A two-stage negotiation strategy based on multi-agent using Q-learning in direct power purchase with large consumers

The negotiation actions of different traders in the negotiation process of direct power purchase with large consumers are simulated using the multi-agent technology. With the Q-learning algorithm based on historical data, an agent can strengthen its own learning capacity and timely adjust its bid price against its opponent's action. Meanwhile, in order to ensure the fairness of market competition, a two-stage negotiation mechanism of 'negotiations+auction' is proposed. It gives one more opportunity to the generator agent who has a lower reserve price but fails to achieve an agreement, due to underestimation of the situation in the negotiations. It also makes the real diversity of different generating costs reflected by contract power price, and can inspire the generators to get the negotiating initiative by lowering their costs. ©2010 State Grid Electric Power Research Institute Press.