부산대학교 도서관

The research proposes a novel approach to enhance the accuracy of Facebook Prophet model for short-term load forecasting (STLF) by encompassing feature selection and discrete wavelet transform (DWT). First, correlation analysis is performed to select the relevant features from the dataset. By identifying the influential variables associated with load demand, the model’s predictive capabilities are improved. To address the non-stationary behaviour of time series, DWT is applied to extract the trend component. This captures long-term patterns and variations in load demand, which are essential for accurate STLF. The Facebook Prophet model, known for its ability to handle time series with seasonality and trends, is then implemented for forecasting. Multiple evaluation metrics are employed to assess the performance of the proposed approach on testing data and comprehensive comparison with the already existing prophet model is done to recognize the importance of DWT in the proposed model.