Dynamic mode decomposition for forecasting and analysis of power grid load data.

摘要: Time series forecasting remains a central challenge problem in almost all scientific disciplines, including load modeling power systems engineering. The ability to produce accurate forecasts has major implications for real-time control, pricing, maintenance, and security decisions. We introduce novel method which observed dynamics are modeled as forced linear system using Dynamic Mode Decomposition (DMD) time delay coordinates. Central this approach is the insight that grid load, like many observables on complex real-world systems, an "almost-periodic" character, i.e., continuous Fourier spectrum punctuated by dominant peaks, capture regular (e.g., daily or weekly) recurrences dynamics. presented takes advantage of property (i) regressing deterministic model whose eigenspectrum maps onto those (ii) simultaneously learning stochastic Gaussian process regression (GPR) actuate system. Our algorithm compared against state-of-the-art techniques not additional explanatory variables shown superior performance. Moreover, its use intrinsic offers number desirable properties terms interpretability parsimony.