The Role of Artificial Intelligence in the Transition from Conventional Power Systems to Modernized Smart Grids
By Soheil Mohseni, Alan C. Brent, and Daniel Burmester
Over the past three decades, research in artificial intelligence (AI) has advanced a wide range of techniques and approaches that can be adapted or employed to solve complex electric power system problems (e.g. power system planning, operation, as well as transient stability and control), previously thought to be unsolvable without making several simplified assumptions. On the other hand, the “smart grid” concept emerged in the early 21st century to take advantage of improvements in information and communication technologies in the electricity industry, in order to address security issues in the centralized power systems, as well as to provide for increasing the penetration of renewable energy by deploying the advanced metering infrastructure to establish a win-win situation for both the electricity consumers and suppliers. This has raised problems in diverse fields of power system analysis and design, which are not amenable to analytical treatment, and need to be solved using metaheuristic/learning algorithms within the field of AI.
Although the application of AI methods in some areas of power system analysis is still tricky (due to their computational expensiveness or even computational intractability), a broad range of AI techniques have successfully been employed, aimed at smartening the whole grid (covering the areas of electricity generation, transmission, distribution, consumption, as well as the deregulation of the electricity market), which can be categorized into the following three main classes:
- optimization of renewable and sustainable energy systems (RSES) – microgrids, energy hubs, virtual power plants, load aggregators, and so forth – in the three levels of transient stability control, day-ahead energy management, and investment planning,
- short- and long-term forecasting of the energy demand, energy prices, and the output powers from weather-driven sources, and
- intelligent state estimation and fault recognition with self- restoration/healing properties to improve the resilience of the systems.
Optimization of RSES:
Proper optimization is of pivotal importance to achieve efficient, yet cost-effective schemes for designing, operating, and controlling the RSES. In this respect, mathematical optimization techniques have traditionally been employed to provide analytical solutions to optimize classic, uncomplicated hybrid renewable energy systems. However, the escalation in the number of decision variables and constraints, as well as the increased nonlinearity and non-convexity of the RSES optimization problems, especially in the case of multi-vector energy networks, made the exact methods incapable of solving such problems without a number of simplifying assumptions that significantly reduce the solution accuracy. Fortunately, the advent of meta-/hyper-heuristic optimization approaches has substantially resolved this issue. Although such stochastic techniques do not secure the global optimality of the solution obtained, their superior efficacy over the accurate methods in the research field of RSES optimization is demonstrated both empirically and conceptually. In this light, a variety of AI-based optimization algorithms have been put to test for the whole-life cost minimization, daily profit maximization, as well as the frequency and/or voltage deviation minimization in three respective levels of equipment capacity planning, operation scheduling, and dynamic control of RSES. Among the investigated algorithms, the following ones have been widely demonstrated to work well on RSES optimization problems:
- the particle swarm optimization,
- the genetic algorithm,
- the ant colony optimization,
- the simulated annealing,
- the evolutionary strategy,
- and the artificial bee colony algorithm.
Prediction of Input Data:
High-precision forecasting of the input variables to the energy system models can lay the foundation for enhancing these models to better approximate the real-world situations. In this context, an assortment of techniques have been developed that advance the quality of forecasts as inputs to numeric models of the energy systems. Depending on the temporal nature of the problem (i.e. the prediction horizon), the forecast analysis can be carried out within four projected time horizons:
- ultra-short-term (seconds-ahead) for dynamic performance and stability assessment applications,
- very short-term (ranging from minutes-ahead to hour-ahead) for energy security analyses, enabling the preventive control of the system,
- short-term (ranging from hour-ahead to 72-hours-ahead) for the economic dispatch, unit commitment, short-term maintenance scheduling, and energy market bidding applications, and
- mid-long- to long-term (up to several years-ahead) for the equipment maintenance planning, energy policy modelling, expansion planning of generation resources/transmission networks, and start-from-scratch energy system planning applications.
