal blending and burning optimization by using the coal blending and burning optimization method of the present application for the segmented planned load and switching mill vector coupling.

[0023] The coal blending and burning process flow of the unit is shown in Figure 1 The maximum rated power generation load of the unit is 350 MW, and it is equipped with 5 coal mills. The deep load (low load) mode refers to running at 30%-50% of the rated load with 2 constant-speed coal mills, which is used for grid peak shaving. The medium-high load mode refers to running at 50%-80% of the rated load with 2 constant-speed coal mills and 1 medium-high load coal mill. The high load mode refers to running at more than 80% of the rated load with 2 constant-speed coal mills and 1 high load coal mill.

[0024] In this embodiment, the coal blending and burning optimization method for the segmented planned load and switching mill vector coupling, as shown in Figure 2 includes the following steps: Step S1: establishing a logical business coupling relationship between the segmented planned load, switching mill vector and coal blending, and determining the objective function of the coupling coal blending and burning optimization model between the segmented planned load and switching mill vector according to the optimal coal blending cost; In this embodiment, a logical business coupling relationship is established between segmented planned load, switching mill vectors, and coal blending. The objective function of the coupled blending optimization model is determined according to the optimal blending cost, as shown in the following formula: (1); In the formula, To minimize the cost of blended coal, For the cutting and grinding vector coefficients, and , This is a segmented load coal distribution function. For the first The number of coal mills added to each planned load segment (starting from zero). For the first The planned load segment The price of each type of coal For the first The first load segment The proportion of blended coal types for each coal type For the first A vector representing the proportion of coal types blended in each planned load segment. To adjust the proportion of coal types in three stages: deep load, medium-high load, and high load. The vector formed by these vectors.

[0025] Step S2: Introduce calorific value constraints, environmental protection constraints, safety constraints, and coal type ratio constraints for the coupled combustion optimization model; To ensure that thermal power units minimize coal blending costs while meeting load demands, the coupled coal blending model needs to introduce calorific value constraints. This means that the calorific value of the coal blended for each load segment of the unit must be greater than or equal to the lower limit of the calorific value determined by the boiler load (including power generation load, heating load, and steam supply load) for each load segment, as shown in the following formula: (2); In the formula, This is the weighted average of the calorific value of the blended coal in the unit. This is the coal blending function based on the calorific value of the load. For the first The planned load segment The calorific value of each type of coal For the first The number of coal mills added to the planned load segment (initially 0 units). This is the lower limit function vector of calorific value corresponding to segmented loads. For the first The lower limit of calorific value for each planned load segment is shown in the following formula: (3); In the formula, For the first The planned load of each planned load segment For the first The first planned load segment corresponding to the planned load. Maximum coal feed rate of the coal mill. For the first Unit coal consumption corresponding to the planned load of each planned load segment, unit coal consumption With the efficiency and load of the unit related, This represents the total number of coal mills operating at deep load, medium-high load, or high load.

[0026] This embodiment established the relationship data between boiler load and coal consumption as shown in Table 1 through field tests. The functional relationship between boiler load and coal consumption can be established by fitting.

[0027] Table 1. Data on the relationship between boiler load and coal consumption; ; Sulfur dioxide (SO2) emitted during coal combustion Sulfur is one of the main pollutants in thermal power plants, and sulfur must be controlled during the coal blending stage. The sulfur content is limited, requiring that the sulfur content of the blended coal in the unit cannot be too high. The sulfur content constraint in the coupled coal blending optimization model is as follows: (4); In the formula, This represents the weighted average sulfur content of the coal blends under different planned load ranges for the coal mill unit, where the sum of the proportions of each blend is 1. This represents the increased sulfur-coal blending function vector under segmented load conditions. For the first The planned load segment The sulfur content of each type of coal, This is the upper limit function vector for sulfur corresponding to the segmented load. For the first The upper limit of desulfurization under each planned load segment is obtained by inverting the capacity of the desulfurization equipment using the following formula: (5); In the formula, For sulfur ( ) is converted into sulfur dioxide ( The conversion efficiency (equivalent to combustion efficiency) of ). For the desulfurization efficiency of desulfurization equipment, The desulfurization equipment removes sulfur dioxide per hour ( The ability to perform ) Figure 3 The functional relationship between coal sulfur conversion efficiency and desulfurization efficiency and load, determined by field tests, can be calculated after fitting. and .

[0028] The sulfur constraint boundary in equation (5) is derived by inversion from the maximum desulfurization capacity of the desulfurization equipment. However, the sulfur conversion efficiency of coal is... Desulfurization efficiency , Desulfurization capacity is subject to random uncertainty, which directly affects the accuracy of the upper limit of desulfurization. In order to strictly ensure that environmental protection standards are not exceeded, in practical applications, equation (5) is often simplified to the following equation: (6); Mill explosions, burner nozzle burnout, and boiler slagging are safety factors to be considered in the operation of thermal power plant units, directly affected by the volatile matter, moisture, and ash content of the blended coal. For example, high volatile matter blended coal increases the release of combustible gases, increasing the risk of mill explosions and making burner nozzles more susceptible to burnout, while low volatile matter blends are less conducive to ignition. Therefore, the following safety constraints are added to the coupled coal blending optimization model: (7); In the formula, For volatiles under segmented load , moisture or ash The weighted average of the values, where the sum of the coal blending ratios is 1. and Increased volatile matter under segmented load , moisture or ash The coal blending function vector, For the first The planned load segment The volatile matter, moisture, or ash content of each type of coal. and These are the upper and lower bound vectors for volatile matter, moisture, or ash content corresponding to the segmented loads.

