Incorporating Machine Learning into Cloud Based HVAC Monitoring System

• What is Eco-Enterprise’s mission?
• What services are offered?
• What’s the importance of sustainability and efficiency?
• What impact do these initiatives have on economics, financials, and business strategy?
• Who are their competitors?
• In general, what does eco-enterprise and eco-system look to accomplish?
• What are the key data points within HVAC?
• How is this data currently collected?
• What are the key metrics and terms used in HVAC?

• What are best practices when working with HVAC data?
• What trends do you see in these reports/data?
• What kind of anomalies occur within HVAC data?
• What tools are available to process HVAC data reports?

Using your data analysis tools and knowledge from the previous milestone, present your findings

• Is there a way to detect anomalies through data analysis tools? yes, ML programs such as Random Forests
• Is there a way to improve data collection?   Measures of pressures could be added
• Can we detect bad sensors?   Yes, this part is easy
• Can we visualize the data to detect anomalies and unexpected events?  yes, using Archetypal Analysis
• What are the best tools in predictive analytics to use in HVAC data processing?  Random Forests, Hierarchical Clustering, Archetypal Analysis are all useful.  
• What’s the best way to implement?   Ideally the data is collected and analyzed on the fly.  Clusters are developed based on historical data, incoming data is first processed to detect gross errors (compressors off, power down, etc), then compared with historical data clusters.  Outliers are potential faults, and are sent for further analysis.
• What’s the strategy to move forward?   Develop library of data clusters seen in usual operation for all months of Macy’s data.

With the information we have about variables in the data set, we developed summary measures for each day of each month, that can be used in by ML algorithm to detect variations in day-to-day and month-to-month performance of HVAC machines.  Examples of results using this data from standard ML algorithms, Random Forest and Hierarchical Clustering are compared with those from Archetypal Analysis, a more flexible clustering algorithm.   Pitfalls and drawbacks are discussed, as well as potential steps forward.