Data-Model Hybrid Driven Method For Short-term Load Forecasting

With the reform of the power market and the improvement of the intelligent level of the power grid,the load connected to the power grid is increasing constantly,which brings great challenges to the load prediction of the power system.Short-term load forecasting can provide strong support and guarantee for the decision-making of unit start-up and power adjustment.It is an important foundation to ensure the safe operation of power system and a necessary precondition for the construction of intelligent power system.The short-term power load is affected by many factors such as temperature and weather,which makes it have strong fluctuation.However,the traditional short-term load forecasting methods have some problems,such as limited ability of nonlinear mapping and weak ability of generalization of unknown data.Ultra-short-term power load has strong non-stationarity and non-linearity,which makes it difficult to mine the time series characteristics of load data.In order to solve the above problems,a data-model hybrid-driven method for short-term load forecasting is proposed.The main contents are as follows:(1)A short-term load forecasting method based on grey relational analysis(GRA),artificial bee colony algorithm(ABC)and support vector machine(SVM)is proposed for short-term load forecasting.Firstly,Pearson correlation coefficient method was used to select the important influencing factors.Secondly,gray relational analysis was used to screen the similar days in the history days and construct the rough set of similar days,K-means clustering is used to construct similar day rough sets.Then,ABC algorithm is used to optimize the penalty coefficient and kernel parameters of SVM.Finally,based on the actual load data of Nanjing City,the above method is applied to simulate and verify,it is also compared with the Support vector machine method of particle swarm optimization,the Support vector machine method of grey wolf algorithm and the long-short-term memory neural network(LSTM)method.(2)Aiming at the problem of ultra-short-term load forecasting,in order to fully exploit the intrinsic characteristics of ultra-short-term load data,in this thesis,a new method based on Kmeans clustering,empirical mode decomposition(EMD)and stochastic configuration network(SCNs)for ultra-short term load forecasting is proposed.Firstly,the load data were decomposed into several intrinsic modal components(IMFs)and residuals(Res)using EMD.Secondly,each IMF was classified using K-means clustering and the same class of IMFs components were added.Then,SCNs is used to forecast the power load based on the classified data.Finally,based on the actual load data of Shenzhen City,the proposed method is applied to simulate and validate the proposed method,it is also compared with the Support vector machine method and the long-short time neural network method.Simulation results show that the proposed data-model hybrid short-term load forecasting method for short-term load forecasting and ultra-short-term load forecasting have high forecasting accuracy.