[0029] The coupled coal blending optimization model, while satisfying the constraints of calorific value, environmental protection, and safety, still needs to further consider the proportion constraints of different coal types in the coal bunker. The coal type proportion constraints after quantifying the blending into one ton of coal are as follows: (8); (9); In the formula, The vector combining the maximum coal feed rate of the coal mill increased by the three load sections. and Indicates the first The maximum coal feed rate of one or two additional coal mills in each load section; The proportion coefficient of the selected coal type in the coal bunker in the coupled combustion optimization model is constrained by the following formula: (10); In the formula, and It is the coal feeding coefficient of the reclaimer. For the first The first load segment The maximum coal feeding rate of the coal mill; the coal blending optimization model requires that the coal blending amount of each coal type be practically meaningful and conducive to the coal feeding of the reclaimer, therefore the coal feeding coefficient of the reclaimer is often taken as... , .

[0030] The above-mentioned coupled combustion optimization model, based on the actual needs of each power generation enterprise, also includes constraints on coal quality characteristics such as slagging characteristics, burnout characteristics, and grindability, as well as other special constraints, such as coal type priority constraints at the coal yard and coal type constraints based on the structure and parameter characteristics of the coal mill unit.

[0031] Step S3: Transform the established coupled firing optimization model into standard and canonical forms; In this embodiment, the inequality constraints in the coupled firing optimization model are transformed into the following standard form (i.e., the standard firing optimization model) by adding relaxation vectors and residual vectors: (11); In the formula, For generalized vectors, For the proportion vector of coal types in coal blending, For the remaining vector, Let be the relaxation vector. , To constrain the target vector, It is a coefficient matrix; , , Both diagonals are 1 ( ) ( identity matrix, Both the main diagonal lines are -1 ( ) ( ) Identity matrix.

[0032] Due to the coefficient matrix Since there is no standard allowable basis, artificial variables need to be introduced to construct the following canonical form: (12); In the formula, , ; , These are artificial variables added to form the initial basis vectors in linear programming. , Let them be the initial basis vectors. Both the main diagonal and the diagonal are 1 ( ) ( ) Identity matrix.

[0033] Step S4: For the canonical form of the coupled combustion optimization model, a two-stage simplex solution is used to obtain the coal blending scheme; The first stage of the simplex method is initiated to solve equation (12). The optimal solution of equation (12) is calculated through the basis removal iteration of artificial variables. If the basis vectors of the optimal solution do not contain artificial variables, the optimal solution of equation (12) is a basic feasible solution of equation (11). The coefficient vector (unit vector) of each artificial variable is all transferred to In the middle, use In the substitution (11) This yields the canonical form of linear programming equivalent to equation (11): (13); In the formula, for The coefficient matrix containing the identity matrix after the first stage transformation using the simplex method. for The constrained target vector after the first stage transformation using the simplex method; The second stage of the simplex method is initiated to solve the canonical form of the linear programming equation (13) to obtain the optimal solution of the sintering optimization model; For the coupled coal blending optimization model involving the typical segmented planned load and the switching mill vector coupling, the classic two-stage simplex method is used for solution. If a solution exists, the coal blending scheme is output, completing this coal blending operation; if no solution exists, the coal feed rate of the mill corresponding to the deep-adjustment load is adjusted. The solution flowchart is as follows: Figure 4 As shown.

[0034] If the deep-load adjustment is relatively small, the calorific value of the coal type blended during the deep-load adjustment stage will be too low. This may cause the calorific value of the blended coal required by the high-load pulverizer to exceed the maximum calorific value of the coal currently in the coal yard, resulting in no solution for the high-load coal blending optimization model. To address this issue, the coal feed rate of the pulverizer corresponding to the deep-load adjustment is gradually reduced by setting a step size. The rate adjustment is as follows: (14); In the formula, To adjust the step size of the coal feed rate, The number of iterations (starting) , is the maximum coal feed rate of the coal mill, is the first stage load (deep regulation load), is the coal mill, is the coal feed rate of the second iteration; At each iteration of the coal feed rate, set and solve equation (11) again until the optimal solution is obtained, and stop the iteration.

[0035] This embodiment gives a comparison example of coupling combustion optimization model and artificial experience coal blending under segmented planning load. This example only considers the upper limit of sulfur constraint using desulfurization equipment inversion calculation, and does not consider the feedforward compensation modeling and iterative learning technology for further dynamic adjustment.

[0036] In order to verify the economy and feasibility of the coupling combustion optimization model established in the foregoing, this embodiment designs a comparison experiment of artificial coal blending and optimized coal blending using the coupling combustion optimization model. For the coal type information of the coal yard in Table 2, the mill parameter settings and coal blending parameter settings in the experiment are shown in Tables 3 and 4, respectively. According to equation (6), the sulfur constraint boundary setting values of the deep regulation, medium-high load and high load segments are 0.68%, 0.45% and 0.45%, respectively. The boiler loads of different load segments are = power generation load + heat supply load + steam supply load , . Therefore, the deep regulation load segment boiler load = 90 MW, the medium-high load segment boiler load = 180 MW, and the high load segment boiler load = 290 MW.

