Fuzzy vector reinforcement learning algorithm for generation control of power systems considering flywheel energy storage

Abstract

To reduce carbon emissions, the proportion of renewable energy in power systems is increased. However, the volatility and uncertainty of renewable energy cause power deviations. To reduce frequency deviations caused by power deviations, a fuzzy vector reinforcement learning (FVRL) is developed for the generation control of power systems considering flywheel energy storage systems (FESSs). The FVRL algorithm consists of two independent fuzzy controls, two independent Q-learnings (QLs), and vector operation. Area control error and control performance standard 1 index are the inputs of one QL; the input of the other QL is the output power of the FESS. Fuzzy control fuzzifies the inputs of QL and provides the row number of the Q-table. Vector operation is acted for the output values of two QLs; the amplitude of the obtained vector is the final power regulation command of automatic generation control. The FVRL, proportional–integral, five reinforcement learnings, and deep Q-network are compared in two cases. Case studies show that the proposed FVRL obtains the lowest frequency deviation, the lowest area control error, the lowest generation cost, and the highest control performance standard 1 index.