數位融合與智慧能源 - 周碩彥特聘教授

DIGITAL FUSION FOR SMART ENERGY 數位融合與智慧能源 Dr. Shuo-Yan Chou Distinguished
Professor of Industrial Management Director of Center for Internet of Things Innovation National Taiwan University of Science and Technology

GREATEST INVENTION 2

EVOLUTION OF CONNECTIVITY INTERNET Computer Computer WORLD WIDE WEB Page
Page SEMANTIC WEB Data Data INTERNET OF THINGS Things Things BLOCKCHAIN Value Value INTERNET OF DISTRIBUTED, AUTONOMOUS THINGS ! +

Copyright © 2020 Center for Internet of Things Innovation, NTUST.
All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 AI EVOLUTION • Expert Systems • Artificial Neural Networks • Fuzzy Logic Systems Traditional AI First Wave Traditional Programming Second Wave Neural Nets – Deep Learning Cognitive Architectures Third Wave • Short- and Long-term Memory • Pattern Matching • Prediction • Prioritization • Reasoning, Planning, and many others • All of the above Artificial General Intelligence (General AI) • Big Data Analytics • Deep Learning Narrow AI Machine Learning: Rather than teaching machines how to do everything …. 4

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 SOURCES OF SMARTNESS “The goal of AI should be to empower humans to be better, smarter and happier, not to create a “machine world” for its own sake. ... People are the strongest component of AI. Smart people, not smart machines, develop the most sophisticated AI systems.” - Gartner AI ARTIFICIAL INTELLIGENCE AUGMENTED INTELLIGENCE IoT Blockchain AI Data Acquisition Data Integrity Data Analytics DATA-DRIVEN TECHNOLOGY-ENABLED USER-CENTERED IoT Blockchain AI

DIGITAL FUSION IoT Physical world data Blockchain AI Trusted data
Data Connectivity Intelligent Internet of Distributed Autonomous Things Trust De-centralization Perception Cognition 6

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 AI IN OPERATIONS AND SERVICES Artificial Intelligence MACHINE LEARNING Predictive Analytics Deep Learning SPEECH Text to Speech Speech to Text VISION Image Recognition Machine Vision LANGUAGE PROCESSING (NLP) Data Extraction Translation Classification EXPERT SYSTEMS PLANNING & OPTIMIZATION ROBOTICS

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 AFFECTIVE COMPUTING Fair AI Algorithms Human+Like Robots Smart Spaces Security Sports Education Healthcare Company Physiological Signals Affective Applications Physiological Sensing Feature Extraction & Classification Emotion Representation … Physiological Data Emotion Estimate Emotion Detection Aspect of Human Physiology Aspect of Emotion Theory

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 Voice AI (One Trick) SERVICE SUPPORT WITH BOTS JIBO, the Family Robot (Social) Customer Support Bots (Shielding) Financial Bots (Optimizer) Poncho, the Weather Cat (Proactive) 11

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 ENERGY-EFFIENCT BUILDING ARCHITECTURAL OR HARDWARE RESOLUTIONS ICT-ENABLED SMART SOLUTIONS HUMAN BEHAVIOR CHANGE • Solar photoelectric glass • BIPV • Ventilation • Green building materials • Natural light utilization • Sun visor / shading • Heat exchange • Renewable energy • Energy Management • Energy harvesting • HVAC • Lighting • Elevator/escalator • Water processing • Demand management • Operation/maintenance efficiency • Consumption data visibility • Persuasive technologies • Facilitation • Gaming • Incentives 12

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 REINFORCEMENT LEARNING FOR HVAC CONTROL Real Building HVAC System Q Neural Network Current State St from sensing data E-Greedy Exploration & Exploitation Control Action At during operation Q value Reward Rt = cost(At-1, St-1 ) + penalty(St ) Q Neural Network Historical transition Storage Store State Transition (St-1 , At-1 , Rt , St ) Last State St , Reward Rt Update Model Q Neural Network Action Value Error (Q value – Q target value) Q target value ( Rt + max Q(St ) ) Optimize Weight (W) using Gradient Descent Update Parameter W Source: Enda Barrett, Stephen Linder, 2015

