Intelligent Dispatch Research For Hydro-PV-PHS Integrated System:A Deep Reinforcement Learning-Based Approach

With the continuous promotion of the energy revolution and transformation,the renewable energy(RE)represented by wind power and photovoltaic(PV)has been rapidly developed and widely used.The RE can alleviate the consumption of traditional fossil energy to some extent and promotes the process of "carbon peak and carbon neutralization",but the feature of randomness and volatility would reduce the security and stability margin of power system,futher aggravates the difficulty of power system operation of dispatcher and increases the pressure on its absorption.Multi-energy complementarity is an integration of many kinds of energy with complementary characteristics into a whole part to make it have high energy efficiency,continuous power supply reliability and optimal operation revenue.For the research of the intelligent dispatching control and application of hydro-photovoltaic –pumped storage multi-energy complementary system,which is supported by the National Key Research and Development Program of China(2018YFB0905200).And the main contribution of this work can be summarized as follows:1)Intelligent economic dispatch for PV-PHS integrated system: a deep reinforcement learning-based approach.The operation of a centralized PV-PHS hybrid system based on the economy and stability is considered in the power market environment.The scheduling strategy aims at maximizing the economic benefits of the PHS and minimizing the penalty for power fluctuation on the tie-line.Taking the active power charge and discharge of PHS as the optimization object,the multiple uncertainties such as photovoltaic output,real-time electricity price and load variation are considered.The deep deterministic policy gradient(DDPG)algorithm is used to solve a real-time optimal operation strategy to maximize the economic benefits of the PV-PHS hybrid system and minimize the power fluctuation on tie-line.2)The opearation of the hydro-PV-pumped storage integrated system in distribution network(DN)based on an improved DDPG.For the distributed PV generation in the DN,the optimization objective is to consider the voltage quality and power fluctuation between the main grid and DN.The power generation of small hydropower station and the active charging/discharging of PHS are defined as the controlled object,and both the uncertainty of PV’s output and load variation are considered.An improved DDPG is used to solve a real-time optimal operation strategy to minimize the power fluctuation on tie-line and node voltage fluctuation of the DN.3)Mechanism analysis and real-time control of PHS based hydro-PV-PHS integrated system oscillation damping: an inproved DDPG approach.This research aims to suppress the power oscillation of hydro-PV-PHS integrated system,and takes the outer loop control of variable speed constant frequency(VSCF)pumped storage grid-side converter as the controll object.Meanwhile,a proportional integral(PI)controller with integral reduction loop is proposed,which is suitable for the interaction between PHS and power grid.PV power supply is used as random disturbance signal to cause power system power oscillation.And then,the improved DDPG algorithm is used to realize the real-time tuning of controller parameters in order to suppress the power oscillation in the shortest time.4)For the above research contents,a detailed simulation test is carried out after the theoretical analysis.Compared with traditional optimization methods.For part 1,the simulation results reveal that the intelligent dispatch strategy can effectively mitigate the power fluctuation and enhance the economic efficiency,that is,the power fluctuation on PCC is reduced by 12.7%and the economic revenue of complementary system is increased by 4.95%,simultaneously.For part 2,the method proposed in this paper can reduce the cumulative average voltage deviation and power fluctuation by 1.4 times and 1.15 times,respectively.For part 3,a modified four-machine two-area system is introduced as the test system,the time-domain simulation results demonstrate the effectiveness of the proposed controller and the superiority of the employed improved DDPG method.