Model Predictive Control System

Level of Intervention:
Canal / Water System

The Dutch regional water boards (Waterschappen) are responsible for maintaining safe and secure water levels within their service region. In rainy periods, most water boards need to pump huge volumes of water from low lying areas in order to safeguard people, houses, roads and other economic infrastructure. Normally, the decision to pump is ‘water level driven’, in order to maintain water levels within an established range (so called ‘feedback control’).

Aim of the Experience:
Water boards can save enormous amounts of energy (and thus reduce CO2 emission) by using a smart decision support system that helps to efficiently steer existing pumping plants, taking into account a variety of parameters beyond water level only.
This kind of decision support system is aimed to pass from ‘feedback control’ to Model Predictive Control (MPC), which aims to:

  • Reduce pumping by optimising water storage capacity in relation to rainfall prediction and by timely pumping in relation to variable ‘outer bounds’ water levels.
  • Define pumping time slots that coincide with abundant electrical energy, preferably ‘green’ electricity.

The purpose of the experience is to 1) develop the Model Predictive Control system, 2) measure the reduction of energy use and cost when applying it in actual water management practices at four pilot sites and 3) extrapolate the results for the national ‘pumping bill’.

Description of the experience:
A Model Predictive Control system, based on a simplified water model, was developed to define pump schedules that make optimal use of:

  • Rainfall forecast
  • Estimates of outer bounds water level (= levels to pump to).
  • Water storage capacity in the low lying areas
  • Variations in energy costs (e.g. day and night prices)
  • Estimates of energy surplus from sustainable energy sources (wind and son energy)

An important input for the MPC are data from the Power Price Scenario Generator, which simulates one-day ahead electricity markets and predicts every hour energy costs and energy mixes. The use of this tool makes it possible to estimate the total reduction of CO2 emission over the year.

Another important data is the total storage capacity of inland waters. Applying a +/- 5 cm water level standard, the national buffer capacity equals 1700 MWh.

The MPC is able to define pumping schedules for a given time slot (e.g. 72 hours), based on the prioritisation of water management goals such as water level, energy reduction, cost reduction, maximum use of green energy, etc. The pumping schedules include detailed prescriptions of number of pumps and rpm.

Summary of measures:

  • Development of Model Predictive Control system to define pumping schedules
  • Application of the MPC on four pilot sites
  • Comparison of three pumping scenarios, and under different flexibility scenarios for the development of the energy market:
    1. Water level driven
    2. Minimization of energy costs
    3. Optimization of CO2 emission

Estimate of CO2 reductions and other outcomes:
The outcomes of the analysis show that the smarter pumping regimes obtained under the Predictive Control Model lead to an average reduction of energy use of 30% and of pumping costs of 20% to 25%. In specific water board regions that count a wider range of variables (tides movement in the target river, locks, etc.), this may increase to a 80% pumping cost reduction.

The outcomes also show a reduction of CO2 emission up to 38 kton, which is mainly due to the replacement of fossil fuel sources by son and wind energy. The CO2 reduction is 80% of the current CO2 emission due to the overall national pumping efforts (46 kton per year = 0,12% of the total national CO2 emission).

The functioning of a Predictive Control Model for water management and pumping plants depends on the availability and reliability of two main components:

  • Data Integration System
  • Telemetric System​

The Data Integration System must collect all required data (weather, water levels, energy costs), validate the data, run the models to produce the required input for the steering model, create the steering output data and send these data to the Telemetric Systems.
The Telemetric System is required to send to and receive data from all stations (sensors, weather stations: rainfall, wind, pumping sites, etc.).

The frequency of data sending and command communication depends on the time resolution needed for every specific application.

Further information:
Slim malen – energie besparen? [Smart pumping – energy saving?], STOWA report 2019-27, STOWA, Amersfoort, 43 pp.
Klaudia Horvath, , Bart P.M. van Esch, Jorn Baayen, Ivo Pothof (2018), Categorization of trapezoidal open channels based on flow conditions for the choice of simple models, La Houille Blanche, Vol. 4, 2018, p. 56-64
Horvath, K. et al. (2018), Model-predictive control of a river reach with weirs, Proceedings of 13th Int. Conf. on Hydroinformatics, HIC2018, 1 – 6 July, Palermo
Horvath, K., van Esch, B., Vreeken, D., Pothof, I., Baayen, J. (2019), Convex modelling of pumps in order to optimize their energy use, Water Resources Research 55, Vol. 3, p. 2432-2445.
Klaudia Horváth, Bart van Esch, Ivo Pothof, Tjerk Vreeken, Jan Talsma, Jorn Baayen (2019), Closed-loop model predictive control with mixed-integer optimization of a river reach with weirs, 1st IFAC Workshop on control methods for water resources systems (CMWRS2019), 19-20 September, Delft.

2022-01-05T15:26:56+00:00January 5th, 2022|Categories: GreenWIN Case Study|Tags: |0 Comments