Graphing issue

[phillb]

The following two graphs show the same time but on different Scales. the 24 hour graph appears to show the boiler coming on between 8 and 9 for25 minutes. The Six hour graph shows the Boiler coming on at about 9. (8.57) which is correct. is the 24 hr graph right I Seem to remember from previous versions that the on time still was reflected reasonably accurately with Longer time sales and the 25 min's would be straddled across the hour. Looks to me like there my be a one hour error when it crept in not Sure but I recon I wound have noticed if last time my heating was starting to come on which was last year!
I am using 5.7.0.3

[Michael McSharry]

Discrete values on the 6 hour scale reflect ON and OFF states. Discrete values on other scales reflect cumulative ON time. It looks to me as if ComputerRoomBoiler was on for 25 minutes and it came ON shortly before 9. I would need to look at the code to confrm, but I believe the transition from OFF to ON occurred during the 8 to 9 hour so the ON time was allocated to this hour. If it had gone off and then on again it would probably be shown as a lower value, but spanning two hours. I would continue to observe and see if this rationale holds.

[phillb]

I will contiune to observe.

I seem to remember that before in the above example the cumlative value would be split between the hours so in this case say 3 minutes between 8 and 9 and 22 minutes between 9 and 10.

If this were not the case and the logic followed through then; if the on event went past the next hour, lets say for 80 minutes (without a return to off), then the cumlative total would be more than 60 which, of course could not be represnented within an hour period between 8 and 9.

Am I explaining myself clearly?

[Michael McSharry]

The quantization is is hourly for this time scale. It may appear to be skewed by up to 1 hour especially when the off to on transition occurs at the end of the hour. If the transition had occurred just after the start of the hour then the skew would appear to less and approaching zero. I could have used midpoint of the hour rather than start of the hour and time may of yielded a better result, but I think you will find that it will be consistent. The algorithm limits the time to 60 minutes per hour and will spillover excess minutes to the next hour. The algorithm is always forward-looking.

[phillb]

I think using the start of the hour is correct. However you could start a new quantization if an hour bondary is crossed and always show that quantization for a boolean type sensor across the whole hour period.

[Michael McSharry]

Based upon the algorithm being used the chart should be viewed as the signal was ON for 23 minutes starting between 8 and 9 AM. If there were multiple ON/OFF transitions between 8 and 10 AM then the time will be split between the two hours as the durations starting in either hour would have been this long. If the signal remained on for 70 minutes then it would show as 60 minutes starting at the 8-9 Hour and 10 minutes starting in the 9-10 hour. The viewing of durations vs. accumulations provide slightly different information. If a user is expecting to see the ON time integrated over the x-axis divisions then he/she will be confused. Had the ON time been presented as a quantized integral then the person looking for duration information would have been confused. When looking at a bigger picture such as over the full span of the x-axis the two presentation methods are equivalent with only a potential skew in the x-axis quantization intervals.

[phillb]

Agreed. Thanks Now i know how it works i can observe over the winter.
PS thanks again for that msjvm reinstal tip it saved me a whole bunch of time.
Phill