Research On Power Generation Cost Optimization Based On Mid-and Short-term Electricity Coal Price Forecast

With the deepening of China’s supply-side reform and electricity market reform,and the continuous production of large-capacity nuclear power and new energy power generation,coal-fired power generation companies are struggling.As the most important cost of coal-fired power generation enterprises,the control of coal cost is of great significance to improve the competitiveness of power generation companies.In this paper,research on the optimization of coal cost in China is carried out,combined with actual optimization objectives and constraints,and a coal cost optimization method is established from the perspective of spot coal price forecasting.Traditional coal cost optimization methods mainly focus on a single power generation process,and seldom coordinate and optimize each coal cost link from the perspective of coal control.Moreover,the study lacks the dynamism of the coal cost optimization process.In the decision-making process,it is necessary to carry out unified planning for multiple decision-making cycles,and also take into account the fluctuation of spot coal price in the coal procurement process.Only considering the unified long-term coal price cannot reflect the changes in the thermal coal market environment.The research content of this paper is as follows:First,from the perspective of coal control of power generation enterprises,the form and relationship of coal cost in the procurement process and blending process are analyzed,as well as the role of coal price forecast in coal cost optimization.Second,in view of the perceptible demand for coal cost optimization for electricity coal prices,a medium and short-term electricity coal price prediction model was constructed.Aiming at the problem of missing feature sequences in predicting future time series,a feature sequence transformation method is established based on the law of historical data price changes.In order to establish a non-linear mapping between features and targets,the main influencing factors of medium and short-term coal prices are determined based on the piecewise chi-square test and correlation analysis.The LSTM and PCA-LSTM methods are introduced to realize the medium and short-term electricity coal price forecasts.The effectiveness of the proposed model is proved by several examples.Third,in view of the problem of predicting performance fluctuations in the rolling forecast of the single-layer thermal coal price prediction model,the Stacking ensemble model is introduced to optimize the thermal coal price forecast.By analyzing the predicted performance and difference of candidate learners in typical months,the framework of ensemble learning model is determined.By comparing the performance of ensemble learning and single learner rolling prediction,and analyzing the structure of ensemble learning model,the effectiveness of the proposed multi-agent ensemble learning coal price prediction model is proved.Finally,in order to optimize the multi-cycle decision of coal burning cost from the perspective of coal burning control,based on the ensemble learning of spot coal price prediction,this paper constructs a double-layer optimization model of coal burning cost according to the procurement and blending processes.In the aspect of solving the double-layer optimization model,the applicability of its neighborhood search method is improved on the basis of the traditional artificial bee colony algorithm,so that it has the ability to solve this problem.The calculation example of the continuous decision-making of planned power generation tasks shows that the external input model of spot coal prediction can achieve good results in multiple scenarios.Moreover,it demonstrated that the double-layer optimization model coordinates procurement optimization and reasonable coal ratios,which reduces the cost of coal-fired power generation to a certain extent and improves the competitiveness of coal-fired power generation companies.