A study of 100kWp PV Plant Output Power Forecasting: A case study

Nowadays solar energy penetration in the local grids is increasing more; this may result in a reduction in reliability, so it requires the future power generation forecasts to improve the grid reliability and to make the grid smarter. This can be possible with Data Analysis; here large volume of time series data are analyzed in python using AI/ML techniques. Tis paper has implemented a DNN (Deep Neural Network) technique using LSTM networks for forecasting the PV Power Generation. For this, the previously recorded generation data of 100kWp Solar Power Plant for past two years (720 day wise data samples) are considered and also tested on 220 hourly data samples for predicting monthly and day ahead power generation. This time-series data processed through the LSTM Networks for forecasting monthly and hourly PV generation. These results are compared with the other forecasting models viz. ARIMA, SARIMA, RNN and fbProphet and observed that LSTM and Prophet are giving better forecasting results. Root Mean Square Error (RMSE) value and the plant’s Capacity Utilization Factor (CUF) was calculated. This study shows that the LSTM networks are more accurate for forecasting the time series data.