Average retrieval

For the time-weighted average (in short: "average") retrieval mode, a time-weighted average algorithm is used to calculate the value to be returned for each retrieval cycle.

For a statistical average, the actual data values are used to calculate the average. The average is the sum of the data values divided by the number of data values. For the following data values, the statistical average is computed as:

(P1 + P2 + P3 + P4) / 4) = Average

For a time-weighted average, values are multiplied by the time difference between the points to determine the time-weighted value. Therefore, the longer a tag had a particular value, the more weight that value holds in the overall average. The overall average is determined by adding all of the time-weighted values and then dividing that number by the total amount of time.

Which values are weighted depends on the interpolation setting of the tag. For a tag that uses linear interpolation, the midpoints between values are weighted. For a tag that uses stair-step interpolation, the earlier of two values is weighted.

For the following data values of a tag that uses linear interpolation, the time-weighted average is computed as:

(((P1 + P2) / 2) x (T2 - T1)) + (((P2 + P3) / 2) x (T3 - T2)) + (((P3 + P4) / 2) x (T4 - T3)) / (T4 - T1) = Average

If the same tag uses stair-step interpolation, the time-weighted average is:

((P1 x (T2 - T1)) + (P2 x (T3 - T2)) + (P3 x (T4 - T3))) / (T4 - T1) = Average

The SQL Server AVG aggregate is a simple statistical average. Using the average retrieval mode with a cycle count of 1 returns data much faster than the AVG aggregate, and usually more accurately due to the time weighting. The Event subsystem also returns a simple statistical average.

Average retrieval returns one row for each tag in the query for each cycle. The number of cycles is based on the specified resolution or cycle count.

The time-weighted average algorithm is only applied to analog and analog summary tags. If you use average retrieval with other tags, the results are the same as when using regular cyclic retrieval.