Particle swarm optimization control method and system for NOx emission and coal consumption of coal-fired boiler

Abstract

This invention relates to a particle swarm optimization control method and system for NOx emissions and coal consumption in coal-fired boilers, belonging to the field of intelligent combustion optimization control technology for boilers. The method includes the following steps: acquiring and preprocessing the operating parameters of the coal-fired boiler to obtain a standard combustion condition parameter set; extracting combustion operating variables to construct a high-dimensional combustion state vector, and performing nonlinear dimensionality reduction through an improved isometric feature mapping to obtain a low-dimensional combustion condition manifold feature space; inputting this feature space into an improved particle swarm optimization algorithm, obtaining combustion parameter control commands through multi-objective Pareto optimization, smoothing them, and then issuing them to the boiler distributed control system to drive the actuators; real-time acquisition of state parameters, using a dynamic time warping algorithm to calculate the time delay deviation between the control commands and the system response, and updating the parameters of the improved particle swarm optimization algorithm accordingly. This invention solves the problems of blind initialization and poor adaptability to changing operating conditions in traditional particle swarm algorithms, achieving real-time collaborative optimization of NOx emissions and coal consumption rate.

Description

Technical Field

[0001] This invention relates to a particle swarm optimization control method and system for NOx emissions and coal consumption in coal-fired boilers, belonging to the field of intelligent combustion optimization control technology for boilers. Background Technology

[0002] As the core equipment of a thermal power generation system, the combustion operation of a coal-fired power plant boiler is crucial to the unit's power generation efficiency, coal consumption, pollutant emissions, and equipment safety. With the increasing demands of the power system on the operation of thermal power units, coal-fired boilers need to frequently switch operating conditions across a wide load range. The operating parameters of various systems exhibit nonlinear and strongly coupled characteristics. Under the requirement of low-NOx combustion control, traditional control methods struggle to achieve synergistic optimization of NOx emission control and coal consumption. Currently, industrial coal-fired boiler combustion optimization mainly employs three technical approaches: manual adjustment based on operational experience, automatic control based on fixed rules or PID feedback, and data-driven optimization based on intelligent optimization algorithms.

[0003] On the one hand, manual adjustment methods based on operational experience lack a unified multi-objective optimization logic and cannot maintain optimal operating conditions in the long term under complex and variable operating conditions. On the other hand, automatic control methods based on fixed rules or PID feedback only target a single objective and do not take into account the synergistic relationship between NOx and coal consumption, which can easily lead to problems such as low NOx with high coal consumption or low coal consumption with excessive emissions. Furthermore, existing boiler combustion optimization technologies based on particle swarm optimization have many defects in engineering applications, such as the tendency of random initialization to generate infeasible particles, the inability to dynamically adjust the search step size, the reliance of multi-objective optimization weights on manual experience and the inability to adapt to different operating conditions, the tendency of the algorithm to get stuck in local optima in the later stages of iteration, and the inability of existing fixed time delay compensation methods to adapt to time delay drift, which leads to a decline in control accuracy over long-term operation.

[0004] Therefore, there is an urgent need for a collaborative control method that integrates high-dimensional feature reduction, physical constraint initialization, load adaptive optimization, multi-objective Pareto decision-making, control command smoothing, and dynamic time delay compensation to improve the accuracy, stability, and engineering practicality of NOx and coal consumption collaborative optimization in coal-fired boilers. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a particle swarm optimization control method and system for NOx emissions and coal consumption in coal-fired boilers.

[0006] The technical solution of the present invention is as follows: On the one hand, this invention proposes a particle swarm optimization control method for NOx emissions and coal consumption in coal-fired boilers, comprising the following steps: The operating parameters of the coal-fired boiler are obtained and standardized preprocessing is performed to obtain a set of standard combustion condition parameters that eliminate the influence of noise and dimensions. Combustion operation variables are extracted from the standard combustion condition parameter set to construct a high-dimensional combustion state vector. Then, nonlinear dimensionality reduction is performed on the high-dimensional combustion state vector through an improved isometric feature map to obtain a low-dimensional combustion condition manifold feature space that characterizes the core characteristics of the combustion condition. The low-dimensional combustion condition manifold feature space is input into the improved particle swarm optimization algorithm. In the particle swarm optimization algorithm, each particle corresponds to a set of candidate combustion control parameters, and the position of the particle in the low-dimensional manifold feature space represents the combustion condition corresponding to the control scheme. With NOx emission and coal consumption rate as optimization objectives, multi-objective Pareto optimization is performed to obtain the optimal combustion parameter control command sequence, which is then sent to the distributed control system of the coal-fired boiler after amplitude limiting and smoothing. The distributed control system for coal-fired boilers drives the actuators to adjust the boiler combustion control parameters according to the optimal combustion parameter control commands. It collects the combustion state parameters after the actuators move in real time, uses a dynamic time warping algorithm to align the optimal combustion parameter control command sequence with the system state response sequence, and calculates the time delay deviation between the two. Based on the time delay deviation, it updates the parameters of the improved particle swarm optimization algorithm online to form a closed-loop rolling optimization.

[0007] Preferably, the method utilizes an improved isometric feature mapping to perform nonlinear dimensionality reduction on the high-dimensional combustion state vector, and the specific steps are as follows: Combustion operation variables are extracted from the standard combustion condition parameter set to construct a high-dimensional combustion state vector; Calculate the Euclidean distance between the high-dimensional combustion state vectors at any two sampling times, set a number of nearest neighbors, and construct... Nearest neighbor graph, as a manifold structure characterizing combustion conditions; The shortest path between high-dimensional combustion state vectors is calculated along the manifold structure to obtain the geodesic distance matrix between the high-dimensional combustion state vectors; The geodesic distance matrix between high-dimensional combustion state vectors is centered by constructing a centered matrix from the identity matrix and all-one column vectors. The inner product matrix is ​​then calculated by squaring the centered matrix and the geodesic distance matrix. Principal component analysis (PCA) is used to perform eigenvalue decomposition on the inner product matrix. The eigenvalues ​​are sorted from largest to smallest, and the eigenvectors corresponding to the top three eigenvalues ​​are selected to construct a low-dimensional combustion condition manifold feature space.

