Digital twin driven energy yield forecasting of staircase based compressed air generators

The energy harvested from pedestrians’ footsteps using staircase compressed air generators (SCAGs) represents a novel source of micro-renewable energy that can be deployed in urban areas. Nevertheless, the randomness of pedestrian behavior renders the accurate prediction of generated energy difficult. In this paper, a methodology for SCAG’s energy generation forecasting using digital twins with physics-based models, machine learning algorithms, and real-time Internet of Things (IoT) data is suggested. The SCAG system was prototypically installed at the staircase of a three-floor academic building equipped with sensors of pressure, flow, and piezoelectric power together with edge-cloud computing system. The digital twin is developed based on Computational Fluid Dynamics (CFD), system-level models utilizing Modelica, and stacked Long Short-Term Memory Network (LSTM) for pedestrians’ behavior prediction updated in real-time by means of Extended Kalman Filtering (EKF). Various signal processing methods, such as wavelet transforms and anomaly detection, increase accuracy of obtained data. The experimental evaluation during 60 days has revealed mean absolute percentage errors of 4.8 to 6.4% outperforming traditional forecasting by 18 to 20%.