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 SMART ENERGY MANAGEMENT SYSTEM Artificial Intelligence Role Demand-driven HVAC Smart Meter Analyzing Data Anomaly Detection Alert Real time recommendation Image Detection Count Number of People

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 ELEVATOR Regenerative Braking System excess energy kinetic recovered to be electrical energy using Generator Stored in the Battery Directly used Energy Produced Energy Harvesting Variable affecting total energy • Battery efficiency • Generator efficiency • Elevator Mass • Amount of time car brakes • Elevator velocity right before braking Energy Generation Scenario

SMART WATER DISPENSER Hot water : 1100 Watt (will heat
every 40 minutes) Icy water : 245 Watt (will cool every 30 minutes) Intelligent Learning Ability to learn Modeling for each dispenser Autonomous dispenser Adaptation Boiling & chilling schedule Complex Based on prediction Better sleep mode Dynamic water processing policy Connected Products 3 2 1 Virtual centralization Drinking Service Intelligent cup Change demand Drinking water service system Sensor User Analysis Data Analytics Predictive Maintenance Inter-machine Communication Learning through Artificial Intelligence Improving or optimizing equipment performance Improving or optimizing system performance Ensuring outcome

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 PERSUASIVE TECHNOLOGY FOR ENERGY CONSERVATION Reduction Tunneling Tailoring Self-Monitoring Suggestion Surveillance Conditioning Embedded Devices Smart phone PC / Laptop ICT Strategies User preferences Energy sensors PERSUASIVE TECHNOLOGIES user’s preference analysis time/location analysis energy consumption analysis Content modelling Content Transformation Database repository Repository Internet service data Users and Performance Evaluation Energy Conservation INPUT OUTPUT Context-Aware Engine

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 INDUSTRIE 4.0 INDUSTRIAL REVOLUTIONS FIRST MECHANICAL PRODUCITON Steam Engine SECOND MASS PRODUCTION Electricity and Division of Labor: Conveyor Line THIRD AUTOMATION Electronics and Info Technology: PLC, MES, ERP INDUSTRIE 4.0 • Cyber Physical System (CPS) • Internet of Things (IOT) • Internet of Services (IOS) INDUSTRIAL INTERNET SMART FACTORY FOURTH Shuo-Yan Chou, 2019, “The Fourth Industrial Revolution: Digital Fusion with Internet of Things,” Journal of International of Affairs, Columbia University, Vol, 72, No. 1. 19

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 SERVICE ORIENTED CPS Semantic Product Memory • Top Shell Selection • Circuit-Top Shell Packaging • RES-COM Engraving • Top and Bottom Shell Assembly Highest Priority Green Production Minimize CO2 Source: Wolfgang Wahlster, DFKI, Germany M2M Communication M2M Communication … … Emerging Product 1 Active Semantic Product Memory Emerging Product N Active Semantic Product Memory Machine 1 Active Semantic Product Memory Machine N Active Semantic Product Memory Production Service Discovery, Matching and Execution … … Workpiece Carrier 1 Active Semantic Product Memory Workpiece Carrier N Active Semantic Product Memory Production Path Planning based on Semantic Product Memory Shuo-Yan Chou and Anindhita Dewabharata, 2018, インダストリー４.０ヘのサイバーフイジカルシフテムのアーキテクチヤの枠組み, Communications of the Operations Research Society of Japan, Vol. 63, No. 4, pp. 226-233. (in Japanese)

