Evaluation of load forecast model performance in Croatian DSO

Abstract

During the revitalization of the Remote Control Systems of four Distribution System Operators in Croatia: Elektra Zagreb, Elektroslavonija Osijek, Elektroprimorje Rijeka and Elektrodalmacija Split, the load forecasting subsystems were implemented as an integral part of the DMS system. Accurate electricity load forecasting presents an important challenge in managing supply and demand of electricity since it cannot be stored and has to be consumed immediately. Electricity consumption forecasting has an important role in the scheduling, capacity and operational planning of the distribution power system. Load forecasting of certain parts or the whole distribution network helps to improve distribution network planning, operation and control which also increases the safety level of the entire distribution system. Although many forecasting methods were developed, none can be generalized for all load patterns. Accurate results of electricity load models are essential to make important decisions in planning and controlling so it is important to keep models as accurate as possible regarding input variables such as historical loads and meteorological data. This article gives a description of the implemented load forecasting subsystems using an artificial neural network with a feedforward multilayer perceptron and backpropagation as a learning strategy. The emphasis is on the simple and systematic use of input and output data as well as on forecasting scenarios of specific measured points where hourly forecasted results for a week ahead are presented and compared for Croatian Distribution Centers.