May, 2023
The client is one of the leading electricity suppliers in the United States. It serves a population of around 15 million across 50,000 square miles of the service area. It is committed to preserving the environment by delivering reliable and affordable power through energy efficiency programs to its consumers. The power utility firm is investing significantly in a clean energy future.
The Challenge
As one of the largest electricity suppliers, the client had to process a large amount of structured and unstructured data from its grid assets as images and videos. For our client, accessing, analyzing, and having near real-time predictions on this data for making data-driven decisions was crucial to ensure a continuous and reliable power supply to its consumers.
Here are some of the challenges it faced:
- The existing on-premises system couldn’t handle the large volume of grid resiliency data.
- Limitations on running model inference at scale led to poor or delayed decision-making.
- Limitations in efficiently and quickly accessing, storing, and analyzing data to make informed decisions.
- Manually intensive data collection, model retaining, and productionalizing the model led to inefficiencies in remote sensing data management.
- Heavy reliance on data operations personnel led to significant data duplication issues and processing backlog.
- There was no retraining mechanism to ensure all the training process was continuous.
- Issues with scalability due to dependency on existing on-premises servers.
- Limitations in sharing and utilizing data across the organization.
- Manually intensive MLOps process.
- Limited capability to visualize remote sensing imagery.
In short, the client lacked an enterprise-grade MLOps platform.
The Solution
The electricity supply company sought a digitization partner to best productionalize its grid resiliency data science platform. It decided to work with Persistent Systems based on its proven track record in setting up enterprise-grade MLOps platforms using Google Cloud Platform (GCP).
Persistent Systems’ Managed Services team implemented a modern and highly scalable enterprise MLOps platform. The objective was to support scalable real-time predictions on images from federated sources that meet the client’s cybersecurity requirements.
Persistent helped the client achieve its goals in the following ways:
- Ensured seamless model migration – from the existing on-premises solution to the Google Cloud Platform Vertex AI.
- Created an MLOps pipeline for continuous training and deployment of the models to predict defects in Images.
- Improved data governance by automating the process of providing near-real-time defect prediction feedback to inspectors at the time of review.
- Set up endpoints with monitoring to provide insights into scaling, errors and performance.
- Provided high scalability to predict defects as uploaded by the vendors 24/7.
- Created Dataflow pipelines to move data between multiple datasources.
- Implemented secure predictions by creating private endpoints that can only be accessed within SCE VPC.
- Provided AI/ML support for inspections by integrating models developed for object detection and defect classification.
- Ensured the platform’s scalability and productivity with Vertex AI.
The Outcome
The electricity supply company has witnessed many benefits since Persistent Systems implemented the GCP-powered data science platform, covering 1200+ users and seven programs for images and 7 models in production.
The client is now able to:
- Analyze over 20 million images to detect potential defects in near real-time as the images are dropped into storage.
- Utilize seven AI/ML models to detect asset defects and data quality issues to help inspectors during the inspection process.
- Boost the grid’s flexibility, reliability, and responsiveness with ML and AI-powered automation.
- Store, manage, and analyze tons of grid resiliency data for better clarity.
- Make better, faster, and more informed decisions to improve its overall capabilities in serving its consumers.
- Reduce the frequency and duration of the service outage.
- Conduct efficient inspections of its infrastructure with drones and LiDAR-equipped aircraft while saving time.
Technology Used
- GCP
- Vertex AI
- BigQuery
