Short-term forecasting using fuzzy neural network accountability air temperature and natural lighting

Abstract

The paper offers a structure forecasting model for short-term forecasting power consumers in Moscow and an algorithm for computing the instrumental and forecast value of air temperature and lighting. The forecasting model uses the Takagi-Sugeno-Kanga fuzzy neural network of five layers. Initial data were time series of power consumption, air temperature and natural lighting for Moscow and cloudiness. When forming forecasting a day graphic of power supply on subsequent days, archive power supply data for 15 days preceding the forecast day and last year were used. An algorithm for formation of input data fuzzy neural model is described. The quality of power supply forecasting for December 2015 and January 2016 is estimated. Examples for a forecast error provided correspond to February 1 and 16, 2016. Results of received forecasting estimation on average values with a relative error are provided. The majority of tests and real forecasting results comply with the error specified for operator of the united power system of Russia.