A review of research on spatial-temporal forecasting of power load based on spatial-temporal data mining

• The first comprehensive review of the current state of research of spatial-temporal forecasting of power loads is presented. • From the perspective of heterogeneity, the reasons for the differences in forecasting performance on datasets were discussed. • We explored the key future technological research directions for spatial-temporal forecasting of power load. Lately, in the context of global energy crisis and economic recovery after the COVID-19 epidemic, the contradiction between power supply and demand has become more prominent, showing a tight balance. Therefore, the accurate forecasting of power load has attracted great attention. Power load forecasting involves the use of reliable methods to accurately forecast future electricity demand, understand the patterns and trends of power load variations, and is an essential requirement for ensuring operation of the power system. However, due to the extensive integration of new factors into the load side, the power load exhibits complex and variable spatial–temporal features in various typical forecasting scenarios, making it difficult for traditional forecasting methods and means to cope. Thanks to the leapfrog development of AI technology, new spatial–temporal forecasting technology provide solutions to the new challenges faced by power load forecasting. In this context, this article comprehensively reviews the latest research progress in the field of spatial–temporal forecasting of power load based on spatial–temporal data mining from multiple perspectives. Firstly, the concept of spatial–temporal data was introduced, and the spatial–temporal features of power load were analyzed. The definition of spatial–temporal forecasting of power load was given. Secondly, from the perspectives of spatial–temporal forecasting technology, mining methods of spatial–temporal forecasting technology, methods of describing spatial correlations, and the step size of spatial–temporal forecasting, this article concentrates on summarizing the application of current spatial–temporal forecasting technology in power load forecasting. Then, in response to the new features and challenges faced during forecasting, a comprehensive and detailed discussion was conducted on the research progress of spatial–temporal forecasting methods for power load based on spatial–temporal data mining from three typical forecasting scenarios: electric vehicle charging load, spatial load, and residential load. Furthermore, from the perspective of inherent heterogeneity in the dataset, a thorough analysis is conducted on the problems and challenges faced by current spatial–temporal data mining based power load forecasting methods. Finally, the potential research directions of spatial–temporal forecasting technology for power load were further explored, in order to provide actionable insights for the advancement of smart grids in the context of massive spatial–temporal big data.