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 Black Tip for Grip Metallic Corkscrew Lifter Power Button Machine Vision Rubber Strip Label & Series Ventilation 360 Rotation Wheels 1’ 2’6” 2’2” Move Pods to warehouse operator’s work station, reduce the 15 miles to 0 for walking for operators. Eliminate the fatigue factor, and increase the efficiency. Precision and highly ordered in the warehouse lower the cost for huge supplies. Simple Mode Highway System The Kiva robots glide along the barcode gridded floor. • Camera underneath to read floor barcode • Camera on the sides to ensure minimum distance between two robots • Barcode sticker placed on the floor in a 40” x 40” grid Charging Station Every 2 or 3 hours, The robot get 15 minutes charging break. At each Packing Station 6-8 boxes can be packed simultaneously as the bots start queueing in a circular position Product to Pick is marked by a Laser Pointer [Pick-to-light] Packer scans Product Light marks designated box [Pack-to-Light] KIVA:AUTONOMOUS AGVS 21

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 AI VERSUS ELECTRICITY “AI is more important than fire and electricity.” Sundar Pichai CEO of Google Inc. ‘AI MORE IMPORTANT THAN ELECTRICITY’ ‘AI THE NEW ELECTRICITY’ “Just as electricity transformed almost everything 100 years ago, today I actually have a hard time thinking of an industry that I don’t think AI will transform in the next several years.” Andrew Ng Former chief scientist at Baidu, Co-founder at Coursera

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 NEEDS FOR AI IN POWER SYSTEMS OPERATION Source: INESC TEC, Center of Power and Energy Systems, 2019 Too much information Reduce cognitive load of humans and exploit memory from repeated situations Need to act fast Increase decision confidence and find complex solutions High uncertainty Process and communicate uncertainty information without increasing stress levels Not enough meaning Find patterns in sparse data and simplify information, and predict future scenarios High Complexity Handle cases where the modelling of the physical system is complex

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 AI APPLICATION FOR POWER SYSTEMS Power Systems 01 02 04 03 Trans- Mission Line Power System Control Predic- tive Mainte- nance Fore- casting Power system stabilizer Security assessment Contingency screening Load forecasting State estimation Load modeling Load frequency Voltage stability assessment Fault diagnosis Protection Stability enhancement Power flow

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 Forecasting Energy Generation Operational-planning Efficiency Fault Outage Diagnosis Wind power prediction and pattern feature based on Deep Learning method Modeling and optimization of NOX emission in a coal-fired power plant using advanced ML Deep learning neural network for power system fault diagnosis Deep Belief Network (DBN) Extreme Learning Machine and Harmony Search Deep Learning Neural Network Effective in improving the prediction accuracy of wind power, with better prediction error Better accuracy and computing time for the modeling of NOX emission. Able to find the relation between fault and data of measurement • Historical wind speed data from wind farm and recorded every 10 mins • Training with 3 months data to predict future 48 hours of wind power • 5 days real data were obtained from Supervisory Information System • Use HS to optimize the operational parameter on the prediction of NOX by ELM • SCADA data: measurement, equipment ledger, health, weather, topology (30 mins) • Measurement involves voltage, current, active power, reactive power of every user Method MRE MSE R Computing Time ANN 2.15 % 13.76 0.965 0.2002 SVR 1.29 % 53.26 0.984 0.6084 ELM 1.13 % 39.40 0.998 0.2643 Average Accuracy Rate Reclosing Failure Fault Reclosing Success Fault Normal Operation Deep Learning Neural Network 65.3 % 63.8 % 71.3 % Back Propagation Neural Network 81.5 % 78.8 % 82.8 % AI APPLICATIONS IN POWER SYSTEM

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 ANN FOR EARLY FAULT DETECTION ON MAIN BEARING Inside Turbine Gearbox Source: P. Bangalore, L.B. Tjenberg, 2015 PCB-A PCB-B ANN-model warning SCADA records multiple attributes with range 10-min average data. Power generated; Gearbox oil temp; Nacelle temp; Rotor rpm; Bearing PCB-B temp ANN estimates next 10-min Bearing PCB-A temp value. Then, error is used by Mahalonobis Distance (MD) to determine anomaly in bearing. Threshold of MD, based normal data (no fault happen) is utilized to detect the anomaly of temp. There was a real-crack in bearing PCB-A at 23rd Nov, alarm rising. ANN could give early fault warning, in 17 Nov, before the real fault happened Alarm Rise

