Research On Residential Distribution Substation-Oriented Regional Level And Multi-Layer Energy Optimal Scheduling Strategy

The rapid development of demand response(DR)technology and active distribution network(ADN)technology has promoted the two-way interaction between the power grid side and user side,and thus it has provided the support for the safe and stable operation of distribution network with high penetration distributed generations intergration.As an important component of loads,residential load is a high quality flexible source with great adjustment ability.On the basis of meeting the household electricity demand,the huge potential of residential load is explored to improve the economy and security of distribution network.Therefore,this paper has carried out the research on the interaction for home energy management system(HEMS),residential distribution substation and distribution network.Firstly,the research on HEMS model and the interaction architecture with residential distribution substation and distribution network is carried out.The energy and information interaction modes of HEMS are analyzed,and the two-tier framework for HEMS is established by considering users’ comfort,electricity cost and household total power limit.The electrical appliances are classified according to their load characteristics and their models are established by taking the comfort requirements and operation characteristics into account.The interaction architecture and process for HEMS,residential distribution substation and distribution network are described.Secondly,a hierarchical optimization strategy for HEMS based on deep reinforcement learning(DRL)is proposed.Based on the HEMS hierarchical model,Markov decision process is used to model the household energy management problem,including state space,action space,reward function and action-value function.Rainbow algorithm is introduced to optimize the strategy with the goal of maximizing the long-term benefits.The simulation is performed on a residential house,which includes photovoltaic,an energy storage,multiple electrical appliances,and an electric vehicle.The case study shows that the DR strategy can reduce electricity cost and ensure that the household total power is within the safe limit.Then,the flexible loads of households under the jurisdiction of the residential distribution substation are aggregated and the response potential for flexible resources is analyzed and assessed.The evaluation system of residental response willingness is constructed.In this system,the annual household income,education level and average age are selected as characteristic indexes.The best-worst method is used to calculate the residental response willingness.Based on the HEMS load model and response coefficient,the regulation potential models of household electrical appliances are established.Combined with the household load operation plan,the reduceable power for power-shiftable loads,time-shiftable loads and electric vehicles is calculated.On this basis,the normal cloud model is utilized to evaluate the grade for the flexible loads of households under the jurisdiction of the residential distribution substation,and the response resource pool is formed to rapid scheduling.Taking the overall controllability into account,the response potential aggregated model of the flexible resources under the residential distribution substation is constructed,and the total regulation capacity is calculated and reported to the distribution network to participate in scheduling.The simulation is performed in the residential distribution substation containing 500 households.The results of case study show that the grade and regulation potential of flexible resources are highly correlated with household types.Finally,an active and reactive power coordinated scheduling strategy in the distribution network based on graph reinforcement learning(GRL)is presented.Aiming at minimuming the comprehensive cost and voltage deviation,the active and reactive power coordinated scheduling model of the distribution network is established.On this basis,the scheduling framework of the distribution network based on GRL is constructed,which is composed of graph attention network(GAT)block to extract the characteristic information of the distribution network and DRL block to formulate real-time scheduling strategy of the distribution network.GRL algorithm is introduced to solve the scheduling problem,including environment perception as well as feature extraction based on GAT and decision process based on deep deterministic policy gradient(DDPG).The simulation is performed in the modified IEEE 33-bus system.The case study illustrates that the proposed scheduling strategy can effectively reduce power loss,interaction power with the transmission network,operation cost as well as voltage deviation,and improve the new energy consumption rate.Meanwhile,the generalization ability of GAT-DDPG algorithm is verified.