Abstract / Description of output
The widespread integration of electric vehicles (EVs) into the electrical grid creates a new opportunity for frequency regulation. In this article, to deal with the penetration of intermittent renewable energy and the time variance of system model, an online evolutionary mechanism is developed for the electrical-grid- electric-vehicle system. With a real-time decoder consisting of the long-short-term memory (LSTM) array, the dispatch center is upgraded from a passive executor to an intelligent analyst, which extracts the rolling features from multiple time scales. Based on the high-dimension decoding information from the LSTM array, a deep neural network (DNN) array is then embedded to provide strategic dispatch commands learning from the evolving memory. The whole decoder-dispatch framework is then upgraded with a unified online adaption technique to achieve gradient optimization and weight evolution. The proposed evolutionary structure is validated on a frequency management system to demonstrate its superior performance.