Research On Energy Management Strategy Of Microgrid With Distributed Energy Storages

With the gradual depletion of traditional energy resources,the development and application of renewable energy become increasingly important.The application of microgrid is important to realize a large-scale utilization of renewable energy.The energy management and scheduling optimization of microgrid are important supports for the stable,reliable and economic operation,which is also of great significance to improve the application rate of renewable energy.Distributed energy storage can not only solve the imbalance of power generation and consumption in microgrid,but also provide a means for the regulation and control of microgrid,conducting research on energy management strategy of distributed energy storage microgrid can effectively improve the economy of microgrid and is also of great research significance.The main research content of this thesis including:Firstly,concerning the day-ahead optimization problems of distributed energy storage microgrid,a modular microgrid model composed of wind power generation,photovoltaic power generation,distributed energy storage,as well as load has been constructed,equality or inequality constraint conditions were established,taking the minimum economic operating cost and voltage deviation of microgrid as multi-objective functions,using the NSGA-Ⅱ to get the non-inferior solution set,then using the fuzzy decision method to select optimal solutions from the non-inferior solution set.This thesis addressed the constraint conditions according to the characteristics of model reversed the generated non-feasible solutions to generate feasible solutions,which improved the probability of the feasible solution by 73%.Secondly,due to the randomness of renewable resources and load,the prescribed day-ahead optimization and scheduling strategy can’t reach the optimum in intra-day online operation.This thesis introduced the random strategy of Soft Actor Critic along with the single agent deep reinforcement learning algorithm to optimize the economic operating cost of distributed energy storage microgrid.The network structure of four 4-layer neurons was established,among them,there are two handling methods of neural network to deal with constraint conditions,the first method is to add the limit value of constraint conditions into the reward function as a penalty directly,the second method is to use the constraint conditions as auxiliary cost functions to help the training of algorithm,by contrast,the second constraint effect is better than the first one,the simulation experiment proved that compared with（1）,the economic operating cost of the proposed algorithm is 637.74 yuan,which decreased by 5.9%.Thirdly,based on the framework of single agent deep reinforcement learning algorithm cannot adapt to the expansion of distributed energy storage microgrid,in order to adapt to the expansion of system,a multi-agent deep reinforcement learning algorithm based on Soft Actor Critic was introduced to optimize the intra-day online operation of objective function of distributed energy storage microgrid.According to the state of current agent,the multi-agent algorithm can set three sub-microgrids as three agents to perform the actions separately during its operation,simplify the complexity of training neural network and finally the convergence speed of the three agents in multi-agent algorithm was improved by 71.4%,85.7% and 71.4% respectively compared with the single agent algorithm.The result of simulation experiment showed that the economic operating cost of the multi-agent algorithm was reduced by 5.7% compared with that of（1）.