Difference between revisions of "On detecting deviations by autonomously discovering district heating control strategy used by buildings"
(Created page with "{{StudentProjectTemplate |Summary=autonomously detect control strategy and building type from district heating data |Keywords=control strategy, heat load shape discovery |Time...") |
|||
| Line 3: | Line 3: | ||
|Keywords=control strategy, heat load shape discovery | |Keywords=control strategy, heat load shape discovery | ||
|TimeFrame=Spring 2018 | |TimeFrame=Spring 2018 | ||
| − | |References=[1] Stefan Byttner, Slawomir Nowaczyk, Rune Prytz, and Thorsteinn Rögnvaldsson. A field | + | |References=[1] Stefan Byttner, Slawomir Nowaczyk, Rune Prytz, and Thorsteinn Rögnvaldsson. A field test with self-organized modeling for knowledge discovery in a fleet of city buses. In 10th IEEE International Conference on Mechatronics and Automation, IEEE ICMA 2013, Takamastu, Japan, August 2013. |
| − | test with self-organized modeling for knowledge discovery in a fleet of city buses. In | + | |
| − | 10th IEEE International Conference on Mechatronics and Automation, IEEE ICMA 2013, | + | |
| − | Takamastu, Japan, August 2013. | + | |
[2] Yizong Cheng. Mean shift, mode seeking, and clustering. IEEE Transactions on Pattern | [2] Yizong Cheng. Mean shift, mode seeking, and clustering. IEEE Transactions on Pattern | ||
| Line 21: | Line 18: | ||
|Prerequisites=basic statistics and programming skills; interest in machine learning and data mining; willingness to learn new topics. | |Prerequisites=basic statistics and programming skills; interest in machine learning and data mining; willingness to learn new topics. | ||
|Supervisor=Shiraz Farouq, Sławomir Nowaczyk | |Supervisor=Shiraz Farouq, Sławomir Nowaczyk | ||
| + | |Level=Master | ||
| + | |Status=Open | ||
}} | }} | ||
In [4], heat load patterns for different building categories is computed for four explicitly defined seasons. Four type of control strategies used in different type of buildings leading to specific heat load patterns have been described in the paper. We briefly state them as follows: | In [4], heat load patterns for different building categories is computed for four explicitly defined seasons. Four type of control strategies used in different type of buildings leading to specific heat load patterns have been described in the paper. We briefly state them as follows: | ||
Revision as of 21:58, 27 September 2017
| Title | On detecting deviations by autonomously discovering district heating control strategy used by buildings |
|---|---|
| Summary | autonomously detect control strategy and building type from district heating data |
| Keywords | control strategy, heat load shape discovery |
| TimeFrame | Spring 2018 |
| References | [[References::[1] Stefan Byttner, Slawomir Nowaczyk, Rune Prytz, and Thorsteinn Rögnvaldsson. A field test with self-organized modeling for knowledge discovery in a fleet of city buses. In 10th IEEE International Conference on Mechatronics and Automation, IEEE ICMA 2013, Takamastu, Japan, August 2013.
[2] Yizong Cheng. Mean shift, mode seeking, and clustering. IEEE Transactions on Pattern Analysis and Machine Intelligence, 17(8):790–799, August 1995. [3] E.W. Forgy. Cluster analysis of multivariate data: efficiency versus interpretability of classifications. Biometrics, 21:768–769, 1965. [4] Werner S. Gadd H. Heat load patterns in district heating substations. Applied Energy, 108:176–183, 2013. [5] J. Flora J. Kwac and R. Rajagopal. Household energy consumption segmentation using hourly data. IEEE Transactions on Smart Grid, 5(1), 2014.]] |
| Prerequisites | basic statistics and programming skills; interest in machine learning and data mining; willingness to learn new topics. |
| Author | |
| Supervisor | Shiraz Farouq, Sławomir Nowaczyk |
| Level | Master |
| Status | Open |
In [4], heat load patterns for different building categories is computed for four explicitly defined seasons. Four type of control strategies used in different type of buildings leading to specific heat load patterns have been described in the paper. We briefly state them as follows:
building type: multi-dwelling residential building. control strategy: continuous operation control. pattern description: the winter heat load pattern is typically different than summer, autumn and spring; see figure 1.
building type: commercial building control strategy: night setback control. pattern description: lower heat loads during nights are followed by high peak heat loads in the mornings, but these peaks vanish quite fast. The peaks are the results of the reheating of the cooled off heating system during the preceding nights; see figure 2.
building type: Commercial building (e.g. offices). control strategy: Time clock operation control 5 days a week pattern description: heat load during nights and weekends is the same. During these periods the ventilation is turned off or reduced and the radiator system is supplying heat to keep the indoor temperature at a desirable level; see figure 3.
building type: Commercial building (e.g. shopping centers). control strategy: Time clock operation control 7 days a week pattern description: the pattern is similar to time clock operation control 5 days a week, but the ventilation is also in operation at the weekends as well and not only during working days; see figure 4.
The patterns and their description comes from expert knowledge. The aim of this project is to not only find these patterns autonomously, but also to discover new ones. One line of thought is to decompose the load profiles based on [5]. The idea is that given daily consumption profile l(t), we decompose it as l(t) = a.s(t), where
a = \sum^{24}{t=1} l(t) and s(t) =l(t) /a
Here, a is the daily total consumption and s(t) is the normalized load profile referred as the load shape. One can then use k-means clustering [3] or mean shift [2] to form clusters in terms of heat load patterns. Other ways also need to be investigated here. However, once we have formed the clusters, we will compare the results with the building type (also geo-location can be used here) labels we have to check if we were able to correctly identify building type and hence its control strategy.
Further analysis can be done, like computing the entropy of buildings heat usage and clustering them accordingly to form peer groups. Once we have figured out the building types, the next step is to apply the COSMO [1] algorithm to different building groups generated by different methods to detect deviating buildings. The final task is to analyze on how the choice of constructing peer groups effects the finding of deviating customers.