[0037] Table 2: Coal yard coal type information ;

[0038] Table 3: Mill settings ;

[0039] Table 4: Coal blending parameter settings ; Based on the working condition parameters in Table 4, this embodiment compares artificial experience coal blending with model optimized coal blending, and the coal blending results are shown in Table 5. The sulfur content of the model coal blending deep regulation, medium-high and high load segment coal blending scheme is controlled between 0.37% and 0.45%, which meets the environmental protection constraints of coal blending. The actual SO2 emission concentration of the mixed coal combustion after desulfurization is 22.84 (the national standard limit is 35 ) Meet environmental standards. Based on the model of coal blending scheme in deep regulation, medium and high load section price respectively reduced 28.29 yuan, 18.87 yuan and 31.69 yuan, the average cost of coal blending reduced 2.68%.

[0040] Table 5 manual and model coal blending scheme comparison; ; According to the coal yard coal data in table 2, the theoretical combination of four operating coal mills in a unit of Dalian power plant can reach 10,000, and if the mixing ratio is considered, the number of coal blending schemes is even larger. Although manual coal blending selects the optimal scheme with sulfur content reaching the constraint boundary from a large number of combinations through experience (table 5 data), its economy is still significantly lower than that of the optimized coal blending result of the blending model. The cost of the blending scheme in deep regulation, medium and high load section is reduced by 28.29 yuan / ton, 18.87 yuan / ton and 31.69 yuan / ton respectively than manual blending, and the average blending cost is reduced by 2.68%. Dalian power plant has four units with daily coal consumption of 5000 tons, according to the annual blending cost = 365 × daily coal consumption × coal price (see table 5 for specific coal price), it can be calculated that based on the blending model of deep regulation, medium and high load section blending, the annual cost can be saved by 5163 million yuan, 3334 million yuan and 5783 million yuan respectively.

[0041] This gap highlights the limitations of manual optimization in complex combinations, even if the manual scheme selected is close to the optimal one, the economic loss is still in the order of tens of millions of yuan.

[0042] Step S5: Calculate the upper limit of sulfur constraint by inversion method, based on historical blending case database, establish improved random configuration network to feed forward compensation for sulfur constraint boundary; The above coupling blending optimization model and solving method, the environmental protection sulfur constraint condition is obtained by inverse calculation based on the desulfurization equipment capacity, in order to ensure the environmental protection, formula (6) is used for the calculation of the reduced boundary, a certain amount of economic sacrifice. This way also does not carry out closed loop detection for each blending, called static coupling blending optimization model, if the enterprise is not optimal to the economy, this model can be used. In order to further improve the economy of blending, the following gives the dynamic coupling blending optimization model and its solving example.

[0043] The above coupling blending optimization model formula (5) or (6) calculates the upper limit of sulfur constraint by inverse calculation of equipment desulfurization capacity. However, the uncertainty of sulfur conversion efficiency of coal, desulfurization efficiency of desulfurization equipment and seasonal variation of seawater desulfurization capacity of equipment directly affects the accuracy of the upper limit of sulfur constraint. If the upper limit of sulfur constraint obtained by inverse calculation is too large, it may lead to environmental protection exceeding the standard; on the contrary, it may lead to too much low sulfur coal blending, which affects the economy of mixed coal blending. Therefore, how to accurately set the upper limit of sulfur constraint is a difficult problem, such asFigure 5 The application further proposes a sulfur constraint boundary feedforward compensation based on an improved stochastic configuration network (ISCN) method.

[0044] The sulfur constraint boundary adjustment amount is related to the load and the sulfur constraint boundary set value in the blending model, and an improved stochastic configuration network (ISCN) is used to establish a two-input-single-output sulfur constraint boundary compensation nonlinear model: (15); In the formula, u is in the real number domain The two input variables of the compensation nonlinear improved stochastic configuration network model are y, which is in the real number domain The single output variable of the compensation nonlinear model is is the input-output function relationship established by the I SCN, is the load, is the sulfur constraint boundary, is the sulfur constraint boundary feedforward compensation amount (sulfur constraint boundary set value deviation).

[0045] The historical blending case database used to train the I SCN network is given Group data , wherein the input , the input of the kth training data and the corresponding output , (k represents the kth training case data in the historical blending case database here and hereinafter). In the network training process, if the I SCN network hidden layer has constructed L-1 (L is the number of hidden nodes, and the initial value is 1) hidden layer nodes, the output of the L-1th hidden layer node can be expressed as: (16); In the formula, is the output of the L-1th hidden layer node, is a hidden layer activation function, represents the output of the L-1th hidden layer node when the kth data input, and are respectively the input weight vector and the bias of the L-1th hidden node, and at this time, the output of the I SCN network based on the training data input is: (17); In the formula, The output predicted by the ISCN network with L-1 hidden nodes according to the input of the training data set, and are the input weight vector and bias of the lth hidden node, respectively, is the output weight of the lth hidden node, which can be calculated by the following formula: (18) ; In the formula, is the output of the lth hidden layer node of the ISCN network, and the network residual error at this time is calculated is: (19) ; The initial residual error of the network , the allowable error limit of the network is set , if the residual error of the ISCN network , the lth hidden node is added to the hidden layer, and the new node parameters are The IPOA algorithm is configured, and the steps are as follows: Step S1. Set the objective function and constraint of IPOA for optimizing the new lth hidden node parameters of the ISCN network as follows: (20) ; In the formula, is the maximum of the objective function, and 0 <1, non-negative sequence , , L is the number of hidden nodes of the ISCN network (initially 1), is the error of the ISCN network with L-1 hidden nodes in the hidden layer, is the output of the lth hidden node; Step S2. Initialize the population: (21) ; (22) ; Let be the position corresponding to the lth pelican optimization parameter vector (each dimension represents an optimization parameter of a new hidden node), be the position of the lth pelican in the lth dimension, , is the number of pelicans in the population, and the position matrix of the pelican population is , and the objective function matrix of the pelican population is , is the objective function value corresponding to the position of the lth pelican individual, It is a chaotic sequence. A random number in the interval [0,1]. and These represent the upper and lower boundaries of the domain of the newly added hidden node parameters. A constant less than 1;

