Inverter-Data-Driven Second-Level Power Forecasting for Photovoltaic Power Plant

Globally, the installed capacity of photovoltaic (PV) power plants is undergoing rapid growth. However, the random output power fluctuation of PV plants has brought great challenge to power system stable operation. One of the most effective approaches to counteract the impact of power fluctuation is to carry out power forecasting under ultra-short-term scales, among which the second-level power forecasting technique is still absent since the pure historical data analysis of PV plant output power cannot deal with the sudden output power change induced by moving cloud and the traditional cloud image detection methods actually fail to achieve accurate PV power forecasting in high temporal and spatial resolutions. This article proposes an inverter-data-driven method to achieve the second-level PV power forecasting. In specific, multilayer feed-forward artificial neural network based on the error back propagation algorithm is applied to first establish the mapping relations from shading conditions of PV array to its corresponding output power. Combining with output power of neighboring PV arrays as well as their geographical location information, the shading conditions of those shaded PV array can be reversely deduced. And then, the deduced shading conditions of all PV arrays will be converted into one virtual cloud image to identify the instant cloud characteristics. In consequence, the consecutively generated virtual cloud images can depict the moving direction, speed, and shape of shading cloud and then help easily realize the second-level power forecasting of PV plant. Case study and experimental results verified the performance of the proposed method.