ILUSTRE

There are currently no vacancies.

If you’d like to join our mailing list, please fill out the form, and we will be happy to keep you informed of all the latest developments regarding recruitment for this lab.

ILUSTRE offers opportunities for PhD candidates who wish to support the acceleration of the energy transition and who have an affinity with the Dutch Caribbean islands. The lab will house five PhD researchers who will work simultaneously on different aspects of the overall applied aim to create accurate and reliable AI solutions to improve the efficiency and effectiveness of water and energy distribution.

Curaçao has a modern National Energy Policy, which sets the objectives and priorities for the development of an effective and sustainable energy system with the goal to achieve 50% renewable energy penetration by 2030. Data technology will be an enabler for this objective which aligns with Curaçao’s ambition to become a smart country that embraces the development of data technology and Artificial Intelligence. Curaçao, therefore, welcomes this initiative of Jheronimus Academy of Data Science (JADS) and LaNubia Consulting.

The Five PhD researchers will be located at JADS and at the utility companies to preserve the academic environment and the transfer of domain-specific expertise, respectively. It is foreseen that the PhD researchers will reside at the utility companies at least once per year for an extended period of time to test their algorithms in real-life settings. To further the dissemination of the acquired knowledge and skills, we foresee short visits to Dutch (e.g. Alliander), European (e.g. E.ON), and other Caribbean (e.g. VEi) utility companies and consortia.

Topic 1

Artificial Intelligence for decision support in water desalination, recycling, and purification

Scientific Challenge

For over 30 years, AI and computational intelligence have been used in the water desalination domain [1] for applications like support in decision-making, prediction, optimization, and control with respect to alarm processing, fault detection, load forecasting, and security assessment. The recent trend to use renewable energy sources for desalination and wastewater treatment makes the decision process more complicated, due to the temporal variability of these sources [2,3,4]. AI systems that go beyond the current point solutions are needed to deal with this complexity while considering a system-wide view and a balanced interaction between the AI systems and human experts. The scientific challenge of this PhD project is to develop hybrid AI solutions that combine advanced prediction methods (e.g. deep learning algorithms), multi-objective optimization, and adaptive models with explainable AI decision-making methods from computational intelligence to realize such a balance.

Methodology

We will use attention-based deep learning techniques for predictive modeling and time series analysis (e.g. [5]) in combination with explainable/transparent deep models (e.g. [6]). The methods will be adaptive to deal with changing conditions and parameter drift (e.g. [7]). For addressing challenging dynamic optimization problems considering multiple perspectives, we will use multi-objective evolutionary computation extended with methods for dealing with soft constraints in such optimization challenges [8]. Probabilistic fuzzy systems [9] will be used for fusing information from multiple sources in a way consistent with models reflecting the experts’ knowledge and preferences. The models will be combined with data-driven process analytics [10] to provide solutions at the business process management level. From this basis, we will develop hybrid approaches that are tailored to the challenges of the desalination, recycling, and purification domain.

References

[1] He, Q., Zheng, H., Ma, X., Wang, L., Kong, H., & Zhu Z. (2022). Artificial intelligence application in a renewable energy-driven desalination system: a critical review. Energy and AI, vol. 7, 100123.
[2] Cabrera, P. & Carta, J.A. (2019). Computational Intelligence in the Desalination Industry. In: Blondin, M., Pardalos, P., Sanchis Sáez, J. (eds). Computational Intelligence and Optimization Methods for Control Engineering. Springer Optimization and Its Applications, vol. 150, Springer, Cham.
[3] Harrou, F., Dairi, A., Sun, Y. & Senouchi, M. (2018). Statistical monitoring of a wastewater treatment plant: a case study. Journal of Environmental Management, vol. 223, 807 – 814.
[4] Cheng, T., Harrou, F., Kadri, F., Sun, Y. & Leiknes, T. (2020). Forecasting of wastewater treatment plant key features using deep learning-based models: a case study. IEEE Access, vol. 8, 2020.
[5] Chen, P., Dong, W., Wang, J., Lu, X., Kaymak, U., & Huang, Z. (2020). Interpretable clinical prediction via attention-based neural network. BMC Medical Informatics and Decision Making, 20 (suppl. 3).
[6] Gegov, A., Kaymak, U., & da Costa Sousa, J. M. (2020). Guest Editorial: Deep Fuzzy Models. IEEE Transactions on Fuzzy Systems, 28(7), 1191-1194.
[7] da Costa, P. R., Rhuggenaath, J., Zhang, Y., Akcay, A., & Kaymak, U. (2021). Learning 2-Opt Heuristics for Routing Problems via Deep Reinforcement Learning. SN Computer Science, 2.
[8] Kaymak, U., & Costa Sousa, da, J. M. (2003). Weighted constraint aggregation in fuzzy optimization. Constraints, 8(1), 61-78.
[9] van den Berg, J., Kaymak, U., & Almeida, R. J. (2013). Conditional density estimation using probabilistic fuzzy systems. IEEE Transactions on Fuzzy Systems, 21(5), 869-882.
[10] Van Zelst, S.J., Mannhardt, F., de Leoni, M. & Koschmider, A. (2021). Event abstraction in process mining: literature review and taxonomy. Granular Computing, 6, 719–736.

