how to calculate a summary value and statistical error in time series I have a set of data that comes for empirical measurements over a number of days.  From the beginning of the experiment to the end of it, every five minutes temperature was measured inside (Ti) and outside (To) a box, and heat flux (Q) was measured between one of the walls of the box and the outside.  For every time point I use these three measures to calculate the thermal resistance, which is
R = (Ti - To)/Q
Obviously each R has a measurement uncertainty due to Ti, To and Q, Obviously there is correlation between measurement, because each measurement is not independent of what has been measured before. 
Hence my questions are:
1) what is the most appropriate way to summaries how the R value changes over time in one single value?
2) given the kind of data I describe, what is the most appropriate way of finding an error value for the value in question (1)?
The time series looks like this (it has one anomalous negative peak because of some other stuff that happened to the box at the time of the peak):

 A: I guess you are knowledgable about the Physics involved here. I am not, so I would ask questions like "is outside temperature controlled" or not? If it is not, then it means that it is affected by the local climate somehow -and this climate would be reasonably expected to exhibit some kind of daily seasonality.
Then I would first explore graphically the relation between the three components, outside and inside temperature, as well as heat flux. Do they move together? To what extent and in what way?
Then I would graph the $R$ series. Does it trend upwards/downwards? Does it have many up-down direction change? Does it seem wildly fluctuating? Does it seem keen on long swings? (If you are experienced some of these questions will already have a suspected answer by the previous graphical analysis). 
In general, visual inspection of time series is crucial in order to obtain important leads about the most appropriate modelling approach.
Now, single values can summarize a time series only if the time series is stationary. So both the visual inspection, and formal tests, should help you decide that, alongside the existence of a seasonal component and/or a deterministic trend (in which case as a while the time series is not stationary but can be made stationary by separating these components). 
As for the error term: I guess the only sources of error here are a) measurement error and b) error because the theoretical physics equation is in reality an approximation (an idealization) of the physical relationship that it describes. I don't think you can separate these sources of error. As for measurement error, I would ask, what are the specs of the measurement instruments that you use? Why should there be a measurement error? For high-precision instruments, usually there exists information about their precision, which should help you "anticipate" some of the properties of the error term.  
I understand that I am not providing a ready-made recipe with my answer -but it would be wrong to, since we know nothing about your actual data. What would you say, create those graphs and share some windows of these graphs here? (I say "windows" because you have 288 observations per day, so the whole time series may look muddy. But then again, a "window" may not be representative if the series does not appear to ~"repeat itself").
ADDENDUM
Based on the graph of the $R$-series posted by the OP, it appears that the series does not have neither a detrministic, nor a stochastic trend(i.e. a unit root). It looks like some seasonality is present (which is reasonable since the OP verified that "outside temperature" is not controlled but follows the daily pattern of the local climate).  
So a seasonal $ARMA$ model (with a non-zero constant term) is a reasonable modeling choice for a start (one could also try a non-seasonal $ARMA$, for contrasting purposes).
As for the negative outlier, since it is known that it has to do with special circumstances that are not connected to what is measured, one could (and should) safely eliminated it, by imputing a "locally reasonable" value in its place.
There is a slight visual suspicion (more of a feeling) that variability may be higher in the second half of the graph compared to the first), but this could be examined at a second stage.
