Photovoltaic Power Generation Forecast by Using Estimator Model and Kalman Filter

Abstract

This paper presents an approach to forecast photovoltaic (PV) power generation by using Kalman filter and Auto Regressive Integrated Moving Average (ARIMA). This method is suitable for real time forecast with high resolution time step and we use it to forecast for five-minute time step in this paper. However, Kalman filter requires real time measurement data to adjust forecast value, hence we propose an estimator model to help this approach to perform reliable forecast even when real time measurement data is unavailable. The dataset for building estimator model is set of historical data of power generation from neighbor PV rooftops and distance between PV rooftops. We use ARIMA model to estimate transition matrix for running Kalman filter. The performance of the test is measured by the Root Mean Square Error (RMSE) and Skill Score (SS). The obtained result shows that ARIMA model has lower accuracy compared to Kalman filter and estimator model. The real time data estimation from the estimator model can be used in Kalman filter to forecast PV power generation with good accuracy.