Machine Learning and AI for Sustainable Power and Energy Systems

The Machine Learning and AI for Sustainable Power and Energy Systems event will focus on the employment of artificial intelligence technologies for power system operation in a competitive environment, integrating renewable energy and storage systems into the power grid, electric vehicles/low-carbon transportation, and the other hot topics of interest in modern power systems.

This event is part of ISAP, a well-established premier forum for exchanging ideas and experiences of academics, industry, and students to discuss intelligent system applications in operation, control, planning, and maintenance of power systems and their particular components.

http://www.isap-power.org/

http://www.isap-power.org/online-event-2022/

Registration & Participation

The registration is free but needed.

Connection link: https://videoconf-colibri.zoom.us/j/99180212342?pwd=OC9YWnBoVno1cTQramluYkRvR2xldz09

Program

Day 1 – 13th October

• 14h00 (20 min) – Opening session | Zita Vale, Polytechnic Institute of Porto, Portugal
• 14h20 (40 min) – Dynamic Security Constrained Unit Commitment for islands with High Renewable Energy Penetration | Nikos Hatziargyriou, National Technical University of Athens, Greece
• 15h00 (20 min) – Distributed Generation control for grid ancillary services using innovative IoT techniques for large-scale implementation and ripple signaling | Dimitrios Tsiamitros, University of West Macedonia, Greece
• 15h20 (20 min) – Heuristic Approaches For Intelligent Optimization of Power Distribution Systems | Oguzhan Ceylan, Marmara University, Turkey
• 15h40 (20 min) – Recurrent artificial neural networks for load forecasting | Samuele Grillo, Politecnico di Milano, Italy
• 16h00 (20 min) – Multi-agent systems and their applications | Jing Xie, Pacific Northwest National Laboratory, United Stated of America
• 16h20 (20 min) – Discussion and break
• 16h30 (10 min) – Closing session day 1 | Zita Vale, Polytechnic Institute of Porto, Portugal
• 16h40 (60 min) – Special session: F1000 research | Jack Brook, F1000Research; and Zita Vale, Polytechnic Institute of Porto

Day 2 – 14th October

• 14h00 (10 min) – Opening session day 2 | Zita Vale, Polytechnic Institute of Porto, Portugal
• 14h10 (60 min) – Students Meeting | Students
• 14h10 (60 min) – [private session] Private session: ISAP BoD Council members | ISAP Council
• 15h10 (20 min) – Photovoltaic power forecasting using Iterative network pruning technique for renewable-based microgrid | Yutaka Sasaki, Hiroshima University, Japan
• 15h30 (20 min) – Decision support in maintenance for electrical components of distribution networks | Hirotaka Takano, Gifu University, Japan
• 15h50 (20 min) – Distributed Artificial Intelligence | Kumar Venayagamoorthy , University of Clemson, United States of America
• 16h10 (20 min) – Marvels of the intelligent power systems | Ioana Pisica, Brunel University, United Kingdom
• 16h30 (20 min) – Computational intelligence applications towards a more sustainable smart power system | João Soares, Polytechnic Institute of Porto, Portugal
• 16h50 (10 min) – Discussion
• 17h00 (30 min) – Doctoral consortium | PhD and MSc Students
• 17h30 (10 min) – Closing session | Zita Vale, Polytechnic Institute of Porto, Portugal

Senior Organizing Committee

• Zita Vale
• João Soares
• Gabriel Santos
• Luis Gomes

Student Organizing Committee

• Akshay Jain
• Cátia Silva
• Brigida Teixeira
• Daniel Ramos

Speakers and talks

Oguzhan Ceylan (Marmara University, Turkey)

Title: Heuristic Approaches For Intelligent Optimization of Power Distribution Systems

Abstract: This talk will be on the implementation of intelligent methods in the operation of power distribution systems. It is known that with the increased integration of photovoltaics, the variable power output has led to increased voltage fluctuations and violation of operating limits. An optimization model based on an intelligent optimization method (grey wolf optimization) will be given. Numerical simulation results of several different test cases (modified 33 and 69 bus test systems) by using tap-changing transformers, capacitors, and PV solar panels will be provided. Then the results of the comparisons to other intelligent methods such as differential evolution and harmony search method will be given.

