Deep Stackelberg heuristic dynamic programming for frequency regulation of interconnected power systems considering flexible energy sources

Abstract

The flexible energy sources (FESs) could refuse to execute the generation commands (GCs) if the benefits fell short of their expectations when applying traditional automatic generation control methods, this problem leads to frequency instability and the waste of scheduling resources. Accordingly, a deep Stackelberg heuristic dynamic programming (DSHDP) is proposed for the frequency regulation and power dispatch of smart generation control (SGC). Firstly, the heuristic deep dynamic programming, whose networks are replaced by three deep neural networks, can be trained offline by historical data to fit the optimal performance index. Then, the GCs can be optimized through an online updating strategy for stabilizing the frequency of interconnected power systems. To maximize the benefits of FESs further, an improved SGC framework based on the Stackelberg game is built to let FESs play games with conventional power plants in a fair environment. Finally, the optimal GCs can be distributed by the Nash equilibrium to obtain higher profits than traditional methods, and the control accuracy is increased indirectly. Two cases, i.e., IEEE two-area power system and six-area power system based on China Southern Power Grid (CSPG), are provided to verify the superiority of the DSHDP. The results indicate that the excellent flexibility and stability of the proposed algorithm in complex power systems containing renewable energy and FESs. Totally, the frequency deviation obtained by DSHDP is reduced by up to 59.38%, while the total profit of FESs rises by up to 67.47% when compared with the other seven representative algorithms.