Research On Intelligent Voltage Regulation Method For Active Distribution Network Based On Deep Reinforcement Learning

According to the "China Energy Development Report(2022)",83.4% of China’s energy demand in 2021 is met by fossil fuels such as coal,oil,and natural gas.With the increasing demand for human energy,non-renewable resources such as fossil fuels are facing the danger of depletion.In addition,the environmental impact of high carbon emissions from fossil fuels cannot be ignored.Facing the energy and environmental crisis caused by excessive dependence on fossil fuels,it is an inevitable choice to strengthen the substitution of renewable energy.The "14th Five-Year Plan for Renewable Energy Development" proposes to vigorously develop renewable energy such as photovoltaics and reduce the use of fossil fuels.By the end of December 2022,the installed capacity of photovoltaic power generation will be about 390 million kilowatts,a year-on-year increase of 28.1%.However,due to the intermittency and uncertainty of renewable energy,a high proportion of renewable energy connected to the distribution network will cause two-way flow of power flow,which will intensify the risk of voltage violation and affect the safe and reliable operation of the distribution network.Therefore,it is of great significance to study the voltage regulation problem of active distribution network.Firstly,the problem of minimizing voltage violations in active distribution network under uncertain environment is studied.Due to the inaccuracy of information such as model and line parameters in the active distribution network,it is challenging to solve the above problem directly.Therefore,it is modeled as a Markov game and an interpretable voltage regulation strategy is proposed.Based on the strategy trained by the attention-based multi-agent proximal policy optimization(AMAPPO)algorithm,the decision tree strategy is extracted from it,which overcomes the lack of interpretability of existing deep reinforcement learning-based voltage regulation methods.The simulation results show that the proposed strategy can achieve similar performance to the strategy based on AMAPPO,and increase the interpretability.Compared with the droop control method,the average voltage deviation and the maximum voltage deviation are respectively reduced 97.90% and 92.80%.Secondly,the joint minimization problem of user comfort,electricity cost and voltage violation is studied when the residential building energy system participates in the voltage regulation of the active distribution network.In this problem,each building energy system needs to minimize its own electricity cost under the premise of ensuring user comfort.At the same time,each building energy system needs to coordinate and participate in voltage regulation.Since it is difficult to obtain an accurate and well-defined model of building thermal dynamics,and each residential building needs to protect its own private information,it is challenging to solve the above problem directly.To this end,the above problem is remodeled as a Markov game and a collaborative control method based on multi-agent proximal policy optimization(MAPPO)algorithm is proposed.This method does not need to know the building thermal dynamic model and uncertain parameter information.The simulation results show that compared with the rule-based method,the proposed method can reduce the average temperature deviation by 96.97%～99.95%,and can reduce the average voltage deviation by 11.17%～77.58% under the condition of equivalent electricity cost,realizing the compromise between user comfort,electricity cost and voltage violation.Finally,the work of the paper is summarized,and the future research work is prospected.