The 40 base models are grouped into four deployment tiers. Cluster specialisations (49 extra DS-019 rows) inherit the tier of their base model and are not counted separately here.
T1 senses and forecasts → T2 schedules and controls → T3 supports design and health → T4 governs and audits the stack.
T1 Foundation — Real-time sensing and forecasting · 12 base models
The live sensing layer. These models clean telemetry, estimate asset state, and produce the short-horizon forecasts that every optimiser and controller consumes.
- Runtime
- Mostly real-time (heat-pump, PCM SoC, IoT fusion); some batch day-ahead forecasts (wind, price, demand).
- TRL band
- Typically TRL 5–8 — closest to operations.
- Criticality
- Critical / High
- In the stack
- Outputs feed Tier 2 schedulers, control policies and the digital twin. Poor T1 quality cascades into every downstream decision.
- Covers
- Heat-pump output/COP/demand, PCM state-of-charge and charge rate, wind and price forecasts, operating-mode classification, IoT quality and multi-sensor fusion.
Registry models
M-HP-001 Heat Pump Thermal Output Predictor (Real-Time · TRL 6→8)M-HP-002 Heat Pump COP Estimator (Real-Time · TRL 6→8)M-HP-003 Heat Pump Demand Pattern Predictor (Batch · TRL 5→8)M-HP-004 Heat Pump Fault & Anomaly Detector (Real-Time · TRL 5→7)M-PCM-001 PCM State-of-Charge Estimator (Real-Time · TRL 5→8)M-PCM-002 PCM Charge/Discharge Rate Predictor (Batch · TRL 5→7)M-WIND-001 Wind Energy Availability Forecaster (Batch · TRL 6→8)M-WIND-002 Renewable Curtailment Predictor (Batch · TRL 5→7)M-PRICE-001 Electricity Price Forecaster (Batch · TRL 6→8)M-SVM-001 Heat Pump Operating Mode Classifier (Real-Time · TRL 5→7)M-IOT-001 IoT Data Quality & Imputation (Real-Time · TRL 5→8)M-IOT-002 Multi-Sensor Fusion Engine (Real-Time · TRL 5→8)
T2 Optimisation, control and coordination — Turn forecasts into setpoints, schedules and market bids · 16 base models
The decision layer. These models convert T1 state and forecasts into thermal schedules, HTHP setpoints, flexibility bids, and closed-loop control.
- Runtime
- Mix of batch (schedulers, VPP, economics, FEM) and real-time (adaptive control, RL agents, digital twin).
- TRL band
- Typically TRL 3–7.
- Criticality
- Critical / High (a few Medium)
- In the stack
- Consumes T1 forecasts and state; writes control actions (DS-014), scheduler output (DS-013) and flexibility bids (DS-011). Seven of these bases also carry 49 cluster specialisations (HACCP, kiln, furnace, etc.).
- Covers
- Industrial thermal scheduler, multi-asset coordination, HTHP setpoint optimiser, carbon-intensity optimiser, VPP portfolio and bid generator, reduced-order FEM, digital twin, RL agents.
Registry models
M-SCHED-001 Industrial Thermal Scheduler (Batch · TRL 4→7)M-SCHED-002 Multi-Asset Coordination Scheduler (Batch · TRL 4→6)M-CARBON-001 Carbon Intensity Optimiser (Batch · TRL 5→7)M-PROC-001 Process Parameter Predictor (Real-Time · TRL 4→7)M-CTRL-001 Adaptive Control Policy Engine (Real-Time · TRL 4→7)M-HP-005 HTHP Setpoint Optimiser (Real-Time · TRL 4→7)M-RL-001 Thermal RL Agent (SAC/PPO) (Real-Time · TRL 3→6)M-VPP-001 VPP Portfolio Optimiser (Batch · TRL 4→7)M-VPP-002 Flexibility Bid Generator (Batch · TRL 4→6)M-ECON-001 Economic Performance Analyser (Batch · TRL 5→7)M-GRID-001 Grid Constraint Forecaster (Batch · TRL 5→7)M-FEM-001 Reduced-Order Thermal FEM Surrogate (Batch · TRL 3→6)M-TWIN-001 Digital Twin State Estimator (Real-Time · TRL 3→6)M-SVM-002 Site Thermal Cluster Classifier (Batch · TRL 5→7)M-PRICE-002 Intraday Price Spike Detector (Real-Time · TRL 5→7)M-RL-002 Multi-Objective RL Coordinator (Batch · TRL 3→5)
T3 Advanced analytics and design support — Design-time and longer-horizon intelligence · 8 base models
Off the live control path. These models support design choices, degradation insight, plant notes, cybersecurity and model-health monitoring rather than second-by-second setpoints.
- Runtime
- Almost all batch; OT anomaly detection is real-time.
- TRL band
- Typically TRL 2–6 — earlier research and design tools.
- Criticality
- Medium / Low (cyber and drift monitors High)
- In the stack
- Informs PCM material selection and maintenance planning; drift and transfer-learning adapters keep T1/T2 models honest across sites.
- Covers
- PCM degradation predictor and material selector, pinch-analysis assistant, work-order and engineering-notes NLP, OT network anomaly detection, model-drift monitor, cross-site transfer learning.
Registry models
M-PCM-003 PCM Degradation Predictor (Batch · TRL 3→5)M-PCM-004 PCM Material Selector (Batch · TRL 3→5)M-FEM-002 Pinch Analysis Assistant (Batch · TRL 3→5)M-MAINT-001 Maintenance Work Order NLP Extractor (Batch · TRL 3→5)M-CYBER-001 OT Network Anomaly Detector (Real-Time · TRL 3→5)M-MLOPS-001 Model Drift Monitor (Batch · TRL 4→6)M-TRANSFER-001 Cross-Site Transfer Learning Adapter (Batch · TRL 2→4)M-NLP-001 Engineering Notes Entity Extractor (Batch · TRL 3→5)
T4 Governance, assurance and MLOps — Trust, auditability and model lifecycle · 4 base models
The assurance layer. These models do not control the plant; they explain, audit, benchmark and check AI-governance compliance for the rest of the registry.
- Runtime
- Batch.
- TRL band
- Typically TRL 4–7.
- Criticality
- High / Medium
- In the stack
- Reads inference logs (DS-012) and the DS-019 registry; produces audit artefacts for operators, auditors and WP7 digital-integration workflows.
- Covers
- Model explainability reporter, sensor-calibration auditor, algorithm benchmark harness, AI governance compliance checker.
Registry models
M-EXPL-001 Model Explainability Reporter (Batch · TRL 4→6)M-CALIB-001 Sensor Calibration Auditor (Batch · TRL 5→7)M-BENCH-001 Algorithm Benchmark Harness (Batch · TRL 4→6)M-GOV-001 AI Governance Compliance Checker (Batch · TRL 4→6)
