Smart city cloud-IoT analysis with spatio-temporal attention and uncertainty-aware spectral signal denoising

The deployment of Internet of Things (IoT) technologies in smart cities has generated massive amounts of energy consumption data, creating strong potential for intelligent energy demand forecasting. Noisy sensor readings, missing data, and complex temporal dependencies have prevented conventional deep learning methods from achieving higher prediction accuracy. To address this challenge, this paper presents an innovative approach for accurate energy demand forecasting: the Smart City Cloud-IoT Spatio-Temporal Attention Network with Uncertainty-Aware Spectral Signal Denoising (SCC-IoT-STAN-USSD). Firstly, IoT energy consumption data are preprocessed through Bayesian Boundary Trend Filtering (BBTF) to remove noise measurement, fill in missing data, and retain significant temporal patterns. Then, an Amplitude-Preserving Adaptive Synchro squeezed S-Transform (APST) is used to acquire time–frequency representations from the filtered signals. The above features are then analyzed using the Spatio-Temporal Attention Network (STAN) to extract spatial correlation and long-term temporal dependencies for accurate forecasting. In addition, the Uncertainty-Aware Spectral Signal Denoising (USSD) method improves prediction accuracy by reducing uncertainty during the process of model optimization. Experimental results demonstrate that the proposed method achieves a prediction accuracy of 98.20%, MAE of 0.31, RMSE of 0.24, MAPE of 2.14%, SMAPE of 1.98%, R 2 score of 0.992, and energy efficiency of 97.84%, thus surpassing the performance of ANN, LSTM, CNN-LSTM, GNN, and recently proposed methods. Additionally, statistical significance testing, regression analysis, and ablation tests show the reliability and effectiveness of the proposed approach.