پيش بيني سري هاي زماني با بهره گيري از فيلتر دو مرحله اي: مطالعه موردي درپيش بيني سرعت باد، توان باد و قيمت نفت خام

Energy an‎d market time series forecasting is inherently challenging due to nonstationarity, noise, abrupt jumps, an‎d regime shifts. Many single-model approaches, an‎d even existing hybrid methods, often suffer from increased error an‎d reduced generalization during turbulent periods, particularly because of insufficient data denoising an‎d static output fusion. Accordingly, this thesis aims to develop a robust an‎d adaptive hybrid framework that performs multi-stage filtering an‎d adjusts the fusion of model outputs in a time-varying manner. This thesis proposes a model based on a Multistage Kalman Filter–Multilayer Perceptron–Long Short-Term Memory (LSTM) architecture. In the first stage, the time series is denoised through multiple passes of a Kalman filter an‎d decomposed into trend an‎d oscillatory components. The trend component is then modeled using a multilayer perceptron, while the oscillatory component is learned using an LSTM network. Finally, a dynamic fusion mechanism is employed to combine the component-level forecasts. The proposed framework is eva‎luated on wind an‎d crude oil datasets using a stan‎dard 80% training / 20% testing split. Numerical results show that the proposed model consistently reduces forecasting errors across all scenarios. For Brent crude oil, the RMSE decreases from 1.387 to 0.714 on the training set an‎d from 2.150 to 1.101 on the test set. The MAE is reduced from 0.878 to 0.438 (training) an‎d from 1.648 to 0.864 (testing). For WTI crude oil, the MAE decreases from 0.957 to 0.400 on the training set an‎d from 2.116 to 0.924 on the test set, while the RMSE improves from 1.527 to 0.634 (training) an‎d from 2.801 to 1.203 (testing). The simultaneous reduction in MAPE on both training an‎d test sets indicates that the improvement is not limited to absolute-scale metrics; rather, the model is also more stable in terms of relative error.

علي زينل همداني , مهدي خاشعي آشياني

صبا صارمي نيا , صفيه محمودي , اميرحسين اميري , علي شاهنده نوك آبادي