Machine Learning and Data Management for Energy Training Course

Participants can map major energy datasets and their relationships to target decisions.

• SCADA tags, telemetry frequencies, engineering units and historian retention policies
• AMI meter intervals, validation estimates and customer-account reference data
• EAM maintenance records, work orders, failure codes and asset hierarchies
• Weather observations, numerical forecasts and spatial joining requirements
• Market prices, bids, dispatch schedules and settlement data structures
• Data lake, warehouse and lakehouse roles for analytical workloads

Participants can establish governance controls for energy data and analytical models.

• Data ownership, stewardship and accountability using a RACI matrix
• Business glossary definitions for demand, availability, outage and curtailment
• Data lineage from source system through feature table and model output
• Access controls for customer, critical-infrastructure and commercially sensitive data
• Retention schedules, archival rules and reproducible dataset versioning
• Governance evidence required for internal audit and model review

Participants can assess energy-data fitness and define practical quality remediation actions.

• Completeness, validity, consistency, timeliness and uniqueness quality dimensions
• Missing SCADA values, communication losses and sensor maintenance events
• Timestamp alignment across UTC, local time and daylight-saving changes
• Range checks using engineering limits, rate changes and physical constraints
• Duplicate meter reads, estimated values and late-arriving data treatment
• Data-quality scorecards with thresholds, owners and remediation actions

Participants can prepare analysis-ready datasets and engineer useful energy-domain features.

• Joining asset, location, weather and operational records using stable identifiers
• Lagged variables, rolling windows and seasonal indicators for time-series models
• Calendar features for holidays, weekdays, billing cycles and peak periods
• Asset-age, maintenance-history and loading features for failure prediction
• Outlier treatment using domain rules, winsorisation and anomaly flags
• Feature documentation covering source, transformation, owner and refresh frequency

Participants can investigate patterns, dependencies and anomalies before model development.

• Load-duration curves revealing demand distribution, peaks and utilisation patterns
• Seasonal decomposition separating trend, seasonal effects and irregular variation
• Correlation analysis interpreted alongside physical causality and operational conditions
• Geospatial plots for feeder events, asset clusters and weather exposure
• Stratified analysis by tariff, region, voltage level and asset class
• Data leakage checks before modelling future operational outcomes

Participants can distinguish core machine-learning problem types and their energy applications.

• Regression, classification, clustering and anomaly detection problem formulations
• Training, validation and test datasets with temporal ordering preserved
• Bias-variance trade-off explained through underfitting and overfitting patterns
• Feature-target relationships, labels and prediction horizons clearly defined
• Baseline models using persistence, seasonal averages and engineering rules
• Interpretability requirements for operational acceptance and human decision-making

Participants can design a demand-forecasting workflow and evaluate forecast performance.

• Forecast horizons for intraday dispatch, day-ahead procurement and network planning
• SARIMA models for seasonal demand patterns and autocorrelated residuals
• Gradient boosting with weather, calendar and lagged demand features
• Hierarchical forecasting across system, region, feeder and customer levels
• MAE, RMSE, MAPE and pinball loss selection by decision purpose
• Prediction intervals supporting reserve planning and procurement decisions

Participants can specify forecasting models for wind and solar generation uncertainty.

• Wind-speed, irradiance and temperature inputs from numerical weather prediction
• Power curves, clipping, curtailment and availability data treatment
• Quantile regression for probabilistic renewable-output forecasting
• Nowcasting compared with day-ahead and week-ahead generation forecasts
• Forecast error decomposition by weather, plant condition and data quality
• Operational use of confidence bands for balancing and dispatch

Participants can design predictive-maintenance use cases for energy asset portfolios.

• Failure modes, effects and criticality analysis for model target selection
• Condition indicators from vibration, temperature, oil analysis and switching operations
• Survival analysis for time-to-failure and maintenance-interval estimation
• Random forest classification for failure-risk prioritisation across asset fleets
• Precision-recall metrics for rare failures and imbalanced maintenance records
• Maintenance intervention costs, false alarms and missed-failure trade-offs

Participants can select anomaly-detection approaches for operational and metering data.

