Graduation Year


Document Type




Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Industrial and Management Systems Engineering

Major Professor

Hadi Charkhgard, Ph.D.

Co-Major Professor

Yu Zhang, Ph.D.

Committee Member

Ankit Shah, Ph.D.

Committee Member

Andrei Barbos, Ph.D.

Committee Member

Changyun Kwon, Ph.D.

Committee Member

He Zhang, Ph.D.


Crowd-sourcing last-mile delivery, Free-floating bike sharing system, Multiplicative programming, Query batching


This dissertation presents four sets of contributions in the field of theory and algorithms for system optimization. In the first set, we introduce a simulation optimization method for redistributing bikes in a free-floating bike sharing system. The second set of contributions is a framework for batching queries in large databases to optimize the data retrieval time. The third set includes two branch-and-bound algorithms to solve minimum multiplicative programming problems and one branch-and-bound algorithm to solve the maximum form of the mentioned problems. At last, the fourth set presents an approach to fairlyassign delivery tasks in an outsourcing last-mile delivery system using Nash Social Welfare solution.

In this dissertations, we propose dynamic hubbing and hybrid rebalancing, i.e. combining user-based and operator-based strategies, along with a novel multi-objective simulation optimization approach to solve bike rebalancing problem in free-floating bike sharing systems as the first set of contributions. Thanks to advances in location and communication technologies, the need for docks and stations in bike sharing systems has been removed. But rebalancing bikes is still a challenging problem in these bike sharing systems. In the proposed model, the users are encouraged to return the bikes to predetermined hubs using an incentive program. Then, for the remaining unbalanced bikes, an operator-based rebalancing operation will be planned. This method determines the number of hubs, their locations, the start time for initiating the user incentive program, and the amount of incentive by considering two conflicting objectives, i.e., level of service and total rebalancing cost. Our results on a real case, Share-A-Bull running on the campus of the University of

South Florida, show that a hybrid rebalancing and dynamic hubbing strategy significantly improves the level of service and reduces the total rebalancing costs. We, also, show this method can entail some promising environmental impacts by decreasing greenhouse gas emissions.

Our second set of contributions presents an approach to optimize query-batching in large database systems. Query batching in database systems is very similar to a well-studied problem called order batching in the warehousing context. However, unlike order batching, the literature of optimization techniques for query batching is outstandingly rare. In this dissertation, we develop a Mixed Binary Quadratic Program (MBQP) to optimize query-batching using predicted query retrieval time. We use a simple regression model for this prediction that can be quickly and dynamically trained. Also, two iterative heuristics are proposed to solve the mentioned problem fast enough. The results on two benchmark databases show that the returning data for batches formed by the proposed method can improve the performance by 19–61.8% compared to returning the required data for all the queries together.

In the third set of contributions, this dissertation presents three branch-and-bound algorithms to solve multiplicative programming problems; two algorithms for minimization form and one for maximization form of the problem. We showed that the algorithms for the minimization form of the problem outperform SCIP, a genetic-purpose solver, by a factor of over 10 on many instances out of 960 test instances. We, also, numerically show that selecting the best algorithm between the proposed algorithms depends on the class of the instances. Specifically, the performance of the algorithms is highly dependant upon the dimension of decision and criterion space of the instances, since the search mechanism in one of the algorithms works over the decision space and the other one over the criterion space. The algorithm for maximization form is limited to only bi-linear maximum multiplicative programs that, yet, has many applications such as finding Nash bargaining solution (Nash social welfare optimization), capacity allocation market, reliability optimization, etc. We proposed several enhancements to tighten the bounds and improve the performance of the algorithm. The result on 400 randomly generated instances is compared with the one obtained by the latest algorithm in the literature and the Second Order Cone Programming solver of CPLEX. We show that the proposed algorithm outperforms the winner of the two mentioned rival methods by the factor of 6.54 on average.

As the last set of contributions in this dissertation, we introduce an equitable crowdsourced last-mile delivery model using Nash social welfare solution for coalition points among different drivers. In this model, the efficiency of the delivery company is guaranteed by putting a cap on the deviation of its cost from the minimum value. Fairness considerations that capture non-monetary performance requirements including equitable service provision to external stakeholders have a long history in routing applications in the public/nonprofit sector. In the private logistics service sector, however, such considerations are new and growing due to public or governmental pressures to improve equity in workload allocation among internal stakeholders, i.e., the drivers or other personnel providing the service. This is more crucial when employing a crowdsourced workforce considering their inherent heterogeneity in terms of skills, availability, and productivity levels. A column generation method is developed to solve and study the behavior of the proposed model. We studied the changes in the company’s cost, drivers’ total profit, and the level of achieved equity among the drivers when the three parameters of the problem vary. We show the superiority of the proposed method to the well-known max-min approach in terms of balancing the workload among the driver and their efficiency. For example, we observe a 50.15% decrease in the average range of profit ratios among drivers only by 5% deviation of the company’s cost from the minimum value.