Research On Combined Forecasting Model In Short-term Energy Consumption Forecasting Of City

The short-term energy consumption of city usually means basing history data to predict future one year's energy amount used. The scientific forecast of the energy consumption, does good to establishing perfect energy program and decreasing production and living cost, and has important meaning for improving energy using effect, optimizing energy using structure and promoting the development of resources economizing society.This paper first analyses short-term energy consumption data's time sequence characteristic and changing rule ,and then points out energy consumption's time sequence is an non-steady time sequence which fluctuates obvious seasonally and exists some increasing trend, and finally come to a conclusion that short-term energy consumption forecasting should mainly use time sequence analyze based method. At the mean time, considering the city energy system as a complex system of multilayer, multifactor and multi-attribute, we combine properly the seasonal ARIMA multiple model and the seasonal RBF Neural Net model to propose a seasonal ARIMA-RBF model to build the short-term energy consumption model. This model has great pertinence on short-term energy consumption prediction's non-linearity, uncertainty and periodicity, so in theory it's a model which suits very much being used to forecast short-term energy consumption.On the other hand, this paper also proposes revised methods on the energy consumption model's effect and precision. First of all, we build a method called"improved seasonal RBF Neural Net ", which is based on self-related analyze to ascertain network model and decrease input dimension, so as to fasting this model's convergence rate. Lastly, starting off by the energy system's non-linear characteristic, we use Neural Net to combine two single prediction model in order to synthesis two models' linear and non-linear information extracted from energy system. The experiment shows that this combination model overcomes the shortcoming of single model and improves the prediction precision.