Short-term Load Temporal-spatial Forecasting Based On Graph Neural Networks

The modern power system is developing in the direction of intelligence,flexibility and networking,presenting a complex form of multiplicity,flexibility and correlation different from before.Load forecasting is of significant importance for the safety of grid dispatching operation and the accuracy of online analysis and decision making.The results can be applied to the analysis of issues such as voltage stability of distribution networks as well as providing a reference basis for the dispatching department to develop reasonable power-side output plans.Compared with the system load,the bus load and the charging load of public charging stations have a lower base.In addition,due to the influence of weather factors and user behavior,they are more volatile,random,and the trend of change is not obvious,which makes it difficult to predict accurately.To improve the accuracy of bus load and public charging station charging load forecasting,this paper proposes a spatial-temporal forecasting method for bus load based on gated graph neural network and a spatial-temporal forecasting method for public charging station based on adaptive graph neural network.In the analysis of spatial-temporal correlation between loads and the construction of multi-node feature set,the coupling correlation between loads is analyzed using Rapid Maximal Information Coefficient(Rapid-MIC)to quantify the spatial-temporal correlation between loads,which provides a basis for considering the coupling correlation between loads in the subsequent forecasting.Meanwhile,in order to reduce the temporal and spatial complexity of the forecasting model,the correlation between loads and natural meteorological factors is quantitatively analyzed to determine the complex and nonlinear coupling relationships between them and different meteorological factors,and based on this,a multi-node feature set containing complex meteorological features and historical load sequences is constructed.In bus load forecasting,in order to solve the problem that the previous Euclidean data are limited in portraying the spatial-temporal coupling correlation of multiple bus loads existing in non-Euclidean space in wide area space,a similarity power time-space diagram with load similarity between buses as edge features is constructed to realize the reconstruction of spatial coupling relationship between non-physical neighboring bus loads independent of geographical distribution and grid structure constraints.Based on this,a load spatial-temporal forecasting model based on gated graph neural network is constructed.The results show that the proposed method can take into account the coupling characteristics between bus loads and has better robustness,noise immunity and generalization ability.As for the charging load forecasting of public charging stations,the coupling association among public charging stations is difficult to be determined in advance due to the irregularity and randomness of the service recipients,and it is difficult to achieve accurate charging load forecasting.To this end,a spatial-temporal short-term charging load prediction method based on adaptive graph generation for multiple public charging stations is proposed.The results show that the new method has higher forecasting accuracy and efficiency,and effectively reduces the worst evaluation index.This study solves the problem that the existing load forecasting research is limited by the sequential input of data in the Euclidean space,which leads to insufficient mining of the spatial-temporal correlation between load data,and proposes the spatial-temporal forecasting method of bus load based on gated graph neural network and the spatial-temporal forecasting method of public charging station based on adaptive graph neural network to realize the spatial-temporal forecasting of multi-bus load and public charging station charging load.The results show that the proposed method has higher forecasting accuracy than other methods,and can effectively reduce the worst evaluation index which has the greatest impact on grid dispatch.