A hybrid TCN-BiLSTM short-term load forecasting model for ship electric propulsion systems combined with multi-step feature processing

The target of load forecasting in this paper is the output power of electric propulsion motor, which can provide reference for the real-time optimization of ship energy management. In electric propulsion ships, the characteristics of high proportion of propulsion load, high dynamic and strong environmental dependence pose challenges to accurate propulsion load prediction. In this paper, a short-term prediction model of hydrogen ship propulsion load based on Temporal Convolutional Network (TCN) and Bidirectional Long Short-Term Memory (BiLSTM) is proposed. Firstly, the temporal features of long historical distance in ship navigation are extracted by TCN. Then, the temporal features are used as the input of BiLSTM, while learning the forward and backward information of the temporal features ,and mine the potential time correlation of the propulsion load. Finally, the prediction result is mapped into the sample space through fully connected layer. In data processing, Variable Modal Decomposition (VMD) is used to separate the multicomponent and nonstationary component features from the original sequence. Then, the data dimension is reduced through Principal Component Analysis (PCA) to retain the effective information characteistics. In model training, K-means clustering is used to achieve automatic data classification to improve the prediction performance. Based on navigational and hydrometeorological dataset of a hydrogen ship, this paper conducts comparative simulation experiments on multiple prediction time scales for three ship conditions: normal, working and in/out-of-port conditions. Compared with other models, the experimental results show the performance superiority of the proposed method in various operational conditions.