What we forecast
- Load / net load
- Renewables
- Zonal / nodal prices
- Congestion risk
Product overview
LAPLACE Core™ + STEWARD™ = Forecast × Decide × Execute
We make market price signals not only visible but actionable and reconcilable end to end.
Forecast engine
LAPLACE Core™ turns volatility into measurable foresight.
Strategy execution
STEWARD™ delivers bids, actions and reconciliation with guardrails.
LAPLACE Core™Insight · blue-violetMultimodal forecasting & calibrated uncertainty
STEWARD™Execution · blue-greenStrategy generation with risk guardrails
Forecast engine for the spot era
See shifts in load, renewables and price ahead of time; translate volatility into reliable forecasts so trading and dispatch stay confident.
How it works
Hover or focus to explore
Forecast demo
Toggle day/week to inspect forecast points, P90/P95 intervals and actuals; hover for error and coverage.
Rolling forecast with intervals
Forecast bands
P90/P95 band tighteningt →
Methods
- Transformer/TCN fusion
- GNN topology encoding
- Diffusion scenario generation
Uncertainty
- Quantile regression
- Ensemble / MC Dropout
- Conformal coverage
Metrics
- WAPE · MAE · RMSE
- Coverage P90/P95
- Interval width
- Extreme weather robustness
Explainability
- SHAP / attention
- Scenario comparisons
- Error attribution
Integration
- API / batch inference
- Low-latency online serving
- Rolling hot updates
Execution platform for spot trading & dispatch
Turn forecasts into playbooks, bids and instructions. Every step becomes executable, settleable and reviewable.
How “seeing” becomes action
Hover or focus to explore
Strategy demo
Switch buyer/seller to visualise bid curves, fulfilment bands and deviation cost estimates with tooltips.
Procurement curve + cost band
Strategy stack
Fulfilment guardrailst →
Buyer playbooks
- Procurement / top-ups
- Peak shifting & price avoidance
- Hedging support
- Billing-window friendly plans
Seller playbooks
- Bid/offer design
- Dispatch scheduling
- Storage arbitrage
- Ancillary service coupling
Constraints internalised
- Caps / deviation / settlement
- SOC / power / thermal limits
- Flow / outage rules
Risk guardrails
- VaR / CVaR
- Max drawdown
- Failed-trade rate
- Exception rollback
Execution links
- Market hub APIs
- EMS / AGC commands
- Ledger reconciliation / A-B
Online learning
- Hot updates
- Provincial parameterisation
- Drift monitoring (KL / PSI)
Platform architecture
Data → Forecast (Core) → Strategy (STEWARD) → Execution (market / EMS / AGC) → Evaluation (KPI / A-B / attribution) → Retrain.
Data
Multi-source ingestion, governance and feature services.
Forecast
LAPLACE Core™ delivers calibrated point/interval/scenario outputs.
Strategy
STEWARD™ optimises decisions under objectives and constraints.
Execution
Submit bids, dispatch commands and receive receipts.
Evaluation
Dashboards, A/B, attribution and retraining loop.
KPI board
Spot-native metrics with hover definitions.
Forecast
- Day-ahead price MAE ↓
- Net-load WAPE ↓
- Coverage P90/P95 proven
- Interval width convergence
Bidding & clearing
- Bid/offer hit-rate ↑
- Spread capture ↑
- Failed trades ↓
- Drawdown ↓
Settlement & compliance
- Reconciliation match ↑
- Deviation penalties ↓
- Fulfilment rate ↑
Risk
- VaR / CVaR within guardrails
- Strategy stability ↑
- Max drawdown bounded
LAPLACE Core™ · Multimodal forecasting model
Multimodal foundation, calibrated uncertainty and diffusion scenarios keep forecasts trustworthy and interpretable.
Model paradigm
- Multimodal: time × topology × weather × rules
- Transformer/TCN + GNN fusion
- Pre-train then provincial fine-tune
Horizons & uncertainty
- Seconds→minutes→hours→DA multi-head
- Quantiles + ensemble + Conformal
- P90/P95 coverage guaranteed
Scenario generation
- Diffusion-based price/load scenarios
- Controlled sampling for extreme weather
- Feeds VaR/CVaR stress testing
Feature engineering
- SCADA / bids / NWP / topology
- Seasonality & rolling stats
- Anomaly detection and clock alignment
Training & inference
- Distributed + mixed precision
- Knowledge distillation for edge serving
- Online latency target <200 ms
Explainability & metrics
- SHAP / attention heatmaps
- Reliability diagram & PIT
- CRPS, OOS drift, robustness sets
Architecture sketch
Core multimodal architecture
Inputs → temporal/topology encoders → fusion → point/interval/scenario heads → calibration layer
- Temporal Transformer + TCN encoders
- Topology / geo-aware GNN encoder
- Multi-head outputs: point / quantile / scenario
- Calibration & monitoring: coverage checks, reliability curves
STEWARD™ · RL + hybrid optimisation agent
Risk-aware RL blended with MPC/MILP and multi-agent coordination keeps strategies executable and controlled.
