artificial intelligence and big data are watching power ... · retrospective • since the early...

16
ARTIFICIAL INTELLIGENCE AND BIG DATA ARE WATCHING POWER SYSTEMS Louis Wehenkel University of Liège EPCC-14, Wiesloch May 14-17, 2017

Upload: others

Post on 15-Sep-2019

1 views

Category:

Documents


0 download

TRANSCRIPT

ARTIFICIAL INTELLIGENCE AND BIG DATA ARE WATCHING POWER SYSTEMS Louis Wehenkel University of Liège EPCC-14, Wiesloch May 14-17, 2017

Retrospective •  Since the early eighties, artificial intelligence and machine

learning have been proposed to address a multitude of relevant power systems problems •   Expert systems, ANNs, Decision trees, Fuzzy Logic, Genetic

algorithms, Data Mining…

•  However, the number of actual real-life implementations has been rather modest •   Up to now, Artificial Intelligence, and in particular Machine Learning,

has not been truly a game changer in the context of power systems operation

Big Data and Modern AI

The main driving applications •   Big Science

•   Biology, Astronomy, Particle Physics, Neurosciences •   The web, e-commerce, social networks

•   Recommender systems, searching, ranking, grouping, people, customers, music, movies,

•   Natural language processing •   Speech generation, speech-understanding, chat-bots, automatic translation

•   Computer vision •   Image classification and annotation, video games, face recognition

•   Autonomous devices •   Cars, drones, robots

•   …

Breakthrough in research •   Graphical models and causality calculus •   Bandit theory and Monte-Carlo Tree Search methods •   Ensemble methods, kernel-methods, Gaussian processes •   Enhanced learning frameworks: RL, SSL, TL, AL •   Deep convolutional networks •   Generative Adversarial Networks The gap between Learning, Optimization and Control Theories is closing and these research fields are progressively merging Since 2015, NIPS has become the place to be !

Sharing the scientific results •   High-quality open-source machine-learning toolboxes

•   (Skikit-Learn, R, Tensorflow, Café…)

•   Public big datasets with high-quality annotations •   (text, images, biology, physics…)

•   Shared machine learnt models, compiling huge data sources •   (general purpose image representation, more to come up)

•   Open publishing movement •   (Google-Scholar, ORBI, …)

Sharing data, software, models, and scientific papers, at ZERO-COST is an irreversible trend and a huge accelerating factor of progress

Other big data sources •  Meteorological, climate and environmental data •  Mobility data •  Human activity data •   Internet of Things: buildings, infrastructures, industry

•  Personalized medicine, teaching, econometrics •   The media industry •  Buildings and homes •  Environment and agriculture

Other big data application fields

AI and Big Data in the Tabloids •   Netflix competition (and several others) •   IBM Deep Blue Chess champion •   IBM Watson Jeopardy champion •   Google Car •   Google’s AlphaGo champion •   CMUs LIBERATUS Poker champion •   Stanford’s Deep learning based skin cancer detection •   ...

Power systems operation

Hmm, What’s the problem,

Bunny?

Weather, Environment

Electric Power System

T&D Operators

Regulators

U Y

U: design/control decisions

Y: technical/economic performance measurements

Consumers Producers

Markets Suppliers

Rest of the society, of the economy

3 different contexts of decision making

•  Grid (re)design: •   New technologies, New needs, More uncertainty •   Towards an ‘agile system design’?

•  Asset management: •   Aging infrastructure, can not be ‘rebooted’ nor rebuilt from scratch,

time budget for maintenance and replacement •   Towards a better modeling of ageing processes and a more effective

prioritization of maintenance according to condition and criticality of asset?

•  Operation and control: •   Uncertainty, new dynamics, new control means •   Towards probabilistic and/or robust optimization methods, exploiting

more measurements, and based on better algorithms?

Some types of applications •   Learning proxies of analytical input/output models •  Modeling exogenous factors and their relations •  Compiling human expertise •   Learning from simulators •  Black-box optimization •   Information condensation for human operators

•   and various combinations of these

Some relevant data sources •  SCADA, Substation automation, EMS, GIS, etc •  Assets (including maintenance operations) •  Cyber-infrastructure

•  Smart-Meters, event-recorders, PMUs, Meteo •  Econometrics and market data •  Environmental data (e.g. remote sensing)

•  Gas and water •  Mobility data, social networks

The future of AI and Big Data in power systems operation General topics for discussion

•   Topic 1: Duties vs Opportunities ? •   Topic 2: Low hanging fruits vs Vision ? •   Topic 3: Open data and open software ? •   Topic 4: Brainpower ? •   Topic 5: New entrants ?

Towards a harmonized combination of model-driven and data-driven approaches