[0046] Step S3. Randomly generate the prey target locations for the pelican population to search within the defined space. The pelican population moves towards its prey location, and the location update is shown in the following formula: (twenty three); In the formula, For the prey in the The position of the dimension For the first After the pelican moved toward its prey, it was in the first... Wei's new position The objective function value corresponding to the prey's location. A random number within the range [0,1] For a random number equal to 1 or 2, update the position of the parameter vector represented by each pelican in the population using the following formula: (twenty four); In the formula, For the Pelicans' new position The objective function value.

[0047] Step S4. Local search using IPOA to find the new position of the parameter vector: (25); In the formula, For the first local search Only the Pelicans in The new position of the dimension parameter, The average value of the population objective function. The optimal position for the current population is at the th position. The coordinates are dimensional, and Cauchy(0,1) represents the standard Cauchy distribution. This represents the current iteration number. The maximum number of iterations, The search space radius is used; after the local search is completed, the position corresponding to the individual optimization parameters is updated in the same way as formula (24).

[0048] Step S5. Set the threshold for changes in the objective function. If the change in the objective function value before updating the optimal position of the IPOA population is less than Then, the optimal position of the population (the optimal parameters of the newly added Lth hidden node) is found. , the Lth implicit node parameter configuration of the ISCN network is completed; Step S6. Calculate the ISCN network output and network residual after adding the node: (26); (27); If the ISCN network residual , the ISCN network training is completed, otherwise the L+1th implicit node is added to the implicit layer, and steps S1-S6 are repeated until the ISCN network residual ; After the ISCN network training is completed, the test data is input , and the sulfur constraint compensation predicted by the ISCN network is: (28); In the formula, is the compensation amount of the sulfur constraint upper bound.

[0049] Step S6: Based on the real-time coal blending data each time, the historical coal blending case database is dynamically and rolling updated by using the iterative learning technology, so as to ensure the accuracy of the modeling case data; To realize the dynamic and accurate setting of the sulfur constraint bound, the application introduces an iterative learning compensation mechanism to form a closed-loop optimization, as shown in Figure 6 . The detection unit detects the actual sulfur dioxide emission concentration after the desulfurization of the boiler mixed coal combustion , and stores the current and previous period detected data into the binary sliding memory unit, and quantifies the deviation from the expected emission concentration of the environmental protection standard (the national standard limit is 35 mg / m 3 ). According to the SO2 emission concentration deviation, the sulfur constraint bound in the coupled coal blending optimization model is iteratively learned and compensated: (29); In the formula, and are the sulfur constraint upper bounds of the coupled coal blending optimization model in the dth and (d+1)th iterations, the initial value is set according to formula (5), is the sulfur constraint upper bound feedforward compensation amount predicted by the ISCN model in the (d+1)th iteration, is the deviation between the actual sulfur dioxide emission concentration after the desulfurization and the environmental protection standard emission concentration in the dth iteration, is the sulfur dioxide emission deviation in the dth iteration, is the actual sulfur dioxide emission concentration after the desulfurization of the mixed coal combustion in the dth iteration, Let d be the flue gas emission flow rate in the d-th iteration. Let be the flue gas density in the d-th iteration. For coal sulfur conversion efficiency, The maximum coal feeding rate of the k-th coal mill in the i-th planned load segment; The historical firing case database of ISCN is updated using the results of the sulfur constraint upper bound iteration, and the ISCN network parameters are retrained to improve the model prediction accuracy. The steps for updating the historical firing case database are as follows: Step C1. Convert the training data in the historical firing case database into triplet case format. Suppose that the case database already stores W source cases, where the w-th source case... The triplet form is as follows: (30); In the formula, Input for the w-th case, These represent the load and sulfur constraint upper bound for the w-th case, respectively. Output the deviation of the sulfur constraint upper bound setting for the wth case. The evaluation metric for the w-th case (the smaller the evaluation metric, the better the case): (31); In the formula, This is the upper bound of the sulfur constraint in the w-th case. The deviation of the upper limit setting value for sulfur constraint in the w-th case.

[0050] Step C2. Case retrieval, for a new case The similarity between a case and a source case in the case library is retrieved using the Euclidean distance shown in the following formula: (32); In the formula, For new cases To the source case European distance, For new cases With source case Based on the overall similarity, the source cases in the case library are traversed according to the above formula to retrieve cases similar to the new cases. With the highest similarity Source Case ,in, In order to new cases The evaluation metric with the highest similarity; Step C3. Evaluate the overall similarity of the new cases; set the overall similarity evaluation threshold. ,like If yes, proceed to step C4; otherwise, proceed to step C6. Step C4. Evaluate new case single dimension similarity; set load evaluation threshold and sulfur constraint boundary threshold If both load similarity and sulfur constraint similarity are within threshold, go to Step C5, otherwise go to Step C6. Step C5. Case evaluation and update, if new case evaluation index , discard this case; otherwise new case assign replacement source case , Step C6. If new case evaluation index , is evaluation threshold, store new case in historical blending case database.

