In this paper,we design a new bidding algorithm by employing a deep reinforcement learning approach.Firms use the proposed algorithm to estimate conjectural variation of the other firms and then employ this variable t...In this paper,we design a new bidding algorithm by employing a deep reinforcement learning approach.Firms use the proposed algorithm to estimate conjectural variation of the other firms and then employ this variable to generate the optimal bidding strategy so as to pursue maximal profits.With this algorithm,electricity generation firms can improve the accuracy of conjectural variations of competitors by dynamically learning in an electricity market with incomplete information.Electricity market will reach an equilibrium point when electricity firms adopt the proposed bidding algorithm for a repeated game of power trading.The simulation examples illustrate the overall energy efficiency of power network will increase by 9.90%as the market clearing price decreasing when all companies use the algorithm.The simulation examples also show that the power demand elasticity has a positive effect on the convergence of learning process.展开更多
基金This work was supported by the National Science Foundation of China(Grant 2014CB249200)the National Natural Science Foundation of China(Grant 61873162)the Shanghai Pujiang Program(Grant 18PJ1405500).
文摘In this paper,we design a new bidding algorithm by employing a deep reinforcement learning approach.Firms use the proposed algorithm to estimate conjectural variation of the other firms and then employ this variable to generate the optimal bidding strategy so as to pursue maximal profits.With this algorithm,electricity generation firms can improve the accuracy of conjectural variations of competitors by dynamically learning in an electricity market with incomplete information.Electricity market will reach an equilibrium point when electricity firms adopt the proposed bidding algorithm for a repeated game of power trading.The simulation examples illustrate the overall energy efficiency of power network will increase by 9.90%as the market clearing price decreasing when all companies use the algorithm.The simulation examples also show that the power demand elasticity has a positive effect on the convergence of learning process.
