Ultra-short-term Distributed Photovoltaic Power Forecasting Based on Cloud Image Feature Extraction

Distributed renewable resources have the characteristics of scattered geographical location, different scale and violent power fluctuation. In order to make full use of the potential of distributed renewable resources, high requirements are put forward for prediction methods. This paper proposes a forecasting method for distributed photovoltaic power prediction. Firstly, this paper establishes a fixed square window with the sun as the center, processes the cloud information and creates an image with only the relationship between the gray and the position of cloud. Secondly, according to the historical data, the position and grayscale picture of the cloud cluster at the next moment are generated. In the third step, proposes the influence factor of cloud cluster to evaluate the influence of the cloud distribution and thickness on the surface illumination, then input it into the improved CNN network as an auxiliary parameter, combined with the current outer atmospheric tangent plane solar radiation to get the predicted illumination at the next time. Finally, outputs the predicted photovoltaic power value, which combine the predicted illumination, current real-time temperature with the efficiency of the inverter. The case study demonstrates that this method has strong adaptability and it is capable of forecasting ultra-short-term changes in photovoltaic power. The method can greatly reduce the amount of data to the prediction network and also accelerate the convergence speed while ensuring the predict accuracy.