Evolutionary gate recurrent unit coupling convolutional neural network and improved manta ray foraging optimization algorithm for performance degradation prediction of PEMFC. (15th April 2023)

• Record Type: Journal Article
• Authors: Tao, Zihan; Zhang, Chu; Xiong, Jinlin; Hu, Haowen; Ji, Jie; Peng, Tian; Nazir, Muhammad Shahzad
• Is Part Of: Applied energy. Volume 336(2023)
• Publication Date: 2023-04-15
• Subjects: PEMFC performance degradation prediction -- Mutual information -- Locally weighted scatterplot smoothing -- Convolutional neural network -- Manta ray foraging optimization -- Gate recurrent unit; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2023.120821
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: MI is employed to select important features as input variables. LOESS is used to smooth the noise and spikes from the degradation data. Three improvement measures are employed to enhance MRFO. The improved MRFO is adopted to optimize the hyperparameters of the GRU model. The proposed model can predict the degradation data of PEMFC effectively. Abstract: Performance degradation prediction is an effective method to improve the durability of proton exchange membrane fuel cell (PEMFC). In this study, a hybrid deep learning model based on two-dimensional convolutional neural network (CNN2D), gate recurrent unit (GRU), and improved manta ray foraging optimization (IMRFO) algorithm is proposed for performance degradation prediction of PEMFC. Firstly, the mutual information (MI) and the locally weighted scatterplot smoothing (LOESS) are used to preprocess the data in order to boost the sample quality and reduce the influence of insignificant and noisy data on the model prediction. Secondly, CNN2D is used to deeply explore the nonlinear degradation characteristics in the data. Thirdly, three strategies including half uniform initialization, exponential weight coefficient and fitness-distance balance (FDB) are added to the algorithm to improve the defect that the optimization algorithm is easy to fall into local optimum. Finally, the GRU model optimized by the improved MRFO algorithm is used to predict the degradation data and obtain the final prediction results. The experimental results show that the prediction accuracy of the proposed prediction model in this study is 99.79%, and the RMSE and MAE are 0.0072 and 0.0042, respectively. Therefore, the method can effectively explore the deep features in the data and improve the accuracy, reliability, and robustness of PEMFC performance degradation prediction.