Currently power series with single time scale and large sampling intervals are generally used in modelingphotovoltaic generation forecast. Simple as the model is, its simulation accuracy of the time-series characteristics of photovoltaic (PV) power series is not high. In order to solve this problem, this paper proposes a local forecasting method for multidimensional time-series based on PV power series with small sampling interval. By constructing the main value series of PV power with different time-scales, the multidimensional time-series with hourly average PV power series as the main series is obtained. The correlation analysis, C-C method and minimum prediction error method of embedding dimension are used to compute the time delay and embedding dimensions of the reconstructed phase space of the multidimensional time series. The 1-hour ahead local forecasting model for PV power is developed by using support vector regression after phase space reconstruction. To demonstrate the effectiveness, the model is applied and tested in a microgrid. Simulation results show that the proposed local forecasting model based on multidimensional time-series outperforms the local forecasting model based on one-dimensional series, hence it has a better application value.