Short term power load forecasting using Deep Neural Networks

Accurate load forecasting greatly influences the planning processes undertaken in operation centres of energy providers that relate to the actual electricity generation, distribution, system maintenance as well as electricity pricing. This paper exploits the applicability of and compares the performance of the Feed-forward Deep Neural Network (FF-DNN) and Recurrent Deep Neural Network (R-DNN) models on the basis of accuracy and computational performance in the context of time-wise short term forecast of electricity load. The herein proposed method is evaluated over real datasets gathered in a period of 4 years and provides forecasts on the basis of days and weeks ahead. The contribution behind this work lies with the utilisation of a time-frequency (TF) feature selection procedure from the actual “raw” dataset that aids the regression procedure initiated by the aforementioned DNNs. We show that the introduced scheme may adequately learn hidden patterns and accurately determine the short-term load consumption forecast by utilising a range of heterogeneous sources of input that relate not necessarily with the measurement of load itself but also with other parameters such as the effects of weather, time, holidays, lagged electricity load and its distribution over the period. Overall, our generated outcomes reveal that the synergistic use of TF feature analysis with DNNs enables to obtain higher accuracy by capturing dominant factors that affect electricity consumption patterns and can surely contribute significantly in next generation power systems and the recently introduced SmartGrid.

I. Introduction

Power load forecasting holds a crucial role in the capacity planning process of power systems scheduling and maintenance as well as end-consumer awareness regarding viewing timely their consumption behaviour and bills. The actual forecasting of the power load distribution is classified into short, medium and long term forecasting. Short term load forecasting (STLF) is associated with load prediction from few hours to few days ahead whereas medium term load forecasting (MTLF) deals with forecasts targeting few weeks to few months ahead. On the other hand, long term load forecasting (LTLF) deals with load prediction from one year to several years. LTLF assists in planning of new power systems setup, MTLF aids in system maintenance, purchasing energy and pricing plans whereas STLF plays a key role in unit commitment, power distribution and load dispatching. STLF is a challenging task due to short time duration as it requires instant and accurate decisions. The errors in STLF can have either leptokurtic or the normal distribution. If the normal distribution is assumed, it represents the tail of distribution insufficiently and leads to under-committing power systems which can cause shortage of energy in market and eventually increases the cost to produce more energy [1].