A coal-fired unit combustion optimization control method for a thermal power plant

Abstract

The application discloses a coal-fired unit combustion optimization control method for a thermal power plant, which comprises the following steps: obtaining a historical data set on a coal mill; obtaining real-time moisture of coal quality according to a heat balance principle of the coal mill and the historical data set; establishing a coal quality calorific value regression model and a coal quality volatile matter regression model based on a BP neural network and using the historical data set; calculating the ash content of the coal quality by taking the output of the coal quality calorific value regression model and the moisture calculation result as inputs; obtaining the volatile matter of the coal quality by using the coal quality volatile matter regression model; obtaining the real-time moisture of the coal quality, the coal quality calorific value, the coal quality ash content and the coal quality volatile matter as the information of the coal fed into the coal-fired unit by performing steady-state processing on the historical data set of the coal-fired unit; establishing a coal quality prediction model, predicting each parameter of the coal-fired unit, calculating an evaluation function by using the prediction results of the parameters, taking the historical optimal value of the current working condition as an initial population, performing genetic algorithm optimization to obtain an optimal evaluation function result and an optimal working condition, and outputting the result for real-time control. The application can improve the control precision.

Description

Technical Field

[0001] This invention relates to the field of thermal power generation technology, specifically to a method for optimizing combustion control of coal-fired units in thermal power plants. Background Technology

[0002] With the rapid development of large-capacity, high-parameter coal-fired power generating units and increasingly stringent air pollutant emission standards, as well as the rapid breakthroughs in the application of advanced measurement and control technologies, there is an urgent need for in-depth development of combustion measurement and closed-loop optimization control technologies for power plant boilers. In-furnace combustion control of coal-fired boilers is quite complex, especially in terms of adaptability to different coal types. Boiler characteristics vary significantly when burning different coal types, such as the calorific value of the coal, ignition distance, in-furnace temperature distribution characteristics, and coking characteristics. These factors all affect the rationality of control outputs such as in-furnace air distribution, excess air coefficient, and desuperheating water distribution, thus impacting the unit's combustion control level and further affecting its operational economy and pollutant emission rates. Therefore, timely and accurate identification of the coal types burned by each burner and real-time combustion optimization based on the identification results are crucial for ensuring stable and economical combustion of coal-fired boilers under conditions of deep peak shaving and multiple coal types.

[0003] For intelligent combustion optimization control systems, existing technologies have proposed the following methods: Patent application CN116241905A discloses a combustion optimization control system based on multi-objective optimization. Its principle is to establish a boiler combustion optimization model based on a large amount of experimental data, and use an intelligent particle swarm optimization algorithm to obtain the optimal input variable data in the boiler combustion model by calculating the objective function, thereby optimizing the combustion control system. However, due to the complexity and variability of boiler coal types, this method cannot achieve refined control based on different coal quality information. Patent application CN116753538A discloses an intelligent combustion optimization control system and method for coal-fired power plants, including a preheating temperature determination system, a pulverized coal optimization power generation preheating system, and a boiler coal circulation system. The preheating temperature determination system is used to identify and distinguish the type of pulverized coal to be heated near the boiler of the coal-fired power plant, match stored data, and determine the preheating temperature of the coal type. The pulverized coal optimization power generation preheating system sets the temperature according to the preheating temperature determination system. This method controls the waste heat system's set temperature based on the coal type's preheating temperature to achieve combustion optimization control. Patent application CN116123561A discloses a method and system for optimizing temperature control in boiler combustion. It acquires images of the target boiler using an image acquisition device, constructs a three-dimensional fitting model based on the image acquisition results and design dimensions, sets temperature measurement points on the furnace wall using the three-dimensional fitting model, deploys an infrared acoustic temperature measurement device, and acquires temperature data using the infrared acoustic temperature measurement device to obtain a set of temperature data and thus obtain real-time control parameters for the target boiler. This patent requires additional image acquisition equipment to obtain the real-time temperature field of the furnace, resulting in high on-site equipment installation costs. Furthermore, due to the harsh working environment of the boiler furnace (high temperature, large airflow disturbance), the equipment reliability is not high. Patent application CN115753886A discloses a method for real-time calculation of coal quality in a power plant boiler. It uses a combination of positive and negative balance to calculate the total calorific value. Based on the linear relationship between the received moisture, ash content, and calorific value of the coal, it achieves the measurement of the coal quality. However, the received moisture, ash content, and calorific value vary significantly with different coal types, making it impossible to obtain accurate component relationships through linear relationships. Summary of the Invention

