深度学习在时序预测中的应用研究综述OA

Time series data is a dynamic data set continuously recorded along the time dimension.Its notable features are reflected in the interdependence among data and the trend of changes over time,which depict the evolution laws of variables in the time dimension.Due to the increasingly complex characteristics of time series data such as nonlinear superposition,high noise interference and strong coupling of multiple variables,it is difficult for traditional machine learning methods to break through the limitations of their own modeling capabilities when dealing with such data.Thus,deep learning,with the advantages of end-to-end autonomous feature learning and flexible network structure design that can adapt to complex data,has gradually developed into the mainstream technology in the field of time series prediction.To promote the continuous innovation and practical application of time series prediction technology,a systematic review is conducted on the application of deep learning in this field in the present paper.Taking the evolution of time as the main line,the mainstream methods and related research progress of deep learning in time series prediction are systematically sorted out.Based on the research status,the core challenges to be solved in this field at present and looks forward to the future research directions are discussed.