Multi-step multivariate forecasting of transmission power in NPPs using operational and meteorological data

As the proportion of renewable energy has increased in the national power grid of Republic of Korea, various efforts are needed to maintain the stability of total power generation. All kinds of power plants, including nuclear power, must notify the grid operation organization of their expected transmission power. Even in NPPs, the accuracy of transmission power forecasting can increase the plant owner's economic benefits as well as the stability of the power grid. The transmission power of a NPP is affected by various plant conditions and environmental conditions, including the temperature of circulating water (sea water). In this study, we explored how to effectively handle the long-term dependence problem and various data characteristics to increase the forecasting accuracy of transmission power in NPPs by introducing a Seq2Seq model with an encoder-decoder structure and an attention mechanism, beyond traditional time series deep learning models, especially LSTM. This approach will improve the accuracy of transmission power forecasting and contribute to a stable power supply. Additionally, the model is expected to provide a realistic and practical solution for the power demand response of power plants.