Full-load dynamic adjustment boiler staged combustion real-time control method and system

Abstract

The application discloses a boiler staged combustion real-time control method suitable for dynamic variable load adjustment, staged combustion, coal consumption and environmental protection coordination optimization, and the method comprises the following steps: when the operation condition of the boiler changes, a combustion variable is taken as a multiple-input variable and is input into a multiple-target optimization module, and multiple multi-dimensional particles are randomly generated; a first optimal adjustable combustion control amount is obtained through one-time traversal iteration optimization of the particles; the particles are judged according to a target function of the multiple-target optimization control, and a historical optimal value group composed of the first optimal adjustable combustion control amount of each particle is taken as a non-inferior solution set; the parameters of each particle are automatically updated; all iterations are cyclically executed, all particles converge to an optimal position set, and a non-inferior solution set of the optimal adjustable combustion control amount is obtained; the weight and adjustment mode of multiple optimization targets in the non-inferior solution set are dynamically changed according to a real-time operation condition, and the optimal adjustable combustion control amount is determined; and the boiler is subjected to staged combustion control.

Description

Technical Field

[0001] This invention belongs to the field of energy and power technology, and in particular relates to a real-time control method and system for staged combustion of boilers with full-load dynamic adjustment. Background Technology

[0002] In recent years, domestic coal-fired power plants have faced production and operational difficulties. On the one hand, coal-fired power plants face enormous pressure to conserve energy and reduce consumption, while air emission standards are becoming increasingly stringent. On the other hand, with the deepening of energy and power system reform, the requirements for flexible peak-shaving in power grid dispatch are becoming increasingly stringent. Therefore, as a key and difficult point in power plant energy conservation and emission reduction, the overall control optimization and performance improvement of coal-fired unit boilers are particularly important and urgent. However, currently, boiler performance deviates significantly from optimal values, and automatic closed-loop combustion optimization has generally not been achieved, with optimization control heavily relying on manual experience. Furthermore, for boiler combustion, a system characterized by large time lags, strong coupling, multiple inputs, and multiple outputs, traditional technologies lack effective means. Therefore, how power plants can improve boiler combustion efficiency and reduce pollutant emissions has become one of their key considerations. Establishing a model of the boiler combustion process and using this model to optimize and adjust the boiler combustion process to improve boiler efficiency and reduce pollutant emissions is a commonly used method.

[0003] Because combustion in the furnace involves complex physical and chemical reactions, there is currently a lack of effective mechanistic models that can accurately reflect the dramatic parameter changes during combustion. The first step is to use artificial intelligence algorithms to establish a dynamic combustion process model for boilers, reflecting how different air and coal distribution methods affect boiler efficiency and NOx generation. Currently, this causal relationship is mainly determined through combustion adjustment experiments, but these experiments are static and discrete, while combustion process models based on artificial intelligence algorithms are dynamic and continuous. Current domestic and international research on furnace combustion process modeling often uses algorithms such as artificial neural networks. These algorithms can reflect the nonlinear relationships of various physical quantities during combustion to a certain extent, but neural networks are a black box for boiler experts, unable to explain the combustion mechanism from the algorithm itself, and in practical applications, they cannot simultaneously meet the requirements for model accuracy and generalization ability.

[0004] Pattern recognition technology has been gradually applied to various industrial sectors. For coal-fired power plants, there is a large amount of data on unit operation, which contains inherent patterns and characteristics. However, due to the large amount of uncertain information in the boiler combustion process, and the difficulty in processing uncertain information using algorithms such as artificial neural networks, the accuracy of boiler combustion process models built using artificial neural network algorithms is not high, and they cannot adapt to the complex and ever-changing operating conditions in power plant production.

[0005] Like other evolutionary algorithms, Particle Swarm Optimization (PSO) is based on the concepts of "population" and "evolution." It searches for optimal solutions in complex spatial environments through cooperation and competition among individuals. However, unlike other evolutionary algorithms that use crossover, mutation, or selection operations, PSO treats individuals in the swarm as massless, volumeless particles in a 3D search space. Each particle moves at a certain speed in the search space and converges towards its historical best position and the overall best position, thus evolving candidate solutions. PSO has a strong biological and social context, making it easy to understand. Its process is simple and easy to implement, with simple parameters, and it possesses strong global search capabilities for nonlinear and multimodal problems. Therefore, since its inception, it has been rapidly applied to the original application areas of genetic algorithms. Continuous-Time Bayesian Networks (CTLs) are ideal models for data mining and representing uncertain knowledge in time series, accurately representing and reasoning about uncertain information. Therefore, there is an urgent need to establish a new method for dynamic identification of boiler operation modes, laying the foundation for the establishment of a real-time closed-loop intelligent combustion control system for boilers that adapts to dynamic load regulation, staged combustion with precise air distribution, and coordinated optimization of coal consumption and environmental protection. Summary of the Invention

