Solution#4: Load forecasting for greenhouses

The load forecast solution for greenhouses is an offline model structure built on data from energy consumption and weather observations.

The load forecast solution for greenhouses, is build on the collection of data from greenhouse consumers and weather observations to derive an offline model structure for load forecasting in greenhouses. Thus, the solution established an online setup for load forecasting where model parameters are adapted to the actual dynamics of consumption and weather.


A model based predictive control of supply temperatures leads to reduced heating loss and pumping costs. This means saving of both resources and money for greenhouse owners applying our solution.

ikon20Solution insights

Greenhouses are due to their construction, highly sensitive to external factors of weather and season such as solar radiation and out-door humidity. All together this implies that the heat load is fluctuating much more for greenhouses than for domestic or factory buildings. Load forecasting including automatic calibration of model parameters is the answer to overcome this challenge. Some dimensions of the project solution includes:

Available data

Greenhouses consumption are recorded hourly as accumulated energy, accumulated volume flow, temperature forward and return. Weather observations such as wind speed, wind direction, and solar radiation radiation is recorded hourly (running average logged every hour, raw data sampled every 1 minute). Forecast are obtained at the same place as the observation, with the same parameters, so here it will be necessary to make the same time adjustment as for the observations.

Time correction for geographic displacement

To compensate for a geographic displacement between observation and a specific greenhouse, it is necessary to calculate a time correction.

Stochastic model for greenhouse demand

A universal model is developed for greenhouses, but for better adaption to single greenhouse parameters must be adapted. Model parameters should be generated from observation obtained over a time horizon of one week (this we don’t know yet). It is important that the model is transparent, and defined by parameters, so it doesn’t need to be a black box model,  and secondly generate a load forecast time series form weather forecast time series (hourly values, 124 hours ahead).

Scenario simulator

Different scenarios are simulated to find the cost minimum of time depending prices, pushing or pulling the load in time.

Forecast services for optimal control of supply temperatures to greenhouses

Forecast services are crucial for optimal decision-making, production planning, trading of power and for control of district heating network. In this section we briefly describe the list of forecast services needed, discuss the statistical forecast characteristics, and illustrate how the forecasts can be used in optimal control and decision making.

It is clear that forecasts are needed both on a day-to-day basis, e.g., in order to provide input for the market clearing, and for the optimal production planning, as well as on a shorter horizon, e.g. in order to use the flexibility of the distributed energy resources to control the electricity load.

The predictive controllers considered are based on forecasts of load, prices, etc. The forecast services needed depend on whether Direct Control (DC) or Indirect Control (IC, also referred to as or control-by-price) is implemented:

  • Load (demand or flexibility) forecasts (for both DC and IC)
  • Price forecasts (only for IC)
  • State forecasts (e.g. room temperature) (only for DC)

In all cases it is assumed the appropriate meteorological forecasts are available.

ikon6Further information

The project of CITIES Research Center is expected to continue on improving the solution until March 2017.

The solution is connected to the project on smart temperature control in greenhouses.


Henrik Madsen   Professor for Applied Mathematics and Computer Science, Center Manger of CITIES Research Project  at DTU Compute

+ 45 45253408
Involved partners
Fjernvarme Fyn

Member since September 2013

Visit the website

Contact person Main Office + 45 65473000

Member since September 2013

Visit the website

Contact person Main Office + 45 45350350
DTU Compute

Member since September 2013

Visit the website
Why is DTU Compute part of CITIES?

As core-partner, DTU Compute is part of CITIES to accelerate its activities in the fields of renewable energy, in particular forecasting methods for wind energy, and demand response in power systems, as well as the use of smart data (data analytics) for the characterisation of buildings and supermarket refrigeration units.

What role does DTU Compute have in the project?

We provide invaluable contributions to CITIES through our expertise in modelling, treatment and management of data, statistics, forecasting and control. Furthermore our Head of Section, Prof. Henrik Madsen is the center leader of CITIES and some of our finest researchers are involved in demonstration projects.

What do DTU Compute expect to gain from being part of the project?

DTU Compute expects to be able to conduct valuable academic research in collaboration with partners.

Contact person Henrik Madsen + 45 45253408
DTU Civil Engineering

Member since September 2013

Visit the website
Why is DTU Civil Engineering part of CITIES?

DTU Civil Engineering (DTU Byg) is a highly active player in CITIES Innovation Center in the research fields of high energy efficiency buildings and systems for energy supply and distribution.

What role does DTU Civil Engineering have in the project?

As a core-partner in CITIES through leadership of projects under Prof. Carsten Rode and Ass. Professor Alfred Heller’s role in the center management, we head research efforts on a variety of areas incl. solar energy and district heating.

What do DTU Civil Engineering expect to gain from being part of the project?

We expect to gain a valuable exchange with partners and hopefully also see cutting-edge scientific results on energy systems integration in the Nordic countries.

Contact person Main Office + 45 45251861
Copyright © 2019 | CITIES Innovation Center
Article location:

Leave a Reply

Your email address will not be published. Required fields are marked *

Popular articles

Continuous Time Stochastic Modelling

 The project on continuous time stochastic modelling (CTSM) is very useful for grey-box modelling of physical systems using data. The grey-box modelling approach bridges the gap between physical and statistical modelling. The solution is widely used for estimating and identifying models for physical systems, like the heat dynamics of a building, the hydraulics in a …