[0051] The uncertainties of coal sulfur conversion efficiency, desulfurization efficiency of desulfurization equipment, and seasonal variation of seawater desulfurization capacity of equipment directly affect the accuracy of the upper bound of sulfur constraint. In order to further reduce the cost of coal blending, this example aims to solve the problem of inaccurate sulfur constraint boundary setting in the coupled blending model under different loads. Based on the 50 sets of experimental data in Table 6, an ISCN sulfur constraint boundary feedforward compensation prediction model is trained. Subsequently, the model is used to predict the compensation amount of the sulfur constraint boundary for the blending conditions in Table 3. According to the prediction results, the sulfur constraint boundary of the blending model is compensated, and finally the sulfur constraint boundary adjustment of the blending model is completed.

[0052] Based on the conditions shown in Table 4, the inputs of the sulfur constraint boundary compensation model under deep regulation, medium-high, and high load segments are = [90, 0.68], = [180, 0.45], = [290, 0.45]. According to the established ISCN model, the sulfur constraint boundary compensation amounts of the three load segments are = 0.21%, = 0.17%, = 0.14%. Based on the sulfur constraint compensation iterative learning process, the sulfur constraint boundary in the blending optimization model is adjusted, and the sulfur dioxide emission concentration deviation ( ) is set as the iterative termination condition. Table 7 shows the changes of various index parameters during the iteration process. Based on the iterative process in Table 7, the sulfur constraint boundaries of the medium-high load and high load segments of the blending model are set to = 0.17% and = 0.14% respectively at the end of the iterative learning.=0.73%. At this point, the sulfur constraint limits for the deep adjustment, medium-high and high load sections were adjusted from 0.68%, 0.45% and 0.45% to 1.09%, 0.73% and 0.73%, respectively. The comparison between the sulfur content after iterative adjustment and the model coal blending scheme without sulfur adjustment is shown in Table 8.

[0053] Table 6. Training Sample Data Case Library; ; Table 7. Iterative Adjustment of Coal Blending Based on Sulfur Constraint Boundaries; ; Table 8 Comparison of coal blending and sulfur regulation schemes before and after model constraints; ; The coal blending results show that, after adjusting the sulfur constraint boundary to model the blending process, the sulfur content of the coal blending schemes for the three load segments (deep adjustment, medium-high, and high) is controlled between 0.59% and 0.73%, meeting the environmental constraints of the blending model. The actual emission concentrations after desulfurization of the blended coal combustion under the three load segments are as follows. They are 30.17 respectively. 30.68 30.24 (National standard limit 35) All meet environmental protection standards and The sulfur content of the blended coal in the deep-load, medium-high, and high-load sections increased by 0.22%, 0.23%, and 0.28% respectively compared to before model optimization. Simultaneously, the blending schemes increased the proportion of high-sulfur, low-priced coal in the coal yard; for example, the blending ratio of high-sulfur coal (Indonesian lignite 2D3) in coal bunker #1 increased from 41% to 53%; and coal bunker #3 was switched to low-priced, high-sulfur coal (3D3), increasing the sulfur content from 0.37% to 0.88%. Correspondingly, the blended coal prices under the three load sections decreased by RMB 64.60, RMB 86.99, and RMB 84.43 respectively, resulting in an average reduction of 8.79% in blending costs. The results indicate that the proposed method of adjusting the sulfur constraint boundary of the blending model under segmented loads significantly reduces the blending cost under various operating conditions, greatly improving the economic efficiency of thermal power plants while meeting environmental protection requirements.

[0054] Step S7: Develop a coal blending optimization software system using a B / S architecture, and realize automatic coal blending based on a coupled coal blending optimization model.

[0055] This embodiment uses ADO.NET to develop a relational database interface and establishes an SQL Server relational database table for coal types in the coal yard, including coal type name, calorific value, sulfur content, moisture, volatile matter, ash content, price, and coal yard number information. Based on the segmented planned load and the coupling relationship between cutting and grinding vectors, a coal blending optimization model running program is developed in C#, including a static and dynamic blending optimization model. The static blending optimization model uses only the inversion calculation method to calculate the upper limit of the sulfur constraint for each coal blending operation. The dynamic blending optimization model uses an inversion calculation method, an improved random configuration network, and an iterative learning integrated technology to dynamically calculate the sulfur constraint boundary throughout the entire cycle for each coal blending operation. A software module is developed in C# to solve the static and dynamic blending optimization models and calculate the coal blending scheme. The system front-end interface is developed using ASP.NET to display multiple coal blending schemes (the optimal scheme and schemes with costs gradually increasing by 10 yuan), such as... Figure 7 As shown, the coal blending personnel select a scheme based on actual needs; based on the selected coal blending scheme, the coal type and proportion s of each coal bunker are transmitted to the material reclaimer control system, coordinating the two material reclaimers to feed coal into the coal bunker; real-time load data of the SIS system is read through the real-time database interface, and the coal mill is automatically switched according to the coal blending optimization model; within each load segment, the coal mill control system adjusts the output of the coal mill to adapt to the load change requirements.

Claims

1. A coal blending and burning optimization method for segment-oriented planned load and switching mill vector coupling, characterized in that, include: Establish the logical business coupling relationship between segmented planned load, switching mill vector, and coal blending; determine the objective function of the coupled combustion optimization model between segmented planned load and switching mill vector according to the optimal coal blending cost. For the coupled combustion optimization model, calorific value constraints, environmental protection constraints, safety constraints, and coal type ratio constraints are introduced. The established coupled firing optimization model is transformed into a standard form and a canonical form; For the canonical form of the coupled coal blending optimization model, a two-stage simplex solution is used to obtain the coal blending scheme; The upper limit of sulfur constraint is calculated using an inversion method. Based on a historical calcination case database, an improved random configuration network is established to perform feedforward compensation on the sulfur constraint boundary. Based on each real-time coal blending data, iterative learning technology is used to dynamically and continuously update the historical coal blending case database to ensure the accuracy of the modeling case data.