[0006] The purpose of this invention is to provide a real-time control method and system for staged combustion of boilers with dynamic adjustment under full load. It uses a particle swarm optimization algorithm based on dynamic adjustment of inertial weights to perform multi-objective optimization control of the boiler combustion process. This realizes an intelligent control strategy for staged combustion of boilers based on the multi-objective optimization algorithm of particle swarm optimization with dynamic adjustment of inertial weights. It dynamically changes the target weights and adjustment modes to make it have the characteristics of adaptive optimization under varying loads. It can quickly converge to the optimal air distribution mode by adjusting multiple variables simultaneously, and realizes multi-objective coordinated optimization to improve boiler efficiency and reduce NOx generation.

[0007] This invention provides a real-time control method for staged combustion in a boiler with dynamic adjustment under full load, comprising:

[0008] S1, When the operating conditions of the boiler change, the combustion variables that cause the change in the operating conditions of the boiler are input as multiple input variables into the multi-objective optimization module. The multi-objective optimization module randomly generates multiple multi-dimensional particles, each of which corresponds to a set of adjustable combustion control values ​​under the current operating conditions.

[0009] S2, based on the inertial weight dynamic adjustment particle swarm optimization algorithm, performs a one-time traversal iteration to find the first optimal adjustable combustion control quantity for each particle.

[0010] S3, evaluate the merits of the first optimal adjustable combustion control quantity according to the objective function of multi-objective optimization control, and take the first optimal adjustable combustion control quantity of each particle as its own historical optimal value, and combine all historical optimal values ​​to form a non-dominated solution set.

[0011] S4 is a particle swarm optimization algorithm based on dynamic adjustment of inertia weight. It introduces an inertia weight factor. After each iteration, each particle automatically updates its own parameters based on its own historical best value and the global best value of the entire particle swarm.

[0012] S5, For each particle after updating parameters, steps S2-S4 are executed cyclically. After all iterations, all particles converge to the optimal position set, thereby obtaining the non-dominated solution set of the optimal adjustable combustion control quantity. According to the real-time operating conditions, the weights and adjustment modes of multiple optimization objectives in the non-dominated solution set are dynamically changed, thereby determining the optimal adjustable combustion control quantity, so that the multiple optimization objectives have variable load adaptive optimization characteristics.

[0013] S6, perform graded combustion control on the boiler based on the optimal adjustable combustion control quantity.

[0014] Preferably, S2 includes:

[0015] S21, each of the multi-dimensional particles is input into the boiler combustion dynamic process model based on a continuous-time Bayesian network. The boiler combustion dynamic process model is obtained by offline learning of the boiler's historical operating data in the full sample space.

[0016] S22 uses continuous-time Bayesian inference to predict the probability distribution of air preheater outlet flue gas temperature, CO content, boiler combustion efficiency, and NOx concentration, which are affected by the adjustment of the adjustable combustion control quantity.

[0017] Preferably, the combustion variables are unit load, coal quality, primary air, and coal mill status.

[0018] Preferably, the adjustable combustion control quantity is the total secondary air volume, the secondary air damper, and the burnout air damper.

[0019] Preferably, the objective function of the multi-objective optimization control in S3 is as follows:

[0020]

[0021] f max (x)=η(x), f min (x) = (NOx(x), Exit flue gas temperature(x)) T

[0022] Wherein, NOx represents the amount of NOx generated corresponding to the particle, η represents the boiler combustion efficiency corresponding to the particle, and the optimal value is a set of adjustable combustion control quantities that minimize the objective function value within the search range; the objective function of the multi-objective optimization control also includes seeking the optimal balance between reducing NOx and improving efficiency, in order to achieve the highest boiler efficiency and the lowest NOx emissions as much as possible.

