#baozhang

The direct-purchasing electricity policy is an attempt to reform the existing power sales mechanism, which is part of the China energy supply reform. The aim is to break the pattern of the monopoly pricing of power grid enterprises and promote the establishment of an open electricity market. This policy allows enterprises with a high electricity demand (such as heavy industry, telecommunications industry, etc.) skip the grid electricity trading platform and directly buying cheaper electricity from power plants. Every month, these enterprises give a prediction number of the electricity they plan to used next month, the power plants produce and sell the corresponding number of electricity with cheaper price to these enterprises, in this process, if the prediction number and the actual electricity consumed number has a number difference with more than plus or minus 2.5%, the enterprise will be punished. So it’s necessary to choose a reliable forecasting method to predict the future electricity consumed data accurately. Based on the historical electricity data of a telecom company from 2014 to 2017 as the training set, this paper establish a system of data forecast model, which selects the ARIMA model and Holt-Winters model as prediction model, then designing a direct-purchasing prediction system, and use this system to predict the future consumption of electricity, finally the paper evaluates the performance of the prediction accuracy of these 2 different model, it comes out that Holt-Winters model has a better performance on the data prediction base on this problem, but still can’t reach the target of lower than 2.5% differences. This paper further analyzes the reason of the error and proposes some possible solutions to reduce the prediction error, also point out some possible further enhancement of the prediction model in the future.

We can see that the Holt-winters model did better than the ARIMA model in this case.