Deep Learning And Intelligent Optimization For Load Forecasting And Allocation In Smart Grids

As the comprehensive construction of uhv ac/dc interconnected power grid,and photovoltaic,wind power and the other new energy,as well as electric vehicles,energy storage system and the other new equipment access,and electric power market reform,which increase the complexity of the smart grid,the resulted huge amounts of electricity data and complex power grid environment make the traditional modeling and optimization methods face many challenges,leading to the deep learning and intelligent optimization and the other artificial intelligence technologies have been increasingly used in smart grid.This thesis studies the load forecasting and allocation problems in smart grid.The main works are as follows:In the first chapter,we introduce the background knowledge of smart grid,as well as the overseas and domestic research status in the field.Then,the motivation and derivation of the main work is presented.In the second chapter,the main tools and methods used in this paper are introduced,including long short-term memory network and estimation of distribution algorithms.In the third chapter,we study the short-term power load forecasting in smart grid.In order to deal with some issues existing in the short-term power load forecasting,such as low prediction ac-curacy,short prediction period and difficult adjustment of prediction parameters,we put forward a long short-term memory network algorithm which integrates with self-organizing maps,chaotic time series and combinatorial intelligence optimization respectively to improve the prediction accuracy,prolong the predictable period,and optimize the network parameters.Finally,the effectiveness of the proposed algorithm is verified by numerical simulation.In the fourth chapter,we study the load allocation and dispatching problem in smart grid.For a class of economic load allocation scheduling problems with a non-convex objective and a non-linear equality constraint,we introduce the variable correlation analysis and dominant probability,and propose the effective correlation estimation of distribution algorithms based on the concept of tracking solution for improving the optimization efficiency.Furthermore,we introduce the con-cept of redundant correlation,and propose the estimation of distribution algorithms with reduced redundancy correlation in combination with manifold learning algorithm to deal with a class of large-scale environmental economic load allocation and dispatching problems with non-convex dual objectives and a linear equality constraint.Finally,the effectiveness of the two algorithms is verified by numerical simulation.In the last chapter,we make a summary and propose the relevant research work in the future.