Papers by Turgut Ozgu
International Journal of Shipping and Transport Logistics
Seaborne transport forecasting has attracted substantial interest over the years because of provi... more Seaborne transport forecasting has attracted substantial interest over the years because of providing a useful policy tool for decision-makers. Although various forecasting methods have been widely studied, there is still broad debate on accurate forecasting models and preprocessing. The current paper aims to point out these issues, as well as to establish the forecasting model of the domestic cargo volumes using SARIMAX, MLP, LSTM and NARX and SARIMAX-ANN hybrid models. Based on the domestic cargo volumes of Turkey, findings suggest that SARIMA-MLP models can be considered as an appropriate alternative, at least for time series forecasting of shipping. Pre-processed data provides a significant improvement over those obtained with unpreprocessed data, with the accuracy of the models found to be significantly boosted with the Fourier term of decomposition. The results indicate that SARIMAX-MLP, with a mean absolute percentage error (MAPE) of 4.81, outperforms the closest models of SARIMAX, with a MAPE of 6.14 and LSTM with Fourier decomposition with a MAPE of 6.52. Findings have implications for shipping policymakers to plan infrastructure development, and useful for shipowners in accurately formulating shipping demand.
SoftwareX, 2022
Energy and power system models represent important insights on the technical operations of energy... more Energy and power system models represent important insights on the technical operations of energy technologies that supply the energy consumption in time steps with hourly resolution. This paper presents the European Model for Power system Investments with Renewable Energy (EMPIRE) that combines short-term operations with the representation of long-term planning decisions including infrastructure expansion. The EMPIRE model has an unique mathematical modelling structure based on multi-horizon stochastic programming, which means investment decisions are subject to short-term uncertainty represented by different realizations of operational scenarios. The model is open source and ready to use to analyse energy transition scenarios towards 2050 and beyond. This paper outlines the building blocks of the model and its software structure. We also present an illustrative example of results from using the software.
Journal of Global Optimization, 2019
The set of all nondominated solutions for a multi-objective integer programming (MOIP) problem is... more The set of all nondominated solutions for a multi-objective integer programming (MOIP) problem is finite if the feasible region is bounded, and it may contain unsupported solutions. Finding these sets is NP-hard for most MOIP problems and current methods are unable to scale with the number of objectives. We propose a deterministic exact parallel algorithm for solving MOIP problems with any number of objectives. The proposed algorithm generates the full set of nondominated solutions based on intelligent iterative decomposition of the objective space utilizing a particular scalarization scheme. The algorithm relies on a set of rules that exploits regional dominance relations among the decomposed partitions for pruning. These expediting rules are both used as part of a pre-solve step as well as judiciously employed throughout the parallel running threads. Using an extensive test-bed of MOIP instances with three, four, five, and six objectives, the performance of the proposed algorithm is evaluated and compared with leading benchmark algorithms for MOIPs. Results of the experimental study demonstrate the effectiveness of the proposed algorithm and the computational advantage of its parallelism.
Procedia Computer Science, 2011
Stochastic multi objective programming problems commonly arise in complex systems such as portfol... more Stochastic multi objective programming problems commonly arise in complex systems such as portfolio analysis, medium-to long-term capacity planning and design applications under uncertainty. The identification of the candidate solution set is a main step in many applications which depends on the nature of uncertainty. This study presents a method to generate Pareto surface for multi-objective integer programs with stochastic coefficients in the objective functions based on minimum expectation and variance criteria. The objective function coefficients are represented through random discrete distributions. The methodology follows a two-phase approach where, in the first phase, the stochastic multiple objectives are converted into deterministic equivalents based on the minimum expectation and variance efficiency concepts. The second phase solves the deterministic multi objective problem, using a Pareto generation methodology which aims at generating the whole Pareto surface of multi objective integer programming problems. We present results of experimental study of applying the proposed method to an assignment problem with three objective functions.
EMPIRE: An open-source model based on multi-horizon programming for energy transition analyses, 2022
Energy and power system models represent important insights on the technical operations of energy... more Energy and power system models represent important insights on the technical operations of energy technologies that supply the energy consumption in time steps with hourly resolution. This paper presents the European Model for Power system Investments with Renewable Energy (EMPIRE) that combines short-term operations with the representation of long-term planning decisions including infrastructure expansion. The EMPIRE model has an unique mathematical modelling structure based on multi-horizon stochastic programming, which means investment decisions are subject to short-term uncertainty represented by different realizations of operational scenarios. The model is open source and ready to use to analyse energy transition scenarios towards 2050 and beyond. This paper outlines the building blocks of the model and its software structure. We also present an illustrative example of results from using the software.
