Mining Temporal Reservoir Data Using Sliding Window Technique

Abstract

Decision on reservoir water release is crucial during both intense and less intense rainfall seasons. Even though reservoir water release is guided by the procedures, decision usually made based on the past experiences. Past experiences are recorded either hourly, daily, or weekly in the reservoir operation log book. In a few years this log book will become knowledge-rich repository, but very difficult and time consuming to be referred. In addition, the temporal relationship between the data cannot be easily identified. In this study window sliding technique is applied to extract information from the reservoir operational database: a digital version of the reservoir operation log book. Several data sets were constructed based on different sliding window size. Artificial neural network was used as modelling tool. The findings indicate that eight days is the significant time lags between upstream rainfall and reservoir water level. The best artificial neural network model is 24-15-3.