[0004] The technical problem to be solved by this invention is to address the issue of low accuracy in controlling the combustion of coal fed into the furnace in coal-fired power units.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A method for combustion optimization control of coal-fired units in a thermal power plant, comprising:

[0007] S10, acquire a historical dataset including coal records and corresponding coal quality information on the coal mill, as well as historical operating data of the coal mill;

[0008] S20, based on the thermal balance principle of the coal mill and combined with historical data sets, obtains the real-time moisture content of the coal.

[0009] S30, based on a multi-layer feedforward neural network, uses historical datasets to establish a coal calorific value regression model to obtain the coal calorific value;

[0010] S40, coal ash content is obtained based on coal calorific value and real-time coal moisture content;

[0011] S50, based on a multi-layer feedforward neural network, uses historical datasets to establish a coal volatile matter regression model to obtain coal volatile matter;

[0012] S60, acquire the historical dataset of the coal-fired unit and perform steady-state processing. Combine the steady-state processed data with real-time coal moisture, coal calorific value, coal ash content and coal volatile matter as the coal quality information for the furnace.

[0013] S70 uses the coal quality information of the furnace feed and establishes a coal quality prediction model based on a multi-layer feedforward neural network to predict the NOx concentration at the SCR inlet, the desuperheating water on the A and B sides of the reheater, the flue gas temperature at the air preheater inlet, the water-cooled wall, the maximum wall temperature of the last stage reheater, the steam temperature of the last stage superheater, and the steam temperature at the reheater outlet.

[0014] S80 uses the NOx concentration at the SCR inlet, the desuperheating water on the A and B sides of the reheater, the flue gas temperature at the air preheater inlet, the water-cooled wall, the last stage superheater, the highest wall temperature of the last stage reheater, and the steam temperature at the reheater outlet as evaluation functions, and sets the weights of each parameter.

[0015] S90 compares the predicted data of each parameter with the historical best data of each parameter to determine the comprehensive evaluation index of the current working condition of each parameter. The genetic optimization algorithm is used to optimize the operating oxygen content, secondary air, and the opening of the burnout damper. During the genetic algorithm optimization, the historical best working condition of the current load range and coal feeding method is selected from the historical data of each parameter as the initial population of the genetic algorithm for optimization calculation to obtain the optimal optimized operating parameters of the current working condition and output them for control.

[0016] In one embodiment of the present invention, the real-time moisture content of coal is obtained using the following formula:

[0017]

[0018] in,

[0019] In the formula, M m The coal feed rate of the coal feeder is expressed in tons per hour (t / h); M ft1 is the primary air volume at the inlet of the coal mill, t / h; t2 is the primary air temperature at the inlet of the coal mill, °C; t3 is the temperature of the air-powder mixture at the outlet of the coal mill, °C; K is the grindability coefficient; W is the power of the coal mill, KW; C1 is the specific heat of the desiccant at the inlet of the coal mill, KJ / kg* °C; C2 is the specific heat of the desiccant at the outlet of the coal mill, KJ / kg* °C; C lk Specific heat of cold air, kJ / kg*℃; M ar For coal receiving basis moisture, C rd Specific heat of coal dryer, KJ / kg*℃; t lk The temperature of the leaking cold air, in °C; t r ΔM is the temperature at which the coal enters the system, in °C; ΔM is the heat q consumed in evaporating moisture. z And the amount of water evaporated per kg of raw coal in the coal mill, K lf R is the air leakage coefficient of the coal mill. 90 Q5 represents the fineness of pulverized coal, in %; Q5 represents the heat loss of the subsystem, in KW; K nm M is the coefficient by which the input power of the coal mill is converted into heat. ar To receive the base moisture content; M mf This refers to the moisture content of pulverized coal.

