Role of Generative AI in Utilities
- 1. Role of Generative AI in advanced scenario planning for enhancing distribution system resilience, reliability, and customer satisfaction Dr. Sayonsom Chanda, Senior Scientist, National Renewable Energy Laboratory (United States Department of Energy) Boulder, Colorado USA 1
- 2. History of Adopting Innovation in the Utility Industry 2
- 3. What is Generative AI? 3 “Prompt”
- 4. History of Generative AI 1950s - Markov Chain, a statistical model that could be used to generate new sequences of data based on input. 1990s - Neural Networks 2014 - Generative Adversarial Network (Ian Goodfellow) 2015-16 - DeepMind, VAE, RNN 2017-18 - NVIDIA Progressive GANs - generates images. 2019, 2020 - GPT-2, GPT-3 2022 - Dall-E, GPT-3.5 (Open AI) 2023 - Google BARD, OpenAI (Chat-GPT powered by GPT-4) 4
- 5. Common Use-cases of Generative AI Tools in the Market 5 Text input in Natural Language Text Generation Multimedia Generation Customer Conversations Synthetic Images, Music and Videos DOCUMENT GENERATION: Email, RFPs, Conversations CODE GENERATION: Models, Data
- 6. Examples of commercially available Generative AI tools today 2,500+ Products available for general public to try/buy. Most Common: ChatGPT 6 Image Credit: RapidOps
- 7. So, what it means for the Utility/Energy Sector Industries? 7
- 9. Key aspects of Resilience and Reliability - review RESILIENCE: - Enable customers to be informed during inevitable power outages - Reduce outage duration during extreme events - Not trigger cascading outages and blackouts - Staff coordination during emergencies - Customer requires faster support 9 RELIABILITY - Avoid faults and power outages as much as possible - Acceptable voltage level and power quality to all customers - Dependability of service - Customer requires justification and support - Documentation
- 10. Generative AI can help keep up with rapid changes in energy landscape. 10
- 11. Traditional AI/ML/Data Science Architecture at Smart Grid Companies 11
- 12. Vegetation Management Situation: In some areas of DISCOM, overhead lines are passing through forest area wherein vegetation on the OH lines are causing frequent interruptions. DISCOM officials are able to check the vegetation encroachments only during regular maintenance. Initiative based on conventional AI: DISCOM is interested to carry out the assessment of vegetation encroachment with integration of IT systems such as Image based Vegetation Management. Past 1 year data is available and any other relevant information → Based on that a AI/ML model will be built. BUT, the limitations of this costly model will be: - will be always data-hungry - satellite or drone imagery is expensive - data for all regions will not be available - Seasonal variations 12
- 13. Consumer Experience Enhancement What Utilities Have Today, and have made significant investments in: Yet, Common Issues in most Utilities Worldwide: - Call Centers - ChatBots - WhatsApp based interaction - Long on-hold time - Incorrect tagging of complaints to officer/locations - Long resolution time, - Non-optimum complaint handling procedure, - Manual intervention etc. 13
- 14. What Generative AI can do? - Automatic generated email responses and explanations to billing questions - Automatically escalate & classify cases using sensitivity & domain expertise analytics - Handle multiple queries at once, analyse huge data & convert it into reports etc. Outcomes: - Optimize agent availability & wait time - Require less number of manual intervention and call center agents - Avoid manual mistakes 14
- 15. Asset Inspection Needs Situation: Live monitoring for all assets is economically infeasible and impractical. At present, the health condition of the major equipments like PTRs and CBs in 33/11kV Substations are analysed by DISCOM officials during the scheduled inspection which happens once-every-2-weeks or once-a-month through thermal imagery by visiting the substation physically. 15 Traditional A.I. (Under Development at many places) Probability of Failure Generative AI What would the thermal imagery be? Engineer inputs a given scenario, or verbally reports somethings that may have happened in the circuit.
- 16. Fill the gap of Primary and Secondary Distribution Feeder Models: Identify ways to reduce T&D losses 16
- 17. Advanced Monte Carlo Simulation for Day-ahead Demand Forecasting Situation: The accuracy of day ahead forecast is very much essential to plan for the requirement of Day ahead Power Purchase and planning. Limitations: Forecasting is influenced by various meteorological and socio-economic factors which can lead to a mismatch between Actual vs Projected demand. It is difficult for a team to think of many different scenarios 17
- 18. Value-Added Services based on Smart Meter Data 18
- 19. Employee Experience Enhancement - Can improve data governance & cybersecurity 19 Generative AI (Custom GPT Model) Managers from different business orgs New Employees Retiring or exiting employees
- 21. Determine Upstream Utility network capacity/loading from incomplete behind-the-meter customer data Outcomes: - re-design overall tariff structures - enhanced incentive to consumers, - better plan power procurement, - upstream NTW capacity design 21
- 22. Caveats & Risks of using Generative A.I 22
- 23. Potential Pitfalls and Risks of using Generative AI 23 Misinformation & Disinformation: Generative AI can produce fake images, videos, or text, leading to spread of falsehoods and challenging the credibility of genuine content. Over-reliance & Loss of Skills: As tasks get automated, there's a risk of reduced human expertise in critical areas, potentially making it harder to detect AI-generated errors or anomalies. Ethical & Privacy Concerns: Generative AI models, particularly those that use vast amounts of data, might unintentionally reproduce, amplify or leak sensitive or biased information, posing risks to privacy and fairness.
- 24. Explainability, Accountability, Ethics, Accuracy - It is not possible (till today) to develop an unbiased AI language model 24
- 25. Getting Started with Generative AI 25
- 26. Some options > Try ChatGPT, Large Language Models and other AI tools shown in slide > For Smart Grid specific Generative AI tasks - consider one or two specific tasks, and choose a dedicated platform for smart grid applications. Example platform: https:/ /GridLeaf.org > Customize and self-host a customized Large Language Model for your company. 26
- 27. Let’s Imagine the Future 27 Generative AI + 3D Manufacturing + Quantum Computing → Create anything we need at record speed and efficiency. “The solution to climate change is innovation, not activism.”
- 28. 28 Thank you for your time. Please stay in touch with me: Email: chanda.sayonsom@gmail.com (Personal) or, sayonsom.chanda@nrel.gov (Official) Whatsapp: +1 (509) 432-9525 If I can add you to my free research update newsletter, please let me know