An exact parallel objective space decomposition algorithm for solving multi-objective integer programming problems
The set of all nondominated solutions for a multi-objective integer programming (MOIP) problem is... more The set of all nondominated solutions for a multi-objective integer programming (MOIP) problem is finite if the feasible region is bounded, and it may contain unsupported solutions. Finding these sets is NP-hard for most MOIP problems and current methods are unable to scale with the number of objectives. We propose a deterministic exact parallel algorithm for solving MOIP problems with any number of objectives. The proposed algorithm generates the full set of nondominated solutions based on intelligent iterative decomposition of the objective space utilizing a particular scalarization scheme. The algorithm relies on a set of rules that exploits regional dominance relations among the decomposed partitions for pruning. These expediting rules are both used as part of a pre-solve step as well as judiciously employed throughout the parallel running threads. Using an extensive test-bed of MOIP instances with three, four, five, and six objectives, the performance of the proposed algorithm is evaluated and compared with leading benchmark algorithms for MOIPs. Results of the experimental study demonstrate the effectiveness of the proposed algorithm and the computational advantage of its parallelism.
Time series forecasting of domestic shipping market: comparison of SARIMAX, ANN-based models and SARIMAX-ANN hybrid model
Seaborne transport forecasting has attracted substantial interest over the years because of provi... more Seaborne transport forecasting has attracted substantial interest over the years because of providing a useful policy tool for decision-makers. Although various forecasting methods have been widely studied, there is still broad debate on accurate forecasting models and preprocessing. The current paper aims to point out these issues, as well as to establish the forecasting model of the domestic cargo volumes using SARIMAX, MLP, LSTM and NARX and SARIMAX-ANN hybrid models. Based on the domestic cargo volumes of Turkey, findings suggest that SARIMA-MLP models can be considered as an appropriate alternative, at least for time series forecasting of shipping. Pre-processed data provides a significant improvement over those obtained with unpreprocessed data, with the accuracy of the models found to be significantly boosted with the Fourier term of decomposition. The results indicate that SARIMAX-MLP, with a mean absolute percentage error (MAPE) of 4.81, outperforms the closest models of SARIMAX, with a MAPE of 6.14 and LSTM with Fourier decomposition with a MAPE of 6.52. Findings have implications for shipping policymakers to plan infrastructure development, and useful for shipowners in accurately formulating shipping demand.
Improved Reinforcement Learning for Preventing Consumer Food Waste in Volatile Food Markets, 2026
This paper presents two novel reinforcement learning (RL) architectures tailored for “online”, i.... more This paper presents two novel reinforcement learning (RL) architectures tailored for “online”, i.e. dynamic, and ”offline”, i.e. tabular, decision support schemas. Uniqueness of the approaches stems from reliance on Operations Research (OR) methods as building-blocks, rather than data sets or actor-critique. Along with benchmark results with fast multi-criteria decision-making heuristic and the standard, i.e. greedy Q-learning schema, outcomes of investigations which aim at answering questions regarding the effect of state space dimensions or size of the control set in problem formulation are also provided. In addition to explaining the theoretical basis of the novel architectures reaching to the rudiments of dynamic programming, the article seeks to exhibit the potential of the proposed methods from the engineering perspective through designing the methods into a practical decision support unit (DSU) to be used in food planning by end-consumer. In this vein, the test framework aims to prevent food waste from household and restaurant kitchens via the DSU which serves to generate menu recommendations in a volatile market while considering three important aspects of food management: the total shopping budget, wasted ingredients, and the nutrition intake resultant by the selected menu. The results reveal the holistic superiority of online architecture which is based on Monte Carlo simulation and mathematical modelling integrated into the Bellman equation. Moreover, taking advantage of wise reduction opportunities of control set or dimension of state space are supported by the results. In addition to the solution quality assessment of each approach, implementation details such as computational requirements and resilience capacity are also discussed from the business perspective.
Machine learning algorithms are spreading into many essential decision-making procedures of intel... more Machine learning algorithms are spreading into many essential decision-making procedures of intelligent systems. These algorithms rely on big data which contain several sensitive features regarding the training instances such as gender, age, and race. In some contexts, these features might be useful and relevant like in health or other physiology-related frameworks. However, there are some fields where these features should be ignored like legal, financial, or educational incentivizing decisions. In complex decision-making procedures, it is important to guarantee indifference to these sensitive features. Choosing suitable algorithms or designing decision-making procedures that take this aspect of fairness into account is of crucial importance; especially from the reliability of upcoming intelligent systems which directly affects the impact of the new technology. In this study sample fairness assessment, experiments are performed in three different relevant contexts by applying two different ensemble building methods. It is shown how fairness can be assessed without hurting the accuracy. Furthermore, guidelines are provided for experimental design in order to derive more generic observations regarding incorporating the fairness notion into major classification algorithms 1 .
