In order to solve the configuration optimization problem of isolated micro-grid, wind driven generator, photovoltaic, diesel generator and energy storage battery optimization model described in the form of energy flow matching is established, which can flexibly form the node connection relation representing the output of equipment at each time. At the same time, in view of the precocious convergence of swarm intelligence algorithm applied to optimal configuration of micro-grid, a random walk optimization algorithm suitable for optimal configuration of micro-grid is proposed. Guided by reducing the annual comprehensive cost of the system, the algorithm realizes synchronous optimization of continuous variable (equipment output) and integer variable (equipment quantity) by randomly increasing or decreasing the hourly output of equipment. By accepting the differential solution mechanism, the algorithm has the ability to jump out of the local optimal solution and better take into account the global search and local search in the capacity optimization process of micro-grid. Applying the random walk algorithm to the simulation example, the annual comprehensive cost is 552 826.39 yuan. Compared with particle swarm optimization algorithm, a better result is obtained. The superiority of the algorithm in optimization accuracy is verified.

A random walk algorithm compulsive evolution with individual reconstruction strategy is proposed to solve the problem that the optimization of mass exchanger network is easily trapped in local extremum due to the weakening of the ability of structural variation and the loss of population diversity. In the process of receiving differential solutions, real-time monitoring is carried out on individuals, and different reconstruction methods are adopted to stimulate the network structure updating and variation of backward individuals, so as to improve the structural variation ability and population diversity of the algorithm. At the same time, according to the characteristics of the unstructured model with shunt nodes, a new individual network structure after cross reconstruction is repaired. Finally, the R2S2 and R4S2 examples are used to verify the effectiveness of the proposed strategy, and the optimization results are all lower than the results in the current literature, which proves that the proposed strategy can effectively enhance the structural variation ability and global optimization ability of the algorithm.