Artificial Intelligence for Smarter Power Systems: Fuzzy logic and neural networks
The urgent need to reduce carbon emissions is leading to growing use of renewable electricity, particularly from wind and photovoltaics. However, the intermittent nature of these power sources presents challenges to power systems, which need to ensure high and consistent power quality. Going forward, power systems also need to be able to respond to changes in loads, for example from EV charging. Neither production nor load changes can be predicted precisely, and so there is a degree of uncertainty or fuzziness. One way to meet these challenges is to use a kind of artificial intelligence - fuzzy logic. Fuzzy logic uses variables that may be any real number between 0 and 1, rather than either 0 or 1. It has obvious advantages when used for optimization of alternative and renewable energy systems. The parametric fuzzy algorithm is inherently adaptive because the coefficients can be altered to accommodate requirements and data availability. This book focuses on the use of fuzzy logic and neural networks to control power grids and adapt them to changing requirements. Chapters cover fuzzy inference, fuzzy logic-based control, feedback and feedforward neural networks, competitive and associate neural networks, and applications of fuzzy logic, deep learning and big data in power electronics and systems.
Inspec keywords: artificial intelligence; smart power grids; fuzzy logic; power engineering computing; neural nets
Other keywords: feedforward neural networks; competitive neural networks; future power systems; fuzzy inference; artificial intelligence; feedback neural networks; big data applications; deep learning; fuzzy sets; fuzzy logic; smart grids; rule based approaches; power electronics; fuzzy-logic-based control; relational approaches; associative neural networks; real-time simulation applications
Subjects: Neural nets; General electrical engineering topics; Power systems; General and management topics; Power engineering computing; Formal logic
- Book DOI: 10.1049/PBPO161E
- Chapter DOI: 10.1049/PBPO161E
- ISBN: 9781839530005
- e-ISBN: 9781839530012
- Page count: 274
- Format: PDF
-
Front Matter
- + Show details - Hide details
-
p.
(1)
-
1 Introduction
- + Show details - Hide details
-
p.
1
–8
(8)
The Introduction to this book begins with a discussion of how renewable-energy-based generation is shaping the future of power systems. It goes on to discuss how power electronics, artificial intelligence and simulations will allow smarter power systems and the optimal operation of renewable systems. Finally the chapter looks at the control of microgrid bidirectional power flow.
-
2 Real-time simulation applications for future power systems and smart grids
- + Show details - Hide details
-
p.
9
–64
(56)
Real-time simulation has been used in various industries for decades now, but it is particularly used in electrical transmission systems. These dynamic simulation tools can be divided into two main categories: (i) transient stability (TS) tools based on fundamental frequency phasors and (ii) electromagnetic transient (EMT) tools allowing simulation of fast transient and higher frequency phenomena.
-
3 Fuzzy sets
- + Show details - Hide details
-
p.
65
–79
(15)
During the twentieth century, many attempts were made for augmenting the intelligence of computer software with further capabilities. Adaptive learning algorithms were developed, making possible the initial developments in neural networks (NNs) in the 1950s. A very innovative learning approach was birthed by L. Zadeh in 1965 with the publication of his paper “Fuzzy Sets.” In that paper, the idea of a membership function based on multivalued logic, allowed a solid theory where technology bundled together thinking, vagueness, and imprecision. An engineering design starts from the process of thinking, i.e., a mental creation, and designers will use their linguistic formulation, with their own analysis and logical statements about their ideas. Then, vagueness and imprecision are considered as empirical knowledge to be incorporate in the model implementation of the system. Scientists and engineers try to remove most of the vagueness and imprecision of the world by making precise mathematical formulation of laws of physics, chemistry, and the nature in general. Sometimes, it is possible to have precise mathematical models, with strong constraints on nonidealities, parameter variation, and nonlinear behavior. However, if the system becomes more complex, the lack of ability to measure or to evaluate features, with a lack of definition of precise modeling, in addition to many other uncertainties and incorporation of human expertise, makes it almost impossible to explore such a very precise model for a complex real-life system. Fuzzy logic (FL) and NNs became the foundation for the newly advanced twenty-first century of smart control, smart modeling, intelligent behavior, and artificial intelligence (AI). This chapter discusses the basics and foundations of FL and NNs, with some applications in the area of energy systems, power electronics, power systems, and power quality.
-
4 Fuzzy inference: rule based and relational approaches
- + Show details - Hide details
-
p.