Topic 2

Artificial Intelligence for power load and renewable energy forecasting in electricity grids

Scientific challenge

Short-term forecasts of (a) power load and (b) renewable energy supply, are crucial for decarbonizing electricity grids: without these forecasts, high-carbon baseload generators must be kept running. The scientific challenge is to achieve accurate and reliable forecasts, in the face of changeable energy demand patterns and external covariates (weather, public events, etc). Deep learning has been shown to perform very well in power-load forecasting and achieves promising results in renewable-energy forecasting (Wang et al., 2019). This PhD plan sets out to develop deep learning algorithms that realize forecasts that are both accurate and reliable, with the flexibility to adapt to local conditions.

Methodology

Time-series forecasting has a rich history: deep learning and Gaussian process regression are two modern paradigms that have shown strong performance, including for energy forecasting (Wang et al. 2019, [2]). We will build on both these paradigms, as well as on promising recent “neural process” models that may offer the best of both worlds. Power load(demand) and renewable energy supply can each be treated within the same framework, but each has different patterns and external covariates. External covariates can have rich features including graph-structured information to improve the accuracy of our predictions and account for power network efficiency losses [3]. We will explore the accuracy and reliability of the described methods for the two forecasting tasks, while iterating with the involved partners Aqualectra, WEB Bonaire, WEB Aruba, and GEBE Sint-Maarten to ensure that what is produced is relevant to deployment needs in Curacao and more broadly. At JADS we have the required expertise and deep learning and probabilistic models [4]. We also have experience applying such methods to the monitoring of critical assets and renewable energy forecasts [2,5,6].

References

[2] al-Lawati, Y., Kelly, J., & Stowell, D. (2020). Short-term prediction of photovoltaic power generation using Gaussian process regression. NeurIPS 2020 Workshop on Tackling Climate Change with Machine Learning.
[3] Güven, Ç., Seipel, D., & Atzmueller, M. (2020). Applying ASP for Knowledge-Based Link Prediction with Explanation Generation in Feature Rich Networks. IEEE Transactions on Network Science and Engineering.
[4] Wilkinson, W.J., Riis Andersen, M., Reiss, J.D., Stowell, D., & Solin, A. (2019). Unifying Probabilistic Models for Time-Frequency Analysis. In Proceedings of ICASSP 2019, 3352-3356.
[5] Albano, M., Abete, J. M., Čuček, V., Inza, I. B., De Brabandere, K., Etxabe, A., Gabilondo, I., Güven, Ç., & 15 others (2019), The MANTIS Book: Cyber Physical System Based Proactive Collaborative Maintenance. Albano, M., Jantunen, E., Papa, G. & Zurutuza, U. (eds.). River Publishers, p. 93-144.
[6] Stowell, D., Kelly, J., Tanner, D., Taylor, J., Jones, E., Geddes, J., & Chalstrey, E. (2020). A harmonised, high-coverage, open dataset of solar photovoltaic installations in the UK. Scientific Data 7 (1), 1-15.

Topic 3

Artificial Intelligence for predictive maintenance in water and electricity infrastructure

Scientific challenge

Predictive maintenance offers great potential value to the energy and water supply industry (cf. SDG7). Timely detection of required maintenance of machines, sensors, or other critical infrastructure can prevent disruptions of service and costly loss of resources. For instance, visual sensors can be used to analyze and detect subtle patterns [1,2,3] and auditory sensors can pick up subtle changes in sounds [4]. More generally, artificial intelligence offers improved prediction performance on predictive maintenance tasks. Recent advances in visual object recognition and auditory analysis allow for continuous and reliable monitoring of system states. In particular, the focus will be on self-supervised and unsupervised learning, see [5]. In the absence of supervisory labels, adequate priors will be acquired using large unlabeled datasets, see [6]. In the context of Industry 4.0, predictive maintenance leads to numerous innovations. One of the main challenges is dealing with real-time-based predictive maintenance [7]. Instead of treating predictive maintenance as simple alert monitoring, real-time-based predictive maintenance offers an estimate of time-to-failure.