[0020] In one embodiment of the present invention, the method for constructing a coal calorific value regression model includes the following steps:

[0021] S31, Establish the input matrix x [1] = [N, M], where N represents the number of historical datasets and M represents the number of input variables;

[0022] S32, Random initialization: Weight matrix w [1] = [M,q], error b [1] =[1,q], a [1] =x [1] w [1] +b [1] Where q is the number of nodes in the first hidden layer; a [1] This is the matrix representing the computation results of the first layer of the neural network;

[0023] S33, Obtain the output z of the first hidden layer. [1] = sigmod(a [1] );

[0024] S34, the output z of the first hidden layer [1] As the input parameters for the second hidden layer, and so on, the output z of the f-th hidden layer is obtained. [4] The output variable is 1, and the output z of the f-th hidden layer is used. [4] As an input variable, repeat step S33 to calculate the predicted value of the target variable output by the output layer. Calculate the predicted value using the L2 norm loss function. The loss relative to the actual value of the target variable y;

[0025] S35, update the hidden layer w according to the L2 norm loss function S. [i] b [i] ;

[0026] S36. After completing the update, repeat S33 to S35, iterating n times, and calculate the predicted value. The root mean square error (MSE) between the actual value of the target variable y and the target variable.

[0027] In one embodiment of the present invention, the method for constructing the coal volatile matter regression model and the coal quality prediction model is the same as the method for constructing the coal calorific value regression model.

[0028] In one embodiment of the present invention, the coal ash content is obtained by the following formula:

[0029]

[0030] Among them, A ar The ash content of the coal received for furnace operation is %, Q net,ar The lower heating value of the coal received into the furnace is given in kJ / kg and M. ar The moisture content of the coal received for furnace operation is %.

[0031] In one embodiment of the present invention, the historical dataset of the coal-fired unit is subjected to steady-state processing, including: determining whether the absolute value of the difference in coal feed rate of each pulverizer of the two operating conditions within one minute is greater than 10 t / h, or whether the change in the opening degree of each secondary air is greater than 20, or whether the change in the opening degree of the electric regulating damper at the inlet of each secondary air box is greater than 10, or whether the change in the opening degree of each SOFA burnout air is greater than 20, or whether the deviation of the speed setpoint of each flue gas recirculation fan inverter is greater than 10; if any of the above conditions are met, the operating condition is considered to be in a fluctuating state, and data within 15 minutes of the operating condition is removed.

[0032] In one embodiment of the present invention, the single parameter evaluation value = |parameter prediction value - historical best value| / historical best value is used as the single parameter evaluation value.

[0033] In one embodiment of the present invention, all evaluation parameters are set to full assist. Where, p j For a single-parameter evaluation value, h j Here, j represents the number of parameters, and J represents the last parameter.

[0034] In one embodiment of the present invention, obtaining the optimal operating parameters for the current working condition and outputting them for control includes:

[0035] S91 provides numerical coding for operating oxygen content, secondary air volume, and the opening degree of each burnout damper.

[0036] S92, Establish the fitness function based on the overall evaluation value;

[0037] S93, Establish the initial population: Select the historical best operating conditions with different load ranges and different coal mill combinations from the coal quality information of the furnace as the initial population of the genetic algorithm;

[0038] S94, Select Individual Replication: In the initial working condition, replicate and evolve the individuals ranked first in a certain proportion, and discard the individuals ranked last.

[0039] S95, Crossover Mutation: Chromosomes A and B cross over at point c, and new individuals A1 and B1 are obtained through a genetic algorithm;

[0040] S96, Individual Evaluation: Calculate the fitness of the new individual to determine whether the new working condition generated after genetic iteration is better than the previous working condition. If the fitness is met, proceed to the next step. If not, repeat step S94 and perform a second round of iteration until the design requirements are met or the maximum number of iterations is reached.

[0041] S97, Output Results: Outputs optimized operating oxygen levels, secondary air damper, and burnout damper opening; realizes real-time combustion control of the boiler.