[0008] Preferably, the improved particle swarm optimization algorithm includes a constraint initialization unit, an evaluation weight construction unit, a particle swarm update unit, and a population perturbation unit, wherein: The constraint initialization unit is used to construct a physical safety domain by combining the preset operating constraints of the current coal-fired unit; the first part of initial particles is generated by uniform sampling within the physical safety domain, and the second part of initial particles is generated by Tent chaotic mapping within the chaotic detection domain; the first and second parts of initial particles are merged to construct the initial particle swarm within the physical safety domain. The evaluation weight construction unit is used to construct the collaborative evaluation weights of NOx emission targets and coal consumption rate targets based on the deviations of real-time NOx emission change rate, target load, real-time load of coal-fired units, and preset environmental protection constraint coefficients. The particle swarm update unit constructs the particle velocity update equation and adjusts the search step size of the initial particle swarm; it updates the particle positions according to the updated velocities to obtain a new generation of candidate particle swarms; where the position of each candidate particle represents a candidate combustion control scheme in the low-dimensional combustion condition manifold feature space. The population perturbation unit calculates the multi-objective fitness value of each candidate particle in the new generation of candidate particle swarm based on preset NOx emission targets, coal consumption rate targets, main steam temperature deviation, and furnace negative pressure fluctuations; performs non-dominated sorting on the new generation of candidate particle swarm based on the obtained multi-objective fitness values, obtains the Pareto non-dominated solution set and calculates the crowding distance; calculates the population distribution entropy in the Pareto non-dominated solution set; when the population distribution entropy is lower than the preset entropy threshold, it indicates insufficient population diversity and triggers the Pareto entropy increase perturbation operation.

[0009] Preferably, the method updates the initial particle swarm, and the specific steps are as follows: Construct the particle velocity update equation and adjust the particle swarm search step size, expressed as a formula: ; ; In the formula, Indicates particle index, Indicates a dimension index. Indicates the first In the nth iteration Individual particles 3D particle velocity, Represents dynamic inertia weight. Indicates the first In the nth iteration Individual particles 3D particle velocity, Indicates the first learning factor. express The first uniformly distributed random number within the interval, Indicates the first Individual particles The optimal position of an individual in dimension. Indicates the first In the nth iteration Individual particles The position of the dimension Indicates the second learning factor. express The second uniformly distributed random number within the interval, Indicates that the particle swarm is The globally optimal position of dimension. Indicates the maximum inertia weight. This represents the minimum inertia weight. This represents the absolute value of the rate of change of unit load. This represents the absolute value of the unit's maximum load change rate; The particle position is updated based on the updated velocity, expressed by the formula: ; In the formula, Indicates the first In the nth iteration Individual particles The position of the dimension; A new generation of candidate particle swarms is obtained, and the position of each candidate particle represents a candidate combustion control scheme in the low-dimensional combustion condition manifold feature space.

[0010] Preferably, the Pareto entropy increase perturbation operation enhances the diversity of the next-generation candidate particle swarm using a Gaussian perturbation method, and the specific steps are as follows: Calculate the multi-objective fitness value for the new generation of candidate particle swarms, perform Pareto undominated sorting to obtain the undominated solution set and calculate the crowding distance; Calculate the population distribution entropy of the non-dominated solution set. When the entropy value is lower than a preset threshold, it is determined that the population diversity is insufficient, triggering the Pareto entropy increase perturbation operation. Sort all particles in the Pareto nondominated solution set in ascending order of crowding distance, select the first preset percentage of particles as the particles to be perturbed, perform Gaussian perturbation, and obtain the perturbed particle swarm after position update. By merging perturbed and undisturbed particle swarms, a new generation of optimized particle swarms with enhanced diversity is formed.

[0011] Preferably, the method uses a dynamic time warping algorithm to update the parameters of the improved particle swarm optimization algorithm, and the specific steps are as follows: Construct a reference sequence for boiler combustion control command response and an observation sequence for actual state feedback; A cost matrix is ​​constructed using the Euclidean distance between each pair of points in the reference sequence and the observation sequence as elements; By traversing the cost matrix using a dynamic time warping algorithm, the optimal warping path that minimizes the cumulative cost value is found, expressed by the formula: ; In the formula, Indicates the reference sequence. Represents the observation sequence. Represents the reference sequence With observation sequence The minimum cumulative cost, Indicates a regular path. This represents the index of matching point pairs on the normalized path. Indicates the first normalized path The local cost of a pair of matching points. This represents the total number of matching point pairs on the regularized path; The time misalignment between the reference sequence and the observed sequence is determined based on the optimal regularization path, and the actual time delay of the current operating condition is calculated in conjunction with the nominal operating condition time base. The time delay deviation between the actual time delay and the nominal time delay of the boiler combustion process is calculated. When the time delay deviation exceeds a preset threshold, the particle swarm learning factor is corrected based on the time delay deviation and the time delay compensation coefficient, expressed by the formula: ; In the formula, This represents the corrected particle swarm learning factor. This represents the particle swarm learning factor before correction. Indicates the time delay compensation coefficient. Indicates the actual time delay under the current operating conditions. Indicates the nominal time delay of the boiler combustion process; The corrected particle swarm learning factor is obtained and fed back to the improved particle swarm optimization algorithm as the calculation parameter for the next round of iterative optimization. The optimal regularization path is a monotonically continuous path that minimizes the cumulative local cost.