[0023] Preferably, the objective function of the multi-objective optimization control in S3 is a combustion dynamic process model based on a continuous-time Bayesian network; through continuous-time Bayesian inference, the corresponding boiler operating state can be derived from each group of candidate combustion control quantities, thereby selecting the optimal combustion control quantity.

[0024] Preferably, step S6 includes: performing load judgment and controlling loop switching based on the load judgment result; wherein the load judgment includes: determining whether the current boiler is operating under steady-state load or variable load conditions; the controlling loop switching based on the load judgment result includes: dynamically changing the weights and adjustment modes of multiple optimization objectives in the non-dominated solution set, so that the multiple optimization objectives have variable load adaptive optimization characteristics. Under steady load conditions, the control logic focuses on improving boiler efficiency; under variable load conditions, the control logic focuses on reducing NOx generation; and quantified expert knowledge is also embedded in the control logic as optimization constraints.

[0025] The present invention also aims to provide a real-time control system for staged combustion of a boiler with full-load dynamic adjustment, comprising:

[0026] The particle generation module is used to input the combustion variables that cause the change in the boiler's operating conditions as multiple input variables into the multi-objective optimization module when the boiler's operating conditions change. The multi-objective optimization module randomly generates multiple multi-dimensional particles, each of which corresponds to a set of adjustable combustion control values ​​under the current operating conditions.

[0027] The iterative module, based on the inertial weight dynamic adjustment particle swarm optimization algorithm, is used to perform a one-time traversal iterative optimization of all the multi-dimensional particles to obtain the first optimal adjustable combustion control quantity for each particle.

[0028] The evaluation module is used to judge the merits of the first optimal adjustable combustion control quantity based on the objective function of multi-objective optimization control, and to take the first optimal adjustable combustion control quantity of each particle as its own historical optimal value, and to form a set of non-dominated solutions by combining all historical optimal values.

[0029] The update module dynamically adjusts the particle swarm optimization algorithm based on inertia weight. It introduces an inertia weight factor so that after each iteration, each particle automatically updates its own parameters based on its historical best value and the global best value of the entire particle swarm.

[0030] The optimization module is used to iteratively execute steps S2-S4 for each particle after updating parameters. After all iterations, all particles converge to the optimal position, thereby obtaining the optimal adjustable combustion control quantity.

[0031] The control module is used to perform graded combustion control on the boiler based on the optimal adjustable combustion control quantity.

[0032] A third aspect of the present invention provides an electronic device including a processor and a memory, the memory storing a plurality of instructions, the processor being configured to read the instructions and execute the method as described in the first aspect.

[0033] A fourth aspect of the present invention provides a computer-readable storage medium storing a plurality of instructions which can be read by a processor and executed as described in the first aspect.

[0034] The method, apparatus, electronic device, and computer-readable storage medium provided by this invention have the following beneficial technical effects:

[0035] After gaining an accurate understanding of the real-time combustion status and boiler operation mode by combining boiler operation data and online furnace temperature measurement data, the inertial weighted dynamic adjustment particle swarm algorithm is used for multi-objective optimization. There are three dimensions of optimization objectives, including reducing flue gas temperature, reducing CO content, and reducing NOx generation (boiler efficiency is improved by reducing flue gas temperature and CO content). The algorithm quickly converges to the current optimal air distribution mode, which is to achieve the simultaneous optimization of these three objectives by simultaneously adjusting the opening of secondary air volume, secondary damper, and burnout damper. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating a smart combustion real-time closed-loop control algorithm based on a particle swarm optimization algorithm with dynamic adjustment of inertial weights, according to a preferred embodiment of the present invention.

[0037] Figure 2 (a) is a schematic diagram showing that the exhaust gas temperature is reduced after optimization according to a preferred embodiment of the present invention;

[0038] Figure 2 (b) is a schematic diagram illustrating the reduction in NOx generation after optimization according to a preferred embodiment of the present invention;

[0039] Figure 3A flowchart of the particle swarm optimization algorithm with dynamic adjustment of inertia weight provided by this invention;

[0040] Figure 4 This is a schematic diagram of the structure of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0041] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0042] Example 1

[0043] like Figure 1 As shown, this invention provides a real-time control method for staged combustion of boilers with dynamic adjustment at full load, belonging to the category of staged combustion control methods for boilers based on inertial weight dynamic adjustment particle swarm multi-objective optimization, including:

[0044] S1, When the operating conditions of the boiler change, the combustion variables that cause the change in the operating conditions of the boiler are input as multiple input variables into the multi-objective optimization module. The multi-objective optimization module randomly generates multiple multi-dimensional particles, each of which corresponds to a set of adjustable combustion control values ​​under the current operating conditions.

