Transcription
Smart Grid Edge Analytics WorkshopGeorgia Tech Global Learning CenterJune 4-5, 2019, Atlanta, GA, USASmart Grid Edge AnalyticsSponsored by the NSF Spoke Smart Grid Data AnalyticsDavid C. PopeUS Energy(Twitter: @David Pope Linkedin: https://www.linkedin.com/in/davidcpope/ )SASC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Smart Grid Edge AnalyticsSponsored by the NSF Spoke Smart Grid Data AnalyticsWhy does Smart Grid Edge Analytics Matter? Prioritized Areas of Analytic Application in Utilities Enterprise Analytics Lifecycle Smart Grid Edge Analytics ExamplesKey DriversWhat is being solvedWhy are we solvingGlobal Customer ExamplesArtificial Intelligence(AI) and Machine Learning(ML) in Utilities Interesting Use Cases across industries Evolve Your Analytics Platform “Don’t let your analytics environment limit the analytic processes you run”SAS EVP COO/CTO Oliver SchabenbergerC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
SAS in the Utilities Industry560 energy customers worldwide100% of Fortune 500 US Utilities are SAScustomers using SAS for an average of 30 years80% of Global Fortune 500 Utilities are SAScustomers1976 SAS founded with 2 utilities among initialcustomersC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Why do Smart Grid Edge Analytics Matter?By 2019, at least 40% of IoT-created data will be stored,processed, analyzed and acted upon close to, or at the edge. Utilities are facing newoperational challenges they havenever seen before.Utility grid leaders are usually“hard-wired” to thinkoperationally, not predictively.Expectations from regulators andcustomers are changing; the baris being raised.Keeping the lights on will remainthe #1 priority for utilities; theywill need new tools to do this.The volumes, velocity, andcomplexity of smart grid/meterdata will require a robustenterprise analytics platform. In addition, a streaming analyticsengine will be required; this canbe and should be deployed at thegrid edge. Edge AI will be critical as Edge Computing becomes mainstream.C opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Prioritized Areas of Analytic Application in UtilitiesSmart GridEdge AnalyticsImpact all ofthese areasdirectlyC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
What is your utility’s current status regarding implementationof an enterprise analytics platform?Source: https://www.sas.com/utility-analytics-2017C opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Enterprise AnalyticsLifecycleDataTRADITIONAL ANALYTICS LIFECYCLE APPROACH:ACCESS - STORE - ANALYZEData StorageETLDeployAlerts / ReportsfTraditional ApproachC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Enterprise AnalyticsLifecycleDataSTREAM IT, FILTER IT, SCORE IT, STORE ITSENSE - UNDERSTAND - ACTData StorageDeployETLAlerts / Reports/DecisioningDeployfData StreamsIntelligent Filter/ TransformStreamingModelExecutionSmart Grid Edge Analytics ApproachOr Recommended IoT Analytical Lifecycle ApproachC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Streaming Analytics EcosystemC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Key DriversSmart Grid Edge AnalyticsReactive Monitoring & ControlProactive OperationsCentral Station GenerationDistributed GenerationTime-Based MaintenancePredictive MaintenanceSiloed Customer OperationsCustomer/Grid NexusC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Key DriversSmart Grid Edge AnalyticsReactive Monitoring & ControlProactive OperationsCentral Station GenerationDistributed GenerationTime-Based MaintenancePredictive MaintenanceSiloed Customer OperationsCustomer/Grid Nexus And do it all in Real to Near Real-Time!!!Photo by Filip Mroz on UnsplashC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Smart Grid Edge Analytics Examples Edge Analytics – through the use of event stream processingAsset Analytics – preventive vs reactivePower Quality – analysis done more often and further down into smaller subsections of the gridUse of PMUs – helping with reliability and fine tuning operationsRenewables & Microgrid Optimization – requires different and/or more load forecasting to bedonePhoto by Nikola Johnny Mirkovic on UnsplashPhoto by Tim Mossholder on UnsplashC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.Photo by RawFilm on Unsplash
What Are We SolvingSmart Grid Edge Analytics Electric power distribution circuits providepower to end users through a network ofcables, transformers, switches, and otherdevicesCircuit loading is limited by the physicalcapacity of devicesWhen load exceeds limits, the circuit is shutdown or damaged or becomes susceptibleto future failureNecessary to identify transformers,switches, devices etc that are of higher riskto avoid interrupting power supplies to endusersC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
