| Reliable and safe operation of distribution network system is the key to ensure people’s daily life and production.Once the reliability of the distribution system has problems,in addition to the impact of power failure on people’s daily life and enterprise production,it also brings high maintenance and repair costs.The purpose of improving the reliability of distribution network is increasingly becoming the focus and difficulty of power system management.In order to improve the reliability of distribution grid,solutions are sought from internal and external aspects.From the external point of view,from the perspective of user demand,through the price to stimulate the supply-demand relationship,adjust the demand to adapt to the supply,and improve the adequacy of the distribution network;from the internal point of view of the distribution network,through the internal control of the scheduling of electric energy,the power can be transferred from the overload area to the area with insufficient electric energy,so as to improve the security of the distribution network.The real-time pricing and control mechanism of distribution network can reduce the congestion and improve the reliability of power grid.In order to realize the above two schemes,deep reinforcement learning algorithm is applied to model construction.The first choice is to use the index of demand price elasticity to simulate the market environment of power supply and marketing.The environment receives the price information and feeds back the user’s electricity consumption information.Then,Deep Q Network(DQN)and related optimization algorithm are applied to the reliability real-time pricing model test.Finally,the reinforcement learning algorithm based on policy is applied to the reliability control model test of distribution network.In the real-time pricing experiment of distribution network reliability,the best action selection rate reaches 86.34%.In the distribution network reliability control experiment,the average revenue index reached 9967 points.Experiments show that the deep reinforcement learning algorithm has two advantages when applied to power grid pricing mechanism and control mechanism.First,it can collect and train the data of the power grid in real time.It is not only fast in calculation,but also has strong adaptive ability.Even if the user’s behavior changes,it can also learn and update synchronously.Secondly,it can choose the appropriate strategy in each strategy selection,and ensure the stability and high reliability of the power grid. |
