Work packages 1: Operation and planning models

Work packages 1: Operation and planning models

Lead: P. S. Mikkelsen, DTU ENV

Identification of suitable modelling approaches across temporal and spatial scales using a range of deterministic and stochastic methods. The goal is to improve the use of forecasting and projections to create longer forecast and optimization time horizons and reduce impacts both for planning and operation. Research will also be directed to automation of the developed methods and procedures in an effort to ensure scalability and transparency.

WP1.1 Planning models

Develop and test fit-for-purpose planning models for generating scenarios of city redevelopment projecting inundation, combined sewer overflow and other relevant decision variables. Focus will be on transparency of deterministic parts, assessment of structural and operational (including real time control) alternatives, integration of states enabling integrated assessments, simulation speed and uncertainty in model forcing, in model structure, and in observations.

WP1.2 Forecast models

Development of prediction models specifically tailored for use in real-time forecasting and control systems. These include models for forecasting flow and pollutant transport in sewer systems, combined sewer overflows and impacts on receiving waters, and flooding. The development will use both surrogate deterministic models and data-based stochastic modelling approaches (grey-box modelling). Forecast models researched and developed will distinguish themselves from state-of-the-art on three main areas: longer forecast horizon, forecast of selected pollutant concentrations and an uncertainty characterisation based on situation specific weather and system conditions.

WP1.3 Model predictive control 

Development of model predictive control (MPC) systems for system-wide control optimisation of storm- and wastewater systems, considering multiple operation criteria: combined sewer overflows, pollutant loads and flooding. The development will follow two routes:

  1. Model-based MPC based on surrogate prediction models derived from high-fidelity urban water system models. The MPC problem will be formulated to enable use of effective optimisation algorithms that rely on approximated linear or piece-wise linear prediction models. Procedures will be developed for inclusion of model prediction uncertainty in the optimisation. The surrogate models will be updated in real-time from high-fidelity model simulations. In this regard, an automatic surrogate model builder will be developed where MPC models are automatically derived and calibrated from existing, high-fidelity models.

  2. Data-based MPC based on a real time assessment of current pollution and flood risks using observed data and grey-box system models to forecast flows, pollutant loads and flooding under a variety of real time operational scenarios. Efforts will be made to generalize the approach through automation of MPC structure identification and calibration based on network structure and observed data. The goal is to obtain an adaptive self-learning scheme that is generally applicable to any network under consideration of its particular structure and operational goals.

The two approaches will be evaluated and compared as part of the testing in WP3.