The Research Training Group (RTG) "Advanced Optimization in a Networked Economy" is jointly hosted by the TUM School of Management , the faculties of Computer Science, and Mathematics at Technical University of Munich. Due to its interdisciplinary focus, the RTG is part of the IGSSE within the TUM Graduate School.
Efficient resource allocation is key to the profitability of firms, and for the prosperity and sustainability of an economy. Resource allocation is therefore a central topic in the management sciences and in economics. Although, there have been substantial algorithmic advances in mathematical optimization in the past two decades, the challenges of a networked economy are fundamental, and they demand for new research at the intersection of management science, informatics, and mathematics. In particular, it requires new approaches, methods, and topics taught in graduate education in operations research and management science.
Future research on advanced optimization in a networked economy will require us to broaden the study and training program of researchers in the following three areas: new computational techniques to solve large problems and handle large data sets, the consideration of incentives in situations with multiple decision makers, and the consideration of dynamism and uncertainty in the parameters of resource allocation problems.
Solving real-world problems has always been a strength of operations research. For the development of new methods and models, it is important to understand requirements of real problems and have real-world data available. This allows synergies across different application domains, and then generalize to more abstract models and theory relevant to management and industry. The research program focuses on important application-oriented projects that would benefit from an interdisciplinary approach, including shared mobility, resource pooling, retail coordination, sales and operations planning, IT management, and airport operations.
Students with a background in one of the three involved disciplines and interest in at least one other discipline will undergo a comprehensive study, training and research program. Specifically designed interdisciplinary projects between Management Science, Informatics, and Mathematics will help develop the ability to model, analyze, solve, and implement solutions for challenging management problems relevant in practice.
In many major cities world-wide, shared mobility has become a major example for a general trend towards a Share Economy. The task of designing and operating such systems poses new scientific challenges (e.g. for robust optimization with stochastic modeling, handling large data, devising incentive based control, and economic evaluation).
In particular, several strategic and operational problems under uncertainty need to be addressed:
- the design of an optimal fleet - including the use of electro mobility;
- the deployment and repositioning of cars (relocation);
- the development of an incentive-based fee system, etc.
The first problem includes the analysis of the occurring (or expected) demand. A prototype application is that of a municipal provider (e.g. road maintenance service, gas works etc.) who needs vehicles of different sizes, capacity, provided with equipment adapted to the relevant requirements. Also, the demand for electro mobility adds additional challenges to the problem.
The second task deals with the local availability of the vehicles. Demands may indeed occur at a location where no appropriate resources are available, while other locations still have resources at hand. Hence transshipment or repositioning of resources is required. Available data has to be analyzed and used for the optimization of car pool sizes, car deployment and repositioning decisions in complex mobility networks. Autonomous driving provides additional opportunities here for achieving a desirable distribution of resources over a geographical area.
The third issue is that of incentives for both actually winning customers and inviting a certain customer behavior. Of course, the general question is, which (positive or negative) incentives should be presented to customers for which behavior. For instance, it is certainly desirable that customers either recharge the batteries of an electro vehicle themselves or deliver it to a charging station. Since recharging currently requires considerable planning and time, one would need to design incentives based on the current status of the batteries. Also, in order to win sustainable business, monthly rates may be offered that come with some sort of guarantee that a vehicle (of appropriate type) is indeed available at any time at a certain location or within a certain range.
Sharing resources in the field of inventory management is well known as pooling. There exist different kinds of inventory pooling, either physically by centralizing inventories at a depot or virtually by sharing local inventories through transshipments. Earlier work mainly focused on the statistical effect of pooling demand and supply risk, whereas more recent work analyzed the problem from a distributed decision making perspective. The increasing availability and visibility of real time data about inventory levels at central and local warehouses and the position of in-transit deliveries in a networked economy provides a significant opportunity to increase fulfillment efficiency and flexibility.
The challenges of inventory pooling in a networked economy are the multiple opportunities to improve performance by available real time information and the willingness to truthfully share it. This includes the value of installed base and condition information in spare parts applications and the stock and delivery status of multiple (decentralized) warehouses in distribution systems. There exist multiple decision makers having distributed private information. Cooperative game theory is only one extreme approach to distribute gains from horizontal co-operations which has often proved to not yield stable negotiation outcomes in repeated decision environments under symmetric information. Therefore, more sophisticated mechanisms to truthfully share installed base and demand forecast information with ex-post satisfactory gain/cost allocations need to be developed.
The research project includes design and operational tasks. The warehouse location of pools and safety stock placement decisions are of a more strategic/tactical nature whereas the inventory deployment needs to be carried out on a short term, in an extreme case, real time fashion when a customer request arrives and the shipping source needs to be confirmed. In a distributed decision environment, a set of stable transfer pricing mechanisms is necessary to simultaneously achieve the goal of truthful initial information sharing when dimensioning the initial pool of capacities (resources) and an in hindsight cost allocation mechanism perceived as being fair and thereby ensuring a stable ongoing horizontal logistics cooperation to pool inventories.
Carriers in retail logistics often face congestion at warehouses at certain times of the day which force them to wait before their trucks can be unloaded. Warehouses suffer from unbalanced utilization due to a lack of coordination between carriers. This has a significant negative impact on logistics efficiency. According to a survey among more than 500 transport companies, 18\% of them have an average waiting time of more than two hours and 51\% have an average waiting time between one to two hours at each warehouse. Another problem is insufficient parking capacity for waiting trucks.
