Proof-of-Concept Project for Implementing AKW’s “Brain”, a Patented Machine Learning System for Forecasting, Predicting, Simulating and Optimizing the Production, Gathering System, and Compression Stations of Range Resources and MarkWest from Pads through to the Houston Plant Sales Gate
Roger Anderson, Artie Kressner, Leon Wu, Principals
[email protected]
What is Machine Learning?
• Machine Learning (ML) is a type of artificial intelligence (AI) that uses computers to recognize and prioritize the importance of correlations among past data in order to forecast and predict future outcomes. – System of Systems, the ML optimizer provides the “brain” – SCADA provides the “nervous system” – Engine rooms provide the “heart”
AKW Specializes in Building Real-Time ML Systems
• At AKW, we specialize in building real-time forecasting and control systems that recognize complex patterns in energy (oil and gas) and transportation (plane, ship, rail, truck, & electric distribution grids) that relate to weather, O&M, and congestion.
AKW Engineers Build Real-World ML Systems
• We are a spinout of one of the top ML laboratories in the world, the Center for Computational Learning Systems at Columbia University. CCLS has active R&D in each of the focus areas above.
AKW Engineers Built ML Systems designed by the Customer
Previous ML predictive systems have been co-developed with: 1. Western Atlas, Baker Hughes & Shell = 4D seismic monitoring of oil and gas production and ML reservoir simulation Chevron = Portfolio Management 2. Boeing/KBR = Energy Services 3. Con Edison = Dynamic Treatment, CAPEX Prioritization, Contingency Analysis Tool, Load and Source Optimizer 4. GE/Fedex = Forecasting Tomorrow’s Package Volumes and EV Charge today 5. Selex = Di-BOSS, a Digital Building Operating System w/ our Total Property Optimizer
The AKW Mission: Produce a “Brain” for all Range Resources (RR) production and MarkWest Energy (MW) gathering, compression, and processing activities in SW PA via Phased Proposals: Phase 0 (completed): We scoped the problem by visiting mission critical RR facilities in the study area, which resulted in…. Phase 1 Proof-of-Concept Proposal to RR/MW for the remainder of 2014: Machine Learning (ML) Data Analysis using past history of production, gathering, compression, trouble tickets, and SCADA systems to and within the MW Houston Plant, as well as past weather so that, if successful….
The AKW Mission: Produce a “Brain” for all Range Resources (RR) production and MW gathering, compression station, and Processing activities in SW PA via Phased Proposals: A Phase 2 Proposal for: a real-time ML forecasting system (the Brain) can be brought online in 2015 and integrated with RR and MW control center systems for day and week ahead predictions of most likely outcomes for all pads, compression stations, and pipelines, up to, and A Phase 3 Proposal, including the MW Houston Plant to the Sales Gate.
Why Do You Need a Brain to Maximize Shale Production and Pipeline Operations? • Understand volatility in performance • Test feasibility of added RR/MW real-time data sharing via the Brain that integrates existing sub-systems • Identify significant problems that can be forecast and prevented from happening • Define unique Machine Learning solutions to the RR/MW collaboration • Evaluate preventive maintenance options • Forecast business performance • More accurately plan growth
AKW ML System Learns an empirically-based Dynamic Hydraulic Model of the combined RR/MW Production & Delivery System from the SCADA data
1. 2. 3. 4.
Wellheads thru pad data Pipeline data Compressor station data SCADA from Houston Plant to sales gate
RR and MW Data Capture
1. Range Resources: Historical production volumes, rates, G/L ratios, pressures, temperatures By well By pad
Trouble reports and actions taken CYGNET 2. MarkWest: Pipeline information
Pressures over time across RR gathering system
Pigging schedules and history
Pigging locations and times Pressures before and after Pigging
Incident reports and choking changes Houston Plant Reports and Historical SCADA Data LNG tank volumes at stations, fractionation towers
Machine Learning Brain Phase 1: Potential Deliverables for RR/MW SW PA 1.
ML production history data analysis for each well, each Pad
2. ML gathering pipeline analysis for each pipeline section by pressure, pigging, liquids ,history
3. ML compressor station downtime history and data analysis for each compressor train
4. ML trunk-line pipeline analysis for each pipeline section by pressure, liquids, pigging history
Potential patterns to be discovered using Machine Learning 5. ML integration for each well, each Pad each compressor stations & Houston Plant, including appropriate SCADA data & available weather history
6. Design and testing of ML system (Apps) to scope phase 2 deliverables
Deliverables: Analyses of Past Data
1. Day-ahead forecasting of pad volumes (actuals versus possible) • Gas and liquids • Gas chromatograph data from pads and compressor stations 2. Analysis of field hydraulics vs time and weather • ML forecasting that learns to predict Actual/Possible KPI’s • NGLs leaving pads • NGL tank levels per compressor station
Deliverables: Analyses of Past Data
3. Rank pipelines section by section forecasting susceptibility to liquids accumulations 4. Use that and measured pressures to dynamically forecast optimal pigging schedules 5. Learn problem events
6. Prove out advantages of real-time ML system: • Determine what mission-critical information should be electronically shared between RR and MWCC’s • MW sees critical data viewed in the RR control room & visa versa • Show what data-loggers with critical gathering systems needs to be live
Mission: Produce a brain for all Range Resources (RR) production and MW gathering and compression activities
EMO D E V I L
• Adapt to variable and dynamic incidences of congestion (traffic jams) and mechanical failures in the pipeline system that might be caused by patterns that the ML system may be able to recognize. • Need to solve such problems now or RR/MW will not be able to grow the field as much as planned.
Benefits • • • • • •
Help RR and MW better understand how to operate the field SCADA data from both companies has not been integrated and studied fully Visibility and situational awareness across RR and MW • Each sees critical operations of the other Even small improvements can result in big volume increases and more profitable operations for both RR and MW Lowering overhead & lowering operating costs Help managing interaction between RR & MW • Pigging schedules • Improved operating efficiency • Preventive instead of reactive maintenance • Prevention of repeated mistakes • Learn from historical data to better predict • Better train new people
Phase 1 Proof-of-Concept Budget AKW$Analytics$Machine$Learning$Proposal$ML$to$Range$Resources$and$MarkWest$Energy Roger$Anderson,$Principal Phase$1:$Proof$of$concept$using$existing$historical$data$from$RR$and$MW$,$including$Machine$Leaarning$(ML)$$ Data$Analysis,$and$scoping$of$potential$for$a$realDtime$system$for$ML$Forecasting$in$Phase$2 Mo
per$Hour
per$Day
Total
Personnel Petroleum$Engineer/ML$Computer$Scientist 6 Chemical/Electrical$Engineer 6 ML$Computer$Scientist/Data$Analyst 6 Weather$Sciences$Consultant 6 RR$PE$Coordinator RR$IT$Coordinator MW$Coordinator AKW$Consultant$Costs Estimated$additional$expenses$billed$seperately$as$acrued Travel$**** 2$at$$12$trips$over$6$mo Estimated$Total$$Costs ****$Travel$includes$Weekly$Onsite$updates
$175 $150 $150 $100
$1,400 $1,200 $1,200 $800
$168,000 $144,000 $144,000 $9,600
$465,600 $1,500
$36,000 $501,600
THANK YOU, AND QUESTIONS?
