Interval Day-Ahead Load Forecast of Micro Grid with Fuzzy Similar Data Selection and Gaussian Process

The load of micro grid usually suffers rapid changes and is greatly affected by meteorological data, which brings great challenges to the load forecast of the micro grid. We present a novel daily load forecast method for the micro grid by using Gaussian process with fuzzy inputs and fuzzy similar data selection to get a confident interval prediction, which is expected to possibly cover the great changes and fluctuations. First, the fuzzification method of the most important features impacting the forecast, i.e., the temperature and the humidity, is given. Then, the similar data selection method based on the fuzzified features and dynamic time warping is proposed for training the forecast model. The Gaussian process with the fuzzified features and similar data is trained and used to obtain the prediction intervals of the daily load. Finally, the performance of the proposed method in higher reliability and accuracy is validated by using two-year practical micro grid load data.