Research Square (Research Square), Apr 15, 2024
Water scarcity is a problem for many regions which requires immediate action and solutions cannot... more Water scarcity is a problem for many regions which requires immediate action and solutions cannot be postponed for a long time. It is known that farming consumes a significant portion of usable water. In this study, a decision support model of bi-objective stochastic linear formulation is proposed. The model is generating annual planting plans together with water consumption projections for each farmer in the region while taking revenue of the overall harvest into account. The structure of the proposed model maintains robustness against the volatilities in precipitation, yield, and market price. The inherent trade-off between water consumption and revenue lends itself to multi-objective planning. This is a perspective especially useful for regional administrations to plan next year's crop pattern together with agricultural incomes and irrigation expenses. Furthermore, it is also shown that how the model can be used to investigate the potential of rainwater harvesting or switching to water-efficient irrigation methodologies. The decision support model is especially unique in the sense that it can generate a set of Pareto optimum solutions as opposed to a single objective counterpart. This property is helpful in terms of not only providing a broader perspective to evaluate and project the possibilities but also increasing the applicability of the results by providing flexible design framework.
Industry is responsible for one-quarter of the global CO 2 emissions. In this study, four differe... more Industry is responsible for one-quarter of the global CO 2 emissions. In this study, four different climate pathways are analyzed with a cost minimizing multihorizon stochastic optimization model, in order to analyze possible realizations of carbon capture and storage (CCS) in the power sector and main industrial sectors in Europe. In particular, we aim to achieve a deeper understanding of the distribution of capture by country and key sector (power, steel, cement and refinery), as well as the associated transport and storage infrastructure for CCS. Results point to the synergy effect of sharing common CCS infrastructres among power and major industrial sectors. The contribution of CCS is mainly found in three industrial sectors, particularly steel, cement and refineries) but also in the power sector to a lesser extent. It is worth noting that retrofitting of CCS in the power sector was not considered in this study. The geographical location for capture and storage, as well as timing and capacity needs are presented for different socio-economic pathways and corresponding emission targets. It has been shown that contributions of the three industry sectors in emissions reductions are neither geographically nor sector-wise homogeneous across the pathways.
ROBUST PLANNING OF IRRIGATION CONSIDERING WATER CONSUMPTION AND REVENUE, 2024
Water scarcity is a problem for many regions which requires immediate action, and solutions canno... more Water scarcity is a problem for many regions which requires immediate action, and solutions cannot be postponed for a long time. It is known that farming consumes a significant portion of usable water. In this study, a decision support model of biobjective stochastic linear formulation is proposed. The model is generating annual planting plans together with water consumption projections for each farmer in the region while taking revenue of the overall harvest into account. The structure of the proposed model maintains robustness against the volatilities in precipitation, yield, and market price. The inherent trade-off between water consumption and revenue lends itself to multi-objective planning. This is a perspective especially useful for regional administrations to plan next year's crop pattern together with agricultural incomes and irrigation expenses. Furthermore, it is also shown how the model can be used to investigate the potential of rainwater harvesting or switching to waterefficient irrigation methodologies. The decision support model is especially unique in the sense that it can generate a set of Pareto optimum solutions as opposed to a single objective counterpart. This property is helpful in terms of not only providing a broader perspective to evaluate and project the possibilities but also increasing the applicability of the results by providing flexible design framework
Expert Systems with Applications, 2012
Voice of the customer (VOC) is a critical analysis procedure that provides precise information re... more Voice of the customer (VOC) is a critical analysis procedure that provides precise information regarding customer input requirements for a product/service output. The ability to conduct a voice of the customer analysis, which could be gained through direct and indirect questioning, will enable engineers and other decision makers to successfully understand customer needs, wants, perceptions, and preferences. The information obtained from the customers is then translated into critical targets that will be used to ultimately satisfy the customer requirements. During this research project, different forms of customer input, including qualitative and quantitative data, were transformed to a common data format to develop a correlation between design input requirements and product/service outputs. We have developed a new method for measuring customer satisfaction ratio (CSR) by considering the following: mining both textual and quantitative data, multiple design parameters, mapping output on a scale of 0-1, and a decision template for means of measure. Previous measures of CSR fail to incorporate the cost implication of fixing customer complaints/issues; however, we include this important and unique measure in our research. The implication of this research will reduce Things Gone Wrong (TGW's) and engineering development time and will achieve improvements in JD Power ratings, quality perception, marketing tools, and customer satisfaction.