[0012] On the other hand, this invention proposes a particle swarm optimization control system for NOx emissions and coal consumption in coal-fired boilers, comprising the following modules: The data acquisition and preprocessing module is used to acquire the operating parameters of the coal-fired boiler and perform standardized preprocessing to obtain a set of standard combustion condition parameters that eliminate the influence of noise and dimensions. The high-dimensional feature extraction and nonlinear dimensionality reduction module extracts combustion operation variables from the standard combustion condition parameter set, constructs a high-dimensional combustion state vector, and performs nonlinear dimensionality reduction on the high-dimensional combustion state vector through an improved isometric feature mapping to obtain a low-dimensional combustion condition manifold feature space that characterizes the core characteristics of the combustion condition. The multi-objective collaborative optimization and command output module is used to input the low-dimensional combustion condition manifold feature space into the improved particle swarm optimization algorithm. In the particle swarm optimization algorithm, each particle corresponds to a set of candidate combustion control parameters, and the position of the particle in the low-dimensional manifold feature space represents the combustion condition corresponding to the control scheme. With NOx emission and coal consumption rate as optimization objectives, multi-objective Pareto optimization is performed to obtain the optimal combustion parameter control command sequence, which is then sent to the distributed control system of the coal-fired boiler after amplitude limiting and smoothing. The actuator drive control module is used in the distributed control system of a coal-fired boiler to drive the actuator to act according to the optimal combustion parameter control command and adjust the boiler combustion control parameters. The time delay evaluation and algorithm parameter correction module is used to collect combustion state parameters after the actuator of the coal-fired boiler operates in real time. It uses a dynamic time warping algorithm to align the optimal combustion parameter control command sequence with the system state response sequence and calculates the time delay deviation between the two. Based on the time delay deviation, the parameters of the improved particle swarm optimization algorithm are updated online to form a closed-loop rolling optimization.

[0013] In another aspect, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method as described in any embodiment of the present invention.

[0014] In another aspect, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any embodiment of the present invention.

[0015] The present invention has the following beneficial effects: 1. This invention proposes a particle swarm optimization control method and system for NOx emissions and coal consumption of coal-fired boilers. By improving the isometric mapping algorithm, nonlinear topological dimensionality reduction is performed on the high-dimensional operating parameters of the boiler, and three-dimensional core combustion features such as combustion intensity, air distribution uniformity, and fuel adaptability are extracted. This effectively eliminates the coupling redundancy interference between high-dimensional variables, reduces the dimensionality of the optimization problem, improves the convergence speed and optimization stability of the particle swarm algorithm, and avoids the defects of traditional particle swarm algorithm in high-dimensional complex working conditions, such as easy divergence or getting trapped in local optima.

[0016] 2. This invention proposes a particle swarm optimization control method and system for NOx emissions and coal consumption in coal-fired boilers. By dynamically and adaptively adjusting the weights based on the NOx emission change rate and the deviation between the target and the actual load, it achieves dynamic and coordinated adaptation between the NOx emission control target and the coal consumption rate optimization target. It can automatically adjust the optimization focus according to the environmental assessment status and unit operating conditions, solving the problem that traditional fixed-weight optimization cannot adapt to the target requirements of different operating conditions. At the same time, the dynamic inertial weight adjustment based on the unit load change rate improves the response speed of the algorithm under deep peak shaving and rapid load change conditions, effectively avoiding control lag and overshoot.

[0017] 3. This invention proposes a particle swarm optimization control method and system for NOx emissions and coal consumption in coal-fired boilers. By using cross-dimensional Pareto entropy increase perturbation, the diversity of the particle swarm population is improved, avoiding premature convergence problems in the later stages of algorithm iteration, and obtaining a better multi-objective non-dominated solution set. At the same time, by smoothing and verifying control commands and using dynamic time delay compensation, the impact of frequent actions of the actuators is reduced, adapting to the time delay drift characteristics caused by coal quality fluctuations and surface ash in the boiler combustion process, ensuring the long-term stability of closed-loop control, and realizing continuous synergistic optimization of NOx emissions and coal consumption rate in coal-fired boilers. Attached Figure Description

[0018] Figure 1 This is a flowchart of the optimized control method proposed in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0021] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0023] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0024] Example 1: See Figure 1 This embodiment proposes a particle swarm optimization control method for NOx emissions and coal consumption in coal-fired boilers, including the following steps: S100 acquires full operational data from the DCS distributed control system, SIS plant-level monitoring information system, CEMS environmental online monitoring system, and boiler thermal measurement points, including: primary air header pressure, primary air volume, secondary air damper opening, burnout damper opening, coal feeder speed, number of operating coal mills, coal mill current, coal mill outlet temperature, furnace negative pressure, furnace outlet oxygen content, flue gas temperature, fly ash carbon content, NOx emission concentration, main steam temperature, main steam pressure, unit real-time load and load change rate, primary air pressure setpoint, and coal mill output distribution ratio. Data preprocessing is performed on data collected from different systems using a unified time reference. This includes timestamp alignment, elimination of time-series bias in multi-source data, suppression of high-frequency measurement noise from sensors using a moving average filtering method, identification and removal of abrupt and distorted values ​​in the data using the Hampel anomaly detection method, compensation for missing data using linear interpolation or forward imputation, and finally, normalization of the cleaned data to eliminate dimensional differences, resulting in a standard combustion condition dataset composed of continuous time samples.

[0025] The data preprocessing method used here is existing technology and will not be described in detail here.

