Short-Term Load Forecasting Based On Chaotic Time Series And Neural Networks

Short-term load forecasting(STLF) is very important for the economical,stable, and secure operation of power systems, and it is a vital common workof distribution operation and load management. It's proved that Short-termload is a multiple dimension nonlinear chaotic system under the influence oftemperature, humidity, wind power, somber day or fine day, rainfall, feast dayetc. and, It's also proved that power load time series is a kind of chaotic timeseries. Under the reciprocity of various influencing factors, load becomecomplexity and it's hard to predict accurately. With the development ofnonlinear theory, especially the development of chaotic theory, it's possiblethat we can get satisfactory forecasting results without considering variousinfluencing factors.Chaotic time series is a burgeoning study, predicting it is a hot study.Neural network is a kind of intellectualized technology. It is excel at dealingwith nonlinear problem. Combining with the both will bring new problems.The paper combines chaotic time series with neural network, and apply it toShort-term load forecasting.Firstly, the paper, studies chaotic time series and phase spacereconstruction technology, computes the best delay time and best embeddingdimension for a practical power system's Short-term load, and reconstructsphase space. Creating a forecasting model based chaotic time series and BPneural network, result analysis of practical examples shows that the proposedmodel is effective and feasible.Secondly, comparability can't weight commendably using Euclid distance, for the shortcoming, best resemble phase points is presented in this paper, andweight commendably using it. Give a sort of method weighting commendablyusing Euclid distance and correlation. Using the method to select neuralnetwork's input data, creating a model based on chaotic time series,correlation and neural network, result analysis of practical examples showsthat the proposed model can improve forecasting precision.Lastly, because general chaotic load forecasting models are one-stepextrapolate, computational complexity and time consumption are mainshortcomings, a new method of short-term load forecasting based ontime-segmented phase spaces reconstruction and Generalized RegressionNeural Networks is presented in this paper. Result analysis of practicalexamples shows that the proposed model can gets good forecasting precision.