Research On Static Voltage Stability Assessment And Reactive Power Cooperative Control Of Power System Based On Machine Learning

The "double carbon" target has become the direction of joint efforts of countries around the world,and China has also made a "30-60" commitment at the World Climate Change Conference in 2020.As a major emitter of carbon emissions,the power industry is striving to increase the proportion of new energy generation and continuously promote low-carbon emission reduction,which is an important means to help achieve the "double carbon" target.Due to the inherent randomness,intermittency and volatility of new energy sources,their large-scale penetration will pose a huge challenge to the safe,reliable,stable and economic operation of the power system.It is of great theoretical and practical value to deal with the impact of large-scale grid connection of new energy sources,to improve the sensitivity of the system to voltage stability perception and to coordinate reactive power distribution to achieve integrated control of voltage stability and network loss reduction.This paper uses machine learning technology to investigate the two core issues of static voltage stability sensing and coordinated reactive power control in power systems under the new situation,with the following main results.(1)The static voltage stability perception problem is defined as a regression problem,and an online intelligent voltage stability perception method based on RRelief F-SDAE-BP-Bagging network is proposed.Firstly,the training sample set is obtained through scenario simulation,power flow calculation and local voltage stability index calculation;then the RRelief F method is used to rank the features and eliminate the attributes with low weights to improve the training efficiency;then the Stacked Denoising Auto Encoder network is built to improve the model data denoising ability.Finally the BP network combined with Bagging integration is used to obtain the mapping relationship between each key feature and voltage Stable mapping relationship.Taking the modified IEEE39 node system as an example,the proposed method is verified to have more satisfactory modelling speed and higher accuracy through algorithm and scenario comparison,and can cope with the new situation of power system voltage stability intelligent sensing requirements.(2)Defining the power system reactive power control problem as a reinforcement learning problem,the power system reactive power cooperative control method based PPO algorithm is proposed.Markov decision process modelling of power system reactive power control is first performed to transform the traditional power system reactive power control problem into a reinforcement learning problem,followed by a PPO algorithm to solve it under continuous state and continuous-discrete hybrid action space.For the reward function setting problem,constraint-objective partitioning and objective preconditioning are proposed to improve the convergence speed of agents.For the multi-objective of reactive power control,two groups of agents are adopted to control separately,and the corresponding agent are intelligently selected for control based on the static voltage stability sensing results.Finally,a case study is carried out with the modified IEEE39 system to verify the effectiveness of the proposed algorithm for cooperative reactive power control of power systems considering the stochastic nature of new energy output through algorithm and scenario comparison.