Photovoltaic Power Forecasting using AI techniques and LSTM Deep Learning Network in Southern West of Algeria

The accurate forecasting of photovoltaic (PV) power refers to the ability to predict the amount of electrical energy that will be generated by a photovoltaic system in response to given irradiation and temperature. This forecast is important for several reasons, including optimizing the system's performance, facilitating energy trade and managing network integration. In this article, Long Short Term Memory (LSTM) neural network are used for forecasting the output power of a photovoltaic power generation station using a set of historical time series data composed of photovoltaic power, irradiation and temperature. The data has been collected by the measuring environmental variables and generated power every 15 minutes from a station located in the city of Adrar in the Southern West of Algeria. The accuracy of the LSTM based model is compared to other classic models such as ANFIS and ANN for the same conditions. The results of the trained LSTM model show a slight difference between the forecasted and measured power with a Mean Absolute Error (MAE) value equal to MAE= 127.10. The training and evaluation of the PV power forecasts by other techniques give an average accuracy MAE= 1.3857 for the ANN model, and MAE=3.0052*10
^-4
for the ANFIS model.