2. The method for coal blending combustion optimization of segment-oriented plan load and switching mill vector coupling according to claim 1, characterized in that, The logical business coupling relationship between segmented planned load, switching mill vectors, and coal blending is established, and the objective function of the coupled blending optimization model is determined according to the optimal blending cost, as shown in the following formula: (1)； In the formula, To minimize the cost of blended coal, For the cutting and grinding vector coefficients, and , This is a segmented load coal distribution function. For the first The number of coal mills added to each planned load segment, For the first The planned load segment The price of each type of coal For the first The first load segment The proportion of blended coal types for each coal type For the first A vector representing the proportion of coal types blended in each planned load segment. To adjust the proportion of coal types in three stages: deep load, medium-high load, and high load. The vector formed by these vectors.

3. The coal blending optimization method for segmented planned load and switching mill vector coupling according to claim 2, characterized in that, The calorific value constraints, environmental protection constraints, safety constraints, and coal type ratio constraints of the coupled combustion optimization model are as follows: The aforementioned calorific value constraint means that the calorific value of the coal supplied to each load segment of the unit must be greater than or equal to the lower limit of the calorific value determined by the boiler load of each load segment, as shown in the following formula: （2）； In the formula, This is the weighted average of the calorific value of the blended coal in the unit. This is the coal blending function for segmented load calorific value. For the first The planned load segment The calorific value of each type of coal For the first The number of coal mills operating in each planned load segment. This is the lower limit function vector of calorific value corresponding to segmented loads. For the first The lower limit of calorific value for each planned load segment is shown in the following formula: （3）； In the formula, For the first The planned load of each planned load segment For the first The first planned load segment corresponding to the planned load. Maximum coal feed rate of the coal mill. For the first Unit coal consumption corresponding to the planned load of each planned load segment, unit coal consumption With unit efficiency and load related, This represents the total number of coal mills operating at deep load, medium-high load, or high load. The environmental constraints, namely sulfur content constraints, are shown in the following formula: （4）； In the formula, This represents the weighted average sulfur content of the coal blends under different planned load ranges for the coal mill unit, where the sum of the proportions of each blend is 1. This represents the increased sulfur-coal blending function vector under segmented load conditions. For the first The planned load segment The sulfur content of each type of coal, This is the upper limit function vector for sulfur corresponding to the segmented load. For the first The upper limit of desulfurization under each planned load segment is obtained by inverting the capacity of the desulfurization equipment using the following formula: （5）； In the formula, sulfur Converted to sulfur dioxide Conversion efficiency For the desulfurization efficiency of desulfurization equipment, Removes sulfur dioxide from the desulfurization equipment per hour Ability; The sulfur constraint boundary in equation (5) is derived by inversion from the maximum desulfurization capacity of the desulfurization equipment. In order to ensure that environmental protection standards are not exceeded, in practical applications, equation (5) is simplified to the following equation: （6）； The safety constraints are shown in the following formula: （7）； In the formula, For volatiles under segmented load , moisture or ash The weighted average of the values, where the sum of the coal blending ratios is 1. and The increase in volatile matter under segmented load , moisture or ash The coal blending function vector, For the first The planned load segment The volatile matter, moisture, or ash content of each coal type. and These are the upper and lower bound vectors for volatile matter, moisture, or ash content corresponding to the segmented loads; The coal type ratio constraint is shown in the following formula: （8）； （9）； In the formula, The vector combining the maximum coal feed rate of the coal mill increased by the three load sections. and Indicates the first The maximum coal feed rate of one or two additional coal mills in each load section; The proportion coefficient of the selected coal type in the coal bunker in the coupled combustion optimization model is constrained by the following formula: （10）； In the formula, and It is the coal feeding coefficient of the reclaimer. For the first The first load segment The maximum coal feeding rate of the coal mill; the coal blending optimization model requires that the coal blending amount of each coal type be practically meaningful and conducive to the coal feeding of the reclaimer, therefore the coal feeding coefficient of the reclaimer is often taken as... , ; The constraints of the coupled combustion optimization model also include coal quality characteristic index constraints, coal type priority constraints at the coal yard, and coal type constraints based on the structure and parameter characteristics of the coal mill unit.