• Isolation Forest detection of unusual meter consumption and operational behaviour
• Control charts for sensor drift, process shifts and threshold breaches
• Autoencoder concepts for multivariate equipment-condition anomaly detection
• Technical losses, non-technical losses and meter-data investigation workflows
• Alert prioritisation using severity, persistence, impact and investigation cost
• Human review queues preventing automatic action from uncertain model outputs

Participants can frame machine-learning applications for outage prediction and restoration decisions.

• Outage event taxonomy, cause codes and restoration-time data quality
• Vegetation, weather and asset-condition predictors for outage-risk models
• Spatial features derived from feeders, substations and weather cells
• Classification thresholds aligned with crew availability and safety constraints
• SAIDI, SAIFI and customer-minutes-lost measures for outcome assessment
• Post-event analysis separating model performance from response-process limitations

Participants can assess analytical applications for customers, tariffs and flexible demand.

• Load-shape clustering for customer segmentation and tariff design
• Propensity models for demand-response participation and programme targeting
• Price elasticity estimation with weather, calendar and customer controls
• Baseline methodologies for measuring demand-response event performance
• Forecasting EV charging demand from adoption, location and behavioural signals
• Fairness checks for customer models affecting offers or service priorities

Participants can evaluate models against statistical, operational and economic acceptance criteria.

• Walk-forward validation for forecasts with changing seasonal operating conditions
• Confusion matrices, ROC curves and precision-recall curves for classifiers
• Backtesting against historical dispatch, maintenance and outage decisions
• Calibration tests comparing predicted probabilities with observed event rates
• Error analysis by season, region, asset class and weather regime
• Acceptance criteria combining accuracy, latency, interpretability and implementation cost

Participants can identify explainability, fairness and accountability requirements for energy models.

• SHAP values explaining feature contributions to individual model predictions
• Partial dependence plots showing model response to key operational variables
• Bias sources in incomplete maintenance, outage and customer-service records
• Fairness assessments for models influencing customer contact or tariff offers
• Human oversight points for safety-critical and high-consequence recommendations
• Model cards documenting purpose, limitations, performance and approved uses

Participants can incorporate cybersecurity and critical-infrastructure controls into analytics designs.

• IEC 62443 concepts for industrial automation and control system security
• IT and OT network segmentation affecting data collection architectures
• Threat modelling for telemetry ingestion, model APIs and data pipelines
• Data poisoning, adversarial inputs and compromised sensor-data risks
• Least-privilege access for analysts, engineers, vendors and service accounts
• Incident response evidence from data pipelines and model-service logs

Participants can plan controlled deployment of machine-learning models into energy workflows.

• Batch, streaming and edge deployment patterns for energy-model inference
• Feature stores, training datasets and reproducible pipeline configuration
• Model registry approvals, version control and release management
• Monitoring data drift, concept drift and prediction-quality deterioration
• Retraining triggers based on performance thresholds and operating changes
• Fallback procedures using rules, forecasts or human operational judgement

Participants can produce an implementation plan for an operational energy data product.

• User stories defining decisions, users, inputs, outputs and acceptance criteria
• Target architecture connecting source systems, pipelines, models and dashboards
• Product backlog prioritised by value, risk, dependencies and data readiness
• Operating model assigning product owner, data steward and model owner
• Vendor evaluation criteria for platforms, integrations, support and portability
• Implementation roadmap covering pilot, validation, rollout and benefits tracking

Participants can review a proposed energy machine-learning project before investment approval.

• Stage-gate reviews from use-case selection through production approval
• Model risk inventory covering materiality, uncertainty and decision consequences
• Independent validation roles, challenge procedures and approval authorities
• Documentation pack including data sheets, model cards and test evidence
• Key risk indicators for data quality, drift, outages and adoption
• Benefits-realisation reviews comparing expected value with realised outcomes

Participants can build, review and present a complete machine-learning use-case package for an energy decision.

• Problem statement linking an energy decision to measurable business outcomes
• Data inventory identifying sources, quality gaps, ownership and access constraints
• Feature plan specifying transformations, refresh cycles and domain assumptions
• Model approach comparing baselines, candidate methods and validation design
• Governance controls covering explainability, security, monitoring and approval evidence
• Implementation roadmap presenting pilot scope, resources, risks and success measures