Hanan Kamal

Solving curve fitting problems using genetic programming

Genetic programming is a branch of genetic algorithms. The main difference between genetic programming and genetic algorithms is the representation of the solution. Genetic programming creates computer programs in LISP computer language as the solution whereas genetic algorithms create a string of numbers that represent the solution (see Holland, J.H., 1975). The new way of representation used in GP encouraged researchers to use it in solving design problems where the size and shape of the solution is unknown (see Koza, J.R., 1992). Curve fitting problems used to be solved by assuming the equation shape or degree then searching for the parameter values as done in regression techniques. This paper demonstrates that curve fitting problems can be solved using GP without need to assume the equation shape. An object oriented technique has been used to design and implement a general purpose GP engine.

Genetic algorithms in solving graph partitioning problem

The Graph Partitioning Problem (GPP) is one of the fundamental multimodal combinatorial problems that has many applications in computer science. Many algorithms have been devised to obtain a reasonable approximate solution for the GP problem. This paper applies different Genetic Algorithms in solving GP problem. In addition to using the Simple Genetic Algorithm (SGA), it introduces a new genetic algorithm named the Adaptive Population Genetic Algorithm (APGA) that overcomes the premature convergence of SGA. The paper also presents a new approach using niching methods for solving GPP as a multimodal optimization problem. The paper also presents a comparison between the four genetic algorithms; Simple Genetic Algorithm (SGA), Adaptive Population Genetic Algorithm (APGA) and the two niching methods; Sharing and Deterministic Crowding. when applied to the graph partitioning problem. Results proved the superiority of APGA over SGA and the ability of niching methods in obtaining a set of multiple good solutions.

A new fuzzy logic technique for handover parameters optimization in LTE

One of the main goals of LTE is to provide seamless and fast handover from one cell to another to meet a strict requirements. Hence, the decision of handover is a critical part in the design process of handover. The design of an efficient and successful handover requires a careful selection of handover (HO) parameters and the optimal setting of these. In this paper a new handover optimization algorithm for long term evolution (LTE) network based on fuzzy logic is presented. It consists of finding the optimum handover margin (HOM) required for handover process and also finding appropriate time-to-trigger (TTT) to perform a success handover using fuzzy logic. The proposed handover optimization technique is evaluated and compared with the four well-known handover algorithms and achieves minimum average number of handover per user and also have maximum throughput than the self-optimization technique.

Optimization of user behavior based handover using fuzzy Q-learning for LTE networks

In LTE networks, handover optimization is necessary to enhance the users’ satisfaction. Specifically, users using real time traffic need to experience continuous connectivity. Hence, radio link failures (RLFs) severely affect their quality of experience. Decreasing the RLFs for non-real time users is not as urgent as the case of real time users. On the other hand, a total network collapse can happen in case of too much unnecessary handovers (ping-pongs). In this work, fuzzy Q-learning is used to optimize the two contradictory handover problems, which are RLFs and ping-pongs. The former needs to decrease Handover Margin (HOM) to reduce the too late handover, and the latter needs to increase the HOM to reduce the unnecessary signaling. In the developed algorithm, the users in the network are divided into four categories, according to their speed and the data traffic used. This increases the satisfaction of some users, while keeping the overall handover problems within acceptable limits. For each category of users, fuzzy Q-learning is applied with a different initial candidate fuzzy actions. The proposed technique shows the best performance for each category of users in terms of the most preferred metric, either decreasing RLF or decreasing ping-pongs, for this category of users in comparison with two other literature techniques, or without using any optimization technique. Moreover, the algorithm is robust against changes in the number of users in the system, as it maintains the best solution when the number of users is halved or even doubled.

An enhanced fuzzy logic optimization technique based on user mobility for LTE handover

In previous cellular networks, changing the parameters of handover manually was time consuming. So, automatic change of the parameters of handover will decrease expenditures, save time and guarantee the best quality of service. A new efficient handover optimization technique for Long Term Evolution (LTE) based on fuzzy logic is proposed in this paper. It is evaluated in terms of average number of handover, system throughput, and system delay based on user equipment (UE) speed. The most important contribution of this work that UE speed is considered in the handover process. The suggested optimization technique will select effective time-to-trigger (TTT) based on the UE speed, and it will also find the optimum handover margin (HOM) required for the handover process. This new handover optimization technique is evaluated for the four well-known handover algorithms then compared together. The proposed algorithm achieves a significantly improve in the handover performance when compared with the standard LTE and self-optimization technique.

Optimal control of fed batch fermentation processes using adaptive genetic algorithm

This paper introduces an application of genetic algorithm (GA) in determining the best feed rate profile for the penicillin G fed-batch fermentation, using a mathematical model based on balancing methods. The proposed algorithm consists of the simple GA with variations in the population representation and a new adaptive mutation mechanism. The results are compared with those obtained by applying simple GA and standard optimal control theory. It has been shown that the developed adaptive mutation is faster in convergence and the obtained solutions are of higher fitness than previous work.

Controller reduction using genetic algorithm

In this paper we introduce a classical method for controller reduction based on the optimal LQG-controller, and modify this method by using the genetic algorithm. The genetic algorithm is used to generate the optimum values of the weighting matrix R of the cost function to minimize the maximum real part of the closed-loop eigenvalues. The closed-loop system is built for both algorithms, and closed-loop, eigenvalues are calculated. Comparison between the two algorithms is discussed from the point of view of stability.

Unified chromosome representation for large scale problems

Genetic Algorithms have been successfully applied to the function optimization problem. However, the main disadvantage of this technique is its large chromosome length and hence long conversion time specially when applied to functions with a large number of parameters. In this paper, a new chromosome representation scheme that reduces the chromosome length is proposed. The scheme is also domain independent and may be used with any function. Results and a comparison between the conventional chromosome representation and the proposed one are presented.