[0026] S200. Extract combustion operation variables from the standard combustion condition parameter set to construct a high-dimensional combustion state vector, expressed by the formula: ; In the formula, Indicates the sampling time index. This indicates the total number of variables involved in combustion. Indicates the first The high-dimensional combustion state vector of the boiler at each sampling time. Indicates the first The normalized result of the first combustion operation variable at each sampling time. Indicates the first The normalized result of the second combustion operation variable at each sampling time. Indicates the first At the sampling time, the first Normalized results of each combustion operation variable; Furthermore, based on the high-dimensional combustion state vector at each sampling time, an improved isometric mapping algorithm is used for nonlinear dimensionality reduction, specifically: The Euclidean distance between the high-dimensional combustion state vectors at any two sampling times is calculated using the following formula: ; In the formula, , These represent different sampling time indices. Indicates the first , The Euclidean distance between the high-dimensional combustion state vectors at each sampling time. Indicates the first A high-dimensional combustion state vector of the boiler at each sampling time; Constructing based on the Euclidean distance of the high-dimensional combustion state vectors at two sampling times Nearest neighbor graph, where the number of nearest neighbors is... Setting the vectors to 8-15, the shortest paths between vectors are calculated along the manifold structure to obtain the geodesic distance matrix between high-dimensional combustion state vectors, expressed by the formula: ; In the formula, This represents the geodesic distance matrix between high-dimensional combustion state vectors. Indicates the first , The shortest path distance along the manifold structure between the high-dimensional combustion state vectors at each sampling time; It should be noted that the geodesic distance matrix between high-dimensional combustion state vectors Dimensions , This represents the total number of sampling times.

[0027] S201. Furthermore, the geodesic distance matrix between high-dimensional combustion state vectors is centered. Specifically: Construct a centralized matrix, expressed by the formula: ; In the formula, Represents a centered matrix. express The identity matrix, express A column vector of all 1s. Indicates the transpose operation; Based on the centered matrix and the geodesic distance matrix, the inner product matrix is ​​calculated, expressed by the formula: ; In the formula, Denotes the inner product matrix. This represents a matrix composed of the squares of each element in the geodesic distance matrix; Principal Component Analysis (PCA) is used to perform eigenvalue decomposition on the inner product matrix. Eigenvalues ​​represent the variance (i.e., information content) in that direction, and eigenvectors represent the principal component directions, which are orthogonal to each other. The eigenvalues ​​are sorted from largest to smallest, and the eigenvectors corresponding to the top three eigenvalues ​​are selected to construct a low-dimensional combustion condition manifold feature space, expressed by the formula: ; In the formula, Represents the manifold characteristic space for low-dimensional combustion conditions. Indicates combustion intensity characteristics, Indicates the characteristics of uniform air distribution. This indicates fuel adaptability characteristics.

[0028] It should be noted that, based on the principal component loading matrix, the first three principal components are defined as combustion intensity characteristics, air distribution uniformity characteristics, and fuel adaptability characteristics, respectively.

[0029] S300: The low-dimensional combustion condition manifold feature space is input into the improved particle swarm optimization algorithm, including a constraint initialization unit, an evaluation weight construction unit, a particle swarm update unit, and a population perturbation unit, wherein: In the constraint initialization unit, the operating condition coupled dual-domain mapping initialization operation is performed. A physical safety domain is constructed based on the current unit load, the number of operating coal mills, and boiler safety operation constraints, expressed as the formula: The physical safety domain is determined by the unit's operating constraints, including preset upper and lower limits for damper opening, upper and lower limits for coal feeder speed, coal mill output range, main steam temperature and furnace negative pressure boundaries, and actuator operation frequency limits.

[0030] ; In the formula, Represents the physical security domain. Indicates control variables, Indicates the lower bound of the control variable. Indicates the upper bound of the control variable. This indicates the constraints for safe boiler operation; It should be noted that the control variables include the primary air pressure setpoint, the opening degree of the secondary air damper at each level, the opening degree of the burnout air damper, the coal feeder speed, the coal mill output distribution ratio, and the oxygen correction setpoint.

[0031] Within the physical security domain, a uniform sampling method is used to generate the first set of initial particles. Simultaneously, within the chaotic detection domain, an improved Tent chaotic mapping is used to generate the second set of initial particles. The mapping relationship is expressed by the formula: ; In the formula, Indicates the iteration count index. Indicates the first Chaotic variables in the next iteration. Indicates the first Chaotic variables in the next iteration; The particles in the physical safety domain and the particles in the chaotic detection domain are merged to form an initial particle swarm, and a boundary projection operation is performed on all particles to ensure that all initial particles meet the boiler's safe operation boundary.

[0032] S301. In the evaluation weight construction unit, the collaborative evaluation weights of the NOx emission target and the coal consumption rate target are dynamically adjusted based on the real-time NOx emission change rate, the deviation between the preset target load and the unit's real-time load, and the preset environmental constraint coefficient, as expressed by the formula: ; ; In the formula, Indicates the sampling time index. This indicates the weighting of the synergistic evaluation between NOx emission targets and coal consumption rate targets. Indicates the basic weighting coefficient. This represents the environmental protection constraint coefficient. Indicates the rate of change in NOx emissions. This represents the load deviation correction factor. Indicates the amount of load change. Indicates the preset target load. This indicates the real-time load of the generating unit.

[0033] S302. In the particle swarm update unit, construct the particle velocity update equation and adjust the particle swarm search step size, expressed by the formula: ; ; In the formula, Indicates particle index, Indicates a dimension index. Indicates the first In the nth iteration Individual particles 3D particle velocity, Represents dynamic inertia weight. Indicates the first In the nth iteration Individual particles 3D particle velocity, Indicates the first learning factor. express The first uniformly distributed random number within the interval, Indicates the first Individual particles The optimal position of an individual in dimension. Indicates the first In the nth iteration Individual particles The position of the dimension Indicates the second learning factor. express The second uniformly distributed random number within the interval, Indicates that the particle swarm is The globally optimal position of dimension. Indicates the maximum inertia weight. This represents the minimum inertia weight. This represents the absolute value of the rate of change of unit load. This represents the absolute value of the unit's maximum load change rate; The particle position is updated based on the updated velocity, expressed by the formula: ; In the formula, Indicates the first In the nth iteration Individual particles The position of the dimension; The result is a new generation of candidate particle swarms, where the position of each candidate particle represents a candidate combustion control scheme in the low-dimensional combustion condition manifold feature space.

