The AI Brain Behind the Well
Built on Deep Sequential Modeling (Deep Sequential Modeling), real-time physics validation, and the proprietary Hive Mind collective intelligence architecture — the system continuously learns from every well in the portfolio.
Why Deep Sequence Modelling Is the Right Approach
Plunger lift data is fundamentally sequential — pressures, velocities, liquid loads, and cycle outcomes evolve over time in patterns where what happened across the last ten cycles directly predicts what will happen in the next one.
A standard model reading each cycle in isolation misses this entirely. Deep Sequential Modeling carries memory forward cycle-over-cycle, capturing trends and deteriorating conditions that single-cycle models cannot detect.
The system tracks performance across 3, 7, and 14-cycle windows, computing both average levels and variability. High variability in arrival velocity signals mechanical or reservoir instability that average values alone would completely hide.
Every Outcome Predicted Before It Happens
Before every cycle begins, the model generates 10 forward-looking predictions — each directly tied to a production or cost outcome.
Plunger Arrival Velocity
Will the plunger arrive within the safe 500–1,000 ft/min window? Predicted before the valve opens. Below 500 ft/min = failed lift. Above 1,000 ft/min = equipment damage. Prevents $200–$300 per plunger replacement and 24–72 hrs of lost production per stuck-plunger event.
Will Plunger Arrive at Surface?
A binary Yes/No prediction. A predicted 'No' automatically extends shut-in before the valve opens — converting a predicted failed lift into a successful cycle with zero manual intervention required.
Optimal Shut-In Duration
How long should the well remain shut before opening? Every saved minute is direct gas revenue. Across 8–12 cycles per day, saving 2 minutes per cycle adds 20 additional production minutes daily.
Optimal Afterflow Duration
Exactly when should the valve close? The system determines the precise moment gas flow is no longer productive and liquid re-accumulation is about to begin — replacing fixed open timers with real-time performance-based control.
Gas Volume This Cycle (MCF)
The primary revenue metric. Predicting volume per cycle enables production managers to track forecast versus actual in real time and intervene when cycles are trending below target.
Flow Rate at Cycle Close
A declining flow rate at the end of a cycle is the earliest available warning that liquid loading is developing. Catching this trend 2–3 cycles early prevents a stuck-plunger event.
Next Cycle Start Time
When should the next cycle begin? Enables proactive field scheduling instead of reactive responses — reducing callout response lag from 4–12 hours to near-zero.
Casing Pressure Change
Predicted casing pressure differentials allow the system to anticipate the energy state of the next cycle before it begins, enabling setpoint adjustments in advance.
Tubing Pressure Change
Predicted tubing pressure differentials validate the lifting energy available, confirming the upcoming cycle will be productive before the valve opens.
Non-Productive Time Next Cycle
Predicting unproductive time gives complete forward visibility into the next cycle's revenue minutes — supporting daily production planning and reporting.
Hive Mind — Collective Intelligence
Rather than treating each well as a closed system, Hive Mind connects all wells in the portfolio into a shared intelligence network where every completed cycle across every well contributes to a shared knowledge base.
Dense Neural Networks
Each well is represented as a node in a connected network. The model learns which wells are operationally similar — based on formation, depth, tubing configuration, and production history — and routes relevant knowledge between them automatically.
Reinforcement Learning
The system learns optimal control strategies through direct interaction with well outcomes — not from rules that engineers write in advance. Over thousands of cycles, it discovers strategies that no fixed rule set could encode.
Transfer Learning
Models trained on data-rich wells are adapted to new or low-data wells, dramatically shortening the learning curve. A new well no longer starts blind — it starts with the inherited understanding of all comparable wells in the portfolio.
Eliminating Unnecessary Methane Emissions
Conventional fixed-timer controllers over-vent wells systematically because they don't monitor actual liquid clearance or real-time afterflow rate. The motor valve remains open beyond the productive flow window, releasing methane that serves no production purpose.
EPA regulations under 40 CFR Part 60 Subpart OOOO (Quad-O) and Subpart OOOOa (Quad-Oa) impose quantitative limits on methane emissions from liquids unloading and mandate Best Management Practices for plunger lift systems.
For a well cycling 10 times per day, reducing unnecessary afterflow by 2 minutes per cycle eliminates 20 minutes of unproductive venting daily. Across a portfolio of 50 wells, this represents 1,000 minutes of daily methane emission reduction.
Precision Afterflow Closure
The valve closes the instant liquid re-accumulation begins — not when a fixed timer expires. Eliminates methane venting that serves no production purpose.
High-Velocity Blowdown Prevention
Excess casing pressure causes disproportionately large gas slugs at valve open. The AI pre-empts this by bleeding excess pressure in the prior shut-in phase.
Automated Compliance Audit Trail
Every cycle is timestamped and exportable. Supports regulatory submissions under 40 CFR Part 60 Subpart OOOO (Quad-O) and OOOOa (Quad-Oa).
Carbon Credit Eligibility
Documented, quantified methane reduction programs qualify for carbon credit programs and strengthen ESG reporting with verifiable emission data.
Ready to see the architecture in action?
View our implementation results or start a 30-day risk-free pilot.