[0042] In one embodiment of the present invention, the fitness function is obtained by the following formula:

[0043] In the formula, F is the fitness function and E is the total evaluation value.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention proposes a combustion optimization control method for coal-fired power units based on real-time monitoring of coal quality entering the furnace without adding measuring points or making technical modifications. Real-time monitoring of coal quality entering the furnace for each coal mill can guide combustion optimization control. The combustion optimization search method based on historical data is closer to actual operation and saves a lot of calculation time. Attached Figure Description

[0045] Figure 1 This is a flowchart of a combustion optimization control method for coal-fired units in a thermal power plant, according to an embodiment of the present invention. Detailed Implementation

[0046] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0047] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0048] Please see Figure 1 As shown, this invention provides a combustion optimization control method for coal-fired power plant units, comprising:

[0049] S10, acquire a historical dataset including coal records and corresponding coal quality information on the coal mill, as well as historical operating data of the coal mill.

[0050] In this embodiment, historical operating data of the direct-fired coal mill is collected, including: current (A), coal feed rate (t / h), inlet air pressure (kPa), coal mill differential pressure (kPa), inlet air temperature (°C), outlet air temperature (°C), inlet air volume (t / h), load (MW), air velocity in 1st pulverizer pipe (m / s), air velocity in 2nd pulverizer pipe (m / s), air velocity in 3rd pulverizer pipe (m / s), and air velocity in 4th pulverizer pipe (m / s). The coal feeding records and coal quality information, i.e., coal quality industrial analysis data, are added to the historical operating data of the coal mill according to the corresponding historical time to establish a historical dataset.

[0051] S20 obtains the real-time moisture content of coal based on the heat balance principle of direct-fired coal mill and historical data.

[0052] In this embodiment, based on the principle of energy conservation:

[0053] q gz +q lf +q nm +q r =q z +q jc +q2+q5;

[0054] In the formula,

[0055]

[0056] q z =ΔM(2491+1884t2-419t) r );

[0057]

[0058] In the formula, q gz q lf q nm q r q z q jcq1, q2, and q5 can be understood as different variables, used to simplify the formula. m The coal feed rate of the coal feeder is expressed in tons per hour (t / h); M f t1 is the primary air volume at the inlet of the coal mill, t / h; t2 is the primary air temperature at the inlet of the coal mill, °C; t3 is the temperature of the air-powder mixture at the outlet of the coal mill, °C; K is the grindability coefficient; W is the power of the coal mill, KW; C1 is the specific heat of the desiccant at the inlet of the coal mill, KJ / kg* °C; C2 is the specific heat of the desiccant at the outlet of the coal mill, KJ / kg* °C; C lk Specific heat of cold air, kJ / kg*℃; M ar To obtain the base moisture, C rd Specific heat of coal dryer, KJ / kg*℃; t lk The temperature of the leaking cold air, in °C; t r ΔM is the temperature at which the coal enters the system, in °C; ΔM is the heat q consumed in evaporating moisture. z And the amount of water evaporated per kg of raw coal in the coal mill, K lf R is the air leakage coefficient of the coal mill. 90 Q5 represents the fineness of pulverized coal, in %; Q5 represents the heat loss of the subsystem, in KW.

[0059] in,

[0060] Where: M ar To receive the base moisture content; M mf This refers to the moisture content of pulverized coal.

[0061]

[0062] Moisture content of pulverized coal (M) mf Generally, it is roughly equal to the moisture content of an air dryer, usually lower, around 15%.

[0063] The real-time moisture content of coal can be obtained using the following formula:

[0064]

[0065] In the formula, K nm The coefficient by which the input power of the coal mill is converted into heat.

[0066] S30, based on a multi-layer feedforward neural network, uses historical datasets to establish a coal calorific value regression model to obtain the coal calorific value.

[0067] In one embodiment of the present invention, historical datasets are used as input to a multilayer feedforward neural network, and the multilayer feedforward neural network is used as the prediction output.

[0068] S31, Establish the input matrix x [1] = [N, M], where N represents the number of historical datasets and M represents the number of input variables;

[0069] S32, Random initialization: Weight matrix w [1] = [M,q], error b [1] =[1,q], a [1] =x [1] w [1] +b [1] Where q is the number of nodes in the first hidden layer; a [1] This is the matrix representing the computation results of the first layer of the neural network;

[0070] S33, Obtain the output z of the first hidden layer. [1] = sigmod(a [1] ); and the sigmod function is shown below:

[0071]

[0072] There are a total of f hidden layers.

