Overview

Focus is on data from buildings and aggregated buildings in districts etc., and on load, solar and wind power forecasting. Finally, the program ends with a day devoted to MPC and optimization.

Some particular examples of applications and techniques:

  • When including the effect of wind and solar in building energy models, without detailed information about the building and its usage, it is important to describe non-linear effects and interactions. For this kernel and basis function (e.g. spline) techniques are indispensable in both discrete and continuous time models.
  • Occupancy in buildings leads to different patterns of stochasticity in observed data. Even though it is “random” there are still systematic patterns (e.g. varying levels of noise), which must be taken into account in models. Included in the course are very useful techniques for modeling: non-parametric diurnal curves (Fourier series basis functions), disaggregating levels of system noise and observation noise from sensors (noise level functions in grey-box models) and non-linear effects of solar radiation (conditional parametric models with base splines).
  • Tracking changing model parameters over time in order to adapt to changes in systems and use, using recursive estimation techniques such as Kalman filters and recursive least squares.
  • Include weather forecasts as model inputs to create optimal forecast models.
  • Optimized control and operation of energy systems using model predictive control. Examples are optimizing building heating with respect to varying prices and optimal charging of batteries.

Each day will be max. 3 hours of lectures (e.g. 2 x 1.5 hours) and computer exercises for the rest. Computer exercises will be given, one for each day. The exercises contain a pdf with description and questions, together with data and pre-made scripts. You will not program anything from scratch!

Schedule with topics

Monday
This day we will start out from where the preparation material ended. Techniques for fitting non-linear models will be introduced:

  • Non-linear versus linear models
  • Non- and semi-parametric models (kernels, splines, etc.)
  • Brief introduction of model identification (cross-validation and information criterion (AIC and BIC))

Tuesday
We will continue with continuous time models (grey-box models) and explain how they are best fitted to data. The models are expressed as stochastic differential equations and the Kalman filter is used to calculate the likelihood function:

  • Grey-box modeling (basics: tests for white noise, test for parameters, SDE’s, …)
  • Model identification with forward selection and likelihood-ratio tests
  • Introduction to the ctsmTMB and how it works (Kalman-filter, examples, …)

Wednesday
Model identification in more details and exercise on “advanced” grey-box modelling techniques:

  • Grey-box modelling (advanced: structural identification, model comparison, model selection, …)
  • Modeling the effect of nonlinear phenomena using splines (wind, solar radiation, humidity, …)
  • We will arrange an event in the afternoon.

Thursday
Models suited for forecasting and tracking changing model parameters over time are introduced. It will be introduced how models can be set up for forecasting using weather forecasts in an effective way:

  • Recursive and adaptive models (RLS)
  • Forecasting (load, wind and solar power forecasting) using non- and conditional-parametric models
  • The R package ‘onlineforecast’, see onlineforecasting.org
  • In the afternoon we will have a workshop event with Center Denmark on data spaces, where data spaces will be introduced and we will discuss how to deal with data in energy systems.

Friday
Using the models for control is a vital application. The basics of model predictive control (MPC) are introduced with some examples for building energy applications. Further, more advanced stochastic control techniques will be presented:

  • Model predictive control
  • Modelling of flexibility
  • Stochastic model predictive control

Time schedule

Below you find the time schedule. There will be five exercises which will be introduced as we go along and we will do wrap-ups of the main points during the exercises.

Monday
08:30 - 09:00 Registration and coffee
09:00 - 09:30 Welcome
09:30 - 11:30 Lecture: Introduction to non-linear time series models and non-parametric estimation techniques
11:30 - 12:00 Exercise on non-parametric models
12:00 - 13:00 Lunch
13:30 - 15:00 Exercise
15:00 - 15:15 Break
15:15 - 17:00 Exercise

Tuesday
09:00 - 11:00 Lecture: Introduction to grey-box modelling
11:00 - 12:30 Exercise on grey-box modelling
12:30 - 13:00 Lunch
13:00 - 14:00 Lecture: Introduction to ctsmTMB
14:00 - 15:00 Exercise
15:00 - 15:15 Break
15:15 - 17:00 Exercise

Wednesday
09:00 - 10:30 Lecture: Model identification
10:30 - 12:30 Exercise on “advanced” grey-box modelling
12:30 - 13:00 Lunch
13:00 - 22:30 Excursion with dinner:
13:00 Bus leaves here: maps bus
14:30 We sail in boats in the harbour from here: maps boats
17:45 We have dinner in the Boathouse restaurant here: maps restaurant

Thursday
09:00 - 10:30 Lecture: Models for forecasting in energy systems
10:30 - 12:30 Exercise on models for forecasting
12:30 - 13:30 Lunch
13:30 - 14:45 Exercise
14:45 - 15:00 Break
15:00 - 17:00 Workshop: TBD

Friday
09:00 - 10:30 Lecture: Model predictive control with a focus on energy systems
10:30 - 12:00 Exercise on model predictive control
12:00 - 12:45 Lecture: Introduction to stochastic programming
12:45 - 13:30 Lunch
13:30 - 16:00 Exercise (you can leave at 12:45 and not miss any new information, but we will stay and help the ones staying and working)