Samuele Grillo (Politecnico di Milano, Italy)

Title: Recurrent artificial neural networks for load forecasting

Abstract: Load forecasting has always played a key role in power systems operation and planning. This has become even more relevant nowadays, as the presence of renewable sources is increasing, and the reliability of power systems is potentially jeopardized. This talk will briefly outline the main advanced techniques, based on artificial intelligence, used for this task, and will focus on a particular structure of neural networks (NNs), the long short-term memory (LSTM) NNs. An application of this technique to forecasting the total load of a building will be also presented.

Nikos Hatziargyriou (National Technical University of Athens, Greece)

Title: TBA

Abstract: TBA

Ioana Pisica (Brunel University, United Kingdom)

Title: Marvels of the intelligent power systems

Abstract: The talk will delve into the combination of communications, information systems, computer intelligence and power systems that led to marvellous applications in electrical engineering. Examples of applications will be presented alongside their importance and relevance to current operation of power systems.

Yutaka Sasaki (Hiroshima University, Japan)

Title: Photovoltaic Power Forecasting using Iterative Network Pruning Technique for Renewable-based Microgrid

Abstract: The author proposes a simple and efficient day-ahead solar power prediction method by improving our previous feedforward neural networks (FNNs) based approach, which is applied to a renewable-base microgrid (MG). The proposed prediction algorithm uses only public available weather data such as day-ahead solar irradiation. Based on weather clustering such as sunny/cloudy/rainy, the multiple FNNs are developed for different weather patterns and time zones. An iterative pruning (IP) algorithm is introduced in the previous FNNs structure to achieve efficient computation. Confidence intervals (CIs) indicating the reliability of the forecasting are displayed in the forecasting result. The values of the CIs are effectively provided for the local energy management and robust power system security in MG. The computation times are improved about 90% against the previously proposed method. At the same time, the prediction accuracy is maintained against previous methods. The rate at which the measured value appears outside the confidence interval is 10% or less.

Joao Soares (Polytechnic Institute of Porto, Portugal)

Title: Computational intelligence applications towards a more sustainable smart power system

Abstract: In this talk, Joao Soares will go through some of the complex issues with the current paradigm of power systems. In reality, standard mathematical methods often simplify the formulations to make optimization problems more manageable but to the detriment of accuracy and better solutions. The application of classical tools is restricted when the models are near to real-world scenarios because of problems with scalability, execution time, memory computation needs, or even uncertainty. Computational intelligence (CI), a group of problem-solving approaches that aim to mimic the intelligence found in nature, has been effectively used to tackle challenging issues in the energy sector. The talk will examines certain CI methodologies, such as Evolutionary Computation (EC), for tackling difficult power system problems while taking into account the potential of these strategies and possible CI research fields in the energy industry. A quick glimpse of some of the tools that GECAD has been working on will be overviewed. Our platform offers a variety of EC optimization algorithms for energy dispatching and vehicle-to-grid/smart charging. These algorithms and platforms are publicly accessible and have been launched in algorithm competitions hosted globally. Further information is available at http://www.gecad.isep.ipp.pt/ERM-Competitions and http://www.gecad.isep.ipp.pt/Meta-ERM.

Jing Xie (Pacific Northwest National Laboratory, United States of America)

Title: Multi-Agent Systems and Their Applications

Abstract: The number of distributed energy components and devices continues to increase globally. As a result, distributed control schemes are desirable for managing and utilizing these devices. In recent years, agent-based technology becomes a powerful tool for engineering applications. As a computational paradigm, multi-agent systems (MASs) provide a good solution for distributed control. In this presentation, MASs and their applications are discussed, covering the system architecture, consensus algorithm, and multiagent platform, framework, and simulator. The distributed consensus part will be highlighted with its applications on load shedding and networked microgrids.

Hirotaka Takano (Gifu University, Japan)

Title: Decision Support in Maintenance for Electrical Components of Distribution Networks

Abstract: To maintain a reliable and stable power supply, operators of electrical power distribution networks regularly inspect electrical components of the distribution networks. Inspection results and necessary measures corresponding to them have been stored electronically, and there is lively discussion how to utilize these electronic records for operations and planning of the distribution networks. We propose a decision support method that utilizes the accumulated records to predict the necessity of replacing electrical components installed in a distribution network. In the proposed method, decision tree learning analyzes the inspection results of the electrical components that have already been replaced and identifies the criteria by which the operators decided to replace them. The resulting tree-like model, which is the decision tree, analyzes the newly input inspection result and determines if the target component needs to be replaced. Using actual inspection and maintenance records, numerical simulations were carried out. In the numerical simulations, our proposal showed reasonable prediction accuracy despite applying a simple decision tree learning.

