ASHRAE Journal - February 2021 - 36

TECHNICAL FEATURE

Using Algorithm Data In
District Heating Systems
BY HANBEI ZHANG; LANBIN LIU; RANG TU; YAMENG LIU

District heating systems produce a large amount of data from temperature sensors,
flow rate meters, pressure sensors, etc., during heating seasons. Collecting this data
is essential to realize smart heating systems, but the presence of collected abnormal
data may cause equipment failures or communication problems. This article proposes
a simplified and efficient detection algorithm (called detection algorithm of abnormal
data based on k-nearest neighbor) to detect and eliminate this data.
At present, China's district heating network covers
an area of about 8.5 billion m2 (91 trillion ft2), nearly
triple that in 2005.1 Both the heated building area and
total pipe length have increased dramatically.2 A large
number of sensors have been installed in district heating system, including thermal substations, thermal
networks and buildings; parameters such as pressure,
temperature, flow rate, energy consumption, indoor
temperature and humidity can be automatically collected, stored and transmitted.3,4
Collecting this data makes it easier to manage district heating systems, improving overall operation
efficiency and eventually reducing energy consumption. However, with a large quantity of data, the existence of abnormal data is inevitable.5 Because sensor
monitoring data is the basis of energy consumption
analysis, operation regulation and characteristic
identification of the heating system, if a small amount
of abnormal data generated by sensors exists, these
functions may fail, and serious safety problems may

occur. It is of great necessity to detect and eliminate
the abnormal data.
Three kinds of faults are defined in the literature:6
delay, offset and stuck-at, including 14 specific types
of faults. This study focuses on processing the outlier
abnormal data and spike abnormal data.
There have been many studies on abnormal data
detection methods, such as distribution based,7,8 distance based9,10 and density based.11 - 13 The distributionbased method assumes that the given data set obeys
some probability distribution model and then determines the abnormal data according to the inconsistency
test on the probability distribution model. It was first
proposed by Grubbs and Frank.7 The Grubbs' test statistic value was calculated for each data on the basis of the
standard t distribution. If the statistical value exceeded
a certain threshold, it was considered abnormal data.
However, using this method requires presupposing the
distribution of data sets, which is not suitable for situations in which the distributions are unknown.

Hanbei Zhang is a postgraduate student, Lanbin Liu is an associate professor, and Rang Tu is an associate professor in the School of Civil and Resource Engineering, University of
Science and Technology Beijing in Beijing, China. Yameng Liu is a HVAC engineer working in the Department of Research and Development, Beijing Nuanliu Technology Co. Ltd, in
Beijing, China.
36

ASHRAE JOURNAL

ashrae.org

FEBRUARY 2021
http://ashrae.org
ASHRAE Journal - February 2021

Table of Contents for the Digital Edition of ASHRAE Journal - February 2021
