Artificial Intelligence and Machine Learning in Smart Energy Systems
ABSTRACT
The course contains the description of best practices of AI & ML integration in up-to-date fuel and energy sector. The major part of the course is industry- oriented practical training, providing necessary knowledge and skills in the sphere of data analysis and machine learning to manage real story problems of the industry.
The main goal of the course is to provide a comprehensive understanding of the theory and practice of intelligent systems and machine learning methods application in the sphere of power industry (in Smart Energy Systems). Within the framework of the discipline students carry out their own theoretical and experimental research in the field of artificial intelligence, machine learning, design of Smart Energy Systems and master the skills of data representation, processing and analysis.
INTENDED LEARNING OUTCOMES
- to assess the basic characteristics of intelligent systems, their basic features and constraints
- to predict the optimal application of Machine Learning and Artificial Intelligence approaches according to the problem under consideration
- to determine and justify the selection of appropriate mathematical apparatus of Machine Learning and Artificial Intelligence in SES-related cases
- to adapt the approaches of Machine Learning and Artificial Intelligence to existing problem formulation in the sphere of SES and various data sources
- to design the structural and functional models of data analytics and expert systems in the sphere of Smart Energy Systems
- to set up up-to-date software packages for Machine Learning and Artificial Intelligence related problems and to organize effective data management
- to be able to identify own role within a multidisciplinary team and explain the roles of the other team members; and be able to act both independently with a little supervision and cooperate with other team members; negotiate and manage conflicts.
- to be able to demonstrate high personal drive, result oriented and service minded work style, as well as the abilities of time and workload management, to act responsibly and account the interests of the larger community in mind; to work in a fast-paced and highly dynamic environment.
TEACHING TECHNIQUES
- Case method
- EduScrum
- Inquiry-based learning
- Design thinking approach
- Simulation-based learning
- Bite-sized learning
- Experiential learning
COURSE OUTLINE
Part I. Introduction: Artificial intelligence and Machine learning
- Artificial intelligence and intelligent data analytics systems
- Introduction to artificial intelligence
- Intelligent data analytics systems: classification and application in power industry
- Data analytics system development: step-by-step guide
- Artificial intelligence methods
- Models of knowledge representation
- Classification of artificial intelligence methods
- Quality metrics of machine learning models
- Artificial intelligence application in power industry: practical cases
- Introduction to machine learning
- Patterns and features: basics of machine learning
- Fitting, underfitting and overfitting: training sample formulation principles
- Supervised learning: Decision trees, regression functions, support vector machine
- Unsupervised learning: Clusterization, k-means approach
- Machine learning application in power industry: practical cases
Part II. Basic methods
- Fuzzy sets theory
- Fuzzy set and fuzzy operators
- Linguistic variables. Fuzzy inference systems
- Fuzzy inference application in power industry
- Artificial neural networks
- Introduction to artificial neural networks
- Types of artificial neural networks
- Artificial neural networks learning methods
- Practical examples of artificial neural networks implementation in power industry
- Bio-inspired approaches
- Genetic algorithms, historical review
- Classification of bio-inspired approaches
- Genetic algorithms basics
- Basic genetic operators
- Main issues of genetic optimization practice
- Practical examples of genetic algorithms application in power industry
- Decision trees and decision trees ensembles
- Pros and cons of decision trees application
- Entropy and information gain as mains characteristics of decision trees
Part III. Decision support systems
- Data mining
- Data mining goals and tasks
- Classification of Data mining approaches
- Data mining practical case studies
- Decision support systems
- Taxonomies, components, classification, development frameworks
Laboratory works
- Neural networks
- Handwritten figures recognition
- Neural network training for equipment technical state recognition
- Genetic algorithms
- Searching extremum of nonlinear function
- Power system loading curve optimization
- Regression models
- Solar power plant output linear regression model training
- Solar power plant output Random forest regression model training
- Fuzzy sets
- Resource consumption fuzzy controller
- Fuzzy system for selecting the primary circuit of the substation
Course project
The course project topics are dealing with development of decision support system model, covering, but not limited to the following Smart Energy Systems topics:
- distributed generation (DG) integration;
- power equipment health-index identification;
- renewable energy sources forecasting.
In the course project the students are provided with problem statement and necessary data. During the course project solution the mathematical model of decision support system should be designed and tested based on the initial data, provided by the course instructor.
The list of methods to be used by students:
- neural networks;
- genetic algorithms;
- decision trees.
LIST OF REQUIRED LABORATORY EQUIPMENT
1. Software, providing basic artificial intelligence and machine learning algorithms, libraries, extension modules, packages, interfaces: MathWorks Matlab (v. 2015b or higher), including: Genetic Algorithm and Direct Search Toolbox; Fuzzy Logic Toolbox; Statistics and Machine Learning Toolbox; Curve Fitting Toolbox.
2. Computer hardware, meeting corresponding software requirements (the number of computers varies depending on the academic group size assuming a maximum of 2 students per 1 computer)
