A multi-model ensemble digital twin solution for real-time unsteady flow state estimation of a pumping station

This paper proposes a digital twin solution for unsteady flow state estimation in a pumping station. Digital twin is expected to accurately estimate the real-time hydraulic parameters of blind spots of the pumping station system even under some adverse conditions including the interference of observation noise and model parameters drift. To solve these challenges, a digital twin framework integrating the model-driven method, control theory and data-driven method is presented. In this framework, an unsteady flow state estimation method combining frequency domain analysis and generalized predictive control theory is developed for the first time, which is superior to traditional time-domain numerical discrete methods in terms of computational efficiency and anti-noise interference. In the model parameter calibration process, the novelty concerns modeling of the optimization problem considering the dynamic operation control of the station and unsteady flow of pipelines. And this process is accomplished through the comprehensive application of the model-free adaptive control algorithm, the transient flow model and the particle swarm optimization algorithm. This work is applied to a pumping station in a product pipeline to verify its effectiveness in estimating the transient flow state of data blind spots and map the dynamic operation behavior under the interference of colored noise and parameter drift.