[0073] S34, the output z of the first hidden layer [1] As the input parameters for the second hidden layer, and so on, the output z of the f-th hidden layer is obtained. [4] The output variable is 1, and the output z of the f-th hidden layer is used. [4] As an input variable, repeat step S33 to calculate the predicted value of the target variable output by the output layer. Calculate the predicted value using the L2 norm loss function. The loss relative to the actual value of the target variable y; where the calculation formula is:

[0074]

[0075] In the formula, S is the L2 norm loss function, n is the number of iterations, and i is the i-th iteration.

[0076] S35, update the hidden layer w according to the L2 norm loss function S. [i] b [i] ;

[0077]

[0078] In the formula, α is the hyperparameter learning rate, which is defined manually.

[0079] S36. After completing the update, repeat S33 to S35, iterating n times, and calculate the predicted value. The root mean square error (MSE) between the actual value of the target variable y and the target variable y.

[0080]

[0081] S37, based on the trained coal calorific value regression model, predicts the calorific value of coal.

[0082] In this embodiment, the network architecture of the multilayer feedforward neural network remains unchanged and is conventional. The only improvement is the addition of L2 norm loss function calculation and prediction value calculation to enhance prediction accuracy. The root mean square error (MSE) between the actual value of the target variable y and the target variable.

[0083] S40, based on the calorific value and real-time moisture content of the coal, obtains the ash content of the coal.

[0084] In this embodiment, the linear relationship between the received ash content of coal and its lower heating value and moisture content, as proposed in the paper "Research on the Relationship between Calorific Value, Moisture, and Ash Content of Coal", is as follows:

[0085]

[0086] Among them, A ar The ash content of the coal received for furnace operation is %, Q net,ar The lower heating value of the coal received into the furnace is given in kJ / kg and M. ar The moisture content of the coal received for furnace operation is %.

[0087] S50, based on a multi-layer feedforward neural network, uses historical datasets to establish a coal volatile matter regression model to obtain coal volatile matter.

[0088] In this embodiment, step S30 is repeated, and the steps for obtaining the model are the same. The input data for the coal volatile matter regression model is also a historical dataset. The difference lies in the output data set. The coal volatile matter regression model predicts and outputs the coal volatile matter.

[0089] S60: Obtain the historical dataset of the coal-fired unit and perform steady-state processing. Combine the steady-state processed data with real-time coal moisture, calorific value, ash content, and volatile matter as the coal quality information for the furnace.

[0090] In this embodiment, the data in the historical dataset of the coal-fired unit includes SCR inlet NOx concentration, desuperheating water on reheater A and B sides, air preheater inlet flue gas temperature, water-cooled wall, last-stage superheater, highest wall temperature of last-stage reheater, reheater outlet steam temperature, and other parameters.

[0091] The steady-state processing method is as follows: Within one minute, determine whether the absolute value of the difference in coal feed rate between each coal mill in the two operating conditions is greater than 10 t / h, or whether the change in the opening degree of each secondary air unit is greater than 20, or whether the change in the opening degree of the electric regulating damper at the inlet of each secondary air box is greater than 10, or whether the change in the opening degree of each SOFA burnout air unit is greater than 20, or whether the deviation of the speed setpoint of each flue gas recirculation fan inverter is greater than 10. If any of the above conditions are met, the operating condition is considered to be in a fluctuating state, and data within a 15-minute interval of this operating condition is discarded. Data within 15 minutes of this operating condition is considered non-steady-state data and cannot be used for training.

[0092] S70 uses the coal quality information fed into the furnace and establishes a coal quality prediction model based on a multi-layer feedforward neural network to predict the NOx concentration at the SCR inlet, the desuperheating water on the A and B sides of the reheater, the flue gas temperature at the air preheater inlet, the water-cooled wall, the maximum wall temperature of the final superheater, the final reheater, and the steam temperature at the reheater outlet.

[0093] In this embodiment, the process of building the coal quality prediction model is the same as the step of obtaining the model in step S30. The difference is that the input data is the coal quality information entering the furnace, and the coal quality prediction model predicts the NOx concentration at the SCR inlet, the desuperheating water on the A and B sides of the reheater, the flue gas temperature at the air preheater inlet, the water-cooled wall, the last stage superheater, the highest wall temperature of the last stage reheater, and the steam temperature at the reheater outlet.