The main reasons for waiting times include shortages of resources (staff and infrastructure) and uncoordinated arrivals of trucks, especially at peak times. These problems are interconnected, since uncoordinated arrivals of trucks make appropriate staffing difficult. In addition, congestion can interfere with other processes due to a high number of trucks at the facility. Proposed remedies for these problems are time slot management, and increased infrastructure capacities. Capacity increases require high investments compared to improved coordination.
We plan to develop new or adapt existing mechanisms for the coordination of retailers in our domain. There are several forms of coordination among the market participants to mitigate the problem: Carriers can make a reservation for a time slot with a retailer via a first-come, first-served system (with or without payments). Alternatively, coordination can be facilitated via an economic mechanism where carriers express their preferences for various routes and respective time slots and the mechanism assigns time slots to carriers such that waiting times are avoided as far as possible while meeting the carriers' preferences, i.e., the routes and times of the day causing the least costs for him.
This project area includes all three major research challenges of a networked economy. Due to the number of retail warehouses and carriers and vehicles involved, already sub-problems to be solved by individual decision makers are large-scale optimization problems under uncertainty. For efficient outcomes, coordination and allocation of resources in a decentralized, multiple decision maker environment is necessary.
The sales and operations planning (S&OP) process involves a multitude of decision makers from different functions of an organization and aims at aligning the demand for products with the supply. This alignment process is carried out at different levels. The interests of the multiple decision makers involved in S&OP are diverse and partly conflicting. On the tactical level, the sales organization is interested in filling all customer requests that may potentially come up, and therefore tends to over-forecast the demand. This leads to a misallocation of production resources, an increase in inventory, and reduced profits. Similar effects also occur on the operational level. Here, a given supply of products is allocated. It is widely known that customers inflate their forecasts in supply shortage situations to game the allocation mechanism of their supplier and to ensure sufficient supply. This, again, leads to an inefficient allocation with decreased overall service levels and an increased inventory. However, with the recent advances in IT tools, companies are now able to monitor the ordering behavior of their customers on the granularity level of individual customers and individual products. The higher transparency of the customers' ordering behavior provides opportunities to increase the efficiency of supply allocations.
Key challenges in S&OP are therefore to address forecast uncertainty and the strategic behavior of multiple decision makers. To provide decision support for practice, we combine two prescriptive methodologies: Mechanism design and optimization. We aim at designing mechanisms for the dynamic S&OP process, which set incentives for the stakeholders to report their forecasts truthfully. We also aim at developing optimization approaches for the resulting large-scale resource allocations problems, which exploit the available information. Another venue is to investigate behavioral aspects in S&OP group forecasting and decision-making behavior. All developed approaches will be tested based on data from the semiconductor industry and the chemical industry.
Information and communication infrastructure provides the backbone of today's networked economy. The fact that service development and provisioning has become highly automated brings down operating costs and facilitates many new business models that were impossible only a few years ago. New technological developments have led to substantial changes in the way how this infrastructure is being provided. The shift to cloud computing and virtualization is arguably the most influential development in IT service management. Infrastructure-as-a-Service (IaaS) providers account for an important part of the cloud service market, hosting virtual machines (VMs) for their customers, but there is no dedicated allocation to a physical server anymore. Live migration allows to move VMs from one server to another during runtime seamlessly.
The consideration of demand complementarities in the demand patterns of VMs allows for a significant reduction in the number of active servers and an increase in the energy efficiency of data centers. Such exogenously given demand patterns lead to new, very high-dimensional bin-packing and scheduling problems, which have received little attention in the literature so far.
Moreover, the order and type of VMs requested by the customers is typically unknown and demands for algorithms that cope with incomplete information. Standard competitive analysis, which evaluates the performance of an online algorithm under an adversarial worst-case arrival sequence, may be too pessimistic and does not always lead to meaningful results. Recent research on an alternative, random-order model has produced significant results---for maximization problems. Minimization problems, such as minimizing cost, loads or waiting times when allocating resources in data centers, appear technically much more challenging since "wrong'' decisions cannot be neglected when appearing with bounded probability but may cause severely increased cost. Thus, different techniques and analytical methods are needed. Moreover, data centers often possess additional distributional information about the VM request sequence which demands for dynamic stochastic scheduling policies tailored to the specific requirements of data centers.
Increasing air travel by about 5.3% per year and increasing freight transportation by more than 4% are demanding sophisticated approaches in order to use existing airport resources and to predict travel and freight demand precisely. We want to address two problems.
1. Operations of Airport Resources
The baggage handling system of airports is a crucial resource, in particular at hub airports. It consists of several sub-systems (such as check-in counters, infeed stations, baggage warehouses, claim carousels, loading carousels) that are connected by a conveyor belt network. The sub-systems are operated by workers (e.g. loading bags into containers at the loading carousel). All baggage traversing at airport is routed through the baggage handling system. Within the airport industry, the baggage handling systems are mainly manually planned and controlled by human dispatchers with the help of IT-based planning systems. These systems provide information but do not apply optimization techniques. Based on existing work (see Frey et al. 2017) we want to develop models and methods in order to optimize the operations of the baggage handling system.
2. Air Cargo Prediction
Accurate prediction of air cargo is required in order to plan required capacity accordingly. Prediction is particularly important as freight is paid for shipped and not for reserved cargo, leading to deviations between reserved and shipped cargo. Furthermore, relevant parameters (such as size, weight, category or type of cargo) are of different data types (continuous vs. discrete, ordered vs. unordered) that need to be unified. Based on previous work (see Brieden and Gritzmann 2012) and its subsequent applications we want to develop appropriate transformations of parameters into a unified space and develop predictive algorithms based on appropriately clustered homogenous substructures.