The ratio between normal data and fault data generated by electric submersible pumps (ESPs) in production is prone to imbalance, and the information carried by the fault data generally as a minority sample is easily overwritten by the normal data as a majority sample, which seriously interferes with the fault identification effect. For the problem that data imbalance under different working conditions of ESPs causes the failure data to not be effectively identified, a fault identification method of ESPs based on unsupervised feature extraction integrated with migration learning was proposed. Firstly, new features were extracted from the data using multiple unsupervised methods to enhance the representational power of the data. Secondly, multiple samples of the source domain were obtained by multiple random sampling of the training set to fully train minority samples. Thirdly, the variation between the source domain and target domain was reduced by combining weighted balanced distribution adaptation (W-BDA). Finally, several basic learners were constructed and combined to integrate a stronger classifier to accomplish the ESP fault identification tasks. Compared with other fault identification methods, our method not only effectively enhances the performance of fault data features and improves the identification of a few fault data, but also copes with fault identification under different working conditions.

Sea Surface Temperature (SST) plays a crucial role in the global meteorological system, particularly as long-term seasonal variations are significant for analyzing SST anomalies and supporting long-term climate decision-making. Current forecasting methods are primarily focused on short-term or fine-grained predictions and often fail to effectively capture long-term seasonal trends. To address this, we introduce a novel Attention Adaptive Temporal Graph Convolutional Network (AA-TGCN) specifically designed for long-term seasonal SST forecasting. Unlike traditional adaptive graph convolutional networks, the AA-TGCN incorporates an attention mechanism to capture inter-node correlations and utilizes a pruning strategy from SGAT to eliminate noisy connections, thereby improving the model's inductive learning capabilities. Moreover, the network employs a TCN-like architecture to expand its receptive field, enhancing its ability to grasp long-term trends, and employs differential embedding to further refine the prediction accuracy of seasonal fluctuations. Practical applications in the Bohai Sea and parts of the South China Sea demonstrate that AA-TGCN outperforms existing technologies on multiple scales, particularly achieving significant improvements in the South China Sea regions.

Accurate prediction of sea surface temperature (SST) is essential for marine economic development, early warning of extreme climate events, and ocean resource management. However, the non-stationarity and nonlinear dynamics of oceanic meteorological data pose significant challenges for prediction. This study proposes a Sampling-based Koopman Interactive Model (SKIM), which integrates Koopman theory to improve SST prediction through interactive feature extraction and linearized modeling. SKIM first employs a downsampling interaction structure to enhance the capture of multi-scale features and to improve the efficiency of temporal representation. Then, a Fourier filter is applied to decompose the complex non-stationary time series into linear and nonlinear components, which are further analyzed by the Koopman operator to enhance the model’s generalization ability. We conducted experiments on real-world datasets from the South China Sea and the Bohai Sea. The results show that SKIM significantly outperforms traditional machine learning methods (RF, SVR) and deep learning models (ConLSTM, TCN, SCINet), achieving the lowest prediction error and the highest accuracy in spatial distribution prediction.