An Integrated WOA-CNN-BiGRU-SE Model and its Application in Medium and Short-Term Time Series Forecasting

Time series contain unknown and valuable information related to time dynamics. Forecasting time series has important significance and application value for production and life. Existing forecasting algorithms have issues such as weak nonlinear processing capabilities and obvious lag in forecasting results. For this reason, this paper proposes a combined forecasting model integrating Whale Optimization Algorithm (WOA), Convolutional Neural Network (CNN), Bidirectional Gated Recurrent Unit (BiGRU) and Squeeze-and-Excitation (SE) block for medium and short-term time series forecasting. Specifically, considering the limitations of BiGRU in terms of feature extraction capability and noise immunity, the CNN module is introduced to strengthen the model's capability to extract data features and to lessen the impact of noise on forecasting accuracy. In addition, to address the problem that the CNN-BiGRU model cannot effectively evaluate the importance of different features and is insufficient in dealing with global information, the SE block is further integrated so that the model is able to weigh different parts of the input data differently and highlight the key information. Finally, to further optimise the model's overall performance, WOA is used to adjust the model's hyperparameters, aiming to improve both the training efficiency and training quality of the model, as well as to improve its generalization capability. The proposed model is tested on real-world scenarios for validation. The experimental results demonstrate that it achieves good forecasting performance, with an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.8906 and an RMSE of 0.0507, thereby verifying the model's validity and applicability. Compared with other combined forecasting models, the proposed model improves the R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> by 0.337%-0.485% and reduces the RMSE by 1.3410/0-1.909%, which verifies the superiority of the model. The constructed WOA-CNN-BiGRU-SE model is characterised by excellent medium and short-term forecasting, robustness, high training efficiency, and can be widely used in various fields.