[0094] S80 uses the NOx concentration at the SCR inlet, the desuperheating water on the A and B sides of the reheater, the flue gas temperature at the air preheater inlet, the water-cooled wall, the last-stage superheater, the highest wall temperature of the last-stage reheater, and the steam temperature at the reheater outlet as evaluation functions, and sets the weights for each parameter.

[0095] In this embodiment, the current optimized operating condition is predicted, and each prediction result is compared with the historical best value. The specific calculation method is as follows: Single parameter evaluation value = |parameter prediction value - historical best value| / historical best value is used as the single parameter evaluation value, and full assist is set for each evaluation parameter.

[0096] Where, p j For a single-parameter evaluation value, h j Here, j represents the number of parameters, and J represents the last parameter.

[0097] S90 compares the predicted data of each parameter with the historical best data of each parameter to determine the comprehensive evaluation index of the current working condition of each parameter. The genetic optimization algorithm is used to optimize the operating oxygen content, secondary air, and the opening of the burnout damper. During the genetic algorithm optimization, the historical best working condition of the current load range and coal feeding method is selected from the historical data of each parameter as the initial population of the genetic algorithm for optimization calculation to obtain the optimal optimized operating parameters of the current working condition and output them for control.

[0098] In this embodiment, the specific steps include:

[0099] S91 provides numerical coding for operating oxygen content, secondary air volume, and the opening degree of each burnout damper.

[0100] S92. Based on the overall evaluation value, establish the fitness function. The fitness function is derived from the changes in the overall evaluation value and evaluates the performance of each individual (single working condition). Since a lower overall evaluation value indicates a better working condition, the fitness function is obtained through the following formula:

[0101]

[0102] In the formula, F is the fitness function and E is the total evaluation value. The larger the fitness function value, the better the optimization of the working condition is considered.

[0103] S93, Establish the initial population: Select the historical best operating conditions of different load ranges and different coal mill combinations from the coal quality information of the furnace as the initial population of the genetic algorithm; according to the current operating conditions, select the operating conditions in the same range as the current operating conditions, and use the operating oxygen content, secondary air damper opening and burnout air damper opening of the best operating conditions as the initial population operating conditions for replication and crossover mutation.

[0104] S94, Individual Replication Selection: Individuals ranked higher in the initial single-condition work are replicated and evolved according to a certain proportion, while those ranked lower are discarded. This invention uses the elite preservation method as the selection criterion, accelerating the convergence speed of the combustion optimization model. The probability of an individual being selected is the ratio of its fitness to the sum of the population fitness, obtained through the following formula:

[0105]

[0106] In the formula, f i Let F be an intermediate variable used to calculate the probability of an individual being selected in the population. i P represents the fitness of the individual. i Let i be the probability of an individual being selected from the population, i be the number of individuals, and N be the population size.

[0107] S95, Crossover Mutation: Chromosomes A and B cross over at point c, and new individuals A1 and B1 are obtained through a genetic algorithm; the implementation process of the genetic algorithm is as follows:

[0108] S951, set chromosome A and chromosome B;

[0109]

[0110] In the formula, a n b is the nth gene in chromosome A. n This is the nth gene in chromosome B.

[0111] S952, obtain chromosomes A and B, and their crossover at point c;

[0112]

[0113] In the formula, α′ c b′ represents the result of the crossover of chromosome A at point c. c This represents the crossover result of chromosome B at point c, where α is the hyperparameter learning rate. c The gene at position c on chromosome A, b c This is the gene at position c on chromosome B.

[0114] S953, if the k-th gene of chromosome A or chromosome B mutates, new individuals A1 and B1 are obtained through a genetic algorithm; specifically, taking the mutation of chromosome A to obtain new individual A1 as an example, the method for obtaining new individual B1 after the mutation of chromosome B is the same. The method for obtaining new individual A1 after the mutation of chromosome A is as follows:

[0115]

[0116] In the formula, a′ k For the k-th gene of the mutated chromosome A, a k a is the k-th gene on chromosome A. max a min Let f be the maximum and minimum gene values ​​of chromosome A, respectively; f be the number of iterations of the genetic algorithm; F be the maximum number of iterations of the genetic algorithm; Q(f) be an intermediate variable for obtaining a new individual; and λ be a random number between 0 and 1.

