Research On Resource Allocation Problem Of Smart Grids Based On Deep Reinforcement Learning

Smart grids can provide automated,safe and cost-effective power services for electrical equipment by integrating communication,sensing,monitoring and control technologies.Efficient grid systems can sense the overload of realtime communication infrastructure and have the advantages of regulating the direction of energy flow,optimizing the use of renewable energy,and reducing user costs.However,with the increase in the number of power equipment access and the tightening of energy supply,the contradiction between the limited power communication energy and the ever-increasing service demand is further intensified.Therefore,reasonably responding to the challenges brought about by business diversification,ensuring the stable supply of power energy,and maximizing the optimization of the communication mode between power equipment are the key issues that need to be solved urgently.Facing the resource allocation problem in smart grids,the thesis focuses on the use of deep reinforcement learning algorithms to solve the problems of task offloading and caching,and energy supply and demand scheduling in smart grids.First of all,this thesis analyzes the advantages of deep reinforcement learning technology in smart grids from the principle and performance aspects based on relevant research at home and abroad.In view of the problems caused by the surge in the number of services and the diversification of service requirements,cloud-edge collaborative technology is used to offload tasks to relieve the pressure of cloud computing.Considering that the deep reinforcement learning algorithm can adapt to the unknown dynamic environment and the huge state and action space in smart grids,as well as the different requirements for the delay of the services in smart grids,a deep reinforcement learning algorithm based on polling update is proposed to jointly optimize the communication,computation,and caching in smart grids.Experiments show that the performance of the target algorithm is better than the baseline algorithm.Secondly,a demand response management method based on price and incentives is proposed to solve the problem that the gradual increase of the penetration ratio of renewable energy in smart grids and power market makes the impact of intermittent output methods on the distribution system increase.Considering the resource allocation problem between the power supply side and the power consumption side of renewable energy,a deep reinforcement learning algorithm based on fair delay is proposed to ensure the real-time and fairness of demand in smart grids.Finally,in order to verify the performance of the proposed system model and algorithm,comparisons with various algorithms and extensive simulations are performed.The simulation results show that the algorithm has greatly improved in terms of algorithm convergence,electricity cost and power supply stability.Finally,the resource allocation problem of using mixed energy to power base station equipment in wireless communication environment is studied.Considering the limitations of single-agent exploration,we propose a multiagent deep reinforcement learning algorithm.Considering the continuity of energy and the communication interference between base stations,the deep reinforcement learning of actors and critics is used to complete the exploration and utilization of continuous states and actions,and reduce the noise interference caused by discrete quantization,and improve the stability of communication.Through the exploration and learning of multi-agents,the power supply of renewable energy,the power supply of non-renewable energy,energy storage,and scheduling and energy distribution of user equipment are optimized collaboratively.Experiments show that the collaborative method of multiple agents has higher algorithm stability,and the target algorithm can balance the power consumption of peaks and valleys,reduce the cost of electricity consumption,and improve the stability of power supply.