Real-time Boiler Fault Detection Using the Multi-modal GCA Model Optimized by GVSAO

The real-time and accurate detection of boiler faults is essential for the optimization of boiler operation and the reduction of energy consumption and carbon dioxide emissions in power generation. However, due to the scarcity and limited diversity of boiler fault data and the complex requirements of feature modeling, significant challenges remain in developing such models. In this study, data representation techniques are applied to the task of real-time boiler fault detection. A CNN-GRU-Attention (GCA) model is designed, where raw time-series data are first transformed using Gramian Angular Field (GAF) to enable effective data representation learning. Subsequently, 1D temporal features and 2D spatial features are extracted through Gated Recurrent Unit (GRU) and Convolutional Neural Network (CNN) branches, respectively. The integration of these features through a self-attention mechanism further improves the model’s performance in terms of fault detection accuracy and robustness. The global variable snow ablation optimization algorithm (GVSAO) has been developed by integrating the good point set initialization method and the periodic mutation strategy within the traditional snow ablation optimization framework. The GVSAO algorithm is employed for the parameter optimization of GCA models. The experimental results on industrial boiler data demonstrate that the fault detection accuracy of this model is 93.1% and that its overall performance and response speed are superior to traditional methods. This method provides an effective solution for real-time fault detection of complex boiler systems.