[0045] S2, based on the inertial weight dynamic adjustment particle swarm optimization algorithm, performs a one-time traversal iteration to find the first optimal adjustable combustion control quantity for each particle.

[0046] S3. The merits of the first optimal adjustable combustion control quantity (i.e., the particle) are evaluated according to the objective function of the multi-objective optimization control, and the first optimal adjustable combustion control quantity of each particle is taken as its own historical optimal value. All historical optimal values ​​are combined to form a set of non-dominated solutions.

[0047] S4 is a particle swarm optimization algorithm based on dynamic adjustment of inertia weight. It introduces an inertia weight factor. After each iteration, each particle automatically updates its own parameters based on its own historical best value and the global best value of the entire particle swarm.

[0048] S5, For each particle after updating parameters, steps S2-S4 are executed cyclically. After all iterations, all particles converge to the optimal position set, thereby obtaining the non-dominated solution set of the optimal adjustable combustion control quantity. According to the real-time operating conditions, the weights and adjustment modes of multiple optimization objectives in the non-dominated solution set are dynamically changed, thereby determining the optimal adjustable combustion control quantity, so that the multiple optimization objectives have variable load adaptive optimization characteristics.

[0049] S6, perform graded combustion control on the boiler based on the optimal adjustable combustion control quantity.

[0050] In this embodiment, S5 is implemented based on the load judgment and control loop switching function. Depending on whether the current boiler is operating under steady-state load or variable load conditions, the weights and adjustment modes of multiple optimization objectives in the non-dominated solution set are dynamically changed to give it variable load adaptive optimization characteristics: under steady load conditions, the control logic focuses on improving boiler efficiency; while under variable load conditions, it focuses on reducing NOx generation to prevent emissions from exceeding standards at the source. This simultaneously adjusts multiple variables such as secondary air volume, secondary damper opening, and burnout damper to quickly converge to the optimal air distribution mode, realizing dynamic following optimization and closed-loop control of the boiler under full load.

[0051] In a preferred embodiment, S2 includes:

[0052] S21, each of the multi-dimensional particles is input into the boiler combustion dynamic process model, which is based on a continuous-time Bayesian network and is obtained by offline learning of the boiler's historical operating data in the full sample space.

[0053] S22 uses continuous-time Bayesian inference to predict the probability distribution of air preheater outlet flue gas temperature, CO content, boiler combustion efficiency, and NOx concentration, which are affected by the adjustment of the adjustable combustion control quantity.

[0054] In a preferred embodiment, the combustion variables are unit load, coal quality, primary air, and coal mill status.

[0055] In a preferred embodiment, the adjustable combustion control quantity is the total secondary air volume, the secondary air damper, and the burnout air damper.

[0056] In a preferred embodiment, the objective function of the multi-objective optimization control in S3 is shown in the following equation:

[0057]

[0058] f max (x)=η(x), f min (x) = (NOx(x), Exit flue gas temperature(x)) T

[0059] Wherein, NOx represents the amount of NOx generated corresponding to the particle, and η represents the boiler combustion efficiency corresponding to the particle. Therefore, the optimal value is the set of adjustable combustion control quantities that minimizes the objective function value within the search range; the objective function of the multi-objective optimization control also includes seeking the optimal balance between reducing NOx and improving efficiency, aiming to achieve the highest possible boiler efficiency and the lowest possible NOx emissions.

[0060] In a preferred embodiment, the objective function of the multi-objective optimization control in S3 is a combustion dynamic process model based on a continuous-time Bayesian network. Through continuous-time Bayesian inference, the corresponding boiler operating state can be derived from each set of candidate combustion control variables, thereby selecting the optimal combustion control variable. This can improve boiler combustion efficiency while reducing NOx emissions. In this embodiment, after determining the model structure and parameters, the Bayesian network-based combustion process model will serve as the objective function in the optimization control, used to evaluate the merits of each set of candidate combustion control variables.