What Are We SolvingSmart Grid Edge Analytics Line distribution transformers: Not monitored by any automated processes- Issues are captured on these devices when customers call in with a problem oroutage is identified- Investigation that follows shows a transformer overload problemSubstation transformers: Annual analysis utilizing SCADA and load forecasting (Five year out) to plan forsubstation transformers- Distribution service transformers typically do not have sensors, instrument transformers, and otherequipment to monitor its health.Some substation transformers may include monitoring systems such as sensorsPreventive Maintenance: All types of equipment on both the generation and transmission sides of the gridC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Why are we solvingSmart Grid Edge Analytics Costs Business Needs Cost of distribution service transformer ranges from 1000 to more than 100,000 basedon its sizeThe power transformer costs can easily exceed 1 millionIncrease proactive prediction of pending transformer failuresFailure factor assessment to reduce transformer maintenance cost (O&M) and increasesystem reliabilityStreamline planning and coordination required for transformer placement and maintenanceBusiness Opportunity Leverage smart meter (AMI) data and big data platform to monitor and proactively managedistribution transformer overloadingC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Why are we solvingSmart Grid Edge AnalyticsValue of Predictive MaintenanceIndependent reports* indicate the following industrialaverage savings resultant from the initiation of afunctional predictive maintenance program: Return on investment:10 times Reduction in maintenance costs:25% - 30% Elimination of breakdowns:70% - 75% Reduction in downtime:35% - 45% Increase in production:20% - 25%*Source: Operations and Maintenance Best Practices Guide. US Department of EnergyC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Global customer examplesPG&EEnterprise platform for analytics;smart meter analytics identify pastand future cost savingsSCEImprove forecasting efficiency withSAS; streamline the load and priceforecasting processEversourcePredict customer payment trends;drive corporate performance reportingSouthern CompanyEstablished predictive model for paymentbehavior to improve collectionsSRPSAS Analytics providesinsights on managing loadand optimizing tradesEDF EnergySAS data analytics improve modeling of churn andpropensity for new products and servicesENELAnalyze smart meter data to determinetrends and seasonal behaviors as well assimulate business scenariosOrigin EnergyImprove demand planning, leading tomore efficient generation and tradingoperationsC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.ScottishPowerMinimize debt across 5M customers byimproving credit risk modelingBritish GasOptimize marketing campaigns tostrengthen customer interactions andimprove asset lifecyclesRWE nPowerImprove short term demand forecastingfor volatile and competitive marketAusgridIntegrate data for a single customer view; providenetwork performance analytics; improve reportingefficiency
Artificial Intelligenceis the science of training systems to emulate human tasksthrough learning and automation.C opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
AI/MLWHAT YOUR FRIENDS AND FAMILIES THINK YOU DO AS A DATASCIENTISTPhoto by Joshua Earle on UnsplashC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
AI/MLWHAT YOU SPEND MOST OF YOUR TIME DOING AS A DATA SCIENTISTPhoto by JESHOOTS.COM on UnsplashC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
AI in UtilitiesGridReliabilityCustomerExperienceC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.EnergyForecasting
C opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Transformer BehaviorOver TimeHISTORICAL DATA USED AS A TARGET FOR PREDICTIVE MODELSC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Target EKG TransformerSignal vs. Decision TreeModelVARIABLE IMPORTANCE INFORMS US WHICH VARIABLES ARE THE MOSTIMPORTANT FOR THIS MODELC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Target EKG TransformerSignal vs. NeuralNetwork ModelIN THIS CASE A NEURAL NETWORK MODEL DOES NOT DO A GOOD JOBIN PREDICTING THIS TYPE OF DATA.C opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Target EKG Transformer Signal vs. Gradient Boost ModelC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Automatic ModelComparison: ShowingGradient Boost as the BestEVEN THOUGH IN THIS CASE IT WAS OBVIOUS THE BEST MODEL IT ISNICE TO HAVE DATA TO BACK US UP.C opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
Transfer ModelDeployed for Real-TimeMonitoringC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
