Sie sind hier: Startseite Aktuelles Aktuelle Nachrichten Pressespiegel

Pressespiegel

Ältere Pressemitteilungen finden Sie im Pressearchiv. 

Publikation im Journal "Computers & Industrial Engineering"

23. September 2021

Unsere Studie "Solving the mixed model sequencing problem with reinforcement learning and metaheuristics" wurde bei der Fachzeitschrift Computers & Industrial Engineering zur Publikation angenommen. Die Studie stellt einen Reinforcement Learning Ansatz für ein kombinatorisches Optimierungsproblem vor, wo die durch Reinforcement Learning erstellte Lösung als Initialisierung für verschiedene Metaheuristiken dient.

Autoren: Janis BrammerBernhard LutzDirk Neumann

 

Abstract:

This study presents a reinforcement learning (RL) approach for the mixed model sequencing (MMS) problem with a minimization of work overload situations. The proposed approach generates the sequence in a constructive way, so that an action denotes the model to be sequenced next. The trained policy quickly creates an initial sequence, which allows us to use the cutoff time to further improve the solution quality with a metaheuristic. Our numerical evaluation based on an existing benchmark dataset shows that our approach is superior to established methods if the demand plan follows its expected distribution from the learning process.

 

Publikation im Journal "OR Spectrum"

31. August 2021

Unsere Studie "Stochastic mixed model sequencing with multiple stations using reinforcement learning and probability quantiles" wurde bei der Fachzeit OR Spectrum zur Publikation angenommen. Die Studie präsentiert einen Reinforcement Learning Ansatz für das Mixed Model Sequencing Problem mit stochastischen Verarbeitungszeiten. Der Zustand wird basierend auf verschiedenen Wahrscheinlichkeitsquantilen modelliert, sodass der Agent lernt, wie sich verschiedene Werte der Verteilung auf die Güte der Lösung auswirken.

Autoren: Janis Brammer, Bernhard Lutz, Dirk Neumann

 

Abstract:

In this study, we propose a reinforcement learning (RL) approach for minimizing the number of work overload situations in the mixed model sequencing (MMS) problem with stochastic processing times. The learning environment simulates stochastic processing times and penalizes work overloads with negative rewards. To account for the stochastic component of the problem, we implement a state representation that species whether work overloads will occur if the processing times are equal to their respective 25%, 50%, and 75% probability quantiles. Thereby, the RL agent is guided towards minimizing the number of overload situations while being provided with statistical information about how fluctuations in processing times affect the solution quality. To the best of our knowledge, this study is the first to consider the stochastic problem variation with a minimization of overload situations.

 

Neue Publikation in "European Journal of Operational Research"

10. August 2021

Unsere Studie "Permutation Flow Shop Scheduling with Multiple Lines and Demand Plans Using Reinforcement Learning" in Kooperation mit Volkswagen wurde beim European Journal of Operational Research zur Publikation angenommen. Die Studie beschreibt einen Ansatz zur Verwendung von Reinforcement Learning für das Permutation Flow Shop Problem mit mehreren Fertigungslinien und Demand Plans.

Autoren: Jannis Brammer, Bernhard Lutz, Dirk Neumann

 

Abstract:

Existing studies on the permutation flow shop problem (PFSP) commonly assume that jobs are produced on a single line. However, manufacturers may speed up their production by employing multiple lines, where each line produces sub-parts of the final product; which must be assembled by a synchronization machine. This study presents a novel reinforcement learning (RL) approach for the PFSP with multiple lines and demand plans. Our approach differs from existing RL-based scheduling methods as we train the policy to directly generate the sequence in an iterative way, where actions denote the job type to be sequenced next. During cutoff time, we follow a multistart approach that generates sequences with the trained policy, which are subsequently optimized by local search. Our numerical evaluation based on 1,050 problem instances with up to three production lines shows that our approach outperforms existing methods on the multi-line problems for short cutoff times, while there is a tie with existing methods for medium and long cutoff times. A further analysis suggests that our approach can also be applied to problems with imbalanced demand plans.

 

Publikation im Journal "Applied Energy"

8. Juli 2021

Unsere Studie "Predicting Residential Electricity Consumption Using Aerial and Street View Images" wurde in das Journal "Applied Energy" aufgenommen.

Autoren: Rosenfelder M, Wussow M, Gust G, Cremades R, Neumann D

 

Abstract:

Reducing the electricity consumption of buildings is an important lever in the global effort to reduce greenhouse gas emissions. However, for privacy and other reasons, there is a lack of data on building electricity consumption. As a consequence, data-driven tools that support decision-makers in this area are scarce. To address this problem, we present an innovative approach to modeling building electricity consumption that relies exclusively on publicly available aerial and street view images. We evaluate our approach in a case study based on real world data from Gainesville, Florida. The results show that our model can predict electricity consumption about as well as conventional models, which are trained on commonly used features that are typically not publicly available at a large scale. Furthermore, our model achieves 68% of the potential accuracy improvements of a model relying on an extensive set of fine-grained tabular features. Spatially aggregating the predictions from the level of buildings to areas of up to 1km² further improves the results.

 

Publikation im Journal "Manufacturing and Service Operations Management" (MSOM)

18. Mai 2021

Unsere Studie "Prescriptive Analytics in Urban Policing Operations" wurde in das Journal "Manufacturing and Service Operations Management" (MSOM) aufgenommen.

Autoren: T. Brandt, O. Dlugosch, A. Abdelwahed, P. van den Berg, D. Neumann

 

Abstract:

Problem definition. We consider the case of prescriptive policing, i.e. the data-driven assignment of police cars to different areas of a city. We analyze key problems with respect to prediction, optimization, and evaluation, as well as trade-offs between different quality measures and crime types.
Academic / Practical Relevance. Data-driven prescriptive analytics is gaining substantial attention in operations management research, while effective policing is at the core of the operations of almost every city in the world. Given the vast amounts of data increasingly collected within smart city initiatives and the growing safety challenges faced by urban centers worldwide, our work provides novel insights on the development and evaluation of prescriptive analytics applications in an urban context.
Methodology. We conduct a computational study using crime and auxiliary data on the city of San Francisco. We analyze both strong and weak prediction methods along with two optimization formulations representing the deterrence and response time impact of police vehicle allocations. We analyze trade-offs between these effects and between different crime types.
Results. We find that even weaker prediction methods can produce Pareto-efficient outcomes with respect to deterrence and response time. We identify three different archetypes of combinations of prediction methods and optimization objectives that constitute the Pareto frontier among the configurations we analyze. Furthermore, optimizing for multiple crime types biases allocations in a way that generally decreases single-type performance along one outcome metric, but can improve it along the other.
Managerial Implications. While optimization integrating all relevant crime types is theoretically possible, it is practically challenging since each crime type requires a collectively consistent weight. We present a framework combining prediction and optimization for a subset of key crime types with exploring the impact on the remaining types to support implementation of operations-focused smart city solutions in practice.