Transformer Load Forecasting Method Based on SARIMA, FBProphet and LSTM Models

Power transformers are an important asset in a power network, and manufacturers are generally asked to carry out routine maintenance work to extend their lives. Further, online and offline testing are carried out to assess the health condition of the transformer to avoid inadvertent failure. In order to carry out routine maintenance (RM) and testing, the transformer needs to be switched off, causing power interruptions for many consumers if alternative arrangements are not possible. Thus, it is important to investigate the possibility of sharing the load of a particular transformer with other transformers in the neighboring transformers. Thus, this research proposed a method to estimate a load of neighboring transformers on a particular day and time, which allowed for the planning of outages with minimum power interruption to the consumers. This study has used loading data of four 132 kV grid substations which can share the distribution loads among and three different machine learning techniques have been used to estimate the load of the transformer at a particular time of the day. This research investigated the performance of load data prediction algorithms such as SARIMA, FB-Prophet, and LSTM. Findings from the study show that the LSTM model enhances the accuracy of power load predictions. This research examines the use of load data prediction algorithms such as SARIMA, FB-Prophet, and LSTM and found that the LSTM model improves the accuracy of power load predictions.