Forecasting the electricity balance of a small manufacturer with photovoltaic production using machine learning

Implementing photovoltaic production capacity at industrial sites is a step forward for smart-grid technologies by enabling improved energy efficiency. The biggest challenge for every renewable energy producer is unpredictability. This paper presents an industry case study and the necessity of energy balance forecasting. Lower forecast errors can improve supply-consumption balance by offering precise information for participating in the power market. Forecasting hourly energy consumption and production of an industrial consumer is difficult because of the high volatility for both processes that need to be independently analyzed. The present paper aims to offer a detailed analysis of the efficiency of algorithms applied to industry data for forecasting the electricity balance, which can help the effectiveness of renewable energy installed at industrial sites. The exogenous variables used in forecasting methods, among others, are temperature, solar irradiation, and cloudiness. Implemented machine learning algorithms for forecasting: Multi-Layer Perceptron and Recurrent Neural Networks, and a traditional Multiple Regression algorithm.