Tower energy, decided every five minutes.

Joule runs the same advisory architecture across all three use cases (telemetry in, decision logic, signed webhook out) and is independently configurable per site. Operators typically adopt source optimisation first, then layer anomaly detection and fleet intelligence as confidence in the platform builds. The shape of the engagement is the same in all three.

Real time source optimisation. A 5 case priority cascade evaluates solar, battery, grid and diesel every 5 minutes against live demand, tariffs and battery state of charge. Anti flap protection enforces minimum hold times of 15 minutes and diesel cool down. Savings thresholds and hysteresis bands prevent micro switching on partly cloudy days. Every recommendation is deterministic, audit trailed, and accompanied by a human readable reason and a confidence score.

Anomaly and security detection. Joule learns a per site, per hour behavioural baseline over 30 days. Readings that deviate by more than 3 standard deviations within a single 5 minute poll cycle are flagged with a probable cause: unauthorised power draw, equipment theft, breaker trip, panel damage, inverter failure, battery cell degradation, or grid tampering. No manual threshold setting required.

Fleet scale energy intelligence. A single agent instance manages 50 or more sites with shared time series storage and a geographic weather cache. At a 10,000 site portfolio, the architecture reduces weather API fetches to approximately 300 per hour. 19 event types cover operational decisions, site level alerts, and cross site patterns such as regional diesel shortage, grid outage cluster, fleet battery aging, and solar underperformance.

Joule is an advisory layer. It assumes the existence of a Remote Monitoring System (RMS) at the tower with a web interface for data access, a working NOC or SCADA system to receive recommendations, and a tariff schedule per site. The combinations below cover the situations most operators present.

Against timer based or rule based switching. Static rules (switch at 06:00, switch back at 18:00) are blind to today's irradiance, today's tariff window, today's battery state, and today's load curve. Joule replaces the timer logic with a per site, per five minute decision. Existing rules can be retained as safety guardrails; Joule's recommendation is constrained to honour them.

Against manual NOC switching. Where a NOC operator manually decides power source switches across a portfolio, Joule provides a recommendation per site per cycle, with reasoning. The operator may auto approve, batch approve, or selectively intervene. Decision throughput moves from operator bound to platform bound; the operator's attention shifts to exceptions and trend management.

Against a full SCADA replacement. Joule does not replace SCADA. It consumes SCADA telemetry where available, emits recommendations to the same NOC, and adds a decision layer above the existing platform. Operators preserve their installed infrastructure, vendor relationships, and the audit history embedded in the SCADA system of record.

Against hardware retrofits. Joule needs no new hardware at the site. Where a retrofit is being considered (additional storage, larger solar, battery chemistry change) Joule can be deployed first as the diagnostic: the savings achievable from intelligence alone, before any capital is committed to new physical assets.

Against the equipment vendor's own AI. Some current generation tower hardware ships AI led power management built in. Huawei being the principal example in the field. Where that capability is present and active, Joule is complementary; where the estate is mixed vendor or older generation, Joule's value is unlocking AI grade optimisation on the legacy footprint a vendor AI cannot reach.

Read the full capability and architecture inventory →