4. The coal blending optimization method for segmented planned load and switching mill vector coupling according to claim 3, characterized in that, The established coupled combustion optimization model is transformed into standard and canonical forms. For the canonical form of the coupled combustion optimization model, the specific method for obtaining the coal blending scheme using a two-stage simplex solution is as follows: For the inequality constraints in the coupled firing optimization model, after adding relaxation vectors and residual vectors, it is transformed into the following standard form, namely the standard firing optimization model: （11）； In the formula, For generalized vectors, For the proportion vector of coal types in coal blending, For the remaining vector, Let be the relaxation vector. , To constrain the target vector, It is a coefficient matrix; , , Both diagonals are 1 ( ) ( identity matrix, Both the main diagonal lines are -1 ( ) ( Identity matrix; Due to the coefficient matrix Since there is no standard allowable basis, artificial variables need to be introduced to construct the following canonical form: （12）； In the formula, , , , These are artificial variables added to form the initial basis vectors in linear programming. , Let them be the initial basis vectors. Both the main diagonal and the diagonal are 1 ( ) ( Identity matrix; The first stage of the simplex method is initiated to solve equation (12). The optimal solution of equation (12) is calculated through the basis removal iteration of artificial variables. If the basis vectors of the optimal solution do not contain artificial variables, the optimal solution of equation (12) is a basic feasible solution of equation (11). The coefficient vectors of all artificial variables are transferred to In the middle, use In the substitution (11) This yields the canonical form of linear programming equivalent to equation (11): （13）； In the formula, for The coefficient matrix containing the identity matrix after the first stage transformation using the simplex method. for The constrained target vector after the first stage transformation using the simplex method; The second stage of the simplex method is initiated to solve the canonical form of the linear programming equation (13) to obtain the optimal solution of the sintering optimization model; For the coupled coal blending optimization model of the typical segmented planned load and the switching mill vector coupling, the classical two-stage simplex method is used to solve it; if there is a solution, the coal blending scheme is output and the coal blending is completed; if there is no solution, the coal feed rate of the pulverizer corresponding to the deep adjustment load is adjusted. The coal feed rate of the pulverizer corresponding to the deep adjustment load is gradually reduced by setting a step size, and the rate adjustment is as follows: （14）； In the formula, To adjust the step size of the coal feed rate, For the number of iterations, For the first phase of load The first coal mill The coal feed rate in the next iteration; At each iteration's coal feed rate, let... Then solve equation (11) again until an optimal solution is found, and stop iterating.

5. The coal blending optimization method for segmented planned load and switching mill vector coupling according to claim 4, characterized in that, Based on the calculation of the upper bound of sulfur constraint using the inversion method, and using a historical calcination case database, the specific method for feedforward compensation of the sulfur constraint boundary using an improved stochastic configuration network is as follows: The sulfur constraint adjustment is related to the sulfur constraint setpoint in the load and combustion model. An improved stochastic configuration network is used to establish a two-input-single-output nonlinear model for sulfur constraint compensation: (15)； In the formula, u represents the number in the real number field. The two input variables of the compensated nonlinear improved stochastic allocation network model are y in the real number domain. Compensation for a single output variable in a nonlinear model The input-output function relationship established for the improved stochastic configuration network. For load, Sulfur confinement boundary, This is the feedforward compensation amount for the sulfur constraint boundary; Given a historical cooking case database for training an improved stochastic configuration network Group data , where input The input of the kth training data And the corresponding output , During network training, assuming the improved randomly configured network has constructed L-1 hidden layer nodes, then the output of the L-1th hidden layer node is... It can be represented as: (16)； In the formula, This is the output of the (L-1)th hidden layer node. Here is the activation function for the hidden layer. This represents the output corresponding to the (L-1)th hidden layer node when the k-th data input is received. and Let be the input weight vector and bias of the (L-1)th hidden node, respectively. Based on the training data input, the output of the improved stochastically configured network is: (17)； In the formula, The output is predicted by an improved randomly configured network with L-1 hidden nodes based on the training dataset as input. and Let be the input weight vector and bias of the l-th hidden node, respectively. The output weight of the l-th hidden node is calculated by the following formula: (18)； In the formula, To calculate the network residual at the output of the l-th hidden layer node in the improved randomly configured network, we need to determine the network residual at this point. for: (19)； Network initial residual Set network tolerance error limit If the improved random configuration network residual Then the hidden layer adds the Lth hidden node, and the parameters of the newly added node are... The IPOA algorithm is used for configuration.

6. The coal blending optimization method for segmented planned load and switching mill vector coupling according to claim 5, characterized in that, Configure new node parameters using the IPOA algorithm The specific process is as follows: Step S1. Define the objective function and constraints for IPOA to optimize the parameters of the newly added Lth hidden node in the improved stochastic network as follows: (20)； In the formula, To maximize the objective function, 0 < <1, non-negative sequence , L represents the number of hidden nodes in the improved randomized network, initially set to 1. The error is for an improved randomly configured network with L-1 hidden nodes in the hidden layer. This is the output of the Lth hidden node; Step S2. Initialize the population: (21)； (22)； set up For the first The pelican seeks the position corresponding to the optimal parameter vector. For the first Only the Pelicans in The position of the dimension , Let be the population size of the pelicans, and the location matrix of the pelican population is as follows: The scalar function matrix of the pelican population is , For the first The objective function value corresponding to the position of each individual pelican. It is a chaotic sequence. A random number in the interval [0,1]. and These represent the upper and lower boundaries of the domain of the newly added hidden node parameters. A constant less than 1; Step S3. Randomly generate the prey target locations for the pelican population to search within the defined space. The pelican population moves towards its prey location, and the location update is shown in the following formula: (23)； In the formula, For the prey in the The position of the dimension For the first After the pelican moved toward its prey, it was in the first... Wei's new position The objective function value corresponding to the prey's location. A random number within the range [0,1] For a random number equal to 1 or 2, update the position of the parameter vector represented by each pelican in the population using the following formula: (24)； In the formula, For the Pelicans' new position The objective function value; Step S4. Local search using IPOA to find the new position of the parameter vector: (25)； In the formula, For the first local search Only the Pelicans in The new position of the dimension parameter, The average value of the population objective function. The optimal position for the current population is at the th position. The coordinates are dimensional, and Cauchy(0,1) represents the standard Cauchy distribution. This represents the current iteration number. The maximum number of iterations, Let be the search space radius; after the local search is completed, update the position corresponding to the individual optimization parameters based on equation (24); Step S5. Set the threshold for changes in the objective function. If the change in the objective function value before updating the optimal position of the IPOA population is less than Then the optimal position of the population is found. The configuration of the parameters of the newly added Lth hidden node in the improved random configuration network has been completed; Step S6. Calculate the improved random configuration network output and network residuals after adding nodes: (26)； (27)； If the improved random configuration network residual If the training is successful, the improved stochastic network training is complete; otherwise, add the (L+1)th hidden node to the hidden layer and repeat steps S1-S6 until the improved stochastic network residual is obtained. ; After the improved randomized network has been trained, given the test data input... The sulfur constraint compensation predicted by the improved random configuration network is as follows: (28)； In the formula, This is the compensation amount for the desired upper limit of the sulfur constraint.

