A featured contribution from Leadership Perspectives, a curated forum for enterprise technology leaders, nominated by our subscribers and vetted by the CIOApplications Editorial Board.

Advanced Energy Economy

Making energy advocacy more predictable - and strategic - using Microsoft Azure's machine learning capabilities

Energy is highly regulated at the state as well as federal levels. That makes the market prospects for advanced energy companies – ranging from wind and solar electricity generation to energy efficiency, demand response, storage, and grid technologies – subject to legislation pending in 50 state legislatures and regulation promulgated by 50 public utility commissions.PowerSuite, the only fully integrated set of online tools for managing policy risks and opportunities related to energy legislation and regulation in all 50 states, the U.S. Congress, and the Federal Energy Regulatory Commission, helps them keep up.

PowerSuite has collected more than 950,000 bills and nearly 400,000 regulatory dockets representing 80 million pages of text, more than twice the size of Wikipedia in terms of content. But making sense of nearly 1 million bills can be a challenge. Besides the basics like the state where it’s filed, bill title, and name of a sponsor, there are potentially hundreds of pages of amendments, partisan characteristics (governor party, legislature control), committees involved, co-sponsors, votes, and other actions. To help tackle this challenge, we worked with Microsoft, which provided advisory support while webuilta predictive analytics feature, powered by Microsoft’sAzure machine learning service.

Predicting Legislative Outcomes

With this new feature, users can assess the probability of a bill’s enactment based on hundreds of data points. Armed with this insight, they can focus resources and adjust strategies to improve the odds of advancing policies that expand markets and remove barriers to business success.

The legislative process is not static, and neither are PowerSuite forecasts. As a bill progresses, the system automatically updates its probability of passage to consider the latest developments, such as passing one chamber, a new amendment, committee referral, new co-sponsors, and more. Actions on a bill can either increase or decrease the probability that a bill will be enacted. Our platform can forecast the outcome of legislation with an accuracy of 87 percent based just on information available at the time of bill filing, and up to 99 percent as legislative action changes the probability of passage over the course of a session.

Machine Learning with 1 Million Bills

Just as important as what PowerSuite can do is how it does it.

Using Azure’s Machine Learning (ML) Studio and connected tools, we were able to sift through the data and identify key features that are most sensitive to prediction.

The ML Studio provides a near single-click option for quickly running coarse-grained sensitivity studies to flag the most obvious variables for making accurate predictions. More fine-grained analysis requires repeated iteration, which is facilitated by the studio tools. When working with so much data, these are crucial steps. They allow you to discard superfluous information that can mislead an algorithm or make the training process more cumbersome and expensive. Each bill contains as many as 500 different data points to consider and, when multiplied by 1 million bills,the sheer amount of data creates a major technical challenge. Our analysis helped us identify approximately 100 data points per bill as critical for prediction. Theremaining data were used to train and validate our final ML model. With the model “trained” on historical data, we were ready to make predictions on future legislation.

Azure’s modular and flexible ML studio allowed us to build a good predictive model quickly. For most projects of this type, more than 90 percent of the effort is spent preparing data for training the machine learning operation. The remaining 10 percent is devoted to algorithm selection, training, and optimization. Existing Microsoft customers have easy options for moving data into ML Studio. For those who are new to the platform, Microsoft’s recent embrace of open source software and database technology also provides several good options for getting data onto their platform, though not quite as quickly and easily. ML studio allows you to run manual or automated “experiments” to explore data, handle clean up, optimize model parameters, and verify results. Once satisfied with the model, a single click launches the service, making your new model instantly ready to make predictions – and help your users put them to use.

A Data Science Roadmap for Energy Advocacy

Our first major project using the Microsoft machine learning tools has given valuable insight to energy advocates and our internal teams. But we are not done.We have new projects in the works to push our analysis and advocacy of policy issues to the next level, with multi-class document classification, sentiment analysis, and entity analysis, giving us X-ray vision to reveal key policy categories and trends, companies, and people involved in making policy happen. All of that will make PowerSuite an even more powerful tool for energy policy advocacy across the country. We can’t wait to show you what is coming next.


Machine learningdrag-and-drop modular workflow (Microsoft Azure)

Graham Richard is CEO of Advanced Energy Economy, a national business organization. In addition to energy technology and service providers, members of AEE include a growing number of corporate energy buyers, including Microsoft.

Check Out: Energy Tech Review
The articles from these contributors are based on their personal expertise and viewpoints, and do not necessarily reflect the opinions of their employers or affiliated organizations.