CN-121998380-A - AI-driven thermal power plant combustion stability-economic benefit dual-target intelligent coal blending method

CN121998380ACN 121998380 ACN121998380 ACN 121998380ACN-121998380-A

Abstract

The invention discloses an AI-driven dual-target intelligent coal blending method for combustion stability and economic benefit of a thermal power plant, which relates to the technical field of coal blending optimization of the thermal power plant, and comprises the steps of constructing a coal-covered type combustion fingerprint characteristic database; the method comprises the steps of inputting combustion fingerprint feature data of candidate coal blending schemes into a boiler combustion risk prediction model based on a weighted support vector machine to obtain corresponding combustion risk probability values, constructing a double-objective optimization function by taking a coal blending proportion as an optimization variable, adopting a penalty function to carry out relaxation treatment on preset constraint conditions, enabling the double-objective optimization function to comprise an economy function taking fuel purchasing cost as a target and a safety function taking the combustion risk probability value as a target, carrying out Pareto optimization on the economy function and the safety function based on a multi-objective evolutionary algorithm to output a Pareto optimal coal blending scheme set, and executing the coal blending scheme. Therefore, the cooperative optimization of the combustion safety of the boiler and the fuel purchasing economy is realized.

Inventors

LI QINGSONG
REN FUQIANG
WANG CHUNHUA
SONG XIN
ZHANG SHUJIN
ZHANG PENG

Assignees

贵州省矿山安全科学研究院有限公司
贵州省煤矿设计研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

An ai-driven thermal power plant combustion stability-economic benefit dual-target intelligent coal blending method is characterized by comprising the following steps: Constructing a coal burning fingerprint characteristic database containing basic coal quality parameters, burning intrinsic characteristic parameters and safety boundary parameters; inputting combustion fingerprint feature data corresponding to a candidate coal blending scheme into a pre-constructed boiler combustion risk prediction model based on a weighted support vector machine to obtain a combustion risk probability value of the candidate coal blending scheme; Constructing a double-objective optimization function by taking a coal blending proportion as an optimization variable, and adopting a penalty function to relax a preset constraint condition, wherein the double-objective optimization function comprises an economic function taking fuel purchasing cost as a target and a safety function taking the combustion risk probability value as a target; performing Pareto optimization on the economic function and the safety function based on a multi-objective evolutionary algorithm, and outputting a Pareto optimal coal blending scheme set; and selecting and executing the coal blending scheme from the Pareto optimal coal blending scheme set according to the actual working condition of the thermal power generating unit.
2. The method of claim 1, wherein constructing a coal combustion fingerprint database comprising base coal quality parameters, combustion intrinsic characteristic parameters, and safety margin parameters comprises: Collecting basic coal quality parameters, safety boundary parameters and combustion intrinsic characteristic parameters obtained through thermogravimetric analysis of single coal and mixed coal, and preprocessing various collected parameters to obtain a standardized characteristic parameter set; If the coal type which does not acquire the thermogravimetric analysis data exists, predicting the combustion intrinsic characteristic parameter by utilizing a pre-trained regression auxiliary model based on the basic coal quality parameter of the coal type, and taking the predicted value as the combustion intrinsic characteristic parameter of the coal type; and integrating the normalized characteristic parameter set and the predicted combustion intrinsic characteristic parameter to form a combustion fingerprint characteristic database.
3. The method of claim 1, wherein the weighted support vector machine based boiler combustion risk prediction model is constructed by: Acquiring a training sample set, wherein a boiler combustion high-risk abnormal working condition is taken as a positive sample, and a normal stable combustion working condition is taken as a negative sample; Setting corresponding weighting factors for the positive class samples and the negative class samples respectively, wherein the weighting factors are determined by the ratio of the number of the negative class samples to the number of the positive class samples; and training to obtain a boiler combustion risk prediction model based on a weighted support vector machine by taking combustion fingerprint characteristic data and boiler operation boundary parameters of a coal blending scheme as input characteristics and combustion working condition types as output labels.
4. A method according to claim 3, wherein the weighting factor is determined by a ratio of the number of negative class samples to the number of positive class samples, comprising: determining the ratio of the weighting factors of the positive class samples to the weighting factors of the negative class samples according to the ratio of the number of the negative class samples to the number of the positive class samples; Based on the ratio of the weighting factors of the positive class samples to the weighting factors of the negative class samples, a specific value of the weighting factor is determined by cross-validation in a range of values from 0 to 1.
5. The method of claim 1, wherein the economic objective function The method comprises the following steps: ; safety objective function The method comprises the following steps: ; Wherein, the Optimizing a variable vector of a coal blending scheme, First of all The blending ratio of the seed coal meets the non-negative constraint And normalizing the constraint ; The total amount of coal types participating in the blended combustion, Is the first Purchasing price per unit mass of seed coal; Scheme for blending coal The high-risk abnormality occurrence probability of (1) the superscript T indicates the row vector Transposed into column vectors.
6. The method of claim 5, wherein relaxing the preset constraint using a penalty function comprises: For a coal blending scheme violating preset constraint conditions, calculating superscalar of each constraint condition and comparing the superscalar with a preset threshold value; If the superscalar exceeds a preset threshold, directly eliminating the coal blending scheme in the population evolution process; If the superscalar does not exceed the preset threshold, determining a penalty term according to the superscalar and a preset penalty coefficient, and calculating the penalty term into an economic function, wherein the corrected economic function is used as an optimization target.
7. The method of claim 1, wherein Pareto optimizing the economic and safety functions based on a multi-objective evolutionary algorithm, outputting a Pareto optimal coal blending solution set, comprising: Initializing to generate a population containing a plurality of coal blending schemes based on a multi-target evolutionary algorithm, wherein each coal blending scheme consists of blending combustion proportions of various coals; calculating fuel purchasing cost values and combustion risk probability values of each individual in the population according to the economic function and the safety function, and correcting the individuals violating constraint conditions based on the penalty function to obtain fitness values of the individuals; non-dominant ranking is carried out on individuals in the population, pareto grades are divided, and crowding degree is calculated on individuals in the same grade; Generating a child population through selection, crossing and mutation operations, merging a parent population with the child population, and then carrying out non-dominant sorting and crowding degree screening again to generate a new generation population; and performing iteration until the preset maximum iteration times or the Pareto front of the population is reached, converging, and outputting a Pareto optimal coal blending scheme set.
8. The method of claim 3, further comprising, after selecting and executing a coal blending scheme from the Pareto optimal coal blending scheme set according to an actual operating condition of the thermal power plant: collecting boiler operation data in real time, and comparing the boiler operation data with a combustion risk probability value to generate an increment labeling sample; When the accumulated number of the increment labeling samples reaches a preset number threshold, merging the increment labeling samples with the training sample set to obtain a merged training sample set; And retraining the boiler combustion risk prediction model based on the weighted support vector machine according to the combined training sample set so as to update the parameters of the optimization model.
9. The method according to claim 1, wherein the method further comprises: When new coal is put in a coal yard, the price fluctuation of a coal market exceeds a preset range or the running condition of a thermal power generating unit is changed greatly, restarting the optimizing flow based on the multi-objective evolutionary algorithm to generate an updated Pareto optimal coal blending scheme set.
10. The method of claim 1, wherein the weighted support vector machine-based boiler combustion risk prediction model is a weighted C-support vector machine or a weighted -A support vector machine.

