A proposed hybrid approach for forecasting daily global horizontal irradiation: Xgboost-LSTM
Abstract
The fast population growth the technological progress and the economic growth, that the world is experiencing today, causes an irremediable reduction of energetic resources in the long term. Finding energy sources to meet the world's growing demand is one of the biggest challenges that faces humanity for the next half-century. This necessity makes solar energy the best option of the future as it is potentially inexhaustible, renewable and clean. In this context, the forecasting of solar irradiation for different time horizons is becoming a very active area of research because it improves the integration of photovoltaic plants into electric grids and allows electricity suppliers to manage the production and the consumption of the installed power plants which leads to minimizing costs and providing high power quality.
This work presents our contribution to the task of forecasting daily global horizontal irradiation one day ahead. For that, we propose a new method that combines the advantages of Xgboost technique in feature selection with those of Deep LSTM neural network in the forecasting task. Different results demonstrate that our proposed model achieved great results and outperformed Support Machine Vector, Random Forest, and Xgboost for the case study Al-Hoceima, Morocco, in terms of minimum Root Mean Square Error (RMSE), Mean Absolute Error, and coefficient of determination R2.
