Machine learning for the way gas networks actually run
GasNet is an AI-enabled industrial intelligence and automation platform for the natural gas industry. Its intelligence layer turns the operating data the platform already collects into anomaly flags, failure-risk scores, sales-loss insight and network recommendations.
How to read this page
- ExistingRunning in the GasNet platform today.
- ProposedProposed ML enhancement, planned or under development.
- DirectionProduct direction for the platform.
One data foundation, four layers of intelligence on top
Each capability uses the same connected station data. The visuals run on simulated data to show how each one is intended to work.
Connected Industrial Data Platform
Collect and consolidate operational data from CNG dispensers, compressors, metering equipment and related station infrastructure.
- Edge gateways for Compac and TGT dispensers, compressors, meter skids and gen-sets
- Live totaliser, flow, pressure, temperature and run-hour readings
- Shift entries, reconciliations and approvals in the same data model
- Structured operating history that machine learning models can learn from
ML-Based Anomaly Detection
Identify unusual dispenser behaviour, abnormal sales patterns, equipment underperformance and deviations from historical operating profiles.
- Learns each dispenser's normal profile by shift, weekday and season
- Flags readings that leave the expected band, with the size of the deviation
- Routes each flag to O&M or marketing with the context needed to act
Predictive Asset Maintenance
Analyse operating trends and equipment history to estimate failure risk and prioritise preventive maintenance.
- Combines run-hours, temperature, pressure and alarm history per asset
- Scores failure risk and ranks assets for the O&M team
- Feeds the existing maintenance schedule rather than replacing it
Intelligent Sales and Loss Analytics
Compare actual and expected sales, identify reconciliation exceptions, and estimate operational or revenue improvement opportunities.
- Builds on GasNet's existing gas and cash reconciliation reports
- Expected sales per station and shift, learned from history and peer stations
- Sizes each gap in kg and ₹ so teams work on the largest first
Network Performance Optimisation
Forecast station demand, benchmark sites and recommend operational improvements across distributed CNG infrastructure.
- Station demand forecasts by time slot
- Benchmarking of comparable sites across the network
- Recommendations for LCV dispatch, capacity and maintenance windows
Four model families, one shared data foundation
What each planned model reads, what it produces and how it is designed to work.
Anomaly detection model
Capabilities 02 · 04- Arm-wise sales and transactions
- Flow, pressure and totaliser readings
- Shift, weekday and season context
- Flag with deviation size
- Ranked drivers
- Probable cause and next step
Planned approachLearned baseline for each dispenser and shift, with deviation scoring against it.
Failure-risk model
Capability 03- Run-hours and maintenance history
- Oil temperature, suction and discharge pressure
- Trips, alarms and communication drops
- 30-day failure-risk score
- Priority in the O&M queue
- Trend behind each score
Planned approachTrend and variance features for each asset, scored for risk and ranked across the fleet.
Expected-sales model
Capability 04- Station and shift sales history
- Price changes and calendar effects
- Peer-station patterns
- Expected volume per station and shift
- Gap in kg and ₹
- Reconciliation exceptions
Planned approachForecast of expected volume, compared with reconciled actuals every shift.
Demand-forecast model
Capability 05- Station sales by time slot
- LCV Filling Post throughput and timing
- Network layout and peer sites
- Utilisation forecast by slot
- Site benchmark ranking
- Dispatch and capacity recommendations
Planned approachTime-slot demand forecasting combined with benchmarking of comparable sites.
From sensor reading to recommendation
How GasNet's machine learning layer is designed to process station data, end to end.
Planned ML pipelineIngest
Telemetry and station entries arrive over MQTT and APIs.
Validate
Gaps, meter jumps and duplicates are cleaned before modelling.
Engineer features
Shift, weekday, season, run-hours and price context.
Score
Models score deviation, failure risk and expected demand.
Explain
Every flag carries its drivers and the size of the deviation.
Learn
Operator feedback on each flag improves the next model.
AI that operations teams can trust
Four design principles behind every GasNet model.
Explainable
Every flag carries its drivers and the size of the deviation, never a bare score.
People decide
Models recommend and people approve. Price changes keep maker–checker approval.
Learns from feedback
Confirmed and dismissed flags become training data for the next model version.
Secure by design
Models run under the same confidentiality, integrity and availability controls as the platform.
Where each capability stands
The automation and data foundation is in production. The machine learning layer builds on it in stages.
Connected data & automation
- Connected industrial data platform
- Live equipment telemetry, alerts and escalation
- Sales, reconciliation and equipment-health analytics
- Remote dispenser rate change with acknowledgement
Detection & prediction
- ML-based anomaly detection
- Predictive asset maintenance
- ML layer for sales and loss analytics
Network optimisation
- Station demand forecasting
- Site benchmarking across the network
- Operational recommendations for distributed CNG infrastructure
Bring AI to your stations
Talk to us about piloting anomaly detection and sales-loss analytics on your own station data.