COATI Map Background

COATI, or Climate-based Outage Analysis and Tracking Intelligence, is a proof-of-concept tool built by CTAC that uses machine learning and predictive artificial intelligence to predict the probability of power outages across US counties. The map contains 3 main reporting views: Probability of a Power Outage, Vulnerability of the County’s Population, and Regional Vulnerability (Heat map).

Data Sources

The data comes from publicly available datasets, including:

• DOE Electrical disturbance reports: https://www.oe.netl.doe.gov/OE417_annual_summary.aspx
• NOAA historical weather data (GHCND): https://www.ncei.noaa.gov/products/land-based-station/global-historical-climatology-network-daily
• NOAA weather forecast data (7-day): https://www.weather.gov/documentation/services-web-api
• Historical power outages (private dataset)
• US Census data
• National hospital information data

Technology Used

We used the latest technologies, including AI and ML, to make it easier for non-technical users to work with big data. The technologies which support the map include:

• AWS Cloud development kit (CDK) used to manage and deploy the application infrastructure and permissions.
• Nodejs web application
• KeplerJS mapping tool used to display the data in a more accessible format than just tables.
• Scikit-learn is used to build and train the machine learning model used to make power outage predictions using a Random forest classifier
• Lambda functions for data ingestion as well as pre and post processing.
• S3 for data storage
• Jupyter notebooks for data exploration and development

Latest Updates to COATI Map

In Spring 2024 we added a Q&A Chatbot to the COATI map interface. A user can query the chatbot and receive an AI generated answer, for example:

• What is COATI?
• How does the tool calculate a vulnerability score?
• What datasets does COATI use?
• What are the top 10 most vulnerable counties over the next 7 days?
• What is the county most likely to lose power today?
• What is the county with the highest vulnerability score in Texas? Include the date, county, state, and vulnerability score in the response. 
• What are the maximum, minimum, and avg outage probabilities for counties on the west coast of the US on 2024-02-18? Include the county, state, outage probability and vulnerability score in the response.

Technology Used

The chatbot uses the following technology stack:

• Google Dialogflow CX
• AWS Lambda 
• AWS API Gateway
• AWS DynamoDB
• Amazon Bedrock 
• Amazon S3 file storage

The chatbot employs Google Dialogflow CX for its conversational AI capabilities, ensuring a robust and interactive user experience. For fulfillment, AWS Lambda functions are utilized, enabling the execution of backend code in response to specific intents identified by Dialogflow CX. This setup allows for dynamic data retrieval and processing, facilitating accurate and contextually relevant responses. AWS Lambda works in tandem with AWS API Gateway, AWS DynamoDB, and other AWS services to provide a seamless and efficient solution for handling user queries and managing data. 

Amazon Bedrock to provide access to the Anthropic Claude Instant 1.2 foundational model and implement Retrieval Augmented Generation through a knowledge base. The knowledge base is configured to use the Amazon Titan Embeddings model to generate the text embeddings stored in OpenSearch serverless.

The COATI Data Exploration Assistant is a custom langchain tool. It was created and deployed as a dockerized Lambda function to explore the COATI prediction and vulnerability score data sets by using natural language queries. The assistant accesses cleaned and processed prediction data sets that power the COATI map. The COATI agent tool implements the ReAct chain of thought process to reason about a user’s request, generates an appropriate data query, and provides an appropriate response to the user.

Read More

The 2023 COATI white paper is available here: https://www.ctacorp.com/case-studies/introducing-ctacs-coati-map-climate-based-outage-analysis-tracking-intelligence/ and the up-to-date COATI demo site is available here to try for yourself: https://coati.ctacorp.com/