Methodology

Building on existing work [7] and our own expertise acquired in the NWO-funded CERTIF-AI project, we will explore unsupervised and self-supervised learning in traditional and generative neural networks to achieve real-time-based predictive maintenance algorithms. The accuracy and reliability of these algorithms will be studied for predictive maintenance of water and energy installations of Aqualectra, and of WEB Bonaire, WEB Aruba, and GEBE Sint-Maarten. Our collaboration with Alliander, E-ON, and KPN facilitates the integration of the electricity grid and telecom infrastructure.

References

[1] Hendrix, N., Scholten, E., Vernhout, B., Bruijnen, S., Maresch, B., de Jong, M., Diepstraten, S., Bollen, S., Schalekamp, S., de Rooij, M., Scholtens, A., Hendrix, W., Samson, T., Ong, S., Postma, E., van Ginneken, B., & Rutten, M. (2021). Development and Validation of a Convolutional Neural Network for Automated Detection of Scaphoid Fractures on Conventional Radiographs. Radiology: Artificial Intelligence, e200260.
[2] van Lieshout, C., van Oeveren, K., van Emmerik, T., & Postma, E.O. (2020). Automated River Plastic Monitoring Using Deep Learning and Cameras. Earth and Space Science, 7(8).
[3] Noord, N. van & Postma, E. (2017). Learning scale-variant and scale-invariant features for deep image classification. Pattern Recognition, 61, 583-592.
[4] Buisman, H. J. & Postma, E. O. (2012). The log-Gabor method: speech classification using spectogram image analysis. Proceedings of the 13th annual conference of the international speech communication association 2012 (Interspeech 2012). Red Hook, U.S.A.: Curran Associates Inc., Vol. 1, p. 518-521.
[5] Olier, J. S., Barakova, E., Rauterberg, M., Marcenaro, L. & Regazzoni, C. (2018). Grounded representations through deep variational inference and dynamic programming. 7th Joint IEEE International Conference on Development and Learning and on Epigenetic Robotics, ICDL-EpiRob 2017.
[6] Ding, Y., Zhuang, J., Ding, P. & Jia, M. (2022). Self-supervised pretraining via contrast learning for intelligent incipient fault detection of bearings. Reliability Engineering & System Safety, 218A, 108126.
[7] Zonta, T., da Costa, C.A., da Rosa Righi, R., de Lima, M.J., da Trindade, E.S., and Li, G.P. (2020). Predictive maintenance in the Industry 4.0: A systematic literature review, Computers & Industrial Engineering, 150, 106889, ISSN 0360-8352.

Topic 4

AI for power grid balancing using recommendation-enhanced Demand Response

Scientific challenge

To improve grid balancing (SDG7), esp. in the case of many renewable energy resources and fluctuating demand, AI forecasting methods (WP2) can be combined with Demand Response (DR) methods. DR motivates energy consumers in some way (e.g. pricing-based) to adjust their energy usage to the available energy resources and demand. Smart grid technology allows DR to be more data-driven and a multitude of AI technologies have already been applied to DR [1], including ML, deep learning, and agent-based approaches. However, little research has investigated the consumer side of DR apart from simple customer segmentation approaches [1]. Rather than having consumers (household or industry) passively follow the DR (pricing) scheme, AI technology such as recommender algorithms could play an active role in recommending to consumers how and when to distribute their energy usage and return-based energy forecasts and DR information. Such tailored interventions improve DR approaches and optimize dynamic grid balancing.

Methodology

In the field of energy advice, most advice is not tailored yet. Recommender technology offers the ability to tailor what actions to take and when. For example, energy recommendations can be more effective when tailored to the users’ ability [2,3] and should be contextualized, for example by using micro-moments [4]. However, these insights are from the domestic domain and have not been researched in a more industrial context. We will extend current AI energy consumption forecasting and DR methods with tailored and explainable recommendations that will provide both household and industry consumers with tailored advice on how to adjust for Demand Response. As many stakeholders are involved, and dynamic changes in energy usage will affect other consumers in the network, a multi-stakeholder perspective on recommendations should be employed [5].

