Malek Sarhani

Hybrid approach-based support vector machine for electric load forecasting incorporating feature selection

Forecasting future electricity demand is very important for the electric power industry. In fact, it has been shown in several research works that machine learning methods are useful for electric load forecasting (ELF) since electric load data are nonlinear in relation and complex. On the other hand, it is important to determine the irrelevant factors as a preprocessing step for ELF. Our objective in this paper is to investigate the importance of applying the feature selection approach to remove the irrelevant factors of electric load. To this end, we introduce a hybrid machine learning approach that combines support vector machine (SVM) and particle swarm optimisation (PSO) in both continuous and binary forms. Specifically, the binary hybridisation is used for feature selection and the continuous one is used for ELF. Experimental results demonstrate the feasibility of applying feature selection by SVM and PSO algorithms without decreasing the performance of the forecasting model for ELF.

Forecasting Demand with Support Vector Regression Technique Combined with X13-ARIMA-SEATS Method in the Presence of Calendar Effect

In order to better manage and optimize supply chain, a reliable prediction of future demand is needed. The difficulty of forecasting demand is due mainly to the fact that heterogeneous factors may affect it. Analyzing such kind of data by using classical time series forecasting methods, will fail to capture such dependency of factors. This paper is released to present a forecasting approach of two stages which combines the recent methods X13-ARIMA-SEATS and Support Vector Regression (SVR). The aim of the first one is to remove the calendar effect, while the purpose of the second one is to forecast the demand after the removal of this effect. This approach is applied to three different case studies and compared to the forecasting method based on SVR alone.

Hidden Markov Model Classifier for the Adaptive Particle Swarm Optimization

Particle swarm optimization (PSO) is a stochastic algorithm based population that integrates social interactions of animals in nature. Adaptive Particle swarm optimization (APSO) as an amelioration of the original one, improve the performance of global search and gives better efficiency. The APSO defines four evolutionary states: exploration, exploitation, convergence, and jumping out. According to the state, the inertia weight and acceleration coefficients are controlled. In this paper, we integrate Hidden Markov Model Particle swarm optimization (HMM) in APSO to have a stochastic state classification at each iteration. Furthermore, to tackle the problem of the dynamic environment during iterations, an additional online learning for HMM parameters is integrated into the algorithm using online Expectation-Maximization algorithm. We performed evaluations on ten benchmark functions to test the HMM integration inside APSO. Experimental results show that our proposed scheme outperforms other PSO variants in major cases regarding solution accuracy and specially convergence speed.

Facing the Feature Selection Problem with a Binary PSO-GSA Approach

Feature selection has become the focus of much research in many areas where we can face the problem of big data or complex relationship among features. Metaheuristics have gained much attention in solving many practical problems, including feature selection. Our contribution in this paper is to propose a binary hybrid metaheuristic to minimize a fitness function representing a trade-off between the classification error of selecting the feature subset and the corresponding number of features. This algorithm combines particle swarm optimization (PSO) and gravitational search algorithm (GSA). Also, a mutation operator is integrated to enhance population diversity. Experimental results on ten benchmark dataset show that our proposed hybrid method for feature selection can achieve high performance when comparing with other metaheuristic algorithms and well-known feature selection approaches.

Simultaneous feature selection and parameter optimisation of support vector machine using adaptive particle swarm gravitational search algorithm

Particle swarm optimisation PSO and gravitational search algorithm GSA are two metaheuristics that have been used to solve both continuous and discrete problems. Furthermore, their hybridisation can enhance the algorithm performance in these two kinds of problems. However, their utilisation for mixed continuous-discrete problems has not been well investigated. One the other hand, feature selection and parameter optimisation are two important issues in machine learning. The aim of this work is to simultaneously explore both issues, proposing a mixed encoded population and variable representation of PSOGSA in order to select the relevant features and to optimise support vector machine parameters which has proved important predictive ability in feature selection. Furthermore, an adaptive mutation operator has been introduced into the hybrid PSOGSA algorithm. Experimental results on 10 benchmark data sets show that this proposed mixed approach can achieve high performance in both training and testing sets when comparing with PSO, GSA, PSOGSA and the genetic algorithm GA.

A metaheuristic approach for solving the airline maintenance routing with aircraft on ground problem

In the airline industry, the Aircraft Maintenance Routing (AMR) problem has been one of the great successes of operations research. The AMR problem is to determine a particular route for each aircraft to undergo different levels of maintenance checks. The objective is to minimize the total maintenance costs. In this study, our aim is to present a mathematical formulation for the AMR problem which takes into account the case of Aircraft On Ground (AOG). We develop solution approaches based on Particle Swarm Optimization algorithm and Genetic algorithm for solving the problem. The results show the effectiveness of this solution in reducing computational time.

Particle swarm optimization with a mutation operator for solving the preventive aircraft maintenance routing problem

Aircraft Maintenance Routing (AMR) is one of the major optimization problems in the airline industry. In this study, we present a mathematical formulation for the daily AMR problem which aims to minimize the risk of both scheduled and non-scheduled maintenance costs. Exact methods may fail to deal with such problems. Our contribution is then to examine the use of an improved particle swarm optimization (PSO) algorithm by a uniform mutation operator for solving this probabilistic problem. Computational results show that our hybrid approach gives competitive results comparing to the native binary PSO.

Particle swarm optimization algorithm for solving airline crew scheduling problem

In air transport, the cost related to crew members presents one of the most important cost supported by airline companies. The objective of the crew scheduling problem is to determine a minimum-cost set of pairings so that every flight leg is assigned a qualified crew and every pairing satisfies the set of applicable work rules. In this paper, we propose a solution for the crew scheduling problem with Particle Swarm Optimization (PSO) algorithm, this solution approach is compared with the Genetic Algorithm (GA) for both crew pairing and crew assignment problems which are the two part of crew scheduling problem.

An extension of X13-ARIMA-SEATS to forecast islamic holidays effect on logistic activities

To better manage and optimize logistic activities, factors that affect it must be determined: The calendar effect is one of these factors which must be analyzed. Analyzing such kind of data by using classical time series forecasting methods, such as exponential smoothing method and ARIMA model, will fail to capture such variation. This paper is released to present a review of the models which are used to forecast the calendar effect, especially moving holidays effect. We adopt the recent approach of X13-ARIMA-SEATS and extend it for being able to forecast the effect of Islamic holidays. Our extension is applied to Moroccan case studies, and aims to give recommendations concerning this effect on logistic activities.

Particle Swarm Optimization for Model Selection of Aircraft Maintenance Predictive Models

Nowadays, predictive models -especially the ones based on machine learning- are widely used to solve many big data problems. One of the main challenges within predictive models is to choose the best model for each problem. In particular, model selection and feature selection are two important issues in machine learning models as they help to achieve the best results. This paper focuses on the restriction of these two problems to &epsis;---SVR (support vector regression) and more specifically the optimization of both problems using the particle swarm optimization algorithm. Our approach is investigated in the estimation of remaining useful life (RUL) of aircrafts which affects their maintenance planning and which is an interesting issue in predictive maintenances. That is, the experiment consists of predicting RUL of aircraft engines using an &epsis;--- SVR optimized by PSO. Experimental results show the efficiency of the proposed approach.