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 SMART GRID The digital technology that allows for two-way communication between the utility and its customers, and the sensing along the transmission lines. Smart Grid will consist of controls, computers, automation, and new technologies and equipment working together Some analysis that can be done such as: • Dimensionality reduction • Load classification • Short-term forecasting • Distributed data mining Digital sensors and smart metering techniques Grid

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 DEEP LEARNING IN POWER LOAD FORECASTING Short-term 24 hours – 2 weeks, climate factor Medium-term 1 month – 3 years, economic factor & land use Long-term 3 years – 10 years, climate & economic transaction Short-term Load Prediction Deep Belief Network (DBN) Built to forecast the hourly load of the power grid Long-term Load Prediction Recurrent Neural Network (RNN) Built to forecast for 3 to 10 years of wind power generation

29 Source: IBM Corporation 2016 USING BLOCKCHAIN: DIFFERENT USE CASES
Blockchain concept Bitcoin blockchain Other blockchain(s) Services using bitcoin Other services Bitcoin Financial transactions Use cases using bitcoin Other use cases

BLOCKCHAIN VALUE PROPOSITION Reducing inefficiency and waste in data verification
Financial inclusion Internet of Distributed, Autonomous Things Collaborative lifestyle time, space, skills, money, and other resources Internet of money, value and ownership in all forms Substantiation of fragmented values Ownership in the digital as well as in the physical world Nano-economy with fractionization of demand and supply 30

WEB 3.0 CHARACTERICS Source: https://medium.com/@matteozago/why-the-web-3-0-matters-and-you-should-know-about-it-a5851d63c949 Pro-privacy, anti-monopoly Web No central
point of control Ownership of data Reduction in hacks and data breach Permissionless blockchains Device agnostic apps Uninterrupted services Web 3.0 Spatial Web P2P Network Enhanced Network Interopera- bility Semantic Web Smarter AI

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 BROOKLYN MICROGRID PROJECT Solar Energy will be read by using Transactive Grid Meter Solar Energy will be used and stored by the producer-consumer (procumer) Communicate with Blockchain by using Transactive Grid Platform Power production is metered on a second by second basis, with information converted to blocks Using microgrid controller to minimize energy loss when transmit the energy from home to home Microgrid Controller All of energy transaction will be paid by using fictitious tokens Energy Flow Information Flow Excess of energy is transferred Transactive Grid Meter read the energy transfer Received power is metered with information converted to blocks Connected by using traditional power grid BLOCKCHAIN Source: Düsseldorf, 2015 32

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 HOW BLOCKCHAIN HELPS IN ELECTRIC VEHICLE CHARGING Get battery information by using QR code, serial production number by using bar code. Use smart contract to store all of battery information in Blockchain Smart Contract for battery Information Blockchain Store all of battery-information Electric vehicle charging Smart Contract for swapping battery Smart Contract for Controlling Battery Usage If there is a price difference between battery, it will charge coin from one side with lower price battery and compensate to the other side Increase actual number of charging /discharging, loading/unloading duration Use smart contract to control process battery swapping Use smart contract to control electric vehicle battery quality Use battery information From Blockchain Source: Song Hua, Ence Zhou, Bingfeng Pi, Jun Sun, Yoshihide Nomura, Hidetoshi Kurihara, 2018 33

Renewable energy generator Power distributor Private Green Power Seller Public
Power Seller Self-Use Generator Settler Green Power Certificate Matching Agency Green Power Certificate Information Platform Green Power User Normal User Free Trade Official Purchase rate ( 躉購費率 ) Direct Indirect Upload information Bundled (電證合一) Unbundled (電證分離) Self Use and Consume User Environment Benefit Demander RENEWABLE GENERATION DEVICE VALIDATION

