Research On Optimal Scheduling Strategy Of Regional Integrated Energy System Based On Joint Source-Load Forecasting

With the growing importance of energy and environmental challenges in recent years,the development of an efficient and low-carbon energy system has become critical.The integrated energy systems can achieve the organic fusion of multiple energy sources,promote the organic consumption of high-proportion renewable energy,meet the diverse energy needs of users,enhance the overall energy utilization efficiency,and have become a new direction for energy system development.As one of the typical scenarios for the integrated application of energy,the regional integrated energy system(RIES)has diverse energy types,varied operating conditions,and random changes in source-load,making it a research hotspot in the field to develop multi-time scale scheduling and operation strategies based on accurate source-load forecasting.However,the randomness of renewable energy output and the uncertainty of multi-dimensional load demand greatly increase the complexity of the optimization and scheduling problem of the RIES,making traditional optimization and scheduling methods inadequate.Therefore,this thesis proposes a multi-scale optimization and scheduling strategy for regional integrated energy systems based on joint prediction of sources and loads.Firstly,to reduce the impact of source and load randomness,the coupling between the "horizontal and vertical" directions of source-load is effectively explored by proposing a joint prediction model for photovoltaic and multi-dimensional load.The model uses the Maximum Information Coefficient(MIC)to select input factors,Temporal Convolutional Networks(TCN)to efficiently store long-term data,Multi-head Attention Mechanism(MHA)to allocate weights to important information,and Multi-task Learning(MTL)to achieve joint prediction.The experimental results show that the MTMH-MTL model that considers the "horizontal and vertical" coupling relationship between source and load has an average forecasting error reduction of 31.38% compared to other traditional models and an average training time reduction of 35.23%.This model effectively solves the problem of inadequate accuracy of traditional forecasting methods and has a faster running efficiency.Secondly,to further reduce the disadvantage of point forecasting error that cannot be completely eliminated,a multi-time scale optimization and scheduling model is established.The model uses short-term forecasting data as input to obtain the scheduling plan for the day;and on the basis of the daily plan,uses ultra-short-term forecasting results to implement rolling scheduling within the day to mitigate the impact of forecasting errors and power fluctuations.Each unit can adjust its output during the day according to the pre-scheduled power output plan.The calculation results show that the proposed multi-time scale optimization and scheduling strategy can more flexibly reduce the impact of randomness on system operation.Compared with traditional optimization and scheduling methods,the proposed method in this thesis reduces the economic index by 6.08%,greatly improves the energy utilization efficiency of RIES,guarantees the dynamic needs of users,and enhances the stability of energy use.Finally,based on the B/S framework and combined with the My SQL database,this thesis designs and develops the intelligent information management platform system for the regional integrated energy system.The platform system includes modules such as user login,source-load forecasting,optimization and scheduling,and personnel management.This platform is applied to a smart demonstration park in Beijing,China,and can facilitate the park management to further understand the operating conditions of the system units and the energy consumption trends of users,thereby improving the management level of the system,promoting the efficient use of energy.