The increasing demand for real-time time series classification, especially in high-stakes industries, underscores the need for predictive explanations and energy-efficient processing. To address these challenges, we propose a sustainable and explainable streaming time series classification model. By incrementally updating a compact time series representation with new arrival batch embeddings, our model enhances classification accuracy while identifying key channels and time steps crucial to predictions. Particularly effective in streaming scenarios, it enables real-time classification while minimizing energy consumption and processing time. Experimental results demonstrate superior classification accuracy, explainability, and efficiency, achieving over \(70\%\) less energy consumption and threefold faster task completion compared to benchmarks. Our work significantly advances real-time responsiveness, result explainability, and energy conservation, offering optimization across diverse applications.