The invention discloses an electrical quantity ultra-short prediction method and system based on data mining, wherein the method comprises the steps: standard normalized processing is performed on a to-be-measured time series, and calculations of non-linear polymerization degree and an optimal embedded dimension E are performed on standard normalized data, so that the non-linear degree of a given dynamic system is inspected; according to the optimal embedded dimension E, delay time [tau] of the to-be-measured time series is calculated; according to the optimal embedded dimension E and the delay time [tau], phase-space reconstruction is performed on the to-be-measured time series; a experience dynamic model is constructed, a simplex projection method is adopted in the reconstructed phase space to predict the given dynamic system, and a prediction result of the to-be-measured time series is obtained.