GreenFuel Intelligence System · Scientific Evidence Website

From black-box prediction to physics-guided methane intelligence.

GFIS exposes every stage: data provenance, chronological splits, model inputs, LSTM epochs, physical constraints, simulator state, uncertainty, limitations, and replayable evidence.

Public-facing scientific package

One project, two intelligence levels, one evidence chain

Level 1 plans biogas infrastructure across villages, districts and states. Level 2 predicts methane yield, estimates stability, tests operating scenarios and records evidence for each plant. Version 3 explains how these layers connect without hiding the model behind a dashboard.

Level 1 planning context
Measured and public data
Physics-guided neural engine
Reactor simulation & decisions
Evidence, memory & feedback
Methane
Core prediction target

GFIS remains centred on methane-yield prediction. Total-biogas datasets are identified separately and are never relabelled as methane.

VFA/ALK
Stability soft sensor

The system exposes the estimated ratio, thresholds, state classification and the future calibration route—rather than returning an unexplained alarm.

48 h
Stateful simulator horizon

Scenario inputs propagate through methane, stability and physics-bound outputs with memory and exportable action/effect records.

Explore the evidence as connected chapters

The former single report is now a navigable website package. Each chapter states what is implemented, what was executed, what data supports it, and what remains to be externally validated.

Transparent method

Inputs, feature engineering, model branches, equations, physical feasibility, soft sensing and decision logic.

Open method

Actual data and training

Evaluator-suggested public data, leakage-safe chronology, rolling folds, epochs, checkpoints and architecture decision.

Open training

Full-scale industrial evidence

500,400 one-minute Muscatine WRRF SCADA records, provenance, quality audit, target boundary and plant-twin integration.

Open industrial evidence

Working simulator

Executed 48-hour traces, Control Room, scenario handoff, VFA/ALK warnings, physics-bound trigger and run memory.

Open simulator evidence

Evaluator guidance and response

The original black-box concern, actions taken, evidence produced and remaining final-phase work appear as an HTML chapter.

Open response

Connected 3D virtual plant

A software-in-the-loop anaerobic-digestion plant whose equipment, telemetry, alarms and 3D state are driven by GFIS.

Open plant architecture

Separate gasification prototype

An interactive high-temperature thermochemical mass/energy calculator kept outside the anaerobic digester, with visible assumptions and outputs.

Open gasification module

GFIS 30+ national infrastructure

Satellite and waste intelligence, plant siting, hubs, gas/energy networks, hierarchical control rooms, economics and 30-year expansion.

Open GFIS 30+

Reproducible artifact library

CSV outputs, epoch histories, split and run manifests, screenshots, model evidence and retained Version 2.

Open retained report

Evidence labels prevent overclaiming

Implemented

Executable source exists

The capability is present in Python, API or browser code and can be inspected directly.

Controlled demonstration

Scenario evidence exists

GFIS executed the capability on synthetic or controlled stress inputs. This proves function, not plant accuracy.

External evaluation

Independent data tested

Measured public observations are evaluated chronologically with provenance, checksums and limitations.

Scientific rule: a visually working simulator is product evidence; a leakage-safe evaluation on measured data is validation evidence. GFIS reports both, but never treats them as identical.