BKV Streamlit Apps - BKV Corporation

Data Scientist/Developer: Manuel Solalinde (manolosolalinde@gmail.com)

A single, unified multi-page Streamlit platform that bundles a growing collection of reservoir and production engineering tools, each selectable from the sidebar. The apps share one data pipeline — daily production, injection and SCADA sensor data live in AWS S3 as parquet and are read on demand by well ID — and one deployment, a containerized Streamlit service running on ECS Fargate. Heavier batch jobs (per-well report generation) fan out across parallel Lambda functions; in-app user documentation is published with MkDocs.

This page covers the engineering toolkit; the platform’s flagship optimization app is documented separately:

• Gas Injection Optimization — a data-driven app that recommends optimal gas-injection rates for gas-lift wells and ranks every injecting well against its recommendation. See the dedicated page →

Applications

• Resolution Visualizer — visualizes SCADA sensor reading frequency per well, broken down by day and channel.
• Salt Water Injection — an infinite-reservoir radial pressure-propagation calculator based on the Theis equation.
• Well Communication — detects inter-well pressure interference from correlated, detrended gas-production residuals.
• Critical Rate Sensitivities — computes the minimum gas rate required to lift liquids out of the wellbore.
• CO2 Runtime History — real-time sensor dashboard for the active CO2 injection sites.
• Horner Plots — pressure-transient analysis on shut-in data, estimating reservoir permeability.
• Hall Plots — injection-well performance analysis from cumulative pressure vs. cumulative injected volume.

Resolution Visualizer

Shows the data-collection frequency (resolution) of a well’s SCADA sensor readings, broken down by day and by sensor channel — a quick way to spot logging gaps or changes in sampling rate.

• Paginated well search across SCADA ID, well name, well ID and API number.
• Plotly line chart of resolution (data points ÷ 86,400 s) per day, one trace per channel (line/tubing/casing pressure, gas rate).
• Dashed reference lines at standard measurement intervals (15 s, 30 s, 1 min, 5 min, 15 min, 1–24 h) for easy comparison.

Salt Water Injection

An Infinite Reservoir Radial Pressure Propagation Calculator that models how pressure spreads outward from an injection well into an assumed infinite, homogeneous aquifer using the Theis equation, ΔP = (ρ·g·Q / 4πT)·E₁(u), with the exponential integral evaluated via scipy.special.exp1.

• Configurable storativity and transmissivity, each derived from rock/fluid properties or supplied directly via an override.
• Point-pressure readout plus four visualizations: pressure-vs-distance, pressure-vs-time, a distance×time contour, and a spatial plan-view (top-down) map.
• Downloadable CSV of the spatial pressure matrix; all inputs converted to SI internally and reported in both psi and Pa.

Well Communication

Detects potential pressure communication between wells by correlating their detrended gas-production residuals, exposed through three sub-applications: Map Selection Analysis (pick wells on an interactive map and run all pairwise correlations), Trend Production Analysis (inspect a single well’s decomposition), and Well Pair Analysis (compare two wells by well ID).

• 7-day centered rolling-median smoothing of daily production.
• Reciprocal decline-curve trend fit (starting 365 days after first production) to isolate residuals.
• Pairwise Pearson correlation with a minimum of 365 overlapping observations; high positive correlation flags likely interference.

Critical Rate Sensitivities

Calculates the minimum gas flow rate needed to lift liquids out of a wellbore (preventing liquid loading), across three modes: an interactive Default sensitivity view, a batch CRS Report, and a single-well Report Preview.

• 2D contour plots of critical rate over a pressure grid versus a chosen sensitivity variable (inclination, specific gravity, bottomhole temperature, or surface tension).
• Turner, Coleman, and Belfroid-Coleman (TNO-Shell, inclination-corrected) velocity models; tubing, casing, or annulus flow paths.
• Batch report fans out Lambda functions in parallel across all wells and writes results to S3, with an optional probe point and CSV export.

CO2 Runtime History

A real-time sensor dashboard for the active CO2 injection sites, remapping raw SCADA tags to human-readable fields.

• Interactive Plotly WebGL chart with four independent stacked Y-axes (temperature, casing pressure, pressure, mass flow rate); legend groups toggle whole sensor categories.
• Adaptive resampling (1 min → 1 day) once a date range grows large, to keep charts interactive, with the active resolution shown above the plot.
• Data-quality warnings for stale sensor fields and a CSV export of the resampled series.

Horner Plots

Performs pressure-transient analysis on well shut-in data: it auto-detects pressure-drop events, fits a Horner regression, and estimates reservoir permeability.

• CSV upload (resampled to 1-minute data) or S3 source, with automatic drop detection by configurable duration/magnitude thresholds.
• Timeseries view with adjustable baseline / shut-in-start / shut-in-end markers for fine-tuning on 1-minute data.
• Robust HuberRegressor fit of bhp vs ln(t/Δt) extrapolated to infinite shut-in time, yielding slope, perm-thickness (kh = 162.6·q/m), and permeability estimate.

Hall Plots

Analyzes injection-well performance via the Hall method — cumulative pressure differential vs. cumulative injected volume — where slope changes reveal injectivity decline or improvement over time.

• CSV upload or direct S3 load of the CO2 runtime-history parquet; data resampled to daily averages with forward/backward fill.
• Configurable flowing-pressure column, assumed reservoir pressure (Pe), tangent window (n_days), and analysis start date.
• Plotly scatter of cumulative ΔP vs. cumulative flow with a fitted tangent line and shaded slope window; rising slope indicates declining injectivity (potential formation damage).

Design decisions & trade-offs

• One multi-page Streamlit app instead of a separate app per tool — shared sidebar navigation, a single deployment to maintain, and a consistent UX across the whole toolkit.
• Read-on-demand S3 parquet by well ID rather than a pre-built cache or index — prioritizes data freshness and low storage cost over the marginal latency a cache would save.
• Parallel Lambda fan-out for batch reports (Critical Rate Sensitivities, Gas Injection) instead of synchronous computation — handles per-well parallelization at fleet scale without blocking the interactive app.
• Robust HuberRegressor for Horner-plot fitting rather than ordinary least squares — tolerates outliers in noisy pressure-transient data for more reliable permeability estimates.

Architecture

All tools run inside one containerized Streamlit application using Streamlit’s multi-page architecture — each tool is a numbered page in the sidebar — deployed to AWS ECS Fargate. Per-well production, injection and sensor data is stored in S3 as parquet and read on demand by well ID; batch report generation (Critical Rate Sensitivities, Gas Injection) fans out across parallel Lambda functions that write results back to S3, and operator decisions persist in DynamoDB. Numerical work is done with Pandas / NumPy / SciPy (exponential integral, robust regression) and Plotly for interactive visualization, drawing on shared in-house engineering libraries (PVT, critical-rate and reservoir calculations). Infrastructure is managed with Pulumi, packaged with Docker, and shipped through GitHub Actions CI/CD. In-app user documentation is built and served with MkDocs.

Note: Screenshots and internal application URLs are omitted, as they display proprietary client data. Available on request, subject to client approval.