[0034] S303. In the population perturbation unit, the multi-objective fitness value of each candidate particle in the next-generation candidate particle swarm is calculated based on the NOx emission target, coal consumption rate target, main steam temperature deviation, and furnace negative pressure fluctuation, expressed by the formula: ; In the formula, Indicates the multi-objective fitness value. Indicates NOx emission targets, Indicates the coal consumption rate target. Indicates the main steam temperature deviation. This indicates fluctuations in the negative pressure within the furnace. Based on the obtained multi-objective fitness values, perform non-dominated sorting on the next generation of candidate particle swarms to obtain the Pareto non-dominated solution set and calculate the crowding distance; The population distribution entropy in the Pareto non-dominated solution set is calculated using the following formula: ; In the formula, Represents the population distribution entropy. Indicates grid index, This indicates the number of grid cells in the Pareto non-dominated solution set. Indicates the first The proportion of particles in each grid to the total number of particles; When the population distribution entropy is lower than a preset entropy threshold, it indicates insufficient population diversity, triggering a cross-dimensional Pareto entropy increase perturbation operation. This operation applies a Gaussian perturbation to the top K% of particles with the lowest crowding distance in the first Pareto non-dominated solution set, expressed by the formula: ; In the formula, Indicates the Gaussian perturbation after the first Individual particles The new particle position in dimension, Indicates the Gaussian perturbation before the first Individual particles The position of the particle in dimension This represents the standard deviation of the Gaussian perturbation. Represents a standard normal random variable; Alternatively, a reverse learning operation can be performed, expressed as a formula: ; In the formula, Indicates the first step after reverse learning Individual particles The new particle position in dimension, Indicates the first Lower bound of dimensional variable Indicates the first Upper bound of a dimensional variable.

[0035] For the Pareto non-dominated solution set, combined with the number of operating coal mills Upper limit of coal supply Environmental frequency regulation constraints and boiler safety boundary Feasibility screening yields a set of feasible constraints, expressed as a formula: ; In the formula, Indicates the number of coal mills in operation. This indicates the upper limit of coal supply. This indicates environmental protection-related frequency regulation constraints. This indicates the boiler's safety boundary.

[0036] upper limit of coal supply Expressed as a formula: ; In the formula, This indicates the upper limit of coal supply. This indicates the number of coal mills currently in operation. This indicates the maximum output of a single coal mill. This indicates the output margin that the coal mill can actually operate stably.

[0037] Furthermore, the distances from each non-dominated solution to the positive and negative ideal solutions are calculated, expressed by the following formula: ; ; In the formula, Indicates the index of non-dominated solutions. Indicates the index of evaluation indicators. Indicates the first The distance from a non-inferior solution to the ideal solution. Indicates the first The distance from a non-dominated solution to a negative ideal solution. This represents the total number of evaluation indicators. Indicates the first The non-inferior solution is at the th . The corresponding value on each evaluation indicator Indicates the first The positive ideal solution for each evaluation index Indicates the first The negative ideal solution of each evaluation index; Calculate the relative proximity based on the distances from each non-dominated solution to the positive and negative ideal solutions: ; In the formula, Indicates the index of non-dominated solutions. Indicates the first The degree of relative proximity of a non-dominated solution index; The non-dominated solution with the largest relative progress is selected as the optimal combustion control instruction set for the current moment; the optimal combustion control instruction set includes the primary air pressure setpoint, the opening degree of the secondary air damper at each level, the opening degree of the burnout air damper, the coal feeder speed, the coal mill output distribution ratio, and the oxygen correction setpoint.

[0038] S400. Input the control commands from the optimal combustion control command set to the second-order rate limiter to limit the rate of change of the control commands. The transfer function of the second-order rate limiter is expressed by the following formula: ; In the formula, This represents the transfer function of a second-order rate limiter. Represents the first time constant. This represents the second time constant.

[0039] The original control commands are filtered using a sliding window smoothing algorithm, expressed by the following formula: ; In the formula, Indicates the smoothed first... Control commands for each control cycle Represents the smoothing coefficient. Indicates the first The original control commands for each control cycle, Indicates the smoothed first... Control commands for each control cycle; A logical consistency check is performed on the smoothed control command, and the final control command value is obtained after passing the check.

[0040] It should be noted that the logic consistency verification includes: damper opening upper and lower limit verification, coal feeder speed upper and lower limit verification, coal mill output distribution verification, burnout damper opening and low-NOx combustion logic verification, main steam temperature and furnace negative pressure safety boundary verification, actuator operation frequency verification, and DCS control authority verification. Taking damper opening verification as an example, when the control command exceeds the equipment's preset allowable range, it is triggered to be truncated and rolled back to the nearest feasible boundary. The other verification items are all performed according to the general verification process of boundary detection, rollback correction, and authority confirmation. When any verification item fails to meet the requirements, the control command is limited, held, or rolled back.

[0041] S401: Based on the OCDA or OPCUA communication protocol, the smoothed and verified control commands are sent to the DCS system. The DCS system drives the damper actuator, coal feeder frequency converter, coal mill control unit, burnout damper actuator, and related auxiliary control equipment according to control commands; it establishes a status feedback loop to collect real-time data on actuator position feedback, actual air volume, actual coal volume, NOx emission concentration, flue gas oxygen content, furnace negative pressure, main steam temperature, and unit load to obtain the actual feedback value of the actuator and calculate the control execution deviation, expressed by the formula: ; In the formula, Indicates control execution deviation. Indicates the final control command value. This indicates the actual feedback value from the implementing agency; When the control execution deviation exceeds the preset threshold, the corresponding actuator is marked as abnormal, and the optimization instructions of that channel are frozen in the next control cycle, while an alarm message is output.

