A two-stage stochastic programing model to optimize logistical decisions in cold supply chain under government regulation
Co-author: Dr Anup Shrestha, Associate Professor Babak Abbasi, Dr Shane Zhang, Professor Alice Woodhead
Increasing awareness of sustainability in supply chain management has prompted organizations and individuals to consider environmental impacts while optimizing supply chain management. The issues concerning environmental impacts is prominent in cold supply chains due to significant carbon emissions arising from storage and distribution of temperature-sensitive products. This research extends the classical supply chain optimization models, which focus on cost efficiency as a single key performance indicator (KPI), to include other sustainability-based KPIs.
More specifically, the impact of accounting for carbon emissions that arise from unique processes of the cold supply chain consisting of a supplier and multiple retailers in the presence of government regulations is investigated. To this end, we developed a two-stage stochastic programing model to determine replenishment policy and transportation schedules aiming to minimize operational and emissions costs considering demand uncertainty. Since the proposed model contains several nonlinear expressions, we apply linearization techniques to develop an equivalent linear mathematical model. A numerical experiment is used to validate the model and explore the trade-off between emissions and operational costs. Finally, several managerial insights on optimizing cold supply chains are offered through parameter analysis.
Key words: Cold supply chain, Two-stage stochastic programing, Inventory decision, Transportation decision, government regulation.
Optimisation of Disaster Waste Management Systems
Co-authors: Professor Russell Thompson, Dr Alysson Costa, Dr Lihai Zhang
The four major stages of disaster management are mitigation, preparation, response, and recovery. Waste management is one of the core activities in the recovery stage and focuses on collecting, reducing or recycling, and final disposal of the remaining waste. The volume of waste generated from a single event can reach 5 to 15 times the annual waste normally produced by affected communities. The clearance, removal, and disposal of such large amounts of debris are costly and time-consuming operations. However, there has been little literature dedicated to the improvement of disaster waste management (DWM) procedures compared to other operations in disaster management. The main objective of this research is to develop an integrated framework to improve DWM. Two sets of models that focus on two topics have been developed, namely reliability analysis of a DWM system and the two-echelon DWM system optimisation. The framework is tested for its validity and capacity for an improved understanding of the challenges in disaster waste clean-up.
Optimisation Design for Energy-Efficient Downlink Cloud Radio Access Networks
Increasing data traffic and reducing the total network energy consumption are listed among of the top priorities of 5G wireless systems. Cloud radio access networks (C-RANs) have been considered as a promising solution to address these challenging objectives. In this work, we aim to maximise the ratio of network throughput and total power consumption of a downlink C-RAN, where the user association plays a key role in network resource allocation. A mixed-integer non-linear problem is formulated under practical constraints on routing, predefined minimum data rates, fronthaul capacity and maximum transmit power. By using convex and continuous relaxation techniques, an iterative algorithm is proposed with guaranteed convergence to a Fritz John solution of the formulated problem. Significantly, each iteration of the proposed algorithm solves only one simple convex program. Numerical examples with practical parameters confirm that the proposed joint optimisation design markedly improves the C-RAN’s energy efficiency compared to benchmark schemes.
Combinatorial optimisation problems often contain uncertainty that has to be taken into account to produce realistic solutions. This uncertainty is usually captured in scenarios, which describe different potential sets of problem parameters based on random distributions or historical data. While efficient algorithmic techniques exist for specific problem classes such as linear programs, there are very few approaches that can handle general Constraint Programming formulations with uncertainty.
Stochastic MiniZinc is an extension to the Constraint Modelling language MiniZinc, that aims to enable operations research practitioners to study the impact of uncertainty on problems they typically solve using deterministic formulations.
Reconciliation of back-casted time series of different aggregation levels using constrained optimisation
Co-author: Steve Xu, Dr Kay Cao, Dr Oksana Honchar
Views expressed in this paper are those of the author(s) and do not necessarily represent those of the Australian Bureau of Statistics. Where quoted or used, they should be attributed clearly to the author.
