The PI System and MATLAB are in several ways complementary whereas the PI System is used to collect and store large amounts of data and MATLAB can perform mathematical operations on large amounts of data. The purpose of this white paper is to provide a case study of machine learning using MATLAB and the PI System.

The learning example uses data from the OSIsoft headquarters in San Leandro, CA. For this facility, a variety of associated data is collected by a number of instruments and archived in a PI Server collective.

- Building power demand (kW)
- Building energy consumption (kWh)
- HVAC performance (temperatures, pressures, energy consumption, etc.)
- Status of networking devices
- Server health (memory, processor usage, disk space, etc.)
- Weather information (temperature, humidity, wind speed & direction, etc.)

Using the data the paper focuses on developing a forecast for building power demand and examine the available data for appropriate predictor variables. This paper we will walk you through using the RegressionTree in MATLAB's powerful ad-hoc computing workspace and show the advantages of using PI ACE for querying data from PI in a .NET environment to sending data to MATLAB using the COM interface. We will then use ProcessBook to visualize the predictive model and show forecast vs actual data.

If you want to give this a try start with what you know and experiment to find what works and what predictor and response variables define your model.