Dimitrios Tsiamitros (University of West Macedonia, Greece)

Title: Distributed Generation control for grid ancillary services using innovative IoT techniques for large-scale implementation and ripple signaling

Abstract: All European Union (EU) countries have the commitment to increase the share or Renewable Energy Sources (RES), which however are fluctuating and uncertain. Therefore, in many EU countries, the laws allow premium access to the grid for RES. However, there are some well known grid power quality indices, which limit the capacity of RES DGs on any medium voltage (MV) grid line: (i) The first has to do with the thermal capacity of the lines (conductors, overhead lines or underground cables). (ii) Another restriction has to do with the lower and the upper limits of the grid’s voltage. (iii) There are also limits of the voltage variation through a complete year of operation. Moreover, many Distribution Network Operators (DNOs) had invested to cross grid lines (interconnected lines) between two main lines and had also invested to remote controlled circuit breakers at the edges, towards their effort not to de-energize significant parts of a grid line or reconfigure part of the grid, during repair or maintenance works, and finally to comply with the SAIDI and SAIFI indices The main objective of this presentation is to show cost-effective solutions that utilize innovative IoT techniques or/and the ripple signaling system of the DNOs, in order to execute operations on the distribution network without worrying about violating the grid parameters and reliability indices, both of the consumers and of the producers. Due to large amounts of DGs connected lately to the LV and MV network, some parameters of the grid can’t be measured accurately, and many load transfer operations are prohibited due to thermal capability or to voltage values. With the help of innovative IoT techniques and ripple control systems, we can control the operation of the DGs. The solution involves already known systems, low cost intervention, and legal compatibility. The simulation results are absolutely encouraging: The voltage level of every node of the MV network is below the limit of 22 kVs, which is not the case without applying these techniques, especially at distant nodes from the substations. Nowadays, embedded IoT devices, acquire vast amounts of data from sensors and metering devices. This acquired data, apart from the need to be stored for further analysis, often needs to be analyzed in real-time, as acquired. More often, this real-time analysis, involves the need of real-time communication (a form of message exchange) between the several IoT devices (consumers or clients) and/or a central server (broker) that make up the system. In small scale networks, this is often accomplished through standard TCP, UDP and websockets communication layer, which are “heavy” implementations and become slower while the system network grows bigger. To over-come this, Asynchronous Message Queuing Protocols (AMQP) are used. The message queue is a form of asynchronous service-to-service communication where the messages are stored on the queue until they are processed by the consumer/broker. Although several commercial and open-source AMQP implementations are available, most of them are predefined client/server implementations, requiring a central communication node (broker) to process and route the messages. These AMQPs make use of the low-level socket networking layer for low latency communication, but in many cases, they lack the ability to be multiplatform and multiarchitecture implemented, support limited message types and of course need a central broker (no support for direct P-2-P communication between the consumers, while adding extra latency, double the network band-width used and adding a Single-Point-of-Failure; if the broker “crashes” the whole network be-comes unavailable). In this presentation we depict the development of a brokerles communication stack on top of a low-level message transport socket library, supporting asynchronous I/O lock-free message passing and automatic reconnections for dynamic modules, providing real-time communication from small scale to large scale networks (8M msg/sec, 3μsec latency).

G Kumar Venayagamoorthy (University of Clemson, United States of America)

Title: Distributed Artificial Intelligence

Abstract: The modern electric power and energy system, referred to as the ‘smart grid’ is complex and one that is expected to be conscious, intelligent, distributed, and flexible. Such an electric power system architecture can facilitate secure and distributed flow of power from renewable energy sources including solar and wind. Furthermore, it can handle flexible loads and energy storage including electric vehicles. This talk will address the potentials and promises of distributed Artificial Intelligence (Distributed AI) for smart grid operations and control. AI has evolved over the last 40+ years to transform operations and control of complex systems. A smart grid with variable power and energy sources, bidirectional power flows, and uncertainty in forecasting and real- time availability of generation, loads, energy storage and other operational resources requires distributed intelligence and computational technologies for its operation and control. Examples of such technologies including Distributed AI for stable, secure, reliable, and efficient operations and control of smart grid will be presented.