API reference & integration guides.

The HelioExpect Forecasting Engine delivers 15-minute forecasts by default and aggregates them to hourly, intraday, day-ahead, and 14-day horizons. Aggregation happens automatically inside the service so trading, operations, and planning teams can consume the cadence they need without rebuilding time-series logic.

Forecast accuracy is evaluated daily by pairing predictions with actual generation inside the prediction-ml analytics jobs. The reconciliation pipeline computes MAPE, MAE, RMSE, bias, and capacity-normalized error, storing the metrics alongside each forecast batch so teams can audit performance over time.

By default the engine refreshes site forecasts every 15 minutes using the latest weather feed and plant telemetry. You can trigger an on-demand run at any time or schedule higher-frequency updates for intraday markets by adjusting the job interval in the orchestration layer.

Each response includes a standard forecast, an ML-adjusted prediction, and an optional conservative track that applies configurable buffers for deviation management. The conservative series is useful for scheduling and compliance workflows where under-generation penalties carry higher risk.

The forecasting dashboard inside HelioExpect visualizes standard, ML-adjusted, and conservative forecasts alongside actual generation, deviation bands, and accuracy trends. Users can toggle sub-arrays, inspect contextual metadata, and drill into any interval without exporting.

Yes. Use the multi-site selector to view combined generation or switch between individual assets. Saved dashboard layouts remember preferred sites, chart groupings, and metric selections so portfolio managers can monitor different regions side-by-side.

Widgets support per-user time ranges, aggregation levels, and normalization (per MW, cumulative, or raw values). Users can pin deviation histograms, intraday ramp charts, probability cones, and accuracy summaries to create views tailored to trading, operations, or executive reporting.

Every chart and table can be exported to CSV, PNG, or PDF. Scheduled deliveries and one-click email sharing allow teams to distribute dashboards without logging into the platform, while secure data feeds provide the same curated datasets for BI tools.

HelioExpect manages licensed irradiance and meteorological datasets on your behalf. Providers can be switched per site configuration, while credential handling remains secured within the platform.

No. Standard deployments include all required weather and irradiance feeds. The platform automatically fetches and reconciles the inputs needed for forecast generation.

Forecasts are produced at 15-minute resolution with optional aggregation to hourly, daily, or custom blocks aligned to each site's timezone.

Each request references HelioExpect site metadata describing capacity, module type, inverter make, number of inverters, and mounting configuration. The engine simulates every active sub-array independently before aggregating to site totals, preserving accuracy for mixed-technology plants.

Yes. Single-axis and dual-axis trackers are modeled using the same algorithms that power our monitoring platform. Provide axis azimuth, tilt range, and tracking limits; the engine computes the real-time plane-of-array irradiance before generating power curves.

Losses are configurable via the `lossDiagram` object. You can specify AC and DC wiring losses, soiling, shading, degradation, auxiliary consumption, and other adjustment factors. These values are applied per sub-array so fleet-level and component-level assumptions stay aligned.

Use the `grid_power_limit` setting to cap exported generation. The engine respects the declared limit after losses and ML corrections, ensuring the final forecast aligns with grid agreements and interconnection contracts.

The forecasting stack pairs physics-based simulations with data-driven corrections. Historical weather, production, and alarm history are featurized inside prediction-ml, which trains gradient boosting and neural models per site. These corrections adjust irradiance-to-power translation while guarding against drift.

The engine applies adaptive damping to avoid unrealistic step changes when weather services update. This protects downstream schedules and makes intraday adjustments smoother, particularly when multiple sub-arrays contribute to the same grid node.

Retraining cadence depends on site maturity and data volume. Production sites typically run weekly calibration jobs, while new assets retrain more frequently until the models stabilize. Triggers are orchestrated through prediction-ml and automatically publish updated weights to the forecasting service.

prediction-ml’s consolidated reporting utilities merge actuals from HelioExpect services with forecast outputs, compute day-level error distributions, and export summaries for energy trading, operations, and compliance stakeholders.

Forecasts are consumable through the dashboard, scheduled email reports, FTP/SFTP drops, and secure data feeds. The delivery layer can publish CSV, JSON, or Excel outputs, and you can enable automated alerts when deviation thresholds are exceeded.

Yes. Use automated file drops or secure feeds to push forecasts into ISO submissions, trading desks, or internal scheduling systems. Forecast metadata includes capacity, block identifiers, and timezone-aware timestamps to streamline downstream ingestion.

Weather provider keys, HelioExpect service tokens, and delivery credentials live in environment variables or managed secret stores. The engine never logs sensitive data, and the documentation includes guidance on rotating keys without downtime.

Structured logs capture request metadata, provider responses, and processing durations. Metrics feed into dashboards so you can monitor run time, error rates, and accuracy in real time. Alerts notify operators of upstream weather issues or missing telemetry.

If a preferred weather feed is unavailable, the engine automatically reuses the last successful run or switches to a secondary provider defined in your configuration, keeping forecasts available without manual intervention.

All forecast requests are timestamped with payload hashes and execution logs. Combined with prediction-ml reports, this provides a traceable history for market regulators or investors without disclosing proprietary data sources.