[0117] In this embodiment, the crossover and mutation process of chromosomes in the genetic algorithm determines the diversity of the population during the iteration process. Setting the crossover probability too low reduces the probability of generating new individuals in the population, while setting it too high reduces population diversity. Similarly, setting the mutation probability too low also reduces the probability of generating new individuals, while setting it too high transforms the global optimization of the genetic algorithm into a random search. Therefore, to ensure population diversity while considering the probability of generating new individuals, the crossover probability is typically between 0.1 and 1, and the mutation probability is between 0.01 and 0.1.

[0118] S96, Individual Evaluation: Calculate the fitness of the new individual to determine whether the new working condition generated after genetic iteration is better than the previous working condition. If the fitness is met, proceed to the next step. If not, repeat step S94 and perform a second round of iteration until the design requirements are met or the maximum number of iterations is reached.

[0119] S97, Output Results: Outputs optimized operating oxygen levels, secondary air damper, and burnout damper opening; realizes real-time combustion control of the boiler.

[0120] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0121] The above embodiments are merely examples of implementation methods of the invention. The scope of protection of the present invention is not limited to the above embodiments. For those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for optimizing combustion control of coal-fired units in a thermal power plant, characterized in that, include: S10, acquire a historical dataset including coal records and corresponding coal quality information on the coal mill, as well as historical operating data of the coal mill; S20, based on the thermal balance principle of the coal mill and combined with historical data sets, obtains the real-time moisture content of the coal. S30, based on a multilayer feedforward neural network and using historical datasets, establishes a coal calorific value regression model to obtain the coal calorific value; the method for building the coal calorific value regression model includes the following steps: S31, Establish the input matrix ,in M represents the number of historical datasets, and M represents the number of input variables; S32, Random Initialization: Weight Matrix =[M q], error =[1 q], , where q is the number of nodes in the first hidden layer; This is the matrix representing the computation results of the first layer of the neural network; S33, Obtain the output of the first hidden layer. ; S34, output of the first hidden layer The output of the f-th hidden layer is obtained by using the input parameters of the second hidden layer and so on. The output variable is 1, based on the output of the f-th hidden layer. As an input variable, repeat step S33 to calculate the predicted value of the target variable output by the output layer. The predicted value is calculated using the L2 norm loss function. The loss relative to the actual value of the target variable y; S35, update each hidden layer according to the L2 norm loss function S. , ; S36. After completing the update, repeat S33 to S35, iterating n times, and calculate the predicted value. The root mean square error (MSE) between the actual value of the target variable y and the actual value of y. S40, coal ash content is obtained based on coal calorific value and real-time coal moisture content; S50, based on a multi-layer feedforward neural network, uses historical datasets to establish a coal volatile matter regression model to obtain coal volatile matter; S60, acquire the historical dataset of the coal-fired unit and perform steady-state processing. Combine the steady-state processed data with real-time coal moisture, coal calorific value, coal ash content and coal volatile matter as the coal quality information for the furnace. S70 uses the coal quality information of the furnace feed and establishes a coal quality prediction model based on a multi-layer feedforward neural network to predict the NOx concentration at the SCR inlet, the desuperheating water on the A and B sides of the reheater, the flue gas temperature at the air preheater inlet, the water-cooled wall, the last stage superheater, the maximum wall temperature of the last stage reheater, and the steam temperature at the reheater outlet. S80 uses the NOx concentration at the SCR inlet, the desuperheating water on the A and B sides of the reheater, the flue gas temperature at the air preheater inlet, the water-cooled wall, the last stage superheater, the highest wall temperature of the last stage reheater, and the steam temperature at the reheater outlet as evaluation functions, and sets the weights of each parameter. S90, compare the predicted data of each parameter with the historical best data of each parameter to determine the comprehensive evaluation index of the current working condition of each parameter, and use the genetic optimization algorithm to optimize the operating oxygen, secondary air, and the opening of the burnout damper. When the genetic algorithm is optimized, the historical best working condition of the current load range and coal feeding method is selected from the historical data of each parameter as the initial population of the genetic algorithm to perform optimization calculations, obtain the optimal optimized operating parameters of the current working condition, and output them for control. Furthermore, the construction methods for the coal volatile matter regression model and the coal quality prediction model are the same as those for the coal calorific value regression model.

