Economic Dispatch Of Virtual Power Plants In The Energy Internet Based On Deep Reinforcement Learning

With the high popularity of large-scale distributed renewable energy generation,the power system is facing huge challenges.In this regard,virtual power plant(VPP)technology can more effectively integrate a large number of distributed generation units and can play a key role in improving the stability of the power system.Due to reliable economic dispatch in virtual power plants requires timely and reliable communication between distributed generation units and the user side.The existing methods have the problems of large communication load and delay,high calculation complexity,and poor reliability of data transmission.We have designed an edge computing framework to meet the real-time communication and calculation requirements of economic dispatch in virtual power plants.In addition,ensuring the authenticity of the data collected by the terminal and accurately describing the characteristics of the data are also key issues for economic dispatch.We use LSTM as the bad data tolerance mechanism to provide true status values for the assessment of renewable energy accommodation capacity,thus ensuring the accuracy of the assessment.Considering the uncertainty and nonlinear characteristics of distributed generation units,as well as the high-dimensional state space problem and the real-time nature of real scenes.We use a deep reinforcement learning algorithm to solve the optimal online economic dispatch strategy.Based on economic dispatch,the assessment of renewable energy accommodation capacity can reasonably assess the largescale renewable energy grid-connected accommodation capacity under random conditions,which is essential to promote the effective use of renewable energy and improve the stability of the power system.This paper proposes an algorithm based on deep reinforcement learning(DRL)to maximize the capacity of renewable energy accommodation in the 5G energy internet.Specifically,the main contributions of this paper include the following four aspects:1.We designed a distributed generation economic dispatch structure based on edge computing using a three-tier architecture,where: the first and second tiers are edge computing tiers,and the third tier is a cloud computing tier.The proposed three-layer edge computing architecture reduces the computational complexity of processing training tasks at the central node.It further reduces the communication load between the virtual power plant operator and the distributed power supply,thus also reducing the response time of industrial users.It also preserves the privacy of industrial users and improves the reliability of data transmission.2.In order to ensure the authenticity and validity of the data in the 5G energy internet,LSTM is used as a tolerance mechanism for bad data,which provides true state values for the solution strategy.Experiments prove the effectiveness of the algorithm.3.Considering that the objective function of economic dispatch is non-linear,an algorithm based on deep reinforcement learning is proposed to realize the optimal economic dispatch problem.Despite the high-dimensional state space(due to the randomness associated with distributed power),our proposed algorithm with low computational complexity.Comparing our proposed method with DPG,our proposed method has better performance.4.In order to reasonably assess the capacity of large-scale renewable energy gridconnected accommodation under random conditions,we propose an L-DRL method to achieve maximum accommodation of renewable energy based on energy storage technology and user demand response.