Project

Edison Project

Goal: Con Edison's & Columbia moonshot initiative to imbue lean energy management and data science into Con Edison's work processes

Updates
0 new
0
Recommendations
0 new
0
Followers
0 new
0
Reads
0 new
47

Project log

Roger N. Anderson
added 2 research items
The Center for Computational Learning Systems –10 years of research in applying machine learning to ranking distribution grid assets to susceptibility to failure for Con Edison. While these rankings are of value in their own right, they have much greater value if they are integrated into decision support systems. Two kinds of decision support: for planning for operations. CAPT Tool Example In a DOE funded smart grid project with Con Edison, we were able to make much of what we developed real time In the "wired enterprise", planning and operations merge.
Albert Boulanger: My interest is in amplifying human intelligence with machines. I am an experimentalist by nature. I hold J.C.R. Licklider’s vision of a Man-Computer Symbiosis as a personal goal in building computer software systems Over 36 years experience in rapid prototyping of systems for R&D. Introducing new technology to industry & moonshots. I have extensive complex system integration experience. My forte is in integrating numerical, intelligent reasoning, human interface, and visualization components into seamless human-oriented systems.
Roger N. Anderson
added 2 research items
Anticipatory Stochastic Control for 3G that integrates all aspects of control to optimize including financial Based on real options & reinforcement learning Simulation of aspects of 3G – both for planning and operation The Integrated Systems Model (ISM) based on DEW from Virginia Tech Machine Learning is used to “drink from the fire hose” Internet scale pub/sub as carrier (Phil Gross’s Thesis) Lean Manufacturing business processes to “leap the tall building” Computer Aided Lean Manufacturing (CALM)
Past pioneering development in advanced 4D (time-lapse) seismic, SeisRes, in a skunkworks with Baker Hughes, IBM Watson Research, and Columbia University Founded by Roger Anderson and Albert Boulanger. 22+ years research in providing intelligence in “last mile” industries, first in Oil and Gas and later in Electric Power Realized there was a need for solutions that would be the arch between situational awareness tools on one hand and the various business and engineering tools used to “make it so” This realization first came at the beginning of the e-Fields (Smart Fields) era in Oil and Gas in the late 90’s driven by extremely large CAPEX (think Boeing scale) exploration and production scale of the ultradeep.
Roger N. Anderson
added 3 research items
The Feeder Susceptibility Rankings are retrained daily at 4 am using Machine Learning algorithms developed specially for Con Edison, and then respond to dynamic changes from continuously running wolf (in Manhattan only a the moment) and Load Pocket Weights summed over each feeder in real time. Currently, the prediction model has indicated 42 O/A's out of the 65 were susceptible to failure at the time of their outage, which is a 63% true positive prediction rate so far this summer. A direct data source from NetRMS will also be introduced to the model in the next few weeks to expand the dynamic monitoring to all Boroughs. The model has continued to learn daily and is improving on the rate of true positive prediction. The susceptibility to failure is displayed on a website, and predicted rankings at the time of each Open Auto are compared with predicted performance accuracies.
2+ years research in providing intelligence in “last mile” industries, first Oil and Gas and later Electric Power, Realized there was a need for solutions that would be the arch between situational awareness tools on one hand and the various business and engineering tools used to “make it so” This realization first came at the beginning of the e-fields (Smart Fields) era in Oil and Gas in the early 2000s.
Goal of Edison Project Plant Model (Distribution Engineering Workstation) initiative Machine Learning initiatives (Feeder & Joint ranking models) Capital Asset Prioritization Tool(CAPT) Engineering planning tool Development of the Contingency analysis tool - Phase II of HUD
Roger N. Anderson
added a research item
In order to meet the requirements of Con Edison applications, Columbia CCLS has significantly advanced basic science as well as software engineering in machine learning. CCLS has innovated in a number of directions that have been required to deal with: • very large data sets • very large numbers of attributes for each data object • ranking of results • extensive pre-processing specific to Con-Edison data mining • data “cleaning” and data with missing or noisy attribute values • joining of several databases based on inferred attribute values, • the programming, testing and deployment within Con Edison of hardened software using our novel algorithms. Several patents on basic algorithms and their applications to power grid tasks have been applied for, and one has recently been issued. Many publications aimed at both academic and practitioner audiences have resulted from the Columbia-Con Edison partnership.
