فصلنامه هوش محاسباتی در مهندسی برق
سال پنجم شماره 1 (بهار 1393)

In this paper a novel unsupervised feature extraction method for detection of melanoma in skin images is presented. First of all، normal skin surrounding the lesion is removed in a segmentation process. In the next step، some shape and texture features are extracted from the output image of the first step: GLCM، GLRLM، the proposed directional-frequency features، and some parameters of Ripplet transform are used as texture features; Also، NRL features and Zernike moments are used as shape features. Totally، 63 texture features and 31 shape features are extracted. Finally، the number of extracted features is reduced using PCA method and a proposed method based on Fisher criteria. Extracted features are classified using the Perceptron Neural Networks، Support Vector Machine، 4-NN، and Naïve Bayes. The results show that SVM has the best performance. The proposed algorithm is applied on a database that consists of 160 labeled images. The overall results confirm the superiority of the proposed method in both accuracy and reliability over previous works.

Dynamic economic dispatch deals with the scheduling of online generator outputs with predicted load demands over a certain period of time so that an electric power system operates most economically. This paper proposes a hybrid methodology integrating Bacterial Foraging Optimization Algorithm (BFOA) with sequential quadratic programming (SQP) for solving dynamic economic dispatch problem of generating units considering valve-point effects. This hybrid method incorporates Bacterial Foraging Optimization Algorithm as a base level search which can give a good direction to the optimal region and sequential quadratic programming as a local search procedure which is used to fine tune that region for achieving the final solution. Numerical results of a ten-unit system have been presented to demonstrate the performance and applicability of the proposed method. The results obtained from the proposed method are compared with those obtained from hybrid of particle swarm optimization and sequential quadratic programming and hybrid of evolutionary programming and sequential quadratic programming.

A new method for model reduction of linear systems is presented based on Laguerre Polynomials using Harmony Search (HS) algorithm. First، a full order system is expanded and then a set of parameters in a fixed structure are determined، whose values define the reduced order system. The values are obtained by minimizing the errors between the first coefficients of Laguerre Polynomials of full and reduced systems using HS algorithm. To satisfy the stability، Routh criterion is used as constraints in optimization problem. To present the ability of the proposed method، three test systems are reduced. The results obtained are compared with other existing techniques. The results obtained show the accuracy and efficiency of the proposed method.

This paper presents new fuzzy sets by imitating from the neuronal structure and the fuzziness and chaotic dynamics in human brain. To constitute the proposed sets، coupled chaotic oscillators are utilized. The major advantage of these sets، in comparison with the existing fuzzy sets، is the bifurcation capability. This property enables the sets to create various fuzzy sets، such as type-1 or type-2 fuzzy sets; convex or non-convex fuzzy sets; etc. To evaluate the proposed sets in modeling uncertainties، a framework is presented to design neuro-chaotic fuzzy systems، and then it is applied to the problem of forecasting Mackey-Glass time series that is corrupted by an additive noise with certain SNRs. Results show that the proposed system، in comparison with type-1 and interval type-2 neuro-fuzzy systems، has lower prediction error for both training and test datasets.

Recognition of handwritten digits is one of the most important problems in Optical Character Recognition (OCR) domain. In this paper a combination of two features، gradient histogram and modified characteristic Loci، are used for Persian handwritten digits. Furthermore، the most important features are selected using improved New Binary Particle Swarm Optimization algorithm (INBPSO) with an appropriate fitness function. SVM is used for classification of the digits. Having the selection pattern found in the training phase، the extracted features of the test samples are reduced using this pattern and then the final vector of the selected features are classified with the trained SVM model. The proposed method is applied on HODA database. Without reducing the features we achieved 99. 40% accuracy and after reducing، the accuracy of 99. 28% is reached. Comparing the results with the previous works، indicates that the proposed method has better performance in feature extraction and feature selection.

In Copy-Move forgery، a part of an image is copied and pasted to another place at the same image. This type of manipulation is performed for add details or to conceal unwanted portions of the image. In this paper، we present an improved method using Discrete Wavelet Transform and Cosine Transform Coefficients Decomposition to detect Copy-Move forgery and its challenges. In the proposed method، with utilizing the nature of Wavelet Transform، Singular Value Decomposition and lexicographically a sorting salient improvement is yielded in comparison with relative algorithms. Although a reduction of detection accuracy occurs، it is negligible because of the increase in speed. Since human eye is more sensitive to lower frequencies، Cosine Transform can successfully detect the forged part even when the copied area is manipulated with additional operation such as retouch. Experimental results show that the proposed scheme accurately detects such specific image manipulations as long as the copied region is not rotated or scaled.

Today modern power systems are operated in lower security level due to power system deregulation and increasing the power transfer capacity. Extensive power systems blackouts in recent years show the remarkable increase of power system vulnerability in contingency situations. Load shedding is one of the last corrective actions for keeping power system stability. In this paper a real time optimal under frequency load shedding by using artificial neural network is presented. This structure contains two offline and online studies. In offline studies according to the values of vulnerability and security margin indices of total power system، minimum frequency، reduction rate of equivalent inertial center frequency (dfc/dt) for