Research On Demand Side Response And Optimal Dispatching Strategy Of Smart Grid Based On Reinforcement Learning Algorithm

With the automation of social production and people’s high requirements for the quality of life,the consumption rate of traditional primary energy has been further increased.Therefore,the society has paid more and more attention to accelerate the replenishment of energy gap and the treatment of environmental pollution.Under the national "2060 carbon neutrality" requirement,it is crucial to improve the utilization rate of traditional primary energy and vigorously develop renewable energy power generation technology.Wind energy and solar energy are abundant in nature and can be used as new energy sources to be connected to the grid.However,due to their strong volatility,the stable operation of the grid is facing challenges.At the same time,with the diversified features of loads,precise regulation of demand-side flexible loads can also be used as an effective means to improve the balance of power supply and demand.In view of the above problems,there have been many literatures at home and abroad for research.On this basis,this article analyzes it from an economic point of view and combines a variety of intelligent algorithms.The research contents of the paper are as follows:Firstly,according to the intermittency and randomness of renewable energy,the wind power generation and photovoltaic power generation modes are analyzed.For the power system on the power generation side,considering the grid connection of wind power and photovoltaic,combined with the energy storage device system,in order to reduce the cost of thermal power generation,the economic dispatching model of energy complement is established.To solve the model,an improved PSO algorithm based on dynamic adjustment of inertia weight parameters is proposed.At the same time,the worst particle elimination strategy is also added into the algorithm.The improved PSO algorithm is used to optimize the output of thermal power units,which can further improve the operation cost and economy.Secondly,the uncertainty of load also brings a lot of difficulty to the grid dispatching,so the shortterm prediction of the user load on the demand side is considered first.In order to improve the accuracy of load forecasting,based on the analysis of load features on the demand side,the demand law of power users under the features of time series is further mastered.On this basis,Long ShortTerm Memory neural network is used to carry out load prediction.According to the time series features of load data,a hybrid optimization method of FA-PSO is proposed to conduct iterative optimization of the parameters of LSTM neural network,and then load data is input into the optimized LSTM to train the weight of LSTM network.Compared with the traditional BP neural network and the classical LSTM neural network,the prediction accuracy of the proposed FA-PSO-LSTM prediction model has been greatly improved.Finally,in view of the uncertainty of the demand side load,the price incentive policy is considered to guide the power users to transfer the flexible load during the time period,forming a power sales company-power user win-win power dispatch demand response model.The model takes into account the income of electricity selling companies and the expenses of electricity users,and defines the cost function of dissatisfaction of electricity users.The solution of the model is based on the principle of reinforcement learning Q-learning algorithm,with the change of retail price as the action set and the state set as the power consumption,power demand and load transfer willingness coefficient of power users.Through the interaction between electricity users and electricity selling companies,the retail price of electric energy can be reasonably established and the decision of users can reach the optimal.Numerical examples show the effectiveness of the demand response model,and it is beneficial for grid load peak shaving and valley filling.