Roger N. Anderson
added a research item
Network 82% of load, 2.3 million customers Second contingency (n-2) design Fed from primary feeders 13kV: Man. / Bx./ West.: 916 27kV: Bklyn / Qns. : 309 33kV: SI: 23 Non-Network: (B/Q, Bx, West, SI) 18% of load, 835,000 customers First contingency (n-1) design 4KV fed from primary feeders 225 unit substations 741 4 KV feeders 134 Auto-Loops fed from primary feeders
Roger N. Anderson
added 26 research items
Success Story: Prediction of Good versus Bad PILC cable sections for the PILC Cable Replacement Program has been developed from state-of-the-art computational algorithms (Machine Learning) and rankings have been delivered to all districts. Collaborators: Cable and Splice Center of Excellence; Bill Fairechio, George Murray, Bill McGarrigle and the Van Nest team. MECC Operations; Mark Mastrocinque and Matt Sniffen. Distribution Engineering; Matt Koenig and Eli Cheblis. The Problem: Failures on remaining PILC cable sections have been increasing over time. The cost of replacement is in the $Millions/yr and will take many years. Open Autos on primary feeders are expensive and increase the risk of customer outage and incidents. The Analysis: With the Cable Center, Columbia University introduced computer learning tools and techniques to the problem of which PILC cable sections to replace next. It is an excellent starting point because an abundance of autopsy date for previous failures exists in the database CAJAK, and the current configuration and history of the topology of the underground network exists in the VISION database. Over 200 attributes were collected about primary feeders and how they fail from databases such as Jeopardy, the Hi-Pot program, PQNodes, VDAMS, WOLF, and the PI database. Only through these next generation machine learning algorithms could some of these attributes be found to contribute to the prediction of cable failure. Unfamiliar techniques such as Support Vector Machines, Boosting, and Martingale Ranking techniques were all used to classify and rank the remaining PILC sections in the ground from Worst to Best. A method of presenting the results to Con Edison engineers was settled on that is called "Receiver Operator Characteristics" or ROC. ROC curves are used widely throughout medical, pharmaceutical, manufacturing, aerospace, and the automotive industries. Below is the ROC curve for a May 3 training run using Martingale Ranking of PILC cable sections (in Red) that predicts their "state-of-health" from manhole to manhole and from Worst (#1) to Best (#43,409). The ROC presents the predictions in an easy-to-compare visual against the chance of getting a failure right. The 80/20 curve shows where failures that fit a model that says 80% of the failures are on the worst 20% of the remaining PILC sections. Similarly, the 60/40 curve would predict that 60 percent of the failures are going to occur in the worst 40% of the remaining sections and so on. The Success: O/A causal information was gathered throughout the summer at the MECC. Blue Diamonds on the figure below were the cable sections that actually failed (less than 30 OA's) and their "susceptibility-to-Failure" ranking from our machine learning done in May. There is excellent agreement between the model (red) and the actual outages (blue). Of particular importance, prediction of remaining GOOD cable sections at 70/30 to 80/20 rates, but the bad sections only at a 60/40 rate. That means the model needs more CAJAK-like autopsy data on the feeders that have not failed, but GOOD sections can immediately be ranked from 20,000 to 43,409 and should be the last PILC sections to be replaced. This new knowledge alone is expected to allow Con Edison to turn around the failure-versus-remaining-PILC cable trend to produce fewer failures in the remaining population.
Roger N. Anderson
added 3 research items
Establish base MTBF from feeder ranking table with regression equation on MTBF vs. Ranks scatter plot. Pick assets to replace – very often all assets on a “run”, a string of PILC sections terminated by stop joints. Change feeder attribute vector based on changes Rerank after fix. Calculate new MTBF using regression equation.
Paper Insulated Lead Cable (PILC) has been dictated by the NY PUC for remeval in all of NYC. Here Bronx Networks are prioritized cable b cable by the CAPT tool.
Roger N. Anderson
added 2 research items
The purpose of the Edison program is to enable operational innovation through development and deployment of adaptive business intelligence software focused on managing business risk and driving out inefficiencies. This corporate-wide, transformational initiative also enhances customer service and reliability, and increases the assurance that new system designs will be effective when introduced to the electric grid. The program is a collaborative effort between Con Edison, Inc. (CE), Electric Distribution Design, Inc. (EDD), and the Center for Computational Learning Systems at Columbia University (CCLS). Three major software systems are being developed and integrated with existing legacy software systems to form one Integrated System Model (ISM) of the business: • Plant Model (PM) – A high-resolution, flexible and integrated model of Con Edison’s entire electrical infrastructure, from transmission, through substations, to distribution, and to every customer. • Business Process Model (BPM) – A flexible model designed to capture and automate tasks & information processes throughout the company. • Machine Learning (ML) – Adaptive dynamic programming that couples to these models and to existing information to create intelligence that optimizes the business in this uncertain world.