Decision modelling
- State: forecast distrib + assets + rules
- Action: bid curve / output / switching
- Reward: revenue − cost − risk − degradation
Algorithm stack
- Distributed PPO/SAC
- CVAR-regularised + distributional RL
- RL chooses windows, MILP lands curves
Multi-agent
- Generators / retailers / loads
- Opponent modelling & coordination
- Conflict resolution and consensus
Training & evaluation
- Monte Carlo simulations
- Off-policy eval (IPS / DR)
- Guardrail policies & black-swan drills
Online deployment
- Market / EMS APIs
- A/B and canary with rollback
- Drift monitoring (KL / PSI)
Explain & audit
- Action attribution & rule alignment
- Billing reconciliation evidence
- Strategy version tracking
Architecture sketch
STEWARD decision pipeline
Forecast distributions → strategy factory (RL + MPC/MILP) → risk guardrails → execution APIs → settlement & review
- Risk-aware RL with multi-objective heads
- MPC / MILP feasibility projection
- VaR / CVaR guardrails & stability monitors
- Execution loop: market hubs / EMS / reconciliation
Stack & engineering
AI foundation
- Time-series foundation models
- Topology-aware GNN
- Diffusion scenario engine
MLOps
- Feature pipelines
- Train/validate/register
- Deploy & monitor/alert
- Automated retraining
DataOps
- Lineage & quality monitors
- Latency/missing/drift watchdogs
- Feature store & DVC
Reliability
- SLA ≥99.5% (99.9% optional)
- Cloud/edge/hybrid deployment
- Blue-green / canary releases
- Performance dashboards
Governance
- Model cards & reports
- Artifact signing
- Change audit & approval flows
Evaluation & benchmarks
Protocol
- Forward-chaining rolling windows
- Cross-province / site splits
- Time-based train/val/test
Metric stack
- Forecast: WAPE/MAE/RMSE/SMAPE/CRPS/coverage/width
- Trading: hit-rate/spread/failed/drawdown
- Settlement: reconciliation/deviation/fulfilment
Baselines & ablations
- Baselines: ARIMA/Prophet/XGB/LSTM/Transformer/GNN
- Ablations: -GNN / -Conformal / -diffusion / point-only
Significance & robustness
- Diebold-Mariano tests
- Bootstrap confidence intervals
- Extreme-weather & OOD stress with missing injection
Data & integration
Input schema
- Timestamp / node / power
- Bid & clearance logs
- NWP grids
- Topology & asset state
Frequencies
- sec/min/15min/30min/hour/day
- Normalised to core sampling rate
Quality control
- Gap fill & anomaly flags
- Clock drift correction
- Geo alignment & tagging
Ingestion modes
- SFTP/HTTPS
- Message bus
- Batch & streaming pipelines
Outputs
- Point / quantile / interval / scenarios
- Bid/offer curves
- Execution signals & reconciliation reports
Security & compliance
Data protection
- TLS everywhere
- KMS encryption
- Column-level masking
- RBAC & audit trails
Regulatory fit
- Spot market basic rules
- Settlement accounting rules
- Provincial hot updates
Resilience
- Disaster recovery with drills
- RPO/RTO targets
- Rollback & fail-safe playbooks
FAQ
Most frequent engineering questions we answer during delivery.
- Cold start & data threshold?
- Minimal datasets + transfer learning + online calibration get the first model ready fast.
- How do you migrate across regions?
- Pre-trained foundation with provincial fine-tuning and rule parameterisation.
- Contract-friendly uncertainty?
- Conformal calibration + fulfilment bands + contract-aligned reporting.
- Integration with market hubs / EMS / AGC?
- Standard APIs & message bus with idempotent receipts and rollback chains.
Download whitepapers
Grab the latest forecasting and strategy playbooks with implementation notes.