The prediction modelling approaches, developed within the context of AI, to estimate the input variables to energy networks with high accuracy, based on the historical data, can be classified into the following main categories:
- artificial neural networks,
- time-series models (e.g. the auto-regression model, the autoregressive (integrated) moving average model,
- the Markov switching model, the triple exponential smoothing technique, persistence models, etc.),
- deep learning methods,
- support vector regression approaches,
- Bayesian estimation algorithms,
- kernel-based machine learning techniques,
- fuzzy prediction interval models, and
- quantile regression analyses.
Power System Protection:
Although the wavelet transform is traditionally adopted to diagnose the locations and types of faults that occur in power systems, the development of big data analytics and AI derived new trends in this research area. In this context, a range of pattern recognition-based data mining techniques have been successfully applied to detect power system faults by clustering the characteristics of each scenario, such as: the k-means algorithm, the k-medoids method, the fuzzy c-means algorithm, the k-nearest neighbours algorithm, and Naïve Bayesian classifiers. At the same time, AI’s potential is exploited to enable the self-healing of power systems under critical contingencies, which has significantly contributed to facilitating the implementation of automated, immediate corrective or reinforcing actions, thereby enhancing the robustness of the systems. In this regard, the multi-agent system concept – i.e. a distributed AI technique combining several collaborative agents in order to perform assigned tasks in accordance with the overall goal(s) of the system – is the most widely utilized strategy.
A number of efforts have also been undertaken to leverage the following AI approaches to provide cutting-edge tools for the self-restoration of power grids:
- knowledge-based expert systems,
- decision support systems,
- reinforcement learning algorithms, and
- Pareto-based multi-objective optimization approaches.
As a future research direction, the effective application of such tools on the situational awareness information obtained from the intelligent wide area monitoring systems or Internet of Things (IoT) systems can be explored to allow for the realization of self-healing objectives in bulk power systems under the smart grid paradigm in the foreseeable future.
In conclusion, the AI methods that can be either adapted or utilized directly to solve intricate electric power system problems, have played, and will continue to play, a significant role in revolutionizing the entire energy sector from generation to distribution, in order to enable the large-scale integration of renewable sources into energy networks.
Soheil Mohseni focuses his research on the optimal equipment capacity planning and design of off-/on-grid renewable energy systems using meta-heuristic optimization algorithms. His research interests include power system operation and planning, smart energy hubs, smart grids, and sustainable energy systems. Soheil holds a BSc in electrical power engineering from Kermanshah University of Technology, Iran. He also holds an MSc with high distinction in electrical power engineering, and power systems from University of Guilan, Iran. He is currently pursuing a PhD at Victoria University of Wellington, in New Zealand at the Chair in Sustainable Energy Systems.
Alan C. Brent is the inaugural holder of the Chair in Sustainable Energy Systems at Victoria University of Wellington, in New Zealand, since 2017. His research revolves around sustainable technology management, with an emphasis on the energy sector. Before joining Victoria University, he was a professor of engineering management and sustainable systems in the Department of Industrial Engineering, and the associate director of the Centre for Renewable and Sustainable Energy Studies, at Stellenbosch University in South Africa. In 2017, he was appointed as an extraordinary professor in the Department of Industrial Engineering at Stellenbosch University. Until 2017 he was also appointed as a part-time professor of sustainable life cycle management in the Graduate School of Technology Management, at the University of Pretoria in South Africa. He holds bachelor degrees in engineering (chemical) and philosophy (sustainable development); master degrees in science (environmental engineering), engineering (technology management), and philosophy (sustainable development); and a PhD in engineering management.
Daniel Burmester currently works at the School of Engineering and Computer Science, Victoria University of Wellington, New Zealand. His research is focused around control within DC nanogrids and nanogrid networks, creating a microgrid.
To have the eNewsletter delivered monthly to your inbox, join the IEEE Smart Grid Community.
IEEE Smart Grid Newsletter Editors
To view archived articles, and issues, which deliver rich insight into the forces shaping the future of the smart grid, please visit the IEEE Smart Grid Resource Center.