Roger N. Anderson
added a research item
Machine Learning is used for cost/benefit analysis of individual components, and then aggregation into Lean Management prioritization of field work on the Smart Electric Grid.
Roger N. Anderson
added 2 research items
The predictive performance of the Columbia Support Vector Machine (SVM) that drives the prediction of susceptibility to failure correlates best with variations in the Arctic Oscillation which controls fluctuations in the Jet Stream, and the magnitude of Temperature, Humidity and Pressure Changes experienced by NYC. Note the phase lag which results in a 10 day heads-up for when the weather will be more extreme nd outages more predictable. otherwise, they are random with mild weather.
Prevention of Summer Outages OA's last summer FOT's PILC Section Failures.
Roger N. Anderson
added 8 research items
Update on accomplishments with the Edison program, formerly the CALM program, that was initiated in April of this year. The program is a collaborative between Columbia University and Con Edison to implement computer-aided lean management principles that are successfully used by companies like Toyota and Boeing. This is program is designed to be transformational for the entire business by driving out inefficiencies in our processes and managing the uncertainties in our business. This transformation is enabled through the rigor of Machine Learning software.
Objectives: Automate Data Quality Control Add Adaptive Intelligent Systems Measure, Analyze, and Improve Performance Optimize Replacement and Preventive Maintenance
Roger N. Anderson
added 5 research items
CAPT is a Machine Learning Tool to help select assets for replacement Uses 3 Tiers: Load Relief, PILC, Reliability Replacement strategies based on sort order e.g. Replace by PILC rank @ 20% Functions: View Logical Feeder Maps with loads & structures Rollup cost/ranks of tier strategies Compare strategies based on cost and MTBF improvement from feeder baseline
Operational for Regions 3 Tiers Load Relief PILC Replacement Reinforcement Overloaded sections shown on topological feeder map Feeder selection by overload and % PILC left in feeder Visual display of selected sections by Manhole Stop joint priorities added to PILC ranking MTBF estimations by cable runs Operational for Regions 3 Tiers Load Relief PILC Replacement Reinforcement Overloaded sections shown on topological feeder map Feeder selection by overload and % PILC left in feeder Visual display of selected sections by Manhole Stop joint priorities added to PILC ranking MTBF estimations by cable runs New Joint and Cable rankings in 2008 2008 Integrated Database populated for versioning and archiving Capital Costs by Network Real-Time CAPT charting “Nick Deliverables” – Cost vs. Benefit charts & Network Cost Summary MTBF by Network and Citywide PILC Replacement/ MTBF improvements by Feeder, Network, Borough OA’s versus Hipots, PQ Events (linked), Time-Between-Failures 2002-2008 Requirements defined for Optimization using Pareto Curves.
We automated within the Capital Asset Prioritization Tool (CAPT) the collection of MTBF statistics for all feeders and networks. The data quality is good back to January 1, 2002. There are online Charts compiled for up to the minute performance analyses since 2002, including linear regressions of yearly MTBF and summer worst performance compiled network-by-network. A system-wide performance histogram is then updated. We then conducted a statistical analysis of the Network-by-Network improvement from 2002 through 2008 using the mean-time-between OA failures as the measure of treatment success from 2002 through 2008, compared against the control group performance in 2002 using Log Rank statistical analysis techniques. Treatment is defined as all interventions that Con Edison did to attempt to improve the MTBF performance of each Network, including all load relief, PILC and stop joint replacement, and reinforcement.
Roger N. Anderson
added 10 research items
We automated within the Capital Asset Prioritization Tool (CAPT) the collection of MTBF statistics for all feeders and networks. The data quality is good back to January 1, 2002. There are online Charts compiled for up to the minute performance analyses since 2002, including linear regressions of yearly MTBF and summer worst performance compiled network-by-network. A system-wide performance histogram is then updated. We then conducted a statistical analysis of the Network-by-Network improvement from 2002 through 2008 using the mean-time-between OA failures as the measure of treatment success from 2002 through 2008, compared against the control group performance in 2002 using Log Rank statistical analysis techniques. Treatment is defined as all interventions that Con Edison did to attempt to improve the MTBF performance of each Network, including all load relief, PILC and stop joint replacement, and reinforcement.
