Adaptive Particle and Kalman Filter for Intra-hour Solar Irradiance Forecasting

With the increasing awareness about the adverse effects of conventional energy resources, policies are being developed to accelerate the penetration of renewable energy resources. Although solar energy has a high potential to meet human energy needs, the variable and uncertain nature of solar energy has impeded the growth of large-scale solar farms. Solar irradiance forecast is a viable solution to cope with this stochastic nature of solar energy and a necessary tool to manage and operate large solar farms in deregulated electricity market. Utilizing the forecasts, the large-scale solar farm operators can bid predicted solar energy in the electricity market and make smart decisions about curtailment and storage systems to avoid sudden ramps in the solar power output. To cover photovoltaic and concentrated solar power, we forecast both Global Horizontal Irradiance (GHI) and Direct Normal Irradiance (DNI) components of solar irradiance individually. In this work, we show the application of particle and Kalman filter to predict GHI and DNI for intra-hour forecast horizons. The main advantage of using particle filter is that the forecast achieved with this filter can approach its Bayesian optimal estimate if the filter is given sufficient sample of stochastic signal inputs. These filters have been widely used in the field of energy forecasting e.g. for electricity load and wind forecasting.  In this work, we investigate their application for intra-hour solar forecasting. Particle and Kalman filters are developed with time varying parameters to predict solar irradiance at 15-minute and 30-minute forecast horizon. The developed models are tested for solar irradiance collected at various locations in California representing different microclimates. Preliminary results are compared to persistence and other benchmark parametric methods. The developed models are apt for online forecasting and real-time control applications.