1
$\begingroup$

I have Room Temperature data(T1) and Outside Temperature data(T2) with me for various houses which are having HVAC system installed. I am building a system which detects faulty HVAC (heating, ventilation, and air conditioning) systems. I am using anomaly detection for this, using (T1 - T2) as a feature. Which is not working completely because when T2 raises above a particular limit system wont be able to maintain the inside temperature anymore so T1 goes up along with it, which is also similar to a faulty case where HVAC is not working properly. In this case the difference between T1 and T2 is low(because both are at high temperatures), Which is similar to the difference between T1 and T2 when operating under normal conditions.

 Normal Case   T1 = 22     T2 = 25
 Normal Case   T1 = 21     T2 = 24
 Normal Case   T1 = 20     T2 = 23
 Normal Case   T1 = 25     T2 = 30

 Normal Case   T1 = 30     T2 = 35 (Outside temperature is beyond the HVAC capacity)
 Normal Case   T1 = 31     T2 = 37 (Outside temperature is beyond the HVAC capacity)

 Faulty case   T1 = 24     T2 = 25
 Faulty case   T1 = 26     T2 = 30

I am looking for a feature which will be able to differentiate normal operation and faulty operation.

Improve this question
$\endgroup$
2
  • $\begingroup$ Missing information: What is the target temperature? Note, this only shows the case for air-conditioning. For example, what would normal functioning look like if T2 were 15? Also, it is unclear what "feature" you are looking for. A physical feature, a look up table? Also, the system capacity in BTU (or watts or whatever) house insulation and heat (cold) loss per degree difference in BTU per h for that house are factors that need to be specified. You might ask this question on site that is more directed towards HVAC problems. This one isn't especially a good choice. $\endgroup$
    – Carl
    Commented Mar 18, 2020 at 13:19
  • $\begingroup$ Do you have a time-series of this data (per house)? If so, what time resolution? $\endgroup$
    – Jon Nordby
    Commented Mar 27, 2020 at 18:44

1 Answer 1

Reset to default
0
$\begingroup$

You may want to take into account whether the system is expected to be cooling or heating. Commonly used in modelling energy use for building is a concept called degree-days. This is based on the assuming that a certain temperature can be considered comfortable (for example 18.5°C / 65°F), and that for this temperature, no heating or cooling is needed.

A simple feature in your case could be T1-Ttarget, where Ttarget could either be estimated from the data (long term median of T1 for example), or you use a standard value like 18.5C.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.