[0061] In a preferred embodiment, S6 includes: performing load judgment and switching the control loop according to the load judgment result; wherein the load judgment includes: determining whether the current boiler is operating under steady-state load or variable load conditions; the control loop switching according to the load judgment result includes: dynamically changing the weights and adjustment modes of multiple optimization objectives in the non-dominated solution set, so that the multiple optimization objectives have variable load adaptive optimization characteristics. Under stable load conditions, the control logic focuses on improving boiler efficiency; under variable load conditions, the control logic focuses on reducing NOx generation, preventing emissions from exceeding standards at the source. This control method can simultaneously adjust multiple variables such as secondary air volume, secondary damper opening, and burnout damper, quickly converging to the optimal air distribution mode, realizing dynamic following optimization and closed-loop control of the boiler under full load; and quantified expert knowledge is also embedded in the control logic as optimization constraints, ensuring the safe, stable, economical, and efficient operation of the unit. Figure 2 As shown in (a) and 2(b), after optimized control, both the flue gas temperature and NOx generation were reduced.

[0062] This embodiment combines boiler operation data and online furnace temperature measurement data to gain an accurate understanding of the real-time combustion status and boiler operation mode. Then, it uses a particle swarm optimization algorithm based on inertial weight dynamic adjustment for multi-objective optimization. There are three dimensions of optimization objectives: reducing flue gas temperature, reducing CO content, and reducing NOx generation (boiler efficiency is improved by reducing flue gas temperature and CO content). The algorithm quickly converges to the current optimal air distribution mode, which is achieved by simultaneously adjusting the opening of secondary air volume, secondary damper, and burnout damper to achieve the simultaneous optimization of these three objectives.

[0063] The Bayesian network structure in step S2 is a directed acyclic graph (DAG). The DAG includes multiple nodes, each of which represents a random variable related to the boiler combustion process. The multiple nodes are divided into cause nodes and effect nodes, where the combustion control quantity is the cause node, the boiler operating state parameter is the effect node, and the directed edges between nodes represent which combustion control quantities affect each boiler operating state parameter.

[0064] Based on the parameters of the boiler combustion process model using the Bayesian network structure, the changes of the boiler combustion process model parameters over time are learned within the Bayesian network structure to establish a boiler combustion process model based on a continuous Bayesian network algorithm for boiler systems with large inertial lag, including:

[0065] Learn the conditional probability density function of a boiler combustion process model with parameters as variables in a Bayesian network structure;

[0066] By studying the Condition Intensity Matrix (CIM) with parameters as variables in the boiler combustion process model, a Continuous Time Bayesian Network (CTBN) structure is obtained to describe the instantaneous dynamic characteristics of the combustion process.

[0067] Based on the continuous-time Bayesian network structure, the boiler combustion effect, including boiler efficiency and NOx generation, is predicted through Bayesian inference based on maximum posterior probability.

[0068] Analysis of the working principle of the method:

[0069] I. Principle Analysis of Particle Swarm Optimization Algorithm

[0070] In practical applications of particle optimization algorithms, the potential solution to each optimization problem can be imagined as a particle in a D-dimensional search space. Each particle has a fitness value determined by the objective function. These particles fly in the search space at a certain speed, the magnitude and direction of which are dynamically adjusted based on the particle's own flight experience and the flight experience of the entire population. Subsequently, all particles will follow the current best particle in the solution space.

[0071] Suppose we are looking for a minimum value in a problem, and we need to find the optimal solution x such that the multidimensional objective function f(x) satisfies the following equation:

[0072] x = argminf(x).

[0073] In a D-dimensional target search space, there is a swarm of N particles, where the i-th particle is represented as a D-dimensional vector. That is, the position of the i-th particle in the D-Wade search space is In other words, the position of each particle is a potential solution to the optimization problem. Substituting the values ​​into the objective function (FitnessFunction) allows us to calculate its fitness value, which is then used to measure fitness. The merits and demerits. Let the fitness value of each particle be Fitness. i (i∈[1,N]). The velocity of the i-th particle is also a D-dimensional vector, denoted as... Let the optimal position found so far by the i-th particle be denoted as . The optimal position found so far by the entire particle swarm is g. best =(g1,g2,...,g D The operational mode of each particle depends not only on its own flight experience (i.e., p) best It is also affected by the flight experience of the entire population (i.e., g). best Therefore, the particle swarm optimization algorithm can guarantee that the final result is the global optimum, rather than being trapped in a local optimum. The process of the particle swarm optimization algorithm is as follows:

[0074] (1) Set the initial values ​​for N particles, each particle x i Let i represent a potential solution to the optimization problem, where i ∈ [1, N]. For the minimum optimization problem, the fitness value of each particle is... i The optimal position of each particle And the optimal position g of the entire population best Set all to infinity.

