For this study to benefit real-world resource management, we have to be mindful of the total process time required to produce the forecasts. Many types of simulation software support real-world applications. Multi-Agent Simulator of Competitive Electricity Markets (MASCEM) [49] and General Electric’s Multi-Area Production Simulation (MAPS) system [85] were created to help power system operators and energy traders perform real-time electricity grid and market operations. The Na-tional Energy Modeling Systems (NEMS) [86] helps power plants plan energy pro-duction strategies based on fuel prices. Long-range Energy Alternatives Planning (LEAP) [87] mitigates climate change effects through continuous real-time energy
policy updates. In other areas, UrbanSim [88] helps urban planners shape cities in a more structural and productive manner. There is even a public transport simula-tion that aids in real-time optimal dispatch [89]. Hence, there are many real-world opportunities for MAS-aided optimization.
In paper I, the amount of time required to produce a journey schedule and simulate it in MATSim depends on the number of drivers. We assume that each car will make a minimal of one journey with a return trip, and some drivers will drive to more than one destination. Hence, we are able to model three different groups of drivers in paperI. Suppose the number of drivers in each group is represented by l,mandn where the summation of them represents the total number of drivers in Singapore. The total amount of journeys,j, that happens in a day can be define as:
j=2l+3m+kn (4.5)
wherekcan be between 4 and 10. The last group includes delivery men, housewives who drive for home purposes and taxi drivers. Increasing any of these variables in-creases the MATSim simulation time and the amount of raw data. This, in turn, increases the data post-processing time. The aggregated time will determine the data synthesis execution time scale; how often it should run for current dataset. If we can execute more often, the data synthesized will be more current for forecast-ing. Hence, managing these variables contributes to managing the data synthesis processing time for this framework. In paper I, the measured time per MATSim simulation is six hours on average. We can produce a day-ahead forecast of the half-hourly energy load on the grid with this approach.
A crucial factor in this optimization framework is the forecast calculation pro-cessing time. In paper III, we studied the time required to calculate the results based on MLP, RNN and LSTM. LSTM was found to have the best accuracy against data size and processing time. However, we still need to make sure the estimation time falls within the Singapore energy market’s bidding time-period. Singapore’s energy market conducts energy transactions on a half-hourly basis. Hence, we need to optimize the forecast accuracy to data amount ratio that falls within the half-hour period. The forecast value degrades for those further away from the point of fore-cast. Hence, we need to boost the forecast values with every half-hour with new energy load data. Therefore, we hope that in addition to building a framework for optimization, it will be practical for real-world adoption. Otherwise, the system may become impractical during the course of use.
5 SUMMARY OF CONTRIBUTIONS
This chapter summarizes the five publications’ contributions that are included in this thesis. The objective of this study is to explore methods of resource manage-ment optimization. The resources managed in this study will be spent to implemanage-ment critical infrastructure and can be tremendous either by monetary means or time.
The dilemma lies in not knowing whether the outcomes are intended and favorable after building the infrastructure, which is made worse by the fact that no known and similar implementations currently exist. In this chapter, we narrate how each publication’s contributions are part of the framework shown in Fig. 2.1. In each of them, we explain their problem statements and corresponding motivations. We detail the methodologies used to solve each problem and how it contributes to the framework.
In paper I, we study large-scale energy resource management, which serves as the initial motivation for this study. Noting that energy resource management for buildings was quite well-studied, we decide to study the effects of massive EV de-ployment in Singapore. We synthesize the required data for the study in paper I as EVs are not massively deployed in Singapore. To find an effective to synthesize human-like data, we study the related literature, including simulated annealing, neural networks, genetic algorithms and MAS. We also compare various MAS used for traffic simulation to find the most suitable MAS, MATSim, for paperIand this study’s optimization framework. The most important contribution in paperIis the methodology formed to model the journey scheduling of all car drivers in Singa-pore. We execute the schedule on a MATSim simulation and compile the raw MAS data into a hourly energy consumption reading for 24 hours. We hope that this method of evaluating the effects of massive EV deployment can be replicated on other cities in similar studies. We found that the highest energy consumption by EVs on the road never rise beyond 10000 Mw and 5000 Mw during charging. These are below Singapore’s installed generation capacity of 10477.5 MW. Hence, Singa-pore could support the complete electrification of cars.
