Introduction to Data Science for PI Data for PI Professionals

1 #OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCI ntroduction to data Science for PI data for PI ProfessionalsAhmad Fattahi Manager, data Science Enablement, OSIsoftKleanthis Mazarakis data Scientist, OSIsoft#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLC Gain a basic understanding of data Science process Understand how PI System fits in the pictureObjective Nomenclature A primer on data Science practice CRISP-DM data Science for PI data : a case studyAgenda#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCA rtificial IntelligenceMachine LearningDeep LearningNomenclatureData Science is an interdisciplinary field of scientific methods, processes, algorithms and systems to extract knowledgeor insights fromdatain various forms. -wikipedia#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCData Science should be a core organizational capabilityas opposed to a technical s all about VALUE!How do you tell myth from fact?

2 #OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCA lways start with a so sharp:how can we improve our bottom line?Sharp:how can we save on energy to cool our building and what factors contribute to that?#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCS tart Small#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCD escriptive/Exploratory What is the number of days in which the set point was never reached? What is the average power usage across all those days?Inferential/Predictive Which features can predictthe power usage on a certain day?Causal/Mechanistic What causespower consumption to fluctuate?How Do You Pick the Right Analysis?#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, predict?.. control?Is the goal of the project #OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLC Architecture, infrastructure, data governanceData Engineer Cleaning, analysis and communicationData Scientist Intuition, what matters and what doesn tSubject Matter Expert Ownership, supportData and Subject Matter ManagementWhat Functions Do You Need For Success?

3 #OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCC ommunication is key.#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCE xplainability#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCThe ExplainabilityTradeoffSource: ResearchGateGmbH#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLC13 Story: Optimize Building Energy Consumption#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCCRISP-DM CRossIndustry Standard Process for data Mining Among most popular methodologies Emphasizes cycles and iterationsSource: KDnuggests#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCA ssume your work is going to be repeated and tweaked frequentlyOver time: Models veer off Physical systems change Priorities evolve New business owners come You get reassigned!Leverage tools such as JupyterNotebooks or other commercial platformsReproducible Work Is the Differentiator#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCThe Cycle Repeats#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCCase study: Interacting with PI System data29#OSIsoftUC#PIWorld 2018 OSIsoft, LLCB usiness and data Understanding What is the Business Objective?

4 What data is available? What data is missing?30#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCI mprove Energy Efficiency of the building, by optimizing the startup of the Variable Air Volume Cooling (VAVCO)unitsCase Study: Facilities OptimizationTimeUnit StartupSetpointreachedOccupancy startCooling DurationWasted Energy7:00 AMCooling Duration If you saved 1 hour a day for 261 Working Days that is a significant amount of energy Subject Matter ExpertValue#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCHow does a VAV unit operate?#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCA vailable data #OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCData understanding Power ConsumptionSaturday and Sunday have different behavior, compared to rest of the weekDaily power usage aligned with Average % CoolingPower ConsumptionCooling rate#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLC Morning startup event Feature extraction with Event Frames Predict Duration of Event FrameData understanding Daily Cooling operationCooling rateTemperatureSetpoint#OSIsoftUC#PIWorl d 2018 OSIsoft, LLCData Preparation and Modelling How do I access PI data from Advanced Analytics platforms?

5 Which features are relevant for model prediction and can I identify any relationships between them? Which model should I use?36#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLC Client libraries for R and Python available in GitHub JupyterNotebooks - data Exploration with Python Asset Analytics -MATLAB integration37 Client libraries for data Science (Python & R)#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCE vent Frames published using PI Integrator for Business AnalyticsWeb based UIData CleansingData shapingAdd / Remove features#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLC TargetAverage end time is 7 AM hoursof wasted Energy Total Lost Hours points to worst performing units Mondayhas significantly higher Average DurationData Exploration Microsoft Power BI#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLC No clear relationships when looking at whole dataset Linear relationship between Duration SetpointOffset on floor level Linear relationship becomes more clear for specific unitsBivariate Analysis#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCC orrelation plot embedded in Power BIConfirms Bivariate Analysis ResultsStrong correlation of Durationwith SetpointOffsetand Outside Temperature#OSIsoftUC#PIWorld 2018 OSIsoft.

6 LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCA zure ML modelResults#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCP roductizing the model Have the model control the wakeup time of the buildingSimpler models can be deployed in PI; some control models are built into the control network Regression or small Decision Trees can be efficiently deployed in AF Larger or more complex models need to go outside PIWork in progressDeployment43#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCData Science delivers ValueLearnthe processIterative process Don t get discouragedLeverage PI System And Subject Matter ExpertiseMain Takeaways#OSIsoftUC#PIWorld 2018 OSIsoft, LLCKeep on learning! PI World presentations Labs and online courses Talk to other users, partners, and us45 List of talks available on PI #OSIsoftUC#PIWorld 2018 OSIsoft, LLCA hmad data Science EnablementOSIsoftKleanthis ScientistOSIsoft#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCQ uestionsPlease wait for the microphonebefore asking your questionsState your name & companyPlease remember the Online Surveyfor this session#OSIsoftUC#PIWorld 2018 OSIsoft, LLC#OSIsoftUC#PIWorld 2018 OSIsoft, LLCT hank YouMerciGrazi