Deep learning approach for one-hour ahead forecasting of energy production in a solar-PV plant

ABSTRACT

Solar power production (SPP) using photovoltaics is one of the most effective ways of solar energy utilization. Prediction of SPP is of great importance due to mitigating the effect of random fluctuations in the incoming solar energy and enabling the operator to access solar power output data in advance. Accurate prediction for SPP is also important for providing high-quality electricity to end-consumers. In the present study, a deep learning approach established on Long Short-Term Memory (LSTM) neural network was introduced. The network aimed to forecast one hour-ahead electrical energy production from the solar-PV power plant with 1.15 MW capacity. In addition to the LSTM neural network, two different data-driven methods, namely, adaptive neuro-fuzzy inference system (ANFIS) accompanied by fuzzy c-means (FCM) and ANFIS with grid partition (GP) were applied. The data obtained from the models were also validated using measured data. The results from the comparison revealed that the LSTM model gives the best results with RMSE, MAE, and R equal to 60.66 kWh, 30.47 kWh, and 0.9777, respectively.

KEYWORDS:

• LSTM
• ANFIS
• deep learning
• solar energy
• solar power production
• forecasting

Acknowledgments

The authors would like to thanks Pilye Energy Construction Industrial and Trading Corporation for allowing us to use the solar power plant’s power production data in our research.