Mid-long term load forecasting of the unstable growth sequence based on Markov chains screening combination forecasting models

In the process of mid-long term load forecasting, it is important to choose right models according to the trend of the historical data. This paper chose the continuous 11 years electricity data of a certain district in Lianyungang to make up the historical data. Five forecasting models were always used in load forecasting field to forecast the electric quantity of the twelfth year. The data, which had been chosen in this paper, had the feature-the medium-term data showed fluctuation and the overall rising tendency was unstable. In order to check the forecasting results whether meet accuracy requirement, the concept of the grey relational degree was introduced. In terms of problems that even the forecast models meet the requirements of the grey relational degree, the forecast results still have great differences. According to the feature that the growth rate of load data is non-after effect property of Markov chains, by analyzing the growth rate of load data, using Markov chain to divide the growth rate to three state intervals, and getting the Markov state-transition matrix. According to the growth rate between the tenth and the eleventh data, combining with the Markov state-transition matrix, judging the rising tendency of the growth rate between the eleventh and the twelfth data. According to the rising tendency, screening two kinds from the models which had met the accuracy requirement, the method of variance-covariance was used to assign weights. Calculating results show that more targeted models could be chosen and combined. By this method of screening, not only can be appropriate for the rising tendency unstable data, but have a high precision.