The Australian Bureau of Statistics (ABS) is going through a transformation period. As part of the transformation, changes to survey processes and estimation methods may result in changes to time series. Methods have been developed to measure statistical impacts that stem from changes for important time series. After statistical impacts have been detected, usually at highly aggregated level, back-casting can be used to adjust for the break in the series before and after method change. Reconciliation is then needed to distribute the impacts to underlying lower level series; while making sure that the lower level series still add up to the higher level series; and time movement pattern (growth rates) of the series are preserved. Using the ABS Labour Force Survey as a case study, analysis has shown that Constrained Optimisation method would be a suitable tool for reconciling series at different levels of aggregation; with time movements preserved, while, at the same time, supporting the incorporation of information about relative impacts to lower level series.
Customized assortment for online retailers considering basket shopping consumer using reinforcement learning
Co-authors: Dr Omar Hussain, Dr Morteza Saberi, Professor Elizabeth Chang
Customized assortment (CA) is a powerful approach to manage demand in e-tailing. CA enhances customer satisfaction as well as e-tailer profit by providing dynamic assortment to the customers. Various studies have been done to optimize the goal of CA by considering different aspects of the problem. However, none consider the fact that most customers in online shopping prefer to buy a basket of items which means they consider the whole utility of a basket and not just one specific item. However, incorporating this fact in CA is important but at the same time is complicated to solve. In this paper, we model the problem as a Markov decision framework and use Reinforcement Learning (RL) to solve the proposed model. RL enable us to tackle curse of modelling by using simulations techniques. To tackle the curse of dimensionality we propose a decomposition-aggregation RL technique. In the decomposition stage, we decompose the problem in to multi sub-problems for each product and solve by RL. State definition for each sub-problem is inventory level of the product as well as an aggregate inventory level of other products. Finally, in the aggregation stage, a total reward is computed by using all the sub problem outcome.
Optimisation in industrial applications: disaster management and signal processing
Co-authors: Dr Julien Ugon, Dr Zahra Roshan Zamir, Dr Behrooz Bodaghi, Mr Aiden Fontes
In this poster, we present two large applications that are based on Optimisation. The first project is disaster management and shift pattern optimisation; and the second one is signal processing. Both projects are based approximation and optimisation (mainly, linear and convex). Some of the results have been published in peer-reviewed optimisation and applied mathematics journals, while some others are the subject of our current research.
Clustering Analysis of Vulnerable Areas in Australia’s Cities Against Disasters
Co-authors: Assistant Professor Mohammad Mojtahedi, Professor Martin Loosemore
The impacts of devastating natural disasters have increased over the last decades due to abundant factors including unplanned urbanisation and population growth. Hundreds of thousands of deaths and billions of dollars of economic loss due to natural disasters highlight the importance of effective disaster operations management (DOM) for reducing the impacts and improving the response to such events. Recognizing the vulnerable area and grouping them is very important for decision-makers to prepare mitigation plans to reduce the disaster impact in vulnerable areas, and response to the post-disaster urgent relief needs through efficient emergency logistics distribution. This study proposes an approach for clustering the identified urban areas in a city into several groups, where the areas with relatively similar vulnerable characteristics are assigned to the same group, and relatively, their preparedness plans, mitigation plans, and post-disaster urgency attributes can be significantly different from those of any other area groups. This study presents an efficient hybrid method algorithm based on combining Metaheuristic algorithms and support vector machine(SVM) for the optimum clustering vulnerable urban areas.
Developing a Comprehensive Disaster Evacuation Planning Model for Urban Areas in Australia
Co-authors: Assistant Professor Mohammad Mojtahedi, Professor Martin Loosemore
Due to an increasing number of disasters, coupling with population growth and the risks of well-unplanned urbanisation, demand for disaster operations management (DOM) has absorbed more attention during the past decade. Evacuation planning is an important component of DOM for protecting lives from disaster and increasing the resiliency of built environment. Although many evacuation studies have been conducted in the past, challenges of evacuating people from cities in response to disaster is still insufficiently explored in Australian urban regions. To fill this research gap, this study introduces a novel mathematical integrated model for evacuation planning in urban regions. Most of the evacuation plans are complex, and frequently characterised by their large-scale sizes and the need for obtaining high-quality solutions in short computing times. Thus, using metaheuristic algorithms to obtain the optimal plan are indispensable. This study also proposes a metaheuristic algorithm which benefit from different random-search strategies to solve the problem in a quicker and optimised way. However, uncertainty is character of a disaster planning, and proposing a hybrid metaheuristic and simulation method can allow decision makers to plan evacuations optimally and finally helping to deal with model uncertainty by integrating simulation into a metaheuristic-driven framework.