Peng Chen

Accurate short-term wind power forecasting is essential for ensuring power system stability and facilitating the large-scale integration of renewable energy. However, existing forecasting models often suffer from complex architectures that hinder practical deployment, insufficient modeling of multivariate dependencies, and limited capability in capturing multi-scale nonlinear temporal patterns. To address these challenges, this study proposes a novel PGA-HDD-Crosslinear framework for short-term wind and photovoltaic power forecasting. Specifically, the Crosslinear architecture is employed to efficiently capture multivariate dependencies with low computational complexity. A Hierarchical Dual Decomposition (HDD) strategy is further introduced, where STL decomposition is first applied to separate the input sequence into trend, seasonal, and residual components, followed by a multi-order Kolmogorov–Arnold Network (KAN) to extract latent nonlinear features from residual signals. In addition, the Plant Growth Algorithm (PGA) is utilized to optimize key hyperparameters and enhance model generalization. Experimental results on real-world wind and photovoltaic datasets demonstrate the effectiveness of the proposed framework. Compared with the baseline Crosslinear model, RMSE, MAE, and MAPE are reduced by 15.9%, 17.4%, and 18.2% on the Xinjiang photovoltaic dataset, respectively, while reductions of 16.0%, 18.9%, and 15.8% are achieved in wind power forecasting tasks.

The rapid expansion of multivariate time series (MTS) data has made deep learning a central tool for forecasting across industrial and scientific domains. As these architectures move into settings such as clinical decision support, industrial monitoring, and financial risk management, researchers need to examine not only predictive accuracy but also computational cost and the kinds of explanations that a model can support. This survey re-examines deep learning models for MTS forecasting through the requirements of efficiency and explainability. Our contributions are threefold: (1) We synthesize the literature through a dual-axis scalability view over sequence length L and variate dimension N , together with a three-question explainability taxonomy covering temporal importance, variate importance, and pattern decomposition. (2) We review dominant architectural families, including Transformer variants, frequency-domain methods, linear and channel-independent backbones, state space models, and foundation-style forecasters, to describe how different designs occupy different efficiency–explainability profiles under specific data and deployment assumptions. (3) Building on this synthesis, we identify key open challenges—including the absence of standardized explainability benchmarks for time series, the interpretability gap in state space models, and the need to advance from correlational to causal explanations—and offer reporting considerations to guide more comparable future research.

Study of early Chinese ceramics is crucial for understanding cultural, economic, and technological developments in Chinese history. With the evolving deep learning techniques, one urgent question would be, whether we can identify early Chinese ceramics by a simple 2D image without further domain knowledge. This work collected a highly diverse dataset for ancient Chinese ceramics from 15 dynasties, with 4 representative glaze colors and 15 shape types. We studied the performance of five state-of-the-art neural networks on two identification tasks: ceramic visual feature recognition and early Chinese ceramic dating. A class-imbalance learning strategy is designed to improve the models’ performance on multi-label tasks. To the best of our knowledge, our work is the first to introduce deep learning into early Chinese ceramic recognition on a large scale. Experiments prove that deep learning can recognize visual features like glaze and most shape types with high accuracy, while ceramic dating is feasible for the main dynasties but remains challenging along the overall history. Further quantitative assessment shows that cultural inheritance and artistic continuity can lead to reasonable false dating by classifying ceramics into adjacent dynasties or periods. Moreover, although domain knowledge is required for interpretation, deep learning shows great potential in recognizing even unlabeled time-relevant features, which can help study the inheritance and evolution of early Chinese ceramic development.