US Utility PMU ScenarioHow SAS supported the processExpected ResultsKey Challenges Phasor Measurement Units (PMUs) take measurementsof the power transmission grid at a much higher speedand fidelity than previous systems providedPMUs take measurements on the power frequency (i.e.60hz), voltage, current, and phasor angle (i.e. where youare on the power sine wave).These units take readings at a speed of 30measurements/second, while the previous systems justtook readings every 3-4 seconds.This more frequent interval provides a much moredetailed view of the power grid and allows detection ofsub-second changes that were completely missed before.To analyze data and learn about the whole grid, we needto synchronize the measurements taken at theselocations.Big data is typically a large amount of data that has beencaptured and stored for analysis. Streaming data isconstantly coming in a high rate of speed and must beanalyzed as it is being received.One of the many interesting things about this project isthat is involves both big data and streaming data. The main purpose of this project is to detect and understand events that are affecting thepower grid, with the objective of keeping the grid stable We have learned there are a number of time-series techniques that are needed for thedifferent aspects of providing the needed answers. The analysis flow breaks down into three areas: event detection (did something happen?),event identification (what happened?), and event quantification (how bad was it?). For event detection, the task at hand is streaming data analysis. 99.99% of the time there isnot event in the streaming data and as such time series model can be used to detect whenthere is a deviation. Event identification is the next order of business. Some events are random, like a lightningstrike or a tree hitting a power line. Others represent some type of equipment failure. As aresult we found these types of events generate similar signatures in the data and through timeseries similarity and time series clustering you can identify what type of event has just happenand ignore those that are non-consequential. Finally there is event qualification. For some events its not that it is taking place, but themagnitude which may make it concerning. For example oscillation which are small in size anddecreasing vs ones that are larger or increasing in size. If there is a need to automatically identify and categorize system events based on any type ofstreaming data patterns, or filter out events that are non-consequential, then thesetechniques will be helpful regardless of the industry.Powered bySAS Restricted DistributionC opy r i gCopyright htSASI nstinc.AllAlrightl r isgreservehts r eserv ed. SASI nsttuteit ut e IInc.d.
Large US UtilityHow SAS supported the processCustomer SuccessExpected Results Key Challenges Distribution grid was under stress fromintegrating renewable energy resources, electricvehicles, and other new technologies.Disruptive technologies had potential to createor extend outages, impacting customersatisfaction.Needed to understand when installed assets areforecasted to exceed design limits because offorecasted changes in load and generation. Use analytics to drive long term assetplanning.Implement data-driven predictivemaintenance programs on distributionsystem.Improve reliability and customersatisfaction, thereby mitigating risk of griddefection.Significant improvements in forecastingaccuracy and speed of performance.Powered bySAS Energy ForecastingRestricted DistributionC opy r i gCopyright htSASI nstinc.AllAlrightl r isgreservehts r eserv ed. SASI nsttuteit ut e IInc.d.
Volvo Trucks and Mack TrucksHow SAS supported the processCustomer SuccessExpected ResultsKey ChallengesVolvo Trucks and Mack Trucks are both subsidiaries of the SwedishManufacturer AB Volvo. Enhance remote diagnostics and monitoring ofcritical engine, transmission and after-treatmenttrouble codes. Minimize unplanned downtime - which creates atremendous toll on fleet operators and theircustomers who depend on timely deliveries. Improve vehicle efficiency and uptime to keep trucksrunning – or ensure the least disturbance to thebusiness if something happens on the road. 175,000 trucks are supported with remote diagnostics. Millions of records are processed instantaneously - reducingdiagnostic time by 70% and repair time 25%. Thousands of sensors on each truck collect streaming IoTdata in real-time to provide the context needed for moreaccurate diagnosis. SAS enables Volvo and Mack to maximize vehicle uptime andminimize the costs of service disruptions by servicingconnected vehicles more efficiently, accurately andproactively. Able to help customers recover from problems faster whilepreventing problems from arising in the first place.Powered bySAS Advanced Analytics and AI“With SAS, we’re working smarter – we’re seeing things that exist in our information that we couldn’t find before, so we can do things moreefficiently and effectively, and drive better results for our customers.” –David Pardue, VP of Connected Vehicle and Uptime Services for Mack Trucks“Our engineers can now see issues before they impact customer operations and change the truck’s design, so we have the best producton the road.” –Conal Deedy, Director of Connected Vehicle Services for Volvo Trucks North AmericaRestricted DistributionC opy r i gCopyright htSASI nstinc.AllAlrightl r isgreservehts r eserv ed. SASI nsttuteit ut e IInc.d.
