Wind power,photovoltaic power and load power in the same area are closely related to meteorological factors such as wind speed,irradiance and temperature,and there is a certain interactive coupling relation in different operation scenarios of power system.In order to capture the difference and correlation between multi-source loads and mine the potential rules contained in high-dimensional data,a multi-task progressive learning model based on deep spatiotemporal fusion network is proposed to realize the wind-photovoltaic-load joint power prediction.Firstly,the shared information and specific information subnet are designed based on deep spatiotemporal fusion network.Then,a multi-task progressive learning model with time-space correlation is constructed to progressively extract the shared and unique spatiotemporal information of wind,photovoltaic and load power from shallow to deep.Finally,the feature vectors obtained from shared information and specific information subnet are fused and mapped to realize the future wind-photovoltaic-load joint power prediction.The actual day-ahead wind-photovoltaic-load power short-term prediction example shows that the proposed model can compensate for the shortcomings of the existing multi-task model with"negative migration"and"seesaw"phenomena,and thus improve the prediction accuracy and robustness.
multi-source loadmulti-task progressive learningwind-photovoltaic-load joint power predictiondeep spatiotemporal fusion networkelectric power systemphotovoltaic power generationwind power generation
国家科技期刊平台
NSTL
万方数据
