Short-Term Reactive Power Forecasting Based on Real Power Demand Using Holt-Winters' Model Ensemble by Global Flower Pollination Algorithm for Microgrid

Abstract

Load forecasting is an integral part of the energy study unit to schedule the generating unit by the load demand. Many studies were conducted on load forecasting based on real power demand; however, very few papers were published on reactive power demand. In this research work, an attempt has been made to predict the requirement of reactive power as a function of demand for real power. Household loads are considered for evaluating the demand for reactive power as a critical load. The attempt has been made based on the data collected from the laboratory experimental setup for one year. The load forecasting requires time series analysis of the data set along with error minimization between predicted values and actual value; therefore, the global flower pollination algorithm along with Holt-Winters' exponential model has been used to predict the reactive power. Autoregressive integrated moving average (ARIMA) and seasonal autoregressive integrated moving average (SARIMA) models have been used as benchmarking models to evaluate the effectiveness of the model under various conditions. Python-developed model has been used to predict the demand for reactive power, and a MATLAB model has been developed to optimize the cost function. A detailed comparative analysis of the proposed model along with some well-established optimized models such as GA, PSO, and FPA has been presented related to evening peak demand for a microgrid architecture in Conclusion. The analysis includes median values of different quantities such as nMBE, nMAE, nRMSE, and RMSE. Normalized MBE, indicating underestimation and overestimation, is negative for ARIMA but 0.42 for HW-GFPA during validation and 0.43 for the testing data set. Normalized RMSE, measuring the variance between actual and forecasted values, is lowest at 0.803 for proposed HW-GFPA during validation and 0.799 for testing.