Comparison of Memory-less and Memory-based Models for Short-Term Solar Irradiance Forecasting

The world's energy consumption is continuously rising due to rapidly growing human population and expanding industrial sector. Integrating Renewable Energy Resources (RERs) with the power system comes up with severe challenges as the nature of these resources is intermittent. Among RERs, solar energy is a viable means of producing power. However, due to the intermittent nature of solar energy, accurate forecasting is necessary for smooth operation of power systems. The operational applications such as load balancing and economic dispatch can be facilitated and made more effective using very short-term forecasting. Therefore, this study proposes a memory-less third-order Markov model for very short-term solar irradiance forecasting. The proposed model uses the Random Forest (RF) for dimensionality reduction and model performance is evaluated using Root Mean Square Error (RMSE), Mean Square Error (MSE) and Normalized Root Mean Square Error (NRMSE). The results are compared with the existing Long-Short Term Memory (LSTM). The 3rd-order MC outperforms LSTM model. In terms of NRMSE, an improvement of 40.5%, 39.9% and 63.4% is recorded for the datasets of three Pakistani cities, Islamabad Karachi and Lahore, respectively.