Research On The Power-load Forecasting Method Of Park Integrated Energy System Considering Complex Correlation

In order to solve the contradiction between the growth of energy demand and the low stock of fossil energy and environmental pollution in the current stage of social development,energy transformation is imminent.How to effectively improve energy utilization is the development problem of current energy transformation.In this context,an integrated energy system integrating various renewable energy sources at the "power" end and providing multiple energy supplies at the "load" end came into being,which greatly improved the energy utilization rate.But at the same time,the uncertainty and complexity of the system have also doubled.On the one hand,the output of renewable energy has the problems of strong randomness and low prediction accuracy.On the other hand,the existing single load forecasting method is difficult to deal with the complex coupling correlation between multiple loads,and the power-load energy management is facing new challenges.In view of this,this paper studies the power-load end photovoltaic power and multivariate load forecasting.The main research contents and conclusions are as follows:(1)This paper analyzes and summarizes the basic concepts and energy consumption characteristics of integrated energy system,and combs and analyzes different energy utilization modes and characteristics with multi energy coupling typical integrated energy system architecture.At the source end,the dynamic correlation between photovoltaic output and meteorological factors is analyzed and quantified by Pearson correlation coefficient.At the charge end,the multi time scale characteristics of multiple loads of electricity,heat and cooling are analyzed,and the coupling correlation of multiple loads is studied.(2)A photovoltaic power prediction model considering multi-stage feature correlation is proposed.Based on the codec,aiming at the dynamic and complex correlation between photovoltaic power output and multi-dimensional meteorological factors,the model introduces three mechanisms:target attention,global attention and time attention to adaptively select the relevant state and obtain the key information.The example simulation shows that the model can mine the complex correlation between photovoltaic output and meteorological factors independently.Compared with the traditional model,it has higher prediction accuracy and application effect.(3)A multivariate load forecasting model based on multi task learning and improved LSTM is proposed.The model selects the input features based on the maximum information coefficient on the input side;Multi task learning mechanism is adopted to integrate different types of energy forecasting tasks,and the coupling information provided by different load forecasting tasks is learned by sharing mechanism;The traditional LSTM structure is improved to improve the ability of the model to extract coupling features.The example simulation shows that the model can effectively improve the information density of input sequence and make full use of coupling information to improve the prediction accuracy.Compared with the traditional model,the model has higher prediction accuracy.