7. The coal blending optimization method for segmented planned load and switching mill vector coupling according to claim 6, characterized in that, The specific method for dynamically updating the historical coal blending case database based on each real-time coal blending data using iterative learning technology is as follows: Real-time monitoring of actual sulfur dioxide emission concentration after desulfurization of mixed coal combustion in boilers The emission concentration expected by environmental standards Deviation quantification is performed to correct the upper bound of the sulfur constraint, and the upper bound of the sulfur constraint is compensated through iterative learning: (29)； In the formula, and The sulfur constraint upper bounds for the coupled sintering optimization model in the d-th and d+1-th iterations are respectively set according to equation (5). This represents the feedforward compensation amount for the sulfur constraint upper bound predicted by the improved random configuration network in the (d+1)th iteration. The deviation between the actual sulfur dioxide emission concentration after desulfurization and the environmental standard emission concentration in the d-th iteration. The sulfur dioxide emission deviation in the d-th iteration The deviation of the sulfur constraint setpoint obtained from the inversion calculation This represents the actual sulfur dioxide emission concentration after desulfurization during the d-th iteration of coal-fired combustion. Let d be the flue gas emission flow rate in the d-th iteration. Let be the flue gas density in the d-th iteration. For coal sulfur conversion efficiency, The maximum coal feeding rate of the k-th coal mill in the i-th planned load segment; The historical calcination case database of the improved stochastic configuration network is updated using the results of the sulfur constraint upper bound iteration, and the parameters of the improved stochastic configuration network are retrained to improve the model prediction accuracy.

8. The coal blending optimization method for segmented planned load and switching mill vector coupling according to claim 7, characterized in that, The historical food preparation case database update process is as follows: Step C1. Convert the training data in the historical firing case database into triplet case format. Suppose that the case database already stores W source cases, where the w-th source case... The triplet form is as follows: (30)； In the formula, Input for the w-th case. These represent the load and sulfur constraint upper bound for the w-th case, respectively. Output for the w-th case. The evaluation metric for the w-th case is: (31)； In the formula, This is the upper bound of the sulfur constraint in the w-th case. The deviation of the upper limit setting value for the sulfur constraint in the w-th case; Step C2. Case retrieval, for a new case The similarity between a case and a source case in the case library is retrieved using the Euclidean distance shown in the following formula: (32)； In the formula, For new cases To the source case European distance, For new cases With source case Based on the overall similarity, the source cases in the case library are traversed according to the above formula to retrieve cases similar to the new cases. Maximum similarity Source Case ,in, Each is a new case Inputs, outputs, and evaluation metrics of the source case with the highest similarity; Step C3. Evaluate the overall similarity of the new cases; set the overall similarity evaluation threshold. ,like If yes, proceed to step C4; otherwise, proceed to step C6. Step C4. Evaluate the one-dimensional similarity of the new cases; set the load evaluation threshold. and sulfur constraint threshold If both the load similarity and the sulfur constraint boundary similarity are within the threshold range, proceed to step C5; otherwise, proceed to step C6. Step C5. Case Evaluation and Updates, if new cases... Evaluation indicators Discard this case; otherwise, create a new case. Assignment substitution source case , ; Step C6. Evaluation metrics for the new case , To evaluate the threshold, new cases will be used. Store in the historical firing case database.

9. The coal blending optimization method for segmented planned load and switching mill vector coupling according to claim 8, characterized in that, The method also employs a B / S architecture to develop a coal blending optimization software system, which automatically blends coal based on a coupled coal blending optimization model. The specific method is as follows: A relational database interface was developed based on ADO.NET, and SQL Server relational database tables for coal types in the coal yard were established, including information such as coal type name, calorific value, sulfur content, moisture content, volatile matter, ash content, price, and coal yard number. Based on the segmented planned load and the coupling relationship between cutting and grinding vectors, a coupled coal blending optimization model running program was developed in C#, including a static and dynamic coal blending optimization model. The static coal blending optimization model uses only the inversion calculation method to calculate the upper limit of sulfur constraint for each coal blending, while the dynamic coal blending optimization model uses the inversion calculation method, an improved random configuration network, and iterative learning integrated technology to dynamically calculate the sulfur constraint boundary throughout the entire cycle for each coal blending. A software module developed in C# is used to solve static and dynamic coal blending optimization models to calculate coal blending schemes. The system's front-end interface, developed using ASP.NET, displays multiple coal blending schemes, allowing coal blending personnel to select a scheme based on actual needs. Based on the selected coal blending scheme, the coal type and proportion of each coal bunker are transmitted to the reclaimer control system, coordinating the two reclaimers to feed coal into the bunkers. Real-time load data from the SIS system is read through a real-time database interface, and the coal mills are automatically switched according to the coal blending optimization model. Within each load segment, the coal mill control system adjusts the coal mill output to adapt to load changes.

Images