[0042] S402. To address the large time delay characteristics of the boiler combustion process, a dynamic time warping algorithm is used to evaluate the dynamic time delay between control commands and system responses. Specifically: Two sets of parameter sequences are constructed, including a reference sequence and an observation sequence, wherein: The reference sequence is expressed by the formula: ; In the formula, This indicates the reference sequence after the control command is issued for a coal-fired boiler under nominal operating conditions. This represents the total number of sampling times in the reference sequence. They represent the first The response value at each sampling time; The observation sequence is expressed by the formula: ; In the formula, This represents the observation sequence after the control command is issued for a coal-fired boiler under the current operating conditions. This represents the total number of sampling times in the observation sequence. They represent The response value at each sampling time; Furthermore, due to the time delay drift under the current operating conditions, the time axes of the observed and reference sequences are misaligned and have different lengths, making it impossible to directly calculate the time delay using a simple time difference. Therefore, the Dynamic Time Warping (DTW) algorithm is used to find the optimal time alignment between the two sets of sequences through dynamic programming, aiming to minimize the cumulative cost and solve for the time delay deviation. Specifically: Construction dimension The cost matrix is ​​obtained by iterating through the point pairs of the reference sequence and the observation sequence respectively, and calculating the Euclidean distance of the point pairs as the elements in the cost matrix. The cost matrix completely covers the similarity of all point pairs in the two sequences. The normalized path is a continuous, monotonic path from the top left corner to the bottom right corner of the cost matrix, where each point on the path represents a point pair aligned between the reference sequence and the observed sequence. It should be noted that the continuous, monotonic path must meet the following conditions: Continuity: You can only move one step to the right, down, or down to the right at a time; Monotonicity: The time sequence is not reversed, ensuring the physical rationality of alignment.

[0043] The optimal regularized path is the path that minimizes the sum of all local costs on the regularized path, expressed by the formula: ; In the formula, Represents the reference sequence With observation sequence The minimum cumulative cost, Indicates a regular path. This represents the index of matching point pairs on the normalized path. Indicates the first normalized path The local cost of a pair of matching points. This represents the total number of matching point pairs on the regularized path; It should be noted that the first step on the regularization path... Local cost of a matching point pair That is, the cost matrix elements are formed by the Euclidean distance between corresponding point pairs in the cost matrix.

[0044] Furthermore, in the optimal regularization path, the time between the reference sequence point and the observed sequence point is used as the time misalignment between the two sets of sequences. Combined with the time base of the nominal operating condition, the actual time delay of the current operating condition is calculated, expressed by the formula: ; In the formula, Indicates the actual time delay under the current operating conditions. This indicates the nominal time delay of the boiler combustion process. Indicates the sampling interval. Indicates the first One reference sequence point, Indicates the first The first One observation sequence point; When the deviation between the estimated actual time delay and the nominal time delay exceeds a preset threshold (e.g., the deviation exceeds ±10%), it indicates that the time delay of the combustion process has drifted significantly, and the original particle swarm learning factor cannot adapt to the current operating conditions, requiring correction. This is expressed by the formula: ; In the formula, This represents the corrected particle swarm learning factor. This represents the particle swarm learning factor before correction. Indicates the time delay compensation coefficient; It should be noted that both the modified and unmodified particle swarm optimization learning factors correspond to the first learning factor. Second learning factor The corrected particle swarm learning factor is fed back to the dynamic particle swarm optimization algorithm as the parameter configuration for the next round of rolling optimization.

[0045] To verify the effectiveness of the method described in this invention, an experiment was conducted, as follows: The experimental platform utilizes historical operating data and a simulation testing platform from a 600MW supercritical coal-fired power unit. The data includes one year's worth of DCS historical operating records, CEMS environmental data, coal mill operating data, flue gas system regulation data, and unit load scheduling data. Test scenarios include steady-state high-load conditions, low-load deep peak-shaving conditions, rapid load increase conditions, rapid load decrease conditions, coal quality change conditions, and environmental assessment conditions. This experiment selected the conventional fixed-parameter PSO algorithm, the adaptive weight PSO algorithm, and the multi-objective genetic algorithm NSGA-II as the baseline. Evaluation metrics included NOx control accuracy, coal consumption optimization magnitude, algorithm convergence generation, and actuator operation frequency. The experimental results are presented in the table below: Table 1 Comparison of Experimental Results

[0046] Experimental results show that the proposed solution outperforms the control solution in terms of NOx control accuracy, coal consumption optimization range, convergence speed, and actuator operation frequency. Specifically, the proposed solution achieves a NOx control accuracy of 94.5%, a coal consumption optimization range of 1.15%, a convergence generation of 45 generations, and reduces the actuator operation frequency to 12 times per hour, fully demonstrating that the proposed solution can achieve synergistic optimization of NOx emissions and coal consumption rate under complex and variable operating conditions.