In paper II, we test the ability and feasibility of replicating the framework for another massive energy load curtailment program deployment in Singapore. It is costly to implement, but the impacts could mean a reduction in peak energy loads in Singapore or a new way to control how it is being consumed. First, in re-applying the framework developed in paperI, we study the load curtailment program’s mech-anisms and payouts. We then model the program with a suitable MAS. The most important contribution in paperIIis the programmed energy user human behavior to meet the curtailment program’s criteria for payout. It involves modeling and sim-ulating human reactions to curtailment program payouts. We create this behavior using a mixture of energy load shifting and the Monte Carlo technique to determine whether a person can successfully alter his or her energy consumption behavior if given incentives. We hope that with the works in paperII, we can help cities with-out an energy load curtailment program realize the benefits of having one. As our
MAS simulation shows that the curtailment program can provide opportunities for retailers to profit by being demand aggregators, Singapore started the "Open Elec-tricity Market" in 2018. Many energy retailers adopted similar demand aggregation strategies presented in paperIIas energy trading participants; this was 4 years after paperIIwas published in 2014 [90].
In paperIII, we investigate the framework’s application to real-time forecasting of EV loads. In papersIandII, we learn that MAS simulations and the data post-processing can be time-intensive. We can aim to run the simulation and forecast the day-ahead EV load. However, as the day progresses, forecast accuracy will degrades due to discrepancies between the forecast and actual energy load data. Hence, in paper III, we study various forecasting techniques, including regression, machine learning and neural network algorithms. We want to maximize the forecast’s ac-curacy while minimizing its execution time. The main contribution in paperIII is determine that LSTM a better forecasting solution than MLP and RNN. All three were widely used at the time of this study due to their high accuracy. We also study the effects of data size on forecast accuracy and execution times. We hope that the processes and relationships between data size, forecast accuracy and execu-tion times discussed in paperIIIwill benefit similar studies of more massive digital technologies prior to deployment. Based on the simulation results reported in paper III, LSTM is the most accurate approach when forecasting the next hourly energy load, with an average error of 1.4% per prediction compared to 3.5% for RNN and 11.9% for MLP.
In paper IV, we investigate the feasibility of using the techniques learned in papers I, II and III to model and study the traffic situation in Joensuu, Finland.
We were obtain Joensuu’s working population data, which include residential and workplace location coordinates. We apply the modeling technique developed in paper I to produce a schedule of many commuting journeys. We simulate it in MATSim to understand Joensuu’s traffic situation. The most important contribution in paperIVis verifying the similarity between synthesized and actual data. We also identify and account for data differences and investigate the outcomes in traffic sit-uations if any of the four bridges crossing the Pielisjoki river in Joensuu, Finland are closed. We hope that city planners can use this work to prepare for possible traffic congestion alleviation when closing roads that are heavily used. We found that the effect of removing the Itäsilta bridge seems to increase traffic by more than six times only on Suvantosilta but has almost no effect on Sirkkalasilta and Pekkalasilta at 8AM.
In paperV, similar to the works in paperIV, we study the effects of applying the framework to another real world situation. In paperV, we study the interactions between lifts and human behaviors in a commercial building. This research is similar to studying traffic situations if we think of lifts as vehicles that travel on vertical tracks. In paperV, we model the arrival of humans in the lift lobbies of a building using collaborator-provided elevator information and Poisson distributions. As a commercial building consists of different businesses, we model the human traffic outcome to be similar to actual data. The most important contribution in paperVis showing that the framework can be applied to model another use case beyond traffic and build an optimized lift scheduling algorithm for a group of elevators. We hope that businesses can use this framework as a cost benefit analysis toolkit to further
their business development, a form of resource management but for companies.
6 CONCLUSION
In this dissertation, an approach to synthesize data related to the real-world was studied. The value in this synthesis lies in creating data that are not yet possi-ble to collect from the real-world, perhaps because the circumstances necessary for generating the data do not yet exist. In the current research, few EVs utilize in Singapore’s road network system, so we needed to estimate the energy consump-tion of EVs in Singapore. The approach involved modeling driver behaviors and letting them interact in a MAS. To make the data approximate real-world data more closely, external factors, such as 24-hour environmental temperatures were incorpo-rated when calculating the estimated energy consumed. The approach can be easily modified as more accurate data becomes available.