Description

AI-driven thermal power plant combustion stability-economic benefit dual-target intelligent coal blending method Technical Field The embodiment of the invention relates to the technical field of coal blending optimization of thermal power plants, in particular to an AI-driven dual-target intelligent coal blending method for combustion stability and economic benefit of a thermal power plant. Background The coal blending of the thermal power plant is a key link of safe and stable operation of the boiler and reduction of fuel cost, and the optimization effect directly influences the safe production and economic benefit of the thermal power plant. The current coal blending technology of a thermal power plant still has a plurality of core defects, and is difficult to adapt to the high-quality development requirements of the power industry, and the technology specifically comprises the following steps: (1) The combustion risk prediction precision is insufficient, the prior art mainly adopts a standard SVM to predict the combustion risk of the boiler, but the working condition data of the thermal power plant has serious class imbalance, the normal combustion sample accounts for more than 95%, high-risk abnormal samples such as slag formation, stable combustion failure and the like are minority, the standard SVM adopts the same misclassification punishment for all samples, the problem that classification is biased to majority is easy to occur, the recognition accuracy of the high-risk abnormal samples is low, the failure report rate is high, and the safety management and control requirements of the power plant cannot be met. (2) The timeliness bottleneck exists in coal quality data acquisition, namely the existing coal blending optimization depends on intrinsic characteristic parameters of coal combustion obtained by thermogravimetric analysis (TGA), but the TGA detection is long in time consumption, the characteristic change of newly-entered coal types cannot be tracked in real time, the model training and the coal blending optimization are poor in timeliness, and dynamic coal blending is difficult to realize. (3) The nonlinear effect is not considered in coal blending prediction, namely the traditional coal blending method derives the combustion characteristic of the mixed coal based on the linear weighted average assumption, and the nonlinear combustion effect after blending combustion of different coal types is not considered, so that the prediction deviation of the combustion characteristic is large, and the actual execution effect of the coal blending scheme is greatly different from the theoretical prediction. (4) The coal blending optimization is difficult to achieve both safety and economy, the existing coal blending optimization is mostly single-target optimization, or a double-target is set but a hard constraint condition is adopted, a non-dominant ranking genetic algorithm (NSGA-II) is easy to cause no feasible solution due to the fact that a feasible region is too narrow during optimizing, meanwhile, the problem of extreme decision making of heavy economy, light safety or heavy safety and light economy exists, and the cooperative optimization of the two cannot be achieved. (5) The model lacks a dynamic adaptation mechanism, namely the existing coal blending model is a static model, is not dynamically updated according to actual operation data of a boiler, coal variety change and market price fluctuation, and the prediction accuracy and the optimization effect of the model can be continuously reduced along with unit aging and coal variety replacement, so that the engineering floor property is poor. In the prior art, although some researches try to apply the weighted SVM to unbalanced data classification and apply NSGA-II to coal blending optimization, the practical working conditions of coal blending of a thermal power plant are not combined for targeted improvement, and the problems cannot be fundamentally solved. Therefore, developing an intelligent coal blending method for a thermal power plant, which gives consideration to combustion risk prediction precision, data instantaneity, safety and economic collaborative optimization and dynamic adaptation worker Cheng Gongkuang, becomes an urgent need in the current field. Disclosure of Invention Aiming at the core defects of the existing coal blending technology of the thermal power plant, the invention provides the AI-driven dual-target intelligent coal blending method for the combustion stability and economic benefit of the thermal power plant, which is used for solving the problems of low recognition accuracy and high missing report rate of few high-risk abnormal samples such as slag formation, stable combustion failure and the like when a standard Support Vector Machine (SVM) model faces to the imbalance of working condition data types in the prior art, and simultaneously breaking through the limitation of linear assumption of the traditional coal ble