An intelligent power management system with active learning prediction engine for pv grid-tied systemsTools Kow, Ken Weng (2020) An intelligent power management system with active learning prediction engine for pv grid-tied systems. PhD thesis, University of Nottingham.
AbstractAn incremental unsupervised neural network algorithm namely time-series self organizing incremental neural network (TS-SOINN) is developed to predict the photovoltaic output power for power fluctuation events detection in photovoltaic micro-grid system. The TS-SOINN is an unsupervised clustering algorithm that identifies the most similar patterns from a data map to predict photovoltaic output power. A novel memory layer and weighted tapped delay line is introduced to establish the time-series learning. By using real-life environment data as input data, the proposed TS-SOINN based real-time prediction engine predicts 97% of power fluctuation events with 10% false acceptance rate. These results outperform three different types of self-organizing incremental neural network, self-organizing map, and nonlinear autoregressive with exogenous input network. The proposed TS-SOINN is then integrated into an intelligent power management system (PMS) to form the novel active learning intelligent PMS. The developed system is tested in simulation and experiment environments. Results show that the developed PMS reduces 89% of power fluctuation events and battery state-of-charge maintains within 30% to 100%. It outperforms hourly rule-based controller and the ramp rate controller by 53.53% and 37.08%, respectively in terms of the number of mitigated power fluctuation events. To conclude, power fluctuation events are mitigated by a novel intelligent PMS with reduced battery energy storage system capacity.
Actions (Archive Staff Only)
|