Overall utility industry analytics maturity has increased substantially when compared to the first Utility Analytics Week nine years ago. This year’s presentations covered the use of machine learning (ML), artificial intelligence (AI), natural language processing (NLP), deep learning techniques, etc. The following highlights a few selected sessions.
Colorado Springs Utilities (CSU) explained how it is using analytics to improve the collections process and optimize resources. We learned that CSU developed a machine learning model in-house and prioritizes collections efforts based on the probability of default and expected past due value. Decision tree analysis is enhanced by random forest and gradient boosting techniques to increase prediction accuracy. The model enables CSU to take proactive action targeting customers at the highest risk of default, and early signs point to a reduction in write-offs.
An Oracle-led session described how its deep learning architecture and machine learning methods detect and disaggregate the power loads of household appliances and electric vehicles (EVs). We viewed graphics that demonstrated remarkable model accuracy including a 98% accuracy rate for EV identification. We learned that sub-metered test homes verified model performance and that the model finds correlations and trains on real data sets rather than relying on rules and/or pre-set assumptions. The speaker emphasized the benefits of Oracle’s pre-built data integration tools, distributed data platform, analytics and utility expertise, and quick time to value for ML applications.
Austin Energy outlined the characteristics of the Electric Reliability Council of Texas (ERCOT) market and wholesale market risk. Key learnings from an unsuccessful time-of-use (TOU) pilot in the 1990s were shared. Customers will adopt TOU plans and change behavior though many factors are at play including program type, incentives, load shape, time intervals, and customer communications. Advanced analytics are being used to evaluate TOU rates for commercial, residential and EV customer classes. The speaker stressed the importance of TOU rates especially as EV adoption increases and described internal and customer challenges.
American Water detailed how it is using AI and NLP to improve dispatch effectiveness. Just beginning its AI journey, the utility achieved a 90% reduction in faulty meter readings and false positives. Through AI, American Water “listens” to the meter and prioritizes meter issues based on risk and revenue loss. The utility credits much of its success to a partnership with App Orchid. American Water sought more than “faster horses”. Advanced analytics, an agile approach, and new processes have been game-changers. Initiatives move from ideation to production in just six to eight months. American Water plans to adopt the same approach for other asset categories.
We heard from American Water again on the event’s final day. The utility is using AI and the Internet-of-Things (IoT) to improve workforce management and safety. AI, ML, and near real-time monitoring of sensor data streams enables the utility to predict problems in advance and a “just in time” approach to work management. Machine capture of incident location, weather, hazard alerts, traffic data, etc. plus a gloves-on app powered by voice-to-text and NLP delivers a complete picture to field engineers via a single pane of glass. The solution has also captured invaluable tribal knowledge from an aging workforce.
To wrap things up, Utiligent shared how live aggregation of AMI data is central to integrated electrical network visualization. The business expectations of AMI were contrasted with practical limitations, largely data related. Utiligent’s Edge Network Services strives to help utilities address these challenges. The CEO of WEL Networks, a New Zealand-based utility and Utiligent client, reviewed the energy trilemma of energy supply, fair pricing and sustainability, and key environmental factors. We learned about the four features of WEL’s strategy: employer of choice, lean operating model, data-driven organization, and sustainability. WEL is asking the right questions: “What should we be doing?”, “What if?”, and “What would we find?” and is assessing the benefits of 5-minute AMI data (e.g. line down detection, impedance mapping, imminent equipment failure detection). Data analytics is seen as the bridge connecting asset management and work delivery to the distribution system operator. WEL is also exploring new revenue streams using data analytics.
In short, a great event and quite a TREAT!