The approach was verified in another use case when Joensuu’s municipal gov-ernment collected traffic usage data on the latest bridge built across the Pielisjoki river. Joensuu’s driver behavior was modeled in the MAS. Synthesized data from the MAS model was compared to the collected data to assess similarity. The compar-ison showed that in most aspects, synthesized and actual data were similar. How-ever, there were some dissimilarities, which could be due to incomplete modeling of actual driver’ behavior. To demonstrate that the approach is useful beyond traf-fic simulations, we also applied it to human behavior using lifts in a commercial building. Finally, to provide additional insights, we explored other situations in Sin-gapore where this approach may be useful.
Regarding the four objectives defined in the introduction, we have made progress in achieving them in this study. The first objective of developing a systematic ap-proach is still in progress. Additional case studies are required to identify common-alities and use them in developing a systematic approach. The second objective, to identify the constraints and processes, was met. This is evident in the methodology sections of papersI, II, IVandV. MAS, in the form of MATSim, was identified as the most appropriate tool to model the scenario and synthesize data. Therefore, the third objective was achieved. Finally, in paperIII, we identified forecast methods to provide look-ahead values for energy consumption. This is the final and crucial step in optimization. Hence, we successfully created a complete framework to optimize a non-existent system virtually, thus achieving the fourth objective.
The developed framework, using MAS as the main technique, is in performing cost and benefit analyses of costly projects prior to deployment. A MAS gener-ates deterministic data that follow the conditions modeled by the simulation. For example, the spread of COVID-19 can be forecasted by modeling people’s move-ments by air, land and sea. Another use for this approach would be in synthesizing data for urban observations when a new stimulus is introduced. The synthesized data would be useful in understanding social reactions when something unexpected happens. Both of these cases require a significant number of agreements between many sovereigns or exorbitant physical infrastructure investment for data
genera-tion. Such boundaries can be more easily overcome with data synthesis using a MAS. The data would also be helpful in understanding the urban population’s re-siliency and measures that can be implemented to strengthen it. Our most recent published paper, titled ’Innovating Services and Digital Economy in Singapore’ [91], describes the various challenges Singapore is acing, which is where the current work could be applied in the future.
BIBLIOGRAPHY
[1] S. Barca, “Energy, property, and the industrial revolution narrative,”Ecological Economics70, 1309–1315 (2011).
[2] C. A. Ristuccia and S. Solomou, “General Purpose Technologies and Economic Growth: Electricity Diffusion in the Manufacturing Sector Before WWII,”
(2010).
[3] W. Knorr, “Is the airborne fraction of anthropogenic CO2 emissions increas-ing?,”Geophysical research letters36(2009).
[4] S. Shafiee and E. Topal, “When will fossil fuel reserves be diminished?,”Energy policy37, 181–189 (2009).
[5] The Business Times, Brunch, "Charged up: Singapore’s journey to the future of cars begins now", created on 02/22/2020, https://www.businesstimes.com.sg/brunch/
charged-up-singapores-journey-to-the-future-of-cars-begins-now (visited on 2020-03-14).
[6] K.-K. Chin and G. Menon, “Transport accessibility and infrastructure in Singapore–pedestrian facilities,” in Proceedings of the Institution of Civil Engineers-Municipal Engineer, Vol. 168 (Thomas Telford Ltd, 2015), pp. 133–139.
[7] P. E. Gill, W. Murray, and M. H. Wright, Practical optimization(SIAM, 2019).
[8] G. E. Davis, “Control system with adaptive process controllers especially adapted for electric power plant operation,” (1976), US Patent 3,939,328.
[9] M. J. Radovich, “Method and apparatus for combined cycle power plant oper-ation,” (2006), US Patent 7,107,774.
[10] C. F. Ridolfo, “Predictive maintenance display system,” (2004), US Patent 6,735,549.
[11] C. Fu, L. Ye, Y. Liu, R. Yu, B. Iung, Y. Cheng, and Y. Zeng, “Predictive mainte-nance in intelligent-control-maintemainte-nance-management system for hydroelectric generating unit,”IEEE transactions on energy conversion19, 179–186 (2004).
