Forecasting energy usage using Amazon machine learning and data lakes
Executives within utilities and energy providers of all types and sizes have multiple ongoing needs to forecast energy usage. For example, as chief customer officer, your teams can use energy forecasts at the household level to proactively engage homeowners with high bill alerts and predict pre-pay or month-end energy charges. As the head of energy efficiency and commercial energy programs, your teams can use forecast to predict potential savings when applying different energy efficiency measures, as well as recommend the best measures to use. You can forecast for each home owner, or in aggregate across multiple commercial buildings for each campus or property manager. As VP of operations you can deploy Amazon Forecast in many different ways, from predicting demand at the circuit or substation level to predicting substation or distribution transformer failures. You can also combine Amazon Forecast with customer demographic data to predict which neighborhoods and streets are most likely to see a rise in solar, battery, and electric vehicle installations.
Most utilities have the data they need to make these types of forecasts including metering, SCADA and customer data, but it’s often locked away in separate incompatible data silos. Utilities often have a variety of tools deployed for forecasting, the most widely being Excel. Usually only a handful of staff are using advanced analytic tools though, and even fewer utilities have data scientists who may have used machine learning (ML) to build artificial intelligence (AI) models for the utility.
Several of our AWS power and utility customers have overcome these issues by leveraging Amazon’s breadth and depth of AI/ML and Big Data & Analytics services to consolidate data into a data lake. These are used to produce analytics insights, predictions, and forecasts in a timely and affordable manner. Utilities have done this without hard-to-find data scientists, using just the skills of their existing IT developers, data engineers, and analytics professionals. One recent AWS customer success story in this area is that of Xylem, a leading global utility technology company.
Our customers can do this because Amazon has taken over a decade of experience in using AI/ML to ensure the smooth running of Amazon’s supply chain and customer engagement processes. That knowledge has been built into a set of easy-to use ML tools, such as Amazon SageMaker and a cadre of fully managed AI/ML services, for functions ranging from image recognition and natural language chat bots, to fraud detection, personalized recommendations, time series data forecasting, and more.
Amazon Forecast is the specific service which utilities would use to predict energy consumption. It’s a fully managed AWS service that uses machine learning (ML) against time series data (exactly what SCADA and metering systems generate) to produce highly accurate forecasts, without requiring any prior ML experience.
This detailed technical blog explores how utilities can use Amazon S3 Data Lakes and the Amazon Forecast ML service to predict energy usage by combining historical interval meter read data with external weather data. Puget Sound Energy used this approach to predict electrical and gas consumption at a typical residence.
Read more about PSE’s proof of concept here.