Research On Spatio-temporal Load Forecasting And Power Flow Optimization Technology Based On Graph Neural Network

With the rapid development of the domestic economy,the people’s living standards continue to rise,the production scale of various enterprises continues to expand,and the demand for electrical energy in all parts of the society is also increasing.At the same time,the management and dispatch of the power system is also facing new challenges.challenge.Load forecasting is the basic work to realize the management and dispatch of power system,and power flow optimization is also an important part of it.Therefore,realizing the accurate prediction of the future load distribution and size and realizing the optimization of the power flow are of great significance to the management and dispatch of the power system.In order to achieve accurate spatiotemporal load forecasting,a spatiotemporal load forecasting model based on graph neural network and recurrent neural network is proposed.The model captures the spatial structure of the power grid through the graph neural network,and captures the time dependence of the load data through the recurrent neural network,and extracts the inherent information of the load from two aspects.In order to better learn the importance of load information at each moment,it is proposed to integrate the attention mechanism into the prediction task to grasp the global change trend of load data.After obtaining the results of the load forecasting,in order to more effectively dispatch the power system and meet the power demand,a power flow optimization model based on graph neural network is proposed to quickly solve the power flow optimization task end-to-end.Considering that power flow optimization needs to be extended to large-scale power grids,and the importance of power system nodes is different,in order to better model the importance of nodes,an attention mechanism is added to form a graph attention network to model the importance of nodes.The proposed spatio-temporal load forecasting model takes into account the spatiotemporal characteristics of load information.It performs best in experiments with different number of grid nodes and different sequence lengths to be predicted,and the accuracy rate can reach more than 86%.The proposed power flow optimization model overcomes the limitations of traditional power flow optimization algorithms.It takes into account the state of the power grid and the topology of the power grid.The experimental results show that the proposed model has lower errors than other baseline models,and the prediction time for a single sample is longer than that of the basic The optimization solution algorithm is more than 1000 times faster.