Search

A novel culture evolution learning is presented to achieve an optimal carbon-energy combined-flow (OCECF). A shared responsibility of carbon emission between electricity producers and electricity consumers is introduced in OCECF, such that a double counting of carbon emission can be eliminated. The proposed algorithm is inspired from the culture construction, culture communication, and cultural inheritance in the human society. First, a culture matrix of each country is constructed by the Q-value matrix of Q-learning, which can be simultaneously updated by its people. Second, each country can learn the superior culture from other more-advanced countries, while this so-called culture communication can improve the quality of the obtained optimal solution. Finally, the culture inheritance can be achieved by transfer learning between different optimization tasks, thus the convergence can be dramatically accelerated. The performance of CEA has been evaluated for OCECF on two IEEE benchmark systems, and a practical urban power grid of southern China, respectively. Simulation results demonstrate that CEA has a high convergence stability and fast convergence, which is around 4.02 to 51.43 times faster than that of conventional heuristic algorithms.

This paper proposes a novel multi-agent bargaining learning (MABL) for the distributed energy hub economic dispatch (EHED) of multiple energy carrier systems (MECS). Distributed EHED is developed by extending the conventional economic dispatch (ED) into MECS in a distributed manner, in which each energy hub is regarded as a learning agent for self-scheduling. The classical Q-learning with associative memory is employed for knowledge learning of each agent, while the non-uniform mutation operator is adopted for handling the continuous control variables. To maximize the total payoff of all the energy hubs, the bargaining game is presented for achieving an effective coordination between the buyer agents and a seller agent, where the slack energy hub is designed as the seller agent and the others are the buyer agents. MABL has been thoroughly evaluated for the distributed EHED on a high-complex 39-hub MECS with 29 energy hub structures and 76 energy production units. Case studies verify the superior performance of MABL for the distributed EHED compared with six centralized heuristic optimization algorithms.