Page 96 - Handbook Of Multiphase Flow Assurance

Machine learning and artificial intelligence in flow network optimization 91

Provision for water injection line and header maintenance scraping to sweep any bacterial
growth should be included, with ensuring that scraped solids do not enter the injection well
or header. Proven technologies for anti-bacterial coating should be considered for water in-
jection lines.
Injection wells should have a provision for hydrate inhibitor injection as hydrocarbons
may migrate up the wellbore and form a solid hydrate blockage near mudline when the in-
jection well is not flowing.

Flow restriction and blockage monitoring
Production control and automation system should be able to monitor for leading indica-
tors of an imminent blockage and to mitigate it as early as is noticed by altering operating
parameters upon approval by operations manager or by solvent / chemical injection. If flow
restriction mitigation was unsuccessful or late, monitoring capability should assist in a safe
and systematic remediation of blockage.
Technologies which could be considered for monitoring of blockage in produced fluids
include:

• pressure differential deviation
• temperature deviation
• flow deviation
• vibration deviation
• valve operability change
• water composition and pH
• chemical residuals
• oil and water quality
• gas dew point and moisture content
• bacteria counts
• asphaltene instability
• solids TDS and TSS monitoring.
Emerging technologies which may be applicable on a case by case basis include gamma-ray
densitometer, ultrasound solids detection, and guided wave deposit detection. The large
number of monitored parameters make it conducive to implement flow and blockage moni-
toring with a machine learning method.

Machine learning and artificial intelligence in flow network optimization

The early development in the use of computer for network flow optimization came in the
form of spreadsheets with multiple runs indicating the hydraulic resistance of individual
components of the network. One early example of such spreadsheet was presented by Lezeau
and Leporcher nearly two decades ago (Lezeau and Leporcher, 2001). The methodology and
logic presented in their work still is generally applicable to a further implementation of the
network flow optimization process.