Learning a fuzzy system from training data using the Münsteraner Optimisation System

Abstract

Abstract-For many classification or controlling problems a set of training data is available. To make best use of this training data it would be ideal to feed the data into a learning algorithm, which then outputs a finished, trained fuzzy controller, that is able to classify or control the original system. For the FUZZ-IEEE 2012 a competition was proposed to predict future volumes sold per day in a certain gas station. The training data includes a collection of gas prices at the current and the competitor's gas station and the according volume sold on every consecutive day in a period of about one year. This training data was analyzed and fit to a fuzzy learning algorithm based on the Münsteraner Optimisation System. As a base point a mean value comparison is used and then different features as fuzzy inputs are tested. Also different fuzzy set widths and and sequence of commands are compared. The final controller chosen shows promising results in the test with left out training data sets. Final results still have to be shown with the test data of the competition.

Efficient Idle Virtual Machine Management for Heterogeneous Cloud using Common Deployment Model

“…The mixed type tasks are analyzed with different parameter. (21) The task allocation in AD is based on the category and PD is based on the idle time of the task which are expressed in the equation (22) and (23) respectively.…”

“…The single dimensional best fit algorithm places the VM by considering the capacity of physical machines, so it leads the performance problem [20], [21]. The volume based best fit algorithm focuses on all dimensions of physical machine and maps to the respective VM [22]. The characteristics of physical machines are represented as a dot product of the vectors for selecting and mapping a best physical machine to VM [23].…”