Automatisk viktning av onlinekorrigering för prognoser
(2013) FMS820 20131Mathematical Statistics
- Abstract (Swedish)
- This report aims at finding methods for reducing the errors
in district heating load forecasts by suggesting a method for
online correction of the base forecast. Online correction is
generally carried out by utilizing patterns in the historic
prediction errors. Traditionally the correction has been
made using a correction method detailed by a few governing
parameters. Values for these parameters are set according
to the experience of energy engineers. One of the major
topics for this report will be to choose these parameters
automatically and in an online fashion using historic error
data. Another approach to the correction problem is made
by modelling the errors from scratch. Starting from a time
series analysis of the error... (More) - This report aims at finding methods for reducing the errors
in district heating load forecasts by suggesting a method for
online correction of the base forecast. Online correction is
generally carried out by utilizing patterns in the historic
prediction errors. Traditionally the correction has been
made using a correction method detailed by a few governing
parameters. Values for these parameters are set according
to the experience of energy engineers. One of the major
topics for this report will be to choose these parameters
automatically and in an online fashion using historic error
data. Another approach to the correction problem is made
by modelling the errors from scratch. Starting from a time
series analysis of the error series all significant linear dependencies
can be identified and different models from the
Auto Regressive Moving Average (ARMA) family of models
are evaluated. It is found that purely Auto Regressive
models will give the best results for these error series. In
the error series there are clearly visible non-linear phenomena
present i.e. a slightly varying mean level as well as a
more heavily varying variance. By putting AR-models with
varying mean and variance on a nonlinear state space form
an attempt is made at handling this. The AR-parameters,
the mean level and the varying variance is then estimated
using an Unscented Kalman Filter. The report is written
in Swedish. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/3994606
- author
- Rosén, Erik
- supervisor
- organization
- course
- FMS820 20131
- year
- 2013
- type
- H2 - Master's Degree (Two Years)
- subject
- language
- Swedish
- id
- 3994606
- date added to LUP
- 2013-08-23 09:14:11
- date last changed
- 2013-08-23 09:14:11
@misc{3994606, abstract = {{This report aims at finding methods for reducing the errors in district heating load forecasts by suggesting a method for online correction of the base forecast. Online correction is generally carried out by utilizing patterns in the historic prediction errors. Traditionally the correction has been made using a correction method detailed by a few governing parameters. Values for these parameters are set according to the experience of energy engineers. One of the major topics for this report will be to choose these parameters automatically and in an online fashion using historic error data. Another approach to the correction problem is made by modelling the errors from scratch. Starting from a time series analysis of the error series all significant linear dependencies can be identified and different models from the Auto Regressive Moving Average (ARMA) family of models are evaluated. It is found that purely Auto Regressive models will give the best results for these error series. In the error series there are clearly visible non-linear phenomena present i.e. a slightly varying mean level as well as a more heavily varying variance. By putting AR-models with varying mean and variance on a nonlinear state space form an attempt is made at handling this. The AR-parameters, the mean level and the varying variance is then estimated using an Unscented Kalman Filter. The report is written in Swedish.}}, author = {{Rosén, Erik}}, language = {{swe}}, note = {{Student Paper}}, title = {{Automatisk viktning av onlinekorrigering för prognoser}}, year = {{2013}}, }