C2SOS: A Military Cyber-Secure & Interoperable System of Systems for the Smart Grid Situational Awareness, Modeling, and Simulation to Power Command and Control and Automated Contingency Management Boeing and Con Edison
ConEd System Performance 2009-2010 as measured By Columbia CCLS.
Roger N. Anderson
added 5 research items
The existing combinations of inputs presently being used within the decision matrix to push recommendations to the operating team needs to be expanded to cover more cases that are likely to occur.
Roger N. Anderson
added 2 research items
MY VITA as of June 1, 2020, with newest Patent Numbers issued for 5 new Continuation Patents.
Roger N. Anderson
added 2 research items
Presentation of Capital Asset Prioritization Tool (CAPT), patented Edison Project, a joint development of Columbia University's Center for Computational Learning and Con Edison.
Contingency Analysis Program (CAP) was written by a joint team from Columbia Univeristy's Center for Computational Learning Systems and Con Edison called the Edison Project. CAP is running in the Distribution Control Centers of Con Ed analyzing the combination of next worst and next most likely events to what Co Ed is dealing with in real-time. CAP uses patented Machine Learning technologies outlined in this patent.
Roger N. Anderson
added a research item
Energy: More and bigger blackouts lie ahead, unless today's dumb electricity grid can be transformed into a smart, responsive and self-healing digital network— in short, an “energy internet”.
Roger N. Anderson
added 2 research items
Ensuring reliability as the electrical grid morphs into the "smart grid" will require innovations in how we assess the state of the grid, for the purpose of proactive maintenance, rather than reactive maintenance; in the future, we will not only react to failures, but also try to anticipate and avoid them using predictive modeling (machine learning and data mining) techniques. To help in meeting this challenge, we present the Neutral Online Visualization-aided Autonomic evaluation framework (NOVA) for evaluating machine learning and data mining algorithms for preventive maintenance on the electrical grid. NOVA has three stages provided through a unified user interface: evaluation of input data quality, evaluation of machine learning and data mining results , and evaluation of the reliability improvement of the power grid. A prototype version of NOVA has been deployed for the power grid in New York City, and it is able to evaluate machine learning and data mining systems effectively and efficiently.
Albert Boulanger
added 27 research items
Power companies can benefit from the use of knowledge discovery methods and statistical machine learning for preventive maintenance. We introduce a general process for transforming historical electrical grid data into models that aim to predict the risk of failures for components and systems. These models can be used directly by power companies to assist with prioritization of maintenance and repair work. Specialized versions of this process are used to produce 1) feeder failure rankings, 2) cable, joint, terminator, and transformer rankings, 3) feeder Mean Time Between Failure (MTBF) estimates, and 4) manhole events vulnerability rankings. The process in its most general form can handle diverse, noisy, sources that are historical (static), semi-real-time, or realtime, incorporates state-of-the-art machine learning algorithms for prioritization (supervised ranking or MTBF), and includes an evaluation of results via cross-validation and blind test. Above and beyond the ranked lists and MTBF estimates are business management interfaces that allow the prediction capability to be integrated directly into corporate planning and decision support; such interfaces rely on several important properties of our general modeling approach: that machine learning features are meaningful to domain experts, that the processing of data is transparent, and that prediction results are accurate enough to support sound decision making. We discuss the challenges in working with historical electrical grid data that were not designed for predictive purposes. The “rawness” of these data contrasts with the accuracy of the statistical models that can be obtained from the process; these models are sufficiently accurate to assist in maintaining New York City’s electrical grid.
Ranking problems arise in a wide range of real world applications where an ordering on a set of examples is preferred to a classification model. These applications include collaborative filtering, information retrieval and ranking components of a system by susceptibility to failure. In this paper, we present an ongoing project to rank the underground primary feeders of New York City's electrical grid according to their susceptibility to outages. We describe our framework and the application of machine learning ranking methods, using scores from Support Vector Machines (SVM), RankBoost and Martingale Boosting. Finally, we present our experimental results and the lessons learned from this challenging real-world application.
Albert Boulanger
added a project goal
Con Edison's & Columbia moonshot initiative to imbue lean energy management and data science into Con Edison's work processes