Extended, Short-Term Neural Prediction Methodology, for European Electricity Production by Type

Abstract

Accurate forecasting in the electrical energy supply sector is essential for cost savings, enhancing power system reliability, decisions on development, expansion, modification or reduction of facilities. It plays a vital role in the long-term development of the electric power industry. Electricity cannot be stored in large quantities, it is difficult to transfer, and requires continual production-consumption balance. The stochastic behavior of electricity consumption makes it challenging to anticipate. It is affected by a number of variables: climate, economy, population increase, pandemic breakout, etc. In this research we conduct experiments with different neural network forecasting topologies and establish the methodology that will most accurately anticipate the trend of the electricity production for various types of sources such as: wind, oil, coal, nuclear power plants, and bioenergy. An approach that incorporates Time Delay Neural Networks is proposed to reduce mistakes and improve forecasting confidence. It is shown that this strategy may significantly increase the forecasting accuracy of the individual networks regardless of their topologies, which improves the applicability of the method. The performance and efficiency of models are assessed using the appropriate performance criteria. Additional forecasting experiments, including ARIMA and Extreme Learning Machine Modeling, have been carried out to quantitatively compare the accuracy of the proposed technique with alternative state-of-the- art forecasting methodologies.