Research On Optimal Scheduling Of Island Integrated Energy System Based On Deep Reinforcement Learning

With the wealth of renewable energy resources and diverse forms of energy demand on islands,constructing an island integrated energy system(IIES)to coordinate and optimize the operation of various heterogeneous energy subsystems has become an essential choice.However,the IIES is highly dependent on renewable energy,and the output of renewable energy fluctuates greatly.The resource requirements on the island are diverse and each has a large time difference.The multiple uncertainties from power supply and load have brought huge challenges to the stable supply of the island’s multiple energy demand.At the same time,the method of deep reinforcement learning can adapt to the changes of the external environment through continuous interaction with the external environment and adjust the strategy dynamically in time,avoiding the modeling of multiple and complex uncertainties.It is an effective way to solve the uncertainty of renewable energy output and load demand by mathematically describing the dynamic economic dispatch problem of IIES and transforming the problem into a deep reinforcement learning framework.In this paper,research is carried out on the use of deep reinforcement learning to solve the problems related to the optimal scheduling of the IIES with multiple uncertainties.The main research contents are as follows:Firstly,an IIES considering various forms of energy demand is constructed.The key components are modeled,including the output model of combined heat and power unit considering the coupling relationship of heat and power,the output model of rapid reaction gas turbine and gas boiler,and the combined water and power unit that utilizes thermal power unit for seawater desalination.To enhance the utilization efficiency of island fresh water,a model of energy and matter co-transfer called "hydrothermal simultaneous transmission" has been developed based on relevant demonstration projects in the Shandong coastal areas.In addition,to address the problem that heat load data are difficult to measure directly,a heat load model of the building is constructed to establish a link between external ambient temperature and heat load.The development of these physical models serves as a foundation for subsequent research on optimization and scheduling.Secondly,taking into account the intermittency of island renewable energy output and the volatility of various load demands,a dynamic economic dispatching optimization model for the IIES with multiple uncertainties was developed,with the objective of minimizing operation costs.To address the multiple uncertainties arising from power supply and load,the scheduling problem of the IIES is transformed into a Markov decision process using the framework of deep reinforcement learning.The optimal scheduling strategy is then obtained by utilizing deep reinforcement learning.This method does not require the predicting and modeling of various complex uncertainties.It can adapt to changes in the external environment and dynamically adjust its strategy through continuous interaction with the environment.The simulation results demonstrate that the proposed method can effectively address the multiple uncertainties arising from power supply and load.It enables the full utilization of renewable energy sources on the island and the stable supply of multiple energy needs while the method has good economy and solving efficiency.Finally,to coordinate the energy interaction among different communities in multi-community island integrated energy system(MCIIES)in an uncertain environment and achieve overall optimal scheduling of the system,a scheduling model is proposed.This model utilizes the multi-agent deep reinforcement learning algorithm to learn the load characteristics of different communities and make decisions based on this information.As the uncertainty dimension of MCIIES increases,the dispatching system’s capacity to handle uncertainty requirements increases exponentially.The simulation results show that the proposed method can well capture the load characteristics of different communities,and use the complementary relationship existing between these load characteristics to coordinate the reasonable energy interaction of different communities,which can realize the complete consumption of renewable energy with good economic and environmental benefits.