[0047] Example 2: This embodiment proposes a particle swarm optimization control system for NOx emissions and coal consumption in a coal-fired boiler, comprising the following modules: The data acquisition and preprocessing module is used to acquire the operating parameters of the coal-fired boiler and perform standardized preprocessing to obtain a set of standard combustion condition parameters that eliminate the influence of noise and dimensions. The high-dimensional feature extraction and nonlinear dimensionality reduction module extracts combustion operation variables from the standard combustion condition parameter set, constructs a high-dimensional combustion state vector, and performs nonlinear dimensionality reduction on the high-dimensional combustion state vector through an improved isometric feature mapping to obtain a low-dimensional combustion condition manifold feature space that characterizes the core characteristics of the combustion condition. The multi-objective collaborative optimization and command output module is used to input the low-dimensional combustion condition manifold feature space into the improved particle swarm optimization algorithm. In the particle swarm optimization algorithm, each particle corresponds to a set of candidate combustion control parameters, and the position of the particle in the low-dimensional manifold feature space represents the combustion condition corresponding to the control scheme. With NOx emission and coal consumption rate as optimization objectives, multi-objective Pareto optimization is performed to obtain the optimal combustion parameter control command sequence, which is then sent to the distributed control system of the coal-fired boiler after amplitude limiting and smoothing. The actuator drive control module is used in the distributed control system of a coal-fired boiler to drive the actuator to act according to the optimal combustion parameter control command and adjust the boiler combustion control parameters. The time delay evaluation and algorithm parameter correction module is used to collect combustion state parameters after the actuator of the coal-fired boiler operates in real time. It uses a dynamic time warping algorithm to align the optimal combustion parameter control command sequence with the system state response sequence and calculates the time delay deviation between the two. Based on the time delay deviation, the parameters of the improved particle swarm optimization algorithm are updated online to form a closed-loop rolling optimization.

[0048] Example 3: This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described in any embodiment of the present invention.

[0049] Example 4: This embodiment proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in any embodiment of the present invention.

[0050] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0051] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0052] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0053] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0054] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A particle swarm optimization control method for NOx emission and coal consumption of a coal-fired boiler, characterized in that, Includes the following steps: The operating parameters of the coal-fired boiler are obtained and standardized preprocessing is performed to obtain a set of standard combustion condition parameters that eliminate the influence of noise and dimensions. Combustion operation variables are extracted from the standard combustion condition parameter set to construct a high-dimensional combustion state vector. Then, nonlinear dimensionality reduction is performed on the high-dimensional combustion state vector through an improved isometric feature map to obtain a low-dimensional combustion condition manifold feature space that characterizes the core characteristics of the combustion condition. The low-dimensional combustion condition manifold feature space is input into the improved particle swarm optimization algorithm. In the particle swarm optimization algorithm, each particle corresponds to a set of candidate combustion control parameters, and the position of the particle in the low-dimensional manifold feature space represents the combustion condition corresponding to the control scheme. With NOx emission and coal consumption rate as optimization objectives, multi-objective Pareto optimization is performed to obtain the optimal combustion parameter control command sequence, which is then sent to the distributed control system of the coal-fired boiler after amplitude limiting and smoothing. The distributed control system for coal-fired boilers drives the actuators to adjust the boiler combustion control parameters based on the optimal combustion parameter control commands. Combustion state parameters after the actuator of the coal-fired boiler is activated in real time. The optimal combustion parameter control command sequence and the system state response sequence are aligned using a dynamic time warping algorithm, and the time delay deviation between the two is calculated. The parameters of the improved particle swarm optimization algorithm are updated online based on the time delay deviation to form a closed-loop rolling optimization.

2. The particle swarm optimization control method for NOx emission and coal consumption of a coal-fired boiler according to claim 1, characterized in that, The method utilizes an improved isometric feature mapping to perform nonlinear dimensionality reduction on the high-dimensional combustion state vector. The specific steps are as follows: Combustion operation variables are extracted from the standard combustion condition parameter set to construct a high-dimensional combustion state vector; Calculate the Euclidean distance between the high-dimensional combustion state vectors at any two sampling times, set a number of nearest neighbors, and construct... Nearest neighbor graph, as a manifold structure characterizing combustion conditions; The shortest path between high-dimensional combustion state vectors is calculated along the manifold structure to obtain the geodesic distance matrix between the high-dimensional combustion state vectors; The geodesic distance matrix between high-dimensional combustion state vectors is centered by constructing a centered matrix from the identity matrix and all-one column vectors. The inner product matrix is ​​then calculated by squaring the centered matrix and the geodesic distance matrix. Principal component analysis (PCA) is used to perform eigenvalue decomposition on the inner product matrix. The eigenvalues ​​are sorted from largest to smallest, and the eigenvectors corresponding to the top three eigenvalues ​​are selected to construct a low-dimensional combustion condition manifold feature space.

3. The particle swarm optimization control method for NOx emissions and coal consumption of a coal-fired boiler according to claim 1, characterized in that, The improved particle swarm optimization algorithm includes a constraint initialization unit, an evaluation weight construction unit, a particle swarm update unit, and a population perturbation unit, wherein: The constraint initialization unit is used to construct a physical safety domain by combining the preset operating constraints of the current coal-fired unit; the first part of initial particles is generated by uniform sampling within the physical safety domain, and the second part of initial particles is generated by Tent chaotic mapping within the chaotic detection domain; the first and second parts of initial particles are merged to construct the initial particle swarm within the physical safety domain. The evaluation weight construction unit is used to construct the collaborative evaluation weights of NOx emission targets and coal consumption rate targets based on the deviations of real-time NOx emission change rate, target load, real-time load of coal-fired units, and preset environmental protection constraint coefficients. The particle swarm update unit constructs the particle velocity update equation and adjusts the search step size of the initial particle swarm; it updates the particle positions according to the updated velocities to obtain a new generation of candidate particle swarms; where the position of each candidate particle represents a candidate combustion control scheme in the low-dimensional combustion condition manifold feature space. The population perturbation unit calculates the multi-objective fitness value of each candidate particle in the new generation of candidate particle swarm based on preset NOx emission targets, coal consumption rate targets, main steam temperature deviation, and furnace negative pressure fluctuations; performs non-dominated sorting on the new generation of candidate particle swarm based on the obtained multi-objective fitness values, obtains the Pareto non-dominated solution set and calculates the crowding distance; calculates the population distribution entropy in the Pareto non-dominated solution set; when the population distribution entropy is lower than the preset entropy threshold, it indicates insufficient population diversity and triggers the Pareto entropy increase perturbation operation.

