Implementation of model predictive control for the sewer system in Kolding, Denmark

The BlueKolding utility in Kolding, Denmark (population ~90,000) has implemented global real time control in 19 basins in the combined sewer network. The objective of the implementation was to reduce combined sewer overflows, minimize the risk of flooding in the city and at the same time reduce the needed extension of basin volume from 5,000 m3 to just 2,000 m3 to fulfill the demands and maintain the same level of service. This system has now been running, tuned and operated with success. BlueKolding has decided to take the next step and introduce cloud-based model predictive control in the system (Jess, 2018). The goal of this is to further reduce CSOs in the system, make it easier to implement and calibrate new control points, maintain the system and prepare for future developments. Background: The control algorithm used in BlueKolding, was developed in its original form during the Storm- and Wastewater Informatics project (Vezzaro and Grum, 2012; Vezzaro et. al., 2013). Since then it has been tested online and gone through several upgrades, which have added more robustness into the algorithm. The updated cloud-based control algorithm in Sewerflex is now part of AQUAVISTA, Veolia Water Technologies' global digital offer for real-time performance optimization of water treatment facilities. Methods: The MPC routine: The Sewerflex MPC routine is performing a coordinated 'smart' usage of the retention basin volumes in order to minimize combined sewer overflows. The control strategy is based on a dynamic risk assessment where the risk of overflow is calculated for every retention basin every 5 minutes, based on:

-- Actual retention basin filling and total retention basin volume
-- Predicted runoff volume from dry weather and rainfall forecast
-- The relative cost due to overflow (depending on i.e. recipient sensitivity)
-- Possible hydraulic emptying capacity of the retention basin equivalent to fixed gate position and/or pump operation
-- The actual hydraulic capacity of the downstream wastewater treatment plant.

Using a simplified model of the sewer system (Figure 1.1) together with a genetic optimization algorithm, the total overflow risk (all retention basins combined) is minimized, by changing the emptying capacities of the retention basins. The simplified model contains the retention basins, the catchment to the retention basins, the sewer connections between the retention basins and a downstream boundary (the wastewater treatment plant). By using a relative cost for overflow (cost/m3) at each location, the optimal solution - predicted optimal flows in connections between retention basins - takes into account which retention basins are more 'expensive' to use than others. This will try to locate any unavoidable combined sewer overflows at 'low cost' locations -- being the most environmentally robust locations. These settings can be changed at any time by the operator, allowing different CSO priority during the year to prioritize e.g. bathing waters and beaches in the summer time. Implementation and Operation: As the first step in the implementation, Sewerflex performance has been evaluated on 120 rain events, in an offline test environment where the MPC interacts with a hydraulic model representing the sewer network of Kolding. Performance of Sewerflex MPC is evaluated based on stability of calculated set-points and from calculated CSO volumes compared to similar calculations based on the current calibrated and tuned Sewerflex 1.0 code implemented in Kolding. In this test environment the Sewerflex MPC has improved the overall performance by approximately 10% based on CSO volume. In order to allow 'proactive' control, where the Sewerflex MPC algorithm can prepare the sewer network for future rain events before they occur (Water Smart Cities 2016), the algorithm will use both online measurements from the system, and weather forecast information. All these input data are visualized through SewerView, see Figure 1.2. As default, the SewerView map gives basic live information about the state of the sewer network, e.g. rainfall, flood risk and overflow. Following the test environment, the algorithm will be launched in 'offline cloud mode', meaning live visualization but no optimization of the sewer network. This phase of the project will allow for everyone to feel comfortable with the information given before online operation. The online operation is expected to be carried out in steps, allowing a slow transformation from the current online control to the cloud-based algorithm. The current RTC control will be kept as a fallback strategy. Further upgrades: The downstream WWTP is also upgraded to a cloud based real-time performance optimization using AQUAVISTA Plant, thereby achieving a holistic solution of a full optimized wastewater system. In dry weather, an ongoing development project (2017-2020) called 'BlueGrid' will add the functionality of flexible power consumption using demand/response at both the WWTP and sewer network, thereby achieving a flexible 'all weather optimization' which minimizes CSO during rain events and reduces costs for power consumption during dry weather (BlueGrid 2017).