Hossein Haghbin

In this study, a hybrid model based on clustering and recurrent neural networks (RNN) is introduced for time series forecasting. One of the main challenges in time series forecasting is the use of general models for all time series, which may reduce prediction accuracy. To address this issue, this research explores clustering methods to categorize similar time series, and for each cluster, a separate forecasting model is designed. In this context, recurrent neural networks are employed for modeling temporal sequences. In this approach, the features of each time series are first extracted, and then clustering algorithms such as \lr{K-Means} and \lr{K-Medoids} are used to group similar time series. After clustering, a separate forecasting model based on recurrent neural networks is designed for each group. The proposed model is tested on real consumption data from electricity users in Bushehr city, showing significant improvement in prediction accuracy compared to traditional univariate models. This model can be applied in various time series forecasting fields, such as demand forecasting, stock market prediction, and system status prediction.