[12] C. P. Salomon, W. C. Santana, E. L. Bonaldi, L. E. de Oliveira, J. G. B. da Silva, G. Lambert-Torres, L. E. B. da Silva, A. Pellicel, M. A. Lopes, and G. C.
Figueiredo, “A system for turbogenerator predictive maintenance based on Electrical Signature Analysis,” in 2015 IEEE International Instrumentation and Measurement Technology Conference (I2MTC) Proceedings(IEEE, 2015), pp. 79–84.
[13] J. L. Asenjo, J. Strohmenger, S. T. Nawalaniec, B. H. Hegrat, J. A. Harkulich, J. L. Korpela, J. R. Wright, R. Hessmer, J. Dyck, E. A. Hill, et al., “Predictive maintenance for industrial products using big data,” (2014), US Patent App.
14/087,730.
[14] R. K. Mobley,An introduction to predictive maintenance(Elsevier, 2002).
[15] B. J. Koch, C. M. Febria, M. Gevrey, L. A. Wainger, and M. A. Palmer, “Nitrogen removal by stormwater management structures: A data synthesis,” JAWRA Journal of the American Water Resources Association50, 1594–1607 (2014).
[16] R. Barth, J. IJsselmuiden, J. Hemming, and E. J. Van Henten, “Data synthe-sis methods for semantic segmentation in agriculture: A Capsicum annuum dataset,”Computers and electronics in agriculture144, 284–296 (2018).
[17] G. Bernstein and K. O’Brien, “Stochastic agent-based simulations of social net-works,”arXiv preprint arXiv:1309.1747(2013).
[18] N. Park, M. Mohammadi, K. Gorde, S. Jajodia, H. Park, and Y. Kim, “Data synthesis based on generative adversarial networks,” Proceedings of the VLDB Endowment11, 1071–1083 (2018).
[19] J.-L. Comeau, D. J. Dalmotas, and A. German, “Event Data Recorders in Toyota Vehicles,” inProceedings of the 21st Canadian Multidisciplinary Road Safety Confer-ence, Halifax, Nova Scotia, Vol. 17 (2011).
[20] P. J. Van Laarhoven and E. H. Aarts, “Simulated annealing,” inSimulated an-nealing: Theory and applications(Springer, 1987), pp. 7–15.
[21] H. Szu and R. Hartley, “Fast simulated annealing,”Physics letters A122, 157–162 (1987).
[22] S. Haykin,Neural networks: a comprehensive foundation(Prentice Hall PTR, 1994).
[23] M. H. Hassoun et al.,Fundamentals of artificial neural networks(MIT press, 1995).
[24] G. R. Harik, F. G. Lobo, and D. E. Goldberg, “The compact genetic algorithm,”
IEEE transactions on evolutionary computation3, 287–297 (1999).
[25] Z. Tu and Y. Lu, “A robust stochastic genetic algorithm (StGA) for global numerical optimization,”IEEE Transactions on Evolutionary Computation8, 456–
470 (2004).
[26] F.-S. Ho and P. Ioannou, “Traffic flow modeling and control using artificial neural networks,”IEEE Control Systems Magazine16, 16–26 (1996).
[27] X. Ma, Z. Tao, Y. Wang, H. Yu, and Y. Wang, “Long short-term memory neu-ral network for traffic speed prediction using remote microwave sensor data,”
Transportation Research Part C: Emerging Technologies54, 187–197 (2015).
[28] D. Teodorovic and K. Vukadinovic, Traffic Control and Transport Planning:: A Fuzzy Sets and Neural Networks Approach, Vol. 13, (Springer Science & Business Media, 2012).
[29] N. J. Gordon, D. J. Salmond, and A. F. Smith, “Novel approach to nonlinear/non-Gaussian Bayesian state estimation,” inIEE proceedings F (radar and signal processing), Vol. 140 (IET, 1993), pp. 107–113.
[30] J. Mun, Modeling risk: Applying Monte Carlo simulation, real options analysis, forecasting, and optimization techniques, Vol. 347, (John Wiley & Sons, 2006).
[31] F. A. Haight,Handbook of the Poisson distribution(Wiley, 1967).
