Transfer Bees Optimizer And Its Application On Power System Optimization

With the continuous development of the scale of the power system,the optimization problem of the system becomes more and more complicated.The traditional numerical optimization method becomes more difficult and even unable to solve such problems.Although the artificial intelligence optimization method opens up new horizons,it is difficult to apply to practical engineering problems because the algorithm takes too long to calculate the large-scale system optimization problem.Thus this paper propose a novel transfer bees optimizer,combining with the decision-making ability of reinforcement learning and the multi-agent coordination mechanism of artificial bee colony based on the idea of transfer learning.First the trial-and-error and the reward mechanism of Q-learning is adopted to construct the learning mode of the bees in order to enhance their information utilizing ability.Then the technology of behavior transfer and knowledge transfer from reinforcement learning is used for transfer learning.Moreover,a space-action chain is proposed to decompose the solution space into several lower-dimensional spaces,thus it can solve the curse of dimension resulted from the multiple variables optimization problem.The coordination mechanism of artificial bee colony is adopted to improve the probability distribution action strategy so as to balance the exploration and exploitation problem.The continuous action space is converted to discrete space by binary code,so that the transfer bees optimizer can solve continuous or hybrid variables problems directly.Inspired by the relationship between the optimization task station and the optimal Q-value,a novel meshing knowledge transfer method is proposed to improve the algorithm's learning ability in random complex environment.Finally,the IEEE 118-bus case and the IEEE 300-bus case simulation have been carried out respectively to verify the algorithm.Simulation results show that TBO can obtain a high-quality optimal solution,while its convergence speed can be accelerated up to 4-67 times faster than that of the conventional heuristic artificial algorithm(AI),which is very suitable for fast optimization of nonlinear programming in a large-scale complex system.