Research On The Method And Application Of Load Forecasting On The User Side Of Smart Grid

With the development of technology and the improvement of users’ living standards,the environmental problems caused by fossil energy are increasingly steep.Renewable energy combined with information technology has developed rapidly.The concept of energy Internet has been proposed and developed rapidly.The power grid is an important part of the energy internet.The traditional power network can no longer meet the needs of users,as a new way of improving information sharing and smart power distribution.Smart grid has been developed vigorously.The key element of smart grid is user-side power distribution,so accurate user-side load prediction is important to the rational operation and economic dispatch of smart grid.This paper takes the smart grid user-side load forecasting as the research background,takes the design of demand-based forecasting model as the research target,and puts forward two ideas to enhance the forecasting performance of the model: data preprocessing framework and integrated model.First,in order to enable the model to obtain information that is not easily explored between data,this paper presents a data preprocessing framework based on empirical mode decomposition and combines it with extreme learning machine to verify its validity.Then,in order to obtain multiple deep belief networks with good performance but different from each other,an unsupervised search for appropriate deep belief networks based on decomposition-based multi-objective optimization method is proposed.Finally,a multi-target deep belief network based on empirical mode decomposition is proposed by integrating multiple deep belief networks into an integrated model and combining with the data preprocessing framework.The specific work done in this paper is as follows:(1)The research progress of the smart grid and its load prediction was described,and the relevant concept and technology involved in the thesis were also introduced.On the basis of consulting a large number of domestic and foreign literature,this paper summarizes the development and research status of smart grid,briefly describes the method of user side load forecasting,and gives the main research work and the framework of the paper.At the same time,this paper also introduces the empirical mode decomposition,sample entropy and other methods for data processing;multi-objective optimization algorithm based on decomposition and fast non dominated multi-objective optimization algorithm with elite strategy;extreme learning machine,deep belief network and other prediction models.(2)A data preprocessing framework based on empirical mode decomposition is proposed and a user prediction experiment platform is designed.In view of the complexity of the current power load data,the prediction model is difficult to capture the hidden information between the data.The data is decomposed by EMD,and the decomposed sub-data is convenient for model prediction;then each sub-data is used as the input of the prediction model for prediction,and each sub-data is established The input and output of the prediction model are combined with the input and output of the original data to form a new data set;finally,the new data set is used as the model input for prediction.Through experimental analysis,the proposed data preprocessing framework can significantly improve the prediction performance of the prediction model.A simulation platform is designed,users can select the month and region that need to be predicted,and the prediction results are displayed in a visual form,making the prediction results more intuitive and convenient for users to observe power load fluctuations.(3)A multi-objective deep belief network is proposed.MOEA / D is used to train DBN parameters to obtain several different DBN prediction models.In order to further select the sub models that meet the requirements of the integrated model,a two-stage strategy is designed to select multiple models generated by MOEA / d.finally,the sub models are combined into an ensemble model for power load forecasting.Experimental results show that the proposed method has good prediction accuracy and generalization ability.This paper not only analyzes the prediction performance of EMD-MODBN,but also makes corresponding comparative experiments in order to select the appropriate model parameters and the aggregation function in MOEA / d.by studying the distribution of the midpoint in the target space,this paper analyzes the objective function selection,function value standardization and two-stage strategy.