Page 124 - Intelligent Digital Oil And Gas Fields

Data Filtering and Conditioning 91

whereasMPFMisaffectedby 102bbl/d(4%error).Thenextstepistoreview
with the production engineer why virtual metering has a mismatch of 8% and
acceptableerror should be below5%. Additionally, the Coriolis meter needs to
be calibrated or checked for the mass and density of the fluid.
We recommend the use of reconciliation processes for daily production
monitoring and equipment surveillance. Table 3.1 shows how a Coriolis
meter can be uncalibrated and a virtual meter untrained, which results in
data mismatch relative to test measurement. The ultimate impact would
be misallocation of the produced fluids from the sales tank back to the wells.

3.3 CONDITIONING

After the data have been validated and replaced, if necessary, they can
be conditioned appropriately for specific workflows. There are two types of
conditioning used in data applications for streaming real-time data: noise fil-
ters and statistical calculations for use in specific workflows. Implementation
of these techniques is the key to keep workflows from becoming over-
whelmed with too much data. It is best to custom-fit the conditioning
methods to the specific workflows of interest. Most SCADA systems, histo-
rians, or databases have statistical tools to condition data. Three types of con-
ditioning are described below, which apply to most DOF projects: down
sampling from the high-frequency data; summation into daily, monthly,
and other specific times; and special status indicators appropriate for DOF
workflows.

3.3.1 The Level of Rate Acquisition (Data Frequency)
Pressure, temperature, and flow rates data can be acquired every second or
more if data storage is available. Often that level of data sampling is not
required, as discussed below. Production, completion, geologist, and reser-
voir engineers are the ultimate end users of the data and must decide on the
requirements. Houze et al. (2017) describe how data can be classified into
low and high frequencies (as described in the points below). However,
we classify data depending on the acquisition rate, with a modified summary
presented in Table 3.2.
• Low-frequency data. Gas, water, and oil rates can be taken daily and pres-
sure can be taken an average of 24h; thereafter, the data can be summa-
rized to weekly, monthly, quarterly, and annually. This category
includes well test processes that measure to the separator and tanks the