DLinear-BiLSTM hybrid gross load estimation model considering the absence of BTM PV generation data

The expanding adoption of distributed photovoltaic (PV) generators introduces uncertainties into power system operation. In particular, as distribution system operators often observe only substation net load, it can be a challenging issue to obtain the gross load by separating the unobserved behind-the-meter (BTM) PV from the net load. This paper proposes a hybrid DLinear BiLSTM,– model to estimate the gross load when BTM PV measurements are unavailable. The proposed model uses a DLinear-based decomposition to split the time series input data into trend and seasonal components. Then, a bidirectional LSTM jointly learns nonlinear correlations and long- and short-term dependencies. In particular, because it uses only accessible exogenous variables such as public meteorological data (solar irradiance, temperature, precipitation/snowfall), calendar factors, and a region identifier, the method is easy to implement and scale without detailed facility-level data. The proposed model is validated by the experiment using one-year dataset obtained from a substation.