Author name cluster

Guodong Wang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

1 author row

EAAI Journal 2025 Journal Article

A short-term load forecasting method considering multiple feature factors based on long short-term memory and an improved temporal convolutional network

Yu Mu
Lingrui Kong
Guoqiang Zheng
Zhonge Su
Guodong Wang

In order to address the problems of multi-factor coupling difficulties and low prediction efficiency of existing short-term electricity load forecasting methods, in this paper a short-term load forecasting method is proposed that combines the maximum mutual information coefficient (MIC) algorithm and the Long Short-Term Memory (LSTM)-Improved Temporal Convolutional Network (ITCN) model. Second, based on the problem of low prediction efficiency of the Temporal Convolutional Network (TCN), the TCN was improved (ITCN) by using the single residual block structure and the parallel activation function structure. Finally, the LSTM-ITCN model is designed to extract the short-term temporal features of the given data using LSTM first, and extract the long-term temporal features of the given data using ITCN and make the final prediction. Comparison experiments with Convolutional Neural Network (CNN)-LSTM, CNN-Bidirectional Gated Recurrent Unit (BIGRU), and other prediction methods on different datasets are conducted, and the findings indicate that the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination ( R 2 ), and Running times values of the proposed method are improved by 10. 56%, 10. 48%, 8. 45%, and 25. 64%, respectively, which significantly improves the prediction accuracy and prediction efficiency.

Details DOI

EAAI Journal 2025 Journal Article

Multisource data-driven intelligent method for detecting surface defects in cold-rolled copper strips

Bowei Duan
Dongcheng Wang
Yuehua Ma
Guodong Wang
Hongmin Liu

Surface defects are critical in determining the quality of cold-rolled copper strips. Although deep learning-based detection techniques are widely used, they rely heavily on large amounts of annotated data, while in actual production, there are practical bottlenecks in sample acquisition and annotation. To address these challenges, a multi-source data-driven lightweight strip surface defect detection method (Ms-LSSD) is proposed. First, a dataset containing nine types of defects was curated, named the Cold-Rolled Copper Strip defect dataset (CRCS9-DET). Then, the architecture, based on Denoising Diffusion Probabilistic Model (DDPM), improves dataset diversity and mitigates issues related to sample size and imbalance. Next, a lightweight real-time object detector is designed based on Real-Time DEtection TRansformer (RT-DETR) to achieve precise detection of surface defects in cold-rolled copper strips. A novel convolution operator, Adaptive Partial Convolution (APConv), is proposed by introducing channel attention mechanism to enhance representational capacity across different tasks. And an Efficient Adaptive Residual network (EARNet) is established as the foundation for achieving a lightweight model. Finally, experiments were conducted on CRCS9-DET dataset and a public dataset (NEU-DET). Ms-LSSD achieves mean Average Precision (mAP50) of 90. 3 % on CRCS9-DET dataset and mAP50 of 78. 7 % on NEU-DET, with a detection speed of 52. 2 Frames Per Second (FPS), striking a balance between detection performance and inference speed. Compared to state-of-the-art algorithms, Ms-LSSD demonstrates more comprehensive performance potential, making it suitable for industrial applications.

Details DOI

AAAI Conference 2025 Conference Paper

Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation

Yajie Liu
Guodong Wang
Jinjin Zhang
Qingjie Liu
Di Huang

Training-free open-vocabulary semantic segmentation aims to explore the potential of frozen vision-language models (VLM) for segmentation tasks. Recent works reform the inference process of CLIP and utilize the features from the final layer to reconstruct dense representations for segmentation, demonstrating promising performance. However, the final layer tends to prioritize global components over local representations, leading to suboptimal robustness and effectiveness of existing methods. In this paper, we propose CLIPSeg, a novel training-free framework that fully exploits the diverse knowledge across layers in CLIP for dense predictions. Our study unveils two key discoveries: Firstly, the features in the middle layers exhibit high locality awareness and feature coherence compared to the final layer, based on which we propose the coherence enhanced residual attention module that generates semantic-aware attention. Secondly, despite not being directly aligned with the text, the deep layers capture valid local semantics that complement those in the final layer. Leveraging this insight, we introduce the deep semantic integration module to boost the patch semantics in the final block. Experiments conducted on 9 segmentation benchmarks with various CLIP models demonstrate that CLIPSeg consistently outperforms all training-free methods by substantial margins, e.g., a 7.8 % improvement in average mIoU for CLIP with a ViT-L backbone, and competes with learning-based counterparts in generalizing to novel concepts in an efficient way.

PDF Details DOI