Solving the Examination Timetabling Problem in GPUs (original) (raw)
Related papers
A GPU Based Genetic Algorithm Solution for the Timetabling Problem
The university course timetabling problem (UCTP) is a combinatorial optimization problem of great importance for every university. This paper proposes the use of a parallel evolutionary algorithm to solve the problem and focuses on accelerating the process for specifically very large sized problems. The problem was solved using the genetic algorithm, and accelerated with the use of the Graphics Processing Units (GPUs) capabilities in order to use very large population sizes and explore the problem solution space in a much exhaustive manner. The genetic algorithm was also enhanced with the use of local search, and allowed to deal flexibly with the incremental changes of the problem's constraints while maintaining the resulting solution with minimal changes. The implementation of the proposed work was tested with the ITC2007 datasets as the benchmark set.
Development of a Hybrid Tabu Search and Genetic Algorithms for the Examination Timetabling Problem
NIPES Journal of Science and Technology Research, 2020
Genetic Algorithms (GA) and Tabu Search Algorithm (TSA) are amongst the leading research approaches for solving the Examination Timetabling Problem (ETP), however, both algorithms are not optimal. GA returns poor solution, uses excessive memory, experience damage to solution during crossover while solving the ETP. TSA consumes much time, can easily miss some regions of the search space since it uses one solution, and may fail to generate some neighborhood candidate solution. TSA also selects best solution based on the current steps without taking future steps into consideration. This research developed a hybrid of GA and TSA, the GATS algorithm, with the aim of mitigating against the GA's and TS weaknesses to produce higher quality results when solving the ETP. The ETP was modeled as an optimization problem, implemented in Java for the three algorithms and experimented with dataset from Bells University of Technology, Ota. The algorithms' performances were evaluated using first Order Conflict Counts (OCC) and second OCC for students and invigilators respectively, as well as with space complexity. The GA, TSA and GATSA yielded average first Order Conflict Counts (OCC) of 0.0, 0.0 and 0.0 for both students and invigilators. They yielded average second OCC of 5228.5, 18.8 and 0.7 for students and, 0.0, 0.0 and 0.0 for invigilators respectively. The Developed GATSA produced higher quality timetables than TSA and GA, and consumes similar amount of memory as the TSA and has an empirical space complexity of O(n).
A Review of Genetic Algorithm Application in Examination Timetabling Problem
Many studies have used different types of genetic algorithm in solving examination timetabling problem. The solutions of the genetic algorithm are found to be efficient and reliable. This study provides a comprehensive review of genetic algorithm application in examination timetabling. It presents many examples of using genetic algorithm in finding optimal solutions to the problem of examination timetabling at universities or institutions. Subsequently, it presents the most used techniques to solve the examination timetabling problem such as tabu search and simulated annealing techniques. The objective of the study is to provide an understanding on what have been achieved in solving examination timetabling problem.
This paper presents a Genetic Algorithm approach to solve a specific examination timetabling problem which is common in Japanese Universities. The model is programmed in Excel VBA programming language, which can be run on the Microsoft Office Excel worksheets directly. The model uses direct chromosome representation. To satisfy hard and soft constraints, constraint-based initialization operation, constraint-based crossover operation and penalty points system are implemented. To further improve the result quality of the algorithm, this paper designed an improvement called initial population pre-training. The proposed model was tested by the real data from Sophia University, Tokyo, Japan. The model shows acceptable results, and the comparison of results proves that the initial population pre-training approach can improve the result quality.
Arranging examination timetable is problematic. It differs from other timetabling problems in terms of conditions. A complete timetable must reach several requirements involving course, group of student sitting the exam in that course, etc. It is similar to the course's timetable but not the same. Many differences between them include the way to create and the requirements. This paper proposes an adaptive genetic algorithm model applied for improving effectiveness of automatic arranging examination timetable. Hard constraints and soft constraints for this specific problem were discussed. In addition, the genetic elements were designed and the penalty cost function was proposed. Three genetic operators: crossover, mutation, and selection were employed. A simulation was conducted to obtain some results. The results show that the proposed GA model works well in arranging an examination timetable. With 0.75 crossover rate, there is no hard constraints appeared in the timetable.
Parallel Genetic Algorithms for University Scheduling Problem
International Journal of Electrical and Computer Engineering (IJECE), 2017
University scheduling timetabling problem, falls into NP hard problems. Researchers have tried with many techniques to find the most suitable and fastest way for solving the problem. With the emergence of multi-core systems, the parallel implementation was considered for finding the solution. Our approaches attempt to combine several techniques in two algorithms: coarse grained algorithm and multi thread tournament algorithm. The results obtained from two algorithms are compared, using an algorithm evaluation function. Considering execution time, the coarse grained algorithm performed twice better than the multi thread algorithm. 1. INTRODUCTION Every institution needs a schedule to represent their functionality. If the institution is small and not so complex, they may choose to draft their schedule manually. Otherwise, if the institution is bigger and needs to represent more complex relations between its functions, they may choose to use an application that generates the schedule automatically. Even then, some of the schedules are harder to create and manage than the others like nurse scheduling problem [1]. One that falls into the hardest ones, is also the university course timetabling problem, which has been proved decades ago, that it represents a NP problem [2]. University schedule needs to represent the location and time of all courses held in a semester, or year, in the university. By doing this, it achieves its main goal: information of the students and professors about the lectures to be held. Another important aspect of the schedule, is trying to accommodate the needs of the professors and students. So, first consideration is on finding the schedule that fits within university working hours and physical space available. Secondly, the schedule can be improved in order to accommodate needs of the professors and students such as not so long working hours, not subsequent lectures, taking into consideration breaks and even personal preferences such as lectures in the morning or in the afternoon. In [3] [4] authors have described proposed usage of the distributed approach for solving the university timetabling problem, mainly due to the emergence of multi core systems. Furthermore, they argue that effectiveness of the parallel processing, need to be studied additionally. Taking this into the account, we have proposed two versions of roulette genetic algorithms: one based on islands and the other one on threads, to improve solution of the university timetabling problem. Therefore, one of the main aims of the paper is to use the parallel processing algorithm to obtain better results. That has been proven also in [5] and [6] but our solution represents first of all a fusion of a model used in [5] and another in [6] and in addition the tournament usage is done in multi thread. According to our experiments the tournament selection method allows for very quick solutions in light-constrained problems, using a fine-grained parallelism method, and roulette selection can be used in a coarse-grained parallel algorithm with slightly slower but better results in even harder problems Therefore, the advantage of our model is in cases of bigger population. This paper is further organized as followed: section 2 describes the related work on the problem with emphasis on genetic algorithms used. Section 3 described in depth the problem and its constraints, while section 4 describes