Deep learning-based energy consumption prediction and anomaly detection of drying system in automotive paint shop

The drying system in automotive spray painting workshops is a major energy consumer, and its complex operations makes accurate energy consumption prediction challenging. This study proposes a data-driven methodology framework (DMFPEC) to effectively predict the energy consumption for such systems. First, data was preprocessed through missing data imputation, anomaly handling, energy calculation, and normalisation. The Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) was then used to decompose energy consumption data into intrinsic mode functions (IMFs) to capture multi-scale fluctuations. A hybrid deep learning model combining CNN and BiLSTM was developed for short-term prediction. An anomaly detection module was also integrated to avoid operational misjudgments caused by prediction distortion. The method was validated using real-world IoT data gathered from a new-energy vehicle painting workshop, achieving 97.52% prediction accuracy. Comparative results confirm its superiority and robustness, even in the face of missing or abnormal data. This approach offers an effective tool for energy management in automotive manufacturing and inspires further research in the field.