As a significant fashion to empolder and utilize the solar energy,the photovoltaic power has gained burgeoning expansion of late years.However,there are randomness in photovoltaic output as it is affected by some meteorological factors.These shortcomings perhaps bring rigorous challenges to power balance,security stability and economic operation of power system.Cloud motion over photovoltaic power station,such as generation,disappearance and deformation,is a direct factor of rapid and dramatic change in surface irradiance,which indirectly results in minute-level fluctuation in photovoltaic power.Therefore,real-time and accurate tracking of cloud is crucial.In this study,block-matching algorithm,optical flow algorithm and feature matching algorithm are three prevailing methods.However,they all have poor robustness.Therefore,this paper optimizes and combines these methods for different cloud motion patterns,and proposes a cloud tracking method suitable for multiple cloud motion patterns.The main contents are as follows.Firstly,sky image preprocessing is used to sharpen image and filter no ise.Secondly,the relevant feature vectors of the image are extracted from the perspectives of texture,chromaticity,information theory and frequency domain.Next,based on the above feature vectors,the k-means algorithm is used to cluster the sky image pairs to obtain different cloud motion patterns,and the cloud motion pattern recognition model is established.Then,based on the above motion pattern recognition model,the particle swarm optimization algorithm is used to optimize and combine the above three traditional algorithms to establish the calculation model of cloud displacement vector.Finally,the model and the traditional algorithms are simulated and compared using the sky images collected by the Yunnan Electric Power Research Institute to ver ify the effectiveness of the model. |