August 2nd, 2017, AUSTIN, TX
SunPower manages 700 power plants from two Remote Operations and Control Centers (ROCC) in Austin, Texas, and the Philippines. These centers operate around the clock and control over 2.5 GW of commercial solar power. Sensors in inverters, trackers, combiner boxes, transformers, weather stations generate enormous streams of data.
Sensors can generate false positives that demand attention even when the systems are working properly. Operators were flooded with alerts, suffering from alert fatigue as they navigated between multiple systems to address each issue.
Our operators were getting too many alerts and had to make too many clicks to get anything done. It was challenging to standardize our operator’s behavior in this kind of environment.
Sarah Herman, Senior Manager of Monitoring Operations at SunPower
SunPower uses “Asset Intelligence as a Service” platform to streamline its operations. It integrates with their existing IBM Maxino (enterprise asset management) and OSIsoft PI (Data Historian) systems. The special data agents were used to integrate data from the enterprise systems as well as SCADA enabled devices. The agents’ distributed architecture makes data accessible and searchable instantly. Both human- and machine-generated data is co-related for better analysis. Decision Engine is used for a real-time stream to process billions of sensor readings daily. This enables SunPower to process a more data against a complex set of alert rules and performance calculations to better identify and resolve issues.
In the first quarter of deployment, SunPower’s Command Center reported half as many false alarms and nuisance alerts. Response times are reduced as performance issues and alerts are processed faster. When an alarm comes in, operators can quickly see relevant time-series data, search for similar prior events to help diagnose and troubleshoot directly from the platform’s interface.
SunPower is now shifting operations and maintenance from reactive to predictive. Machine learning algorithms are being applied to SunPower’s historical dataset of system performance, to better predict and diagnose failures.