2. The combustion optimization control method for coal-fired power plant units according to claim 1, characterized in that, The real-time moisture content of coal can be obtained using the following formula: in, ; ; In the formula, The coal feed rate of the coal feeder is expressed in tons per hour (t / h). The primary air volume at the inlet of the coal mill is expressed in tons per hour (t / h). The primary air temperature at the inlet of the coal mill, in °C; The temperature of the air-coal mixture at the coal mill outlet, in °C; Where W is the grindability coefficient and W is the power of the coal mill (kW). The specific heat of the desiccant at the coal mill inlet is expressed in kJ / kg*℃. The specific heat of the desiccant at the coal mill outlet is expressed in kJ / kg*℃. Specific heat of cold air, kJ / kg*℃; The moisture content of the coal is as follows. Specific heat of coal dryer, KJ / kg*℃; The temperature of the leaking cold air, in °C; The temperature at which the coal enters the system, in °C; Heat consumed to evaporate water And the amount of water evaporated per kg of raw coal in the coal mill. The air leakage coefficient of the coal mill; For coal powder fineness, % For heat dissipation losses of the subsystem, KW; The coefficient by which the input power of the coal mill is converted into heat; To receive the base moisture content; This refers to the moisture content of pulverized coal.

3. The combustion optimization control method for coal-fired power plant units according to claim 1, characterized in that, The ash content of coal is obtained using the following formula: ； in, The ash content of the coal received for furnace operation is %. The net calorific value of the coal fed into the furnace is given in kJ / kg. The moisture content of the coal received before being fed into the furnace is %.

4. The combustion optimization control method for coal-fired power plant units according to claim 1, characterized in that, Steady-state processing is performed on the historical dataset of coal-fired power units, including: determining whether the absolute value of the difference in coal feed rate of each pulverizer in the two sets of operating conditions within one minute is greater than 10 t / h, or whether the change in the opening degree of each secondary air is greater than 20, or whether the change in the opening degree of the electric regulating damper at the inlet of each secondary air box is greater than 10, or whether the change in the opening degree of each SOFA burnout air is greater than 20, or whether the deviation of the speed setpoint of each flue gas recirculation fan inverter is greater than 10; if any of the above conditions are met, the operating condition is considered to be in a fluctuating state, and data within 15 minutes of the operating condition are removed.

5. The combustion optimization control method for coal-fired power plant units according to claim 1, characterized in that, Obtain the optimal operating parameters for the current working condition and output them for control, including: S91 provides numerical coding for operating oxygen content, secondary air volume, and the opening degree of each burnout damper. S92, Establish the fitness function based on the overall evaluation value; S93, Establish the initial population: Select the historical best operating conditions with different load ranges and different coal mill combinations from the coal quality information of the furnace as the initial population of the genetic algorithm; S94, Select Individual Replication: In the initial working condition, replicate and evolve the individuals ranked first in a certain proportion, and discard the individuals ranked last. S95, Crossover Variation: Chromosome A and chromosomes B ,exist c Point crossover, using a genetic algorithm to obtain new individuals. A1 and B1 ; S96, Individual Evaluation: Calculate the fitness of the new individual to determine whether the new working condition generated after genetic iteration is better than the previous working condition. If the fitness is met, proceed to the next step. If not, repeat step S94 and perform a second round of iteration until the design requirements are met or the maximum number of iterations is reached. S97, Output Results: Outputs optimized operating oxygen levels, secondary air damper, and burnout damper opening; realizes real-time combustion control of the boiler.

6. The combustion optimization control method for coal-fired power plant units according to claim 5, characterized in that, The fitness function is obtained using the following formula: ； In the formula, Let E be the fitness function. .

7. The combustion optimization control method for coal-fired power plant units according to claim 6, characterized in that, Set all evaluation parameters to full assist. ;in, For single-parameter evaluation values, Here, j represents the number of parameters, and J represents the last parameter.

8. The combustion optimization control method for coal-fired power plant units according to claim 7, characterized in that, The single parameter evaluation value is calculated as: |parameter prediction value - historical best value| / historical best value as the single parameter evaluation value.

Images