Metro

Abstract

Multivariate time series forecasting has been drawing increasing attention due to its prevalent applications. It has been commonly assumed that leveraging latent dependencies between pairs of variables can enhance prediction accuracy. However, most existing methods suffer from static variable relevance modeling and ignorance of correlation between temporal scales, thereby failing to fully retain the dynamic and periodic interdependencies among variables, which are vital for long- and short-term forecasting. In this paper, we propose METRO, a generic framework with multi-scale temporal graphs neural networks, which models the dynamic and cross-scale variable correlations simultaneously. By representing the multivariate time series as a series of temporal graphs, both intra- and inter-step correlations can be well preserved via message-passing and node embedding update. To enable information propagation across temporal scales, we design a novel sampling strategy to align specific steps between higher and lower scales and fuse the cross-scale information efficiently. Moreover, we provide a modular interpretation of existing GNN-based time series forecasting works as specific instances under our framework. Extensive experiments conducted on four benchmark datasets demonstrate the effectiveness and efficiency of our approach. METRO has been successfully deployed onto the time series analytics platform of Huawei Cloud, where a one-month online test demonstrated that up to 20% relative improvement over state-of-the-art models w.r.t. RSE can be achieved.

Multi-Granularity Spatio-Temporal Correlation Networks for Stock Trend Prediction

“…Time series forecasting has received extensive attention in academia and industry, including financial markets [1]- [3],retail sales [4], transportation [5], [6], and the energy sector [7]. In the financial markets, stock prediction plays a pivotal role in guiding the investment decisions of market participants.…”

Deep Learning for Time Series Forecasting: Advances and Open Problems

“…Time series forecasting estimates values that a time series takes in the future, allowing the implementation of decision-making strategies, e.g., abandonment of fossil fuels to reduce the surface temperature of the Earth. Specifically, time series forecasting is very relevant for the energy domain (e.g., electricity load demand [7,8], solar and wind power estimation [9,10]), meteorology (e.g., prediction of wind speed [11], temperature [12,13], humidity [12], precipitation [13,14]), air pollution monitoring (e.g., prediction of PM 2.5 , PM 10 , NO 2 , O 3 , SO 2 , and CO 2 concentrations [12,15,16]), the finance domain (e.g., stock market index and shares prediction [17,18], the stock price [19,20], exchange rate [21,22]), health (e.g., prediction of infective diseases diffusion [23], diabetes mellitus [24], blood glucose concentration [25], and cancer growth [26]), traffic (e.g., traffic speed and flow prediction [27][28][29][30]), and industrial production (e.g., petroleum production [31], remaining life prediction [23,32,33], industrial processes [34], fuel cells durability [35], engine faults [36]). Deep learning algorithms are currently the leading methods in machine learning due to their successful application to many computer science domains (e.g., computer vision, natural language processing, speech recognition).…”

One for All: Unified Workload Prediction for Dynamic Multi-tenant Edge Cloud Platforms

“…However, MT-ECP constitutes a dynamic system due to the heterogeneous application mode, differentiated infrastructure attributes, and frequent application deployments, unlike the traditional cloud service. Workload prediction, from Markov models [18] and moving averages [13] to neural networks [5] and complex hybrid models [1,11,29], has grown accurate and efficient. Although these models effectively predict workloads in stable, static deployment systems, they struggle in dynamic systems like MT-ECP.…”