Abstract:
The intermittency of solar energy sources is a major constraint in integration of solar PV generation to the power grid. The dependency of these sources on the meteorological conditions poses a challenge in balancing the power generation and load demand. In order to enhance the reliability of the system, it is crucial to forecast solar photovoltaic power. Various machine learning algorithms with different input weather parameters have been developed in the past to address this problem. In this study, all possible 63 combinations of six input parameters i.e. temperature, dew point, wind speed, cloud cover, relative humidity and pressure are applied one by one to Artificial Neural Network (ANN) to forecast 24 hours ahead PV generation. This work presents the analysis of these results aiming to find out the most effective combination to be used as input to the forecast model. Historical weather and power data of one year is obtained for an actual site in Lahore to train the model. The power forecast results are obtained based on weather forecast data of 21 days sampled from the recorded forecasted data of 180 days. To quantify the error between predicted and measured solar PV generation, Root Mean Squared Error (RMSE) is used and results of different input combinations are also compared on basis of this statistical matrix. The analysis shows that the generation is best predicted on two combinations with the first comprising of temperature, dew point, relative humidity and cloud cover while the second consisting of all six parameters. Both of these combinations predicted the power with an RMSE of 4.44.While, some of the three input combinations also resulted in RMSEs as in close proximity of this value. The most accurate predictor among these was the combination of temperature, cloud cover and relative humidity.