[0075] (2) When the number of iterations t reaches the set maximum number of iterations t max Previously, or if a certain termination condition was not met, the following steps were repeated in each iteration:

[0076] Calculate the fitness value for each particle. i =f(x) i );

[0077] Update the best position found so far for each particle.

[0078] Update the optimal position found so far for the entire particle swarm.

[0079] The particle is moved according to the following formula.

[0080] xi,t+1 =x i,t +u i,t+1 ,

[0081] Where, u i,t+1 Defined as

[0082]

[0083] Among them, u i,t u represents the velocity of the i-th particle during time interval t. i,t+1 ω represents the velocity of the i-th particle in the next time interval t+1, where ω is a constant less than 1, used to reflect the influence of the particle's velocity in time interval t on its velocity in the next time interval t+1. i,t This represents the current position of the i-th particle. Learning factors c1 and c2 are the weights of these variables in determining the flight speed. r1 and r2 are random constants between [0,1], adding a random element to the algorithm.

[0084] t = t + 1.

[0085] (3) When the iteration ends, the optimal solution x that satisfies the multidimensional objective function f(x) can be obtained.

[0086] The flowchart of the particle swarm optimization algorithm is as follows: Figure 3 As shown.

[0087] II. Principle Analysis of Particle Swarm Optimization Algorithm Based on Dynamic Adjustment of Inertia Weight

[0088] After extensive research and experimentation, in order to improve the convergence performance of the basic particle swarm optimization algorithm and avoid the algorithm getting trapped in local optima, in u i,t+1 Introducing inertia weighting factors w and u into the evolutionary equation i,t+1 The first part of the evolutionary equation is the particle's previous velocity, which ensures the global convergence of the algorithm. The second and third parts are the social factors that cause changes in particle velocity, enabling the algorithm to perform local search. Therefore, w plays a role in balancing global and local search capabilities; a larger w value indicates stronger global search capability and weaker local search capability, and vice versa. An appropriate w value can improve algorithm performance, enhance optimization ability, and reduce the number of iterations.

[0089] The introduction of inertia weights has greatly promoted the development of particle swarm optimization (PSO) algorithms, significantly expanding the scope for algorithm improvement. However, achieving optimal algorithm performance still presents several drawbacks. A larger w value favors global search, resulting in faster convergence but making it difficult to obtain an exact solution. Conversely, a smaller w value facilitates local search and yields more accurate solutions, but leads to slower convergence and sometimes getting trapped in local maxima. To find a suitable w value that strikes a balance between search accuracy and speed, several inertia weight adjustment strategies have been proposed, primarily linear decreasing, nonlinear decreasing, and adaptive adjustment strategies. This embodiment employs an adaptive, dynamically changing inertia weight adjustment method.

[0090] The adaptive adjustment strategy determines the change in inertia weight based on the precocity convergence of the population and the individual fitness value. Let particle p... i The fitness value is f i The fitness value of the optimal particle is f. m The average fitness of the particle swarm is It will be better than f avg The fitness values ​​are averaged to obtain f′. avg And define Δ = |f m -f′ avg |. According to f i f avg and f′ avg The population is divided into three subgroups, each undergoing different adaptive operations. The inertia weights are then adjusted as follows:

[0091] f i Better than f′ avg but

[0092]

[0093] f i Better than f avg But second only to f′ avg Then the inertial weight remains unchanged;

[0094] f i Second to f avg but

[0095]

[0096] The first type of particle represents the better particles in the swarm, which are close to the global optimum. They are given a smaller inertia weight to enhance their local search ability. The second type of particle represents general particles with good global and local search capabilities, and their inertia weight does not need to be changed. The third type of particle represents the worse particles in the swarm, and their inertia weight is adjusted using a method similar to adaptively adjusting the control parameters of the genetic algorithm. k1 and k2 are control parameters; k1 controls the upper limit of w, and k2 mainly controls the adjustment capability of the above formula. When the algorithm stops, if the particle distribution is relatively dispersed, Δ is larger, and the particle's w is reduced using the above formula to strengthen the local search ability and make the swarm converge. If the particle distribution is relatively clustered, Δ is smaller, and the particle's w is increased using the above formula to give the particles a stronger detection ability, thus effectively escaping local optima.