References

[1] Antonopoulos, I., Robu, V., Couraud, B., Kirli, D., Norbu, S., Kiprakis, A., Flynn, D., Elizondo-Gonzalez, S., & Wattam, S. (2020). Artificial intelligence and machine learning approaches to energy demand-side response: A systematic review. Renewable and Sustainable Energy Reviews, 130, 109899.
[2] Starke, A.D., Willemsen, M.C., & Snijders, C. (2017). Effective User Interface Designs to Increase Energy-efficient Behavior in a Rasch-based Energy Recommender System. In Proceedings of the Eleventh ACM Conference on Recommender Systems (pp. 65–73). New York, NY, USA.
[3] Starke, A. D., Willemsen, M.C., & Snijders, C. C. P. (2020). Beyond “one-size-fits-all” platforms: Applying Campbell’s paradigm to test personalized energy advice in the Netherlands. Energy Research and Social Science, 59.
[4] Alsalemi, A., Himeur, Y., Bensaali, F., Amira, A., Sardianos, C., Varlamis, I., & Dimitrakopoulos, G. (2020). Achieving Domestic Energy Efficiency Using Micro-Moments and Intelligent Recommendations. IEEE Access, 8, 15047–15055.
[5] Abdollahpouri, H., Adomavicius, G., Burke, R. et al. (2020). Multistakeholder recommendation: Survey and research directions. User Model User-Adap Inter 30, 127–158.

Topic 5

Social support for the real-world introduction of AI in critical infrastructure

Scientific challenge

The social impact of technology on its users has been vastly proven to be enormous [1] since it might pose organizational and social obstacles. Acceptance of new technologies, especially in the energy field, has been recognized as one of the primary barriers to implementing technological innovations [2]. In the context of the ILUSTRE project, the impact of IT technology on the energy and water supply domains is twofold. First: the impact on the partners’ industry managers and employees. The implementation creates disruption, and unless managers support the innovation and workers understand and comply with the new infrastructure, they might actively oppose the implementation. Second: the impact on the broader society since AI technology will affect the quality and the features of the services provided to the population. The scientific challenge addressed in this Ph.D.-project is to employ group model building (GMB) and social network analysis (SNA) to monitor the extent to which the employees and the larger public (together called the stakeholders) receive and respond to the implementation. GMB is a widely used approach to collect data and monitor (and influence) the opinions and sentiments of groups of stakeholders [3]. SNA has been used extensively to analyze the group dynamics at the roots of technological implementation reception [4]. These techniques can well be used in conjunction with agent-based simulation models.

Methodology

The research team will involve the stakeholders in GMB activities (such as one seminar per year, either online or in-person, where the researchers coordinate activities) with two main purposes: 1) collect relevant data about stakeholders’ opinions, demographics, and relationships, 2) increase the stakeholder involvement in the implementation and their perception of being an essential part of the process.

References

[1] King, W. R., & He, J. (2006). A meta-analysis of the technology acceptance model. Information & management, 43(6), 740-755.
[2] Huijts, N. M., Molin, E. J., & Steg, L. (2012). Psychological factors influencing sustainable energy technology acceptance: A review-based comprehensive framework. Renewable and sustainable energy reviews, 16(1), 525-531.
[3] Peck, S. (1998). Group model building: facilitating team learning using system dynamics. Journal of the Operational Research Society, 49(7), 766-767.
[4] Sasovova, Z., & Leenders, R. T. A. (2009). A corporate social capital view on E-HRM implementation. In Encyclopedia of Human Resources Information Systems: Challenges in e-HRM (pp. 210-215). IGI Global.

Eric Postma

Scientific Director

Eric is a professor in Artificial Intelligence at the Cognitive Science & AI department at Tilburg University and at the Jheronimus Academy of Data Science in ‘s-Hertogenbosch, a joint initiative of Eindhoven University of Technology and Tilburg University. His research focuses on pattern recognition in humans and machines. Although the term “pattern recognition” is not in vogue anymore, it does capture the human capability to perceive patterns in images, signals, and data in general. Nowadays, deep learning algorithms excel in pattern recognition on narrow domains.

Connect

Renato Calzone

Program Director

Renato is a Program Manager Data Science for Social Good at JADS (a program inspired by the UN sustainable development goals). His objective is to create a research, support and user community around AI in core areas such as clean energy, water and human centricity. Renato obtained MSc (Aeronautical Engineering) from Delft University and completed a MSc in Business Administration from Nyenrode University. He recently also graduated from the JADS Data Science for Professionals program.

Connect

Rigo Selassa

Industry Director

Rigo is a transformation manager with more than 15 years of experience helping organizations to transform using technological innovations. As a Global Executive MBA candidate he had the privilege to travel around the globe to different countries and follow courses at 6 recognized Business Schools (EGADE, GIBS, FGV, RSM, XMU and UNC). Based on his experience and network he is motivated to inspire organization to adopt human centric AI solutions and create education opportunities for talents to support the energy & water transition in the Caribbean and Latin American region.

Connect