81
–97
(17)
A fuzzy modeling system or a fuzzy controller can be implemented by understanding sampled valued variables, translated from the real-life world into a fuzzy domain. The model or the controller will estimate an output, either for a decision-making action or to impose a set-point in a closed-loop control system. Those processes are defined as “fuzzification,” i.e., going through the membership functions of the input data into fuzzy sets and their corresponding pertinence degrees to those sets, through a “fuzzy processing” or also called inference engine; and “defuzzification,” i.e., it is a transformation from such a fuzzy evaluation into crispy variables, ready for control, or modeling analysis (Yen, 1999). Such a methodology could be compared to have a phasor domain analysis, or a Laplace domain analysis. Fuzzy data processing requires to be done in the fuzzy domain with proper direct and inverse transformations.
-
5 Fuzzy-logic-based control
- + Show details - Hide details
-
p.
99
–115
(17)
We consider a simple fuzzy logic control methodology that is applicable for most systems. It was proven and utilized for the first time in renewable energy systems and wind turbine control systems. We also look at fuzzy control preliminaries, fuzzy controller heuristics, fuzzy logic controller design, and industrial fuzzy control supervision and scheduling of conventional controllers.
-
6 Feedforward neural networks
- + Show details - Hide details
-
p.
117
–138
(22)
The chapter considers aspects of feedforward neural networks including backpropagation algorithm, binary classifier; artificial neural network architecture, neuron activation transfer functions, data processing, and neural network based computing.
-
7 Feedback, competitive, and associative neural networks
- + Show details - Hide details
-
p.
139
–160
(22)
The chapter discusses feedback, competitive and associative neural networks. It specifically considers linear vector quantization networks, counterpropagation networks; probabilistic neural networks and industrial applicability of artificial neural networks.
-
8 Applications of fuzzy logic and neural networks in power electronics and power systems
- + Show details - Hide details
-
p.
161
–190
(30)
There are typically three frameworks with some generalization of functionalities, i.e., three paradigms, that can be used for energy conversion systems, with artificial-intelligence-based computation: a function approximation or input/ output mapping, a negative feedback control, and a system optimization. The first one is the construction of a model, using either heuristic or numerical data, the second one is the comparison of a set point with an output that can be either measured or estimated with a function that minimizes the error of the set point with the output, and the third one is a search for parameters and system conditions that will maximize or minimize a given function. Fuzzy logic and neural network techniques make the implementation of such three paradigms possible, robust, and reliable in practical cases. The integration of modern power electronics, power systems, communications, information, and cyber technologies with a high penetration of renewable energy resources has been at the edge and at the frontier for the design and implementation of smart-grid technology. The emergence of AI techniques in past industrial applications is allowing smart-grid technology to be an interdisciplinary field with multiple dimensions of complexity. This chapter will present some background and established applications of AI in power electronics, power systems, and renewable energy systems.
-
9 Deep learning and big data applications in electrical power systems
- + Show details - Hide details
-
p.
191
–227
(37)
Power systems are massive and complex electrical engineering systems. The power system analysis and decision-making has been dependent only on physical modeling, numerical calculations, and some statistical inferences. Contemporary smart grids have bidirectional power flow, and uncertainty on the random nature of renewable energy availability, also a geographical dispersion of mobile loads, with a partial observability of power quality issues. A new generation of power-electronics-enabled power systems hardware, electrical circuits instrumentation, communications, intelligent control, and real-time performance is shaping the present and future development of smart-grids. Engineers must develop the technology for smarter power systems in order to build smart-grids, and big data applications are a requirement for such modernization. The analysis of transmission and distribution has been traditionally conducted as completely decoupled infrastructures, in which the design engineer will select a section and apply a model. Integrated co-simulation and analysis on both transmission and distribution sides of the grid can be conducted using machine learning (ML) and other AI methods. Aggregation techniques can be used with new instantaneous power theories and advanced signal processing. Deep learning models can be used in updating generator and load setpoints, based on the load forecasts, as well as incorporating online estimation algorithms about weather, storage, net-metering, possibilities of natural disasters, approximate generation schedules over the next hour, and information about transmission line faults that may severely disrupt the currents flowing on the distribution side. Load control algorithms can be tuned whether the distribution grid depends on the transmission grid or on dispersed generation.
-
Back Matter
- + Show details - Hide details
-
p.
(1)