[0097] Example 2

[0098] This embodiment also provides a real-time control system for staged combustion of a boiler with full-load dynamic adjustment, including:

[0099] The particle generation module is used to input the combustion variables that cause the change in the boiler's operating conditions as multiple input variables into the multi-objective optimization module when the boiler's operating conditions change. The multi-objective optimization module randomly generates multiple multi-dimensional particles, each of which corresponds to a set of adjustable combustion control values ​​under the current operating conditions.

[0100] The iterative module, based on the inertial weight dynamic adjustment particle swarm optimization algorithm, is used to perform a one-time traversal iterative optimization of all the multi-dimensional particles to obtain the first optimal adjustable combustion control quantity for each particle.

[0101] The evaluation module is used to judge the merits of the first optimal adjustable combustion control quantity (i.e., the particle) according to the objective function of multi-objective optimization control, and to take the first optimal adjustable combustion control quantity of each particle as its own historical optimal value, and to form a set of non-dominated solutions by combining all historical optimal values.

[0102] The update module dynamically adjusts the particle swarm optimization algorithm based on inertia weight. It introduces an inertia weight factor so that after each iteration, each particle automatically updates its own parameters based on its historical best value and the global best value of the entire particle swarm.

[0103] The optimization module is used to iteratively execute steps S2-S4 for each particle after updating parameters. After all iterations, all particles converge to the optimal position, thereby obtaining the set of non-dominated solutions for the optimal adjustable combustion control quantity. According to the real-time operating conditions, the weights and adjustment modes of multiple optimization objectives in the set of non-dominated solutions are dynamically changed, thereby determining the optimal adjustable combustion control quantity, so that the multiple optimization objectives have variable load adaptive optimization characteristics.

[0104] The control module is used to perform graded combustion control on the boiler based on the optimal adjustable combustion control quantity.

[0105] The present invention also provides a memory that stores a plurality of instructions for implementing the method as described in Embodiment 1.

[0106] like Figure 4 As shown, the present invention also provides an electronic device, including a processor 301 and a memory 302 connected to the processor 301. The memory 302 stores a plurality of instructions, which can be loaded and executed by the processor to enable the processor to perform the method as described in Embodiment 1.

[0107] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A real-time control method for staged combustion in boilers with full-load dynamic adjustment, used for implementing intelligent closed-loop control of boiler combustion in real time to achieve precise air distribution, coordinated optimization of coal consumption and environmental protection in staged combustion, characterized in that, include: S1, When the operating conditions of the boiler change, the combustion variables that cause the change in the operating conditions of the boiler are input as multiple input variables into the multi-objective optimization module. The multi-objective optimization module randomly generates multiple multi-dimensional particles, each of which corresponds to a set of adjustable combustion control values ​​under the current operating conditions. S2, based on the inertial weight dynamic adjustment particle swarm optimization algorithm, performs a one-time traversal iteration to find the first optimal adjustable combustion control quantity for each particle. S3, evaluate the merits of the first optimal adjustable combustion control quantity according to the objective function of multi-objective optimization control, and take the first optimal adjustable combustion control quantity of each particle as its own historical optimal value, and combine all historical optimal values ​​to form a non-dominated solution set. S4 is a particle swarm optimization algorithm based on dynamic adjustment of inertia weight. It introduces an inertia weight factor. After each iteration, each particle automatically updates its own parameters based on its own historical best value and the global best value of the entire particle swarm. S5, For each particle after updating parameters, steps S2-S4 are executed cyclically. After all iterations, all particles converge to the optimal position set, thereby obtaining the non-dominated solution set of the optimal adjustable combustion control quantity. According to the real-time operating conditions, the weights and adjustment modes of multiple optimization objectives in the non-dominated solution set are dynamically changed, thereby determining the optimal adjustable combustion control quantity, so that the multiple optimization objectives have variable load adaptive optimization characteristics. S6, perform staged combustion control on the boiler based on the optimal adjustable combustion control quantity; S2 includes: S21, each of the multi-dimensional particles is input into the boiler combustion dynamic process model, which is based on a continuous-time Bayesian network and is obtained by offline learning of the boiler's historical operating data in the full sample space. S22 uses continuous-time Bayesian inference to predict the probability distribution of air preheater outlet flue gas temperature, CO content, boiler combustion efficiency, and NOx concentration, which are affected by the adjustment of the adjustable combustion control quantity.

