Abstract
Photovoltaic (PV) generation estimation plays a crucial role in the operation and planning of distribution networks (DNs). Existing literature primarily relies on model-based and data-driven methods for estimating PV generation in low voltage (LV) feeders. However, model-based techniques require accurate information regarding PV panel orientation, dust, shading, and cloud effects, which are often limited in DNs. On the other hand, supervised and semi-supervised data-driven approaches necessitate a ground-truth dataset for feeder or customer level estimation. Consequently, unsupervised data-driven techniques have gained traction in estimating behind-the-meter (BtM) PV generation in recent years. This paper focuses on feeder-level estimation, which is more practical and useful for utilities, as it eliminates the need for netload measurements at the customer level. We propose a new data-driven method using a signal-processing approach to estimate aggregated PV at the feeder-level. The PV scaling factor is estimated through first-order differencing and a moving average (MA) filter. The filter is designed to reduce the impact of quantization noise after employing the differentiator, based on the signal-to-noise ratio (SNR) on sunny days. By applying these techniques, the proposed approach offers an accurate estimation of BtM PV generation in LV feeders, enabling efficient operation and planning. To demonstrate the effectiveness of our method, we present comprehensive comparisons using a real dataset provided by an Australian distribution utility.