[32] R. K. Freeland and B. P. McCabe, “Forecasting discrete valued low count time series,”International Journal of Forecasting20, 427–434 (2004).
[33] A. Chojaczyk, A. Teixeira, L. C. Neves, J. Cardoso, and C. G. Soares, “Review and application of artificial neural networks models in reliability analysis of steel structures,”Structural Safety52, 78–89 (2015).
[34] H. M. Hasanien and S. Muyeen, “Design optimization of controller parameters used in variable speed wind energy conversion system by genetic algorithms,”
IEEE Transactions on Sustainable Energy3, 200–208 (2012).
[35] J. Ferber and G. Weiss,Multi-agent systems: an introduction to distributed artificial intelligence, Vol. 1, (Addison-Wesley Reading, 1999).
[36] E. L. Karfopoulos and N. D. Hatziargyriou, “A multi-agent system for con-trolled charging of a large population of electric vehicles,”IEEE Transactions on Power Systems28, 1196–1204 (2012).
[37] W. Li, T. Logenthiran, and W. L. Woo, “Intelligent multi-agent system for smart home energy management,” in2015 IEEE Innovative Smart Grid Technologies-Asia (ISGT ASIA)(IEEE, 2015), pp. 1–6.
[38] R. A. Flores-Mendez, “Towards a standardization of multi-agent system frame-work,”XRDS: Crossroads, The ACM Magazine for Students5, 18–24 (1999).
[39] P. Wavish, “Exploiting emergent behavior in multi-agent systems,” ACM SIGOIS Bulletin13, 16 (1992).
[40] H. Yang and J. Zhou, “Optimal traffic counting locations for origin–destination matrix estimation,” Transportation Research Part B: Methodological 32, 109–126 (1998).
[41] S. Tisue and U. Wilensky, “Netlogo: A simple environment for modeling complexity,” inInternational conference on complex systems, Vol. 21 (Boston, MA, 2004), pp. 16–21.
[42] N. Minar, R. Burkhart, C. Langton, M. Askenazi, et al., “The swarm simulation system: A toolkit for building multi-agent simulations,” (1996).
[43] C.-h. JIANG, W. HAN, and Y.-h. HU, “REPAST-A multi-agent simulation plat-form,”Journal of System Simulation18, 2319–2322 (2006).
[44] S. Luke, C. Cioffi-Revilla, L. Panait, K. Sullivan, and G. Balan, “Mason: A multiagent simulation environment,”Simulation81, 517–527 (2005).
[45] M. Behrisch, L. Bieker, J. Erdmann, and D. Krajzewicz, “SUMO–simulation of urban mobility: an overview,” inProceedings of SIMUL 2011, The Third Interna-tional Conference on Advances in System Simulation(ThinkMind, 2011).
[46] J. Miller and E. Horowitz, “FreeSim-a free real-time freeway traffic simulator,”
in2007 IEEE Intelligent Transportation Systems Conference(IEEE, 2007), pp. 18–23.
[47] A. Horni, K. Nagel, and K. W. Axhausen, The multi-agent transport simulation MATSim(Ubiquity Press London, 2016).
[48] P. Oliveira, T. Pinto, H. Morais, and Z. Vale, “MASGriP—a multi-agent smart grid simulation platform,” in2012 IEEE Power and Energy Society General Meet-ing(IEEE, 2012), pp. 1–8.
[49] I. Praça, C. Ramos, Z. Vale, and M. Cordeiro, “MASCEM: a multiagent system that simulates competitive electricity markets,”IEEE Intelligent Systems18, 54–
60 (2003).
[50] E. W. Dijkstra et al., “A note on two problems in connexion with graphs,”
Numerische mathematik1, 269–271 (1959).
[51] Y. Niu, Z. Yang, P. Chen, and J. Xiao, “Optimizing the green open vehicle routing problem with time windows by minimizing comprehensive routing cost,”Journal of cleaner production171, 962–971 (2018).
[52] Y. Niu, Z. Yang, P. Chen, and J. Xiao, “A hybrid tabu search algorithm for a
[52] Y. Niu, Z. Yang, P. Chen, and J. Xiao, “A hybrid tabu search algorithm for a