Expert Systems with Applications, 2012
Voice of the customer (VOC) is a critical analysis procedure that provides precise information re... more Voice of the customer (VOC) is a critical analysis procedure that provides precise information regarding customer input requirements for a product/service output. The ability to conduct a voice of the customer analysis, which could be gained through direct and indirect questioning, will enable engineers and other decision makers to successfully understand customer needs, wants, perceptions, and preferences. The information obtained from the customers is then translated into critical targets that will be used to ultimately satisfy the customer requirements. During this research project, different forms of customer input, including qualitative and quantitative data, were transformed to a common data format to develop a correlation between design input requirements and product/service outputs. We have developed a new method for measuring customer satisfaction ratio (CSR) by considering the following: mining both textual and quantitative data, multiple design parameters, mapping output on a scale of 0-1, and a decision template for means of measure. Previous measures of CSR fail to incorporate the cost implication of fixing customer complaints/issues; however, we include this important and unique measure in our research. The implication of this research will reduce Things Gone Wrong (TGW's) and engineering development time and will achieve improvements in JD Power ratings, quality perception, marketing tools, and customer satisfaction.
Conference Presentations by Turgut Ozgu
Generating Pareto Surface for Multi Objective Integer Programming Problems with Stochastic Objective Coefficients
Stochastic multi objective programming problems commonly arise in complex systems such as portfol... more Stochastic multi objective programming problems commonly arise in complex systems such as portfolio analysis, medium-to long-term capacity planning and design applications under uncertainty. The identification of the candidate solution set is a main step in many applications which depends on the nature of uncertainty. This study presents a method to generate Pareto surface for multi-objective integer programs with stochastic coefficients in the objective functions based on minimum expectation and variance criteria. The objective function coefficients are represented through random discrete distributions. The methodology follows a two-phase approach where, in the first phase, the stochastic multiple objectives are converted into deterministic equivalents based on the minimum expectation and variance efficiency concepts. The second phase solves the deterministic multi objective problem, using a Pareto generation methodology which aims at generating the whole Pareto surface of multi objective integer programming problems. We present results of experimental study of applying the proposed method to an assignment problem with three objective functions.
Fair Classification with Ensembles, 2023
Machine learning algorithms are spreading into many essential decision-making procedures of intel... more Machine learning algorithms are spreading into many essential decision-making procedures of intelligent systems. These algorithms rely on big data which contain several sensitive features regarding the training instances such as gender, age, and race. In some contexts, these features might be useful and relevant like in health or other physiology-related frameworks. However, there are some fields where these features should be ignored like legal, financial, or educational incentivizing decisions. In complex decision-making procedures, it is important to guarantee indifference to these sensitive features. Choosing suitable algorithms or designing decision-making procedures that take this aspect of fairness into account is of crucial importance; especially from the reliability of upcoming intelligent systems which directly affects the impact of the new technology. In this study sample fairness assessment, experiments are performed in three different relevant contexts by applying two different ensemble building methods. It is shown how fairness can be assessed without hurting the accuracy. Furthermore, guidelines are provided for experimental design in order to derive more generic observations regarding incorporating the fairness notion into major classification algorithms 1 .
Books by Turgut Ozgu
Empowering Smallholder Farmers with Digital Supply Chain Management Tools
Smallholders are small-scale farmers who manage areas varying from less than one hectare to 10 he... more Smallholders are small-scale farmers who manage areas varying from less than one hectare to 10 hectares and they are an important portion of the farming population especially in developing countries. However, they are usually abandoned in a market of big players or lack government support. Their poor conditions do not only affect well-being of these families, but also-as it can be observed from Turkey example-the isolated position of these people cause emergence of more than necessary intermediaries. This-in turn-results in hyperinflation where neither the customer nor the producer is financially satisfied. By the same token the wasted products due to poor logistical conditions increase significantly. Based on these observations, a digital market model together with an AI and optimization integrated decision support framework is proposed in this study. Analytical (AI algorithm based predictive and Mathematical Modelling based prescriptive) components of the digital tool are explained rigorously. Also, a case study is provided to elaborate on the potential contributions of the tool.
Uploads
Papers by Turgut Ozgu
Conference Presentations by Turgut Ozgu
Books by Turgut Ozgu