2. The real-time control method for staged combustion of a boiler with full-load dynamic adjustment according to claim 1, characterized in that, The combustion variables are unit load, coal quality, primary air, and coal mill status.

3. The real-time control method for staged combustion of a boiler with full-load dynamic adjustment according to claim 1, characterized in that, The adjustable combustion control parameters are the total secondary air volume, the secondary air damper, and the burnout air damper.

4. The real-time control method for staged combustion of a boiler with full-load dynamic adjustment according to claim 1, characterized in that, The objective function of the multi-objective optimization control in S3 is shown in the following equation: Wherein, NOx represents the amount of NOx generated corresponding to that particle. This represents the boiler combustion efficiency corresponding to the particle. The optimal value is a set of adjustable combustion control quantities that minimize the objective function value within the search range. The objective function of the multi-objective optimization control also includes seeking the optimal balance between reducing NOx and improving efficiency, aiming to achieve the highest boiler efficiency and the lowest NOx emissions possible.

5. The real-time control method for staged combustion of a boiler with full-load dynamic adjustment according to claim 4, characterized in that, The objective function of the multi-objective optimization control in S3 is a combustion dynamic process model based on a continuous-time Bayesian network. Through continuous-time Bayesian inference, the corresponding boiler operating state can be derived from each group of candidate combustion control quantities, thereby selecting the optimal combustion control quantity.

6. The real-time control method for staged combustion of a boiler with full-load dynamic adjustment according to claim 5, characterized in that, S6 includes: performing load judgment and switching the control loop according to the load judgment result; wherein the load judgment includes: determining whether the current boiler is operating under steady-state load or variable load conditions; the control loop switching according to the load judgment result includes: dynamically changing the weights and adjustment modes of multiple optimization objectives in the non-dominated solution set, so that the multiple optimization objectives have variable load adaptive optimization characteristics. Under steady load conditions, the control logic focuses on improving boiler efficiency; under variable load conditions, the control logic focuses on reducing NOx generation; and quantified expert knowledge is also embedded in the control logic as optimization constraints.

7. A real-time control system for staged combustion of a boiler with full-load dynamic adjustment, used to implement the real-time control method for staged combustion of a boiler with full-load dynamic adjustment as described in any one of claims 1-6, characterized in that, include: The particle generation module is used to input the combustion variables that cause the change in the boiler's operating conditions as multiple input variables into the multi-objective optimization module when the boiler's operating conditions change. The multi-objective optimization module randomly generates multiple multi-dimensional particles, each of which corresponds to a set of adjustable combustion control values ​​under the current operating conditions. The iterative module, based on the inertial weight dynamic adjustment particle swarm optimization algorithm, is used to perform a one-time traversal iterative optimization of all the multi-dimensional particles to obtain the first optimal adjustable combustion control quantity for each particle. The evaluation module is used to judge the merits of the first optimal adjustable combustion control quantity based on the objective function of multi-objective optimization control, and to take the first optimal adjustable combustion control quantity of each particle as its own historical optimal value, and to form a set of non-dominated solutions by combining all historical optimal values. The update module dynamically adjusts the particle swarm optimization algorithm based on inertia weight. It introduces an inertia weight factor so that after each iteration, each particle automatically updates its own parameters based on its historical best value and the global best value of the entire particle swarm. The optimization module is used to iteratively execute steps S2-S4 for each particle after updating parameters. After all iterations, all particles converge to the optimal position set, thereby obtaining the non-dominated solution set of the optimal adjustable combustion control quantity. According to the real-time operating conditions, the weights and adjustment modes of multiple optimization objectives in the non-dominated solution set are dynamically changed, thereby determining the optimal adjustable combustion control quantity, so that the multiple optimization objectives have variable load adaptive optimization characteristics. The control module is used to perform graded combustion control on the boiler based on the optimal adjustable combustion control quantity.

8. An electronic device, characterized in that, It includes a processor and a memory, the memory storing multiple instructions, and the processor being used to read the instructions and execute the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions, which can be read by a processor and executed as described in any one of claims 1-6.

Images