4. The particle swarm optimization control method for NOx emissions and coal consumption of a coal-fired boiler according to claim 1, characterized in that, The method updates the initial particle swarm, and the specific steps are as follows: Construct the particle velocity update equation and adjust the particle swarm search step size, expressed as a formula: ； ； In the formula, Indicates particle index, Indicates a dimension index. Indicates the first In the nth iteration Individual particles 3D particle velocity, Represents dynamic inertia weight. Indicates the first In the nth iteration Individual particles 3D particle velocity, Indicates the first learning factor. express The first uniformly distributed random number within the interval, Indicates the first Individual particles The optimal position of an individual in dimension. Indicates the first In the nth iteration Individual particles The position of the dimension Indicates the second learning factor. express The second uniformly distributed random number within the interval, Indicates that the particle swarm is The globally optimal position of dimension. This represents the maximum inertia weight. This represents the minimum inertia weight. This represents the absolute value of the rate of change of unit load. This represents the absolute value of the unit's maximum load change rate; The particle position is updated based on the updated velocity, expressed by the formula: ； In the formula, Indicates the first In the nth iteration Individual particles The position of the dimension; A new generation of candidate particle swarms is obtained, and the position of each candidate particle represents a candidate combustion control scheme in the low-dimensional combustion condition manifold feature space.

5. The particle swarm optimization control method for NOx emissions and coal consumption of a coal-fired boiler according to claim 3, characterized in that, The Pareto entropy increase perturbation operation enhances the diversity of the next-generation candidate particle swarm using Gaussian perturbation. The specific steps are as follows: Calculate the multi-objective fitness value for the new generation of candidate particle swarms, perform Pareto undominated sorting to obtain the undominated solution set and calculate the crowding distance; Calculate the population distribution entropy of the non-dominated solution set. When the entropy value is lower than a preset threshold, it is determined that the population diversity is insufficient, triggering the Pareto entropy increase perturbation operation. Sort all particles in the Pareto nondominated solution set in ascending order of crowding distance, select the first preset percentage of particles as the particles to be perturbed, perform Gaussian perturbation, and obtain the perturbed particle swarm after position update. By merging perturbed and undisturbed particle swarms, a new generation of optimized particle swarms with enhanced diversity is formed.

6. The particle swarm optimization control method for NOx emissions and coal consumption of a coal-fired boiler according to claim 1, characterized in that, The method uses a dynamic time warping algorithm to update the parameters of the improved particle swarm optimization algorithm. The specific steps are as follows: Construct a reference sequence for boiler combustion control command response and an observation sequence for actual state feedback; A cost matrix is ​​constructed using the Euclidean distance between each pair of points in the reference sequence and the observation sequence as elements; By traversing the cost matrix using a dynamic time warping algorithm, the optimal warping path that minimizes the cumulative cost value is found, expressed by the formula: ； In the formula, Indicates the reference sequence. Represents the observation sequence. Represents the reference sequence With observation sequence The minimum cumulative cost, Indicates a regular path. This represents the index of matching point pairs on the normalized path. Indicates the first normalized path The local cost of a pair of matching points. This represents the total number of matching point pairs on the regularized path; The time misalignment between the reference sequence and the observed sequence is determined based on the optimal regularization path, and the actual time delay of the current operating condition is calculated in conjunction with the nominal operating condition time base. The time delay deviation between the actual time delay and the nominal time delay of the boiler combustion process is calculated. When the time delay deviation exceeds a preset threshold, the particle swarm learning factor is corrected based on the time delay deviation and the time delay compensation coefficient, expressed by the formula: ； In the formula, This represents the corrected particle swarm learning factor. This represents the particle swarm learning factor before correction. Indicates the time delay compensation coefficient. Indicates the actual time delay under the current operating conditions. Indicates the nominal time delay of the boiler combustion process; The corrected particle swarm learning factor is obtained and fed back to the improved particle swarm optimization algorithm as the calculation parameter for the next round of iterative optimization. The optimal regularization path is a monotone continuous path that minimizes the cumulative local cost.

7. A particle swarm optimization control system for NOx emissions and coal consumption in a coal-fired boiler, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to acquire the operating parameters of the coal-fired boiler and perform standardized preprocessing to obtain a set of standard combustion condition parameters that eliminate the influence of noise and dimensions. The high-dimensional feature extraction and nonlinear dimensionality reduction module extracts combustion operation variables from the standard combustion condition parameter set, constructs a high-dimensional combustion state vector, and performs nonlinear dimensionality reduction on the high-dimensional combustion state vector through an improved isometric feature mapping to obtain a low-dimensional combustion condition manifold feature space that characterizes the core characteristics of the combustion condition. The multi-objective collaborative optimization and command output module is used to input the low-dimensional combustion condition manifold feature space into the improved particle swarm optimization algorithm. In the particle swarm optimization algorithm, each particle corresponds to a set of candidate combustion control parameters, and the position of the particle in the low-dimensional manifold feature space represents the combustion condition corresponding to the control scheme. With NOx emission and coal consumption rate as optimization objectives, multi-objective Pareto optimization is performed to obtain the optimal combustion parameter control command sequence, which is then sent to the distributed control system of the coal-fired boiler after amplitude limiting and smoothing. The actuator drive control module is used in the distributed control system of a coal-fired boiler to drive the actuator to act according to the optimal combustion parameter control command and adjust the boiler combustion control parameters. The time delay evaluation and algorithm parameter correction module is used to collect combustion state parameters after the actuator of the coal-fired boiler operates in real time. It uses a dynamic time warping algorithm to align the optimal combustion parameter control command sequence with the system state response sequence and calculates the time delay deviation between the two. Based on the time delay deviation, the parameters of the improved particle swarm optimization algorithm are updated online to form a closed-loop rolling optimization.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 6.

Images