Machine Learning Infrastructure for Operational Intelligence

Many organizations rely on operational intelligence systems to reduce the risk of critical failures and the likelihood of incidents that disrupt market systems. Typically, organizations with complex operational systems develop and operate specific dashboards to monitor their status. For example, the dashboards of an electrical grid operator provide key information on the security and reliability of the power system, including component overloads and failures, voltage-related problems, frequency deviations, electromagnetic disturbances, and possible security incidents. Often systems are monitored centrally by a supervisory control and data acquisition (SCADA) system, with alarms that trigger action when conditions exceed defined thresholds.

Machine learning (ML) is increasingly being deployed to augment human oversight. Incorporation of predictive intelligence is on the rise, with predictive indicators also being exposed to operational dashboards. ML models help identify previously unknown problems in simple-to-understand ways, e.g., clustering of substation measurements can identify overloads or detection of unexpected signals in voltage measurements provides early warnings against magnetic fields and frees up engineering resources to assess the cause of the alerts rather than generating alarms with little operational relevance. A growing number of organizations have developed and deployed ML solutions for operational intelligence. Such solutions require careful consideration of their entire lifecycle and specialized tooling to accommodate the challenges of such systems.