Here are three short videos that demonstrate some recent machine learning projects by OSIsoft interns. Like anyone who wants to apply big data analytics on their real-time data, they need to format it in a way that their favorite big data tools can accept. Then clean it of outliers. Customers tell us with traditional tools that can take 90% of their project time. But these interns made short work of it. They used the PI Connector for UFL to bring the data that has been retrieved from a variety of sources into the PI System. They used PI DataLink to inspect their data in Excel and identify the outliers that require filtering. Finally they used the PI Integrator for Business Analytics to transfer the right set of data and hierarchical structures from the PI System into their favorite analysis tool--without the outliers.
Have a look. They make it look easy.
Here intern Thomas Hanh gives an overview of the components he used to predict electricity demand using regression analysis in Microsoft Azure Machine Learning Studio and data retrieved from the PI System.
Intern Ryan Amaudruz describes two Azure machine learning projects, one to predict electricity demand in the state of New South Wales in Australia and a second to predict energy price fluctuation. For the former he was able to assemble all of the data pertinent to the prediction and his results were good. The price fluctuation prediction however involved variables that weren't available to him and so it was not as successful.
Customer support intern Miwa Teranishi also created an electricity price prediction model, this one based on Falkonry's pattern recognition machine learning. See this post on the PI Square community for more details on her project.