WARGAMINGCustomer Successhttps://www.sas.com/en m source TWITTER&utm medium social-sprinklr&utm content 1036305675C opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
WARGAMINGCustomer SuccessEvery day, millions of people play World of Tanks to clash with other tanks in virtual battlefields.Big Data Wargaming scales its analytics to understand terabytes of daily customer dataAlex Ryabov – Head of Business Intelligence Data Services Wargaming. “Once we understood the need for in-depth data analysis and data mining, we started doing some initial, advanced analytics modeling in R, Spark, Python andall the other open source solutions.” “The biggest issue for us was scalability. Our data scientists come up with a model concept, do some data wrangling, some data extraction and then we needto automate the results. It was all manual. It was a lot of work for our developers.” According to Ryabov, the first models his team created took three to six months to implement. “Once we realized that we’re going to be running hundreds oreven thousands of models for all of our games, all of our regions and in all of our time frames, we started looking for the solution that could make it scalablefor us.” “SAS Factory Miner and SAS Model Manager were perfect for our use cases,” he says, “because we can take the same model and multiply it by time frames,regions and by different products. So a model is virtually the same, but we can put it into the production environment, where we run, maintain and promote itover and over in an industrial sort of way. In our research, SAS was the only viable option.”Business Results: Transitioned most coding to a point-and-click based workflow for model building efficiencies. Reduced the amount of time needed to develop and deploy models by 60 percent. Reduced the need for data warehouse administration in the deployment and automation of models by 80 percent. Most importantly, the players benefit too. “SAS helps increase overall satisfaction and make the player experience even better.” Taken steps to realize an ROI of US 20 million to 30 million by identifying player microsegments and extending the next best offer to individual gamers.https://www.sas.com/en m source TWITTER&utm medium social-sprinklr&utm content 1036305675C opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
ENTERPRISE ANALYTICS PLATFORM: HIGH LEVELDataStreamingDataAutomatedResponsesC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
EVOLVING YOUR ENTERPRISE ANALYTICS PLATFORM: CURRENT STATEUSEREXPERIENCESStudioEGMIPECMSSAS FarmRetail/CardOpen PlatformSAS OnlyShared FileSystemC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
EVOLVING YOUR ENTERPRISE ANALYTICS PLATFORM: ADD SAS VIYAUSEREXPERIENCESStudioEGVisualsViya Platform – Storage ComputeSAS OnlyShared FileSystemC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
EVOLVING YOUR ENTERPRISE ANALYTICS PLATFORM: REFACTORUSEREXPERIENCESStudioViya Platform – Storage ComputeC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.EGVisuals
EVOLVING YOUR ENTERPRISE ANALYTICS PLATFORM: REFACTORUSEREXPERIENCESStudioViya Platform – Storage ComputeC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.Visuals
EVOLVING YOUR ENTERPRISE ANALYTICS PLATFORM: END STATE 1ON-PREMISEUSEREXPERIENCESStudio VisualsViya Platform – Storage Compute
EVOLVING YOUR ENTERPRISE ANALYTICS PLATFORM: END STATE 2HYBRID ON-PREMISE WITH AWSUSEREXPERIENCESStudioViya Platform – Storage ComputeVisualsViya Platform – Storage ComputeAmazonEMRAmazon S3
Smart Grid Edge Analytics WorkshopGeorgia Tech Global Learning CenterJune 4-5, 2019, Atlanta, GA, USASmart Grid Edge AnalyticsSponsored by the NSF Spoke Smart Grid Data AnalyticsDavid C. PopeUS Energy(Twitter: @David Pope Linkedin: https://www.linkedin.com/in/davidcpope/ )SASC opy r i g ht SAS I nsti tute I nc. Al l r i g hts r eser v ed.
smart meter analytics identify past and future cost savings RWE nPower Improve short term demand forecasting for volatile and competitive market ENEL Analyze smart meter data to determine trends and seasonal behaviors as well as simulate business scenarios Origin Energy Improve demand plann