DIGITAL TWIN https://authentise.com/news/tag/digital-twin/ Adjustable Assets Predictive Maintenance Outcome-based Business 35

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 Demand Response deals with curtailing/shifting power consumption during peak periods. In detecting patterns, making decisions and taking appropriate reactions, a holistic view of information across multiple domains is needed in DR. To solve DR crucial information integration problem, Semantic and Complex Event Processing (CEP) can be adopted. Application Processing Collection Semantic Information Integration Complex Event Processing Situation Mining, Situation Reasoning, Situation Prediction Real time sensor and instrument data Data Sources Ontology language (CIM, dbPedia, etc) Information Infrastructure Load Predictio n Load Curtailmen t Knowledge Information Data Application Real time consumptio n Schedule information Weather Customer behavior Thermal Storage Storage Electrical Equipment Generator Meter HVAC unit Emergency appliance Appliance Decoration appliance Meter Report Occupancy Change Appliance event Sensor Event Campus Event Event Temporal Concept Instant Building Element Building Office Meeting Room Room Interval Measurements rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type rdf:type Preceds/follows/inConcurrentWith Implies/causes eventHasSource eventHasTimestamp eventHasLocation eventHasSource composes eventHasSource ElectricalEquipment.owl Event.owl Time.owl Building.owl SEMANTIC SMART GRID EVENT MODEL CEP is complex event detection and reaction to complex events • Track & trace, senses & respond • Efficient processing of large number of events Semantic enhances CEP • More agile • More flexible for dynamic change SEMANTIC IN SMART GRIDS

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 Context analyzer Service discoverer Space Type Status Architecture for the implemented reasoning system An example of reasoning process for energy waste context based on policy ontology 606-1 Light_S1 DateTimeValue_1 hasTime hasBooleanStatus 606-1 On Subclass Of SubProperty Of Type Type (after reasoning) R D B Building Data Ontology Model Source: Han, 2015 SEMANTIC IN BUILDING ENERGY MANAGEMENT O_Summer Dawn Time Vacancy Lighting On Policy Building: Date Time Description Building: Summer Season Upper: Time Context Space Vacancy Light On Dawn Office Enclosed Space Contain Belong to Context Service Context Synch Service Synch Model Synchronizer Preprocessor Ontology Translator Reasoner Gathering Rules Conversion Rules Monitoring UI SQL Interface A rule-based ontology reasoning system for context-aware building energy management 38

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 TOWARDS EXPLAINABLE AI (XAI) Source: Dan McCreary Programmatical Approach Machine Learning Knowledge Graph Raw Data Rules Raw Data Answers Raw Data Machine Learning Programs Machine Learning Knowledge Graph Data Answers Rules Predictions Knowledge/Answers Explanations

All rights reserved. 版權所有 © 2020 國立臺灣科技大學物聯網創新中心 For Enabling AI-based Performance Improvements PRAGMATIC RECOMMENDATIONS Get a grasp of what AI can do, prioritize use cases, and don’t lose sight of the economics – without a business case no innovation survives Develop core analytical capabilities internally but also leverage third-party resources – trained people are scarce Store granular data where possible and make flat or unstructured data usable – it is the fuel for creating value Leverage domain knowledge to boost the AI engine – specialized know-how is an enabler to capture AI’s full potential Make small and fast steps through pilots, testing, simulation – AI transformation does not require large up-front investments, but agility

數位融合與智慧能源 - 周碩彥 特聘教授

數位融合與智慧能源 - 周碩彥 特聘教授

More Decks by learnenergy

Other Decks in Education

Featured

Transcript

數位融合與智慧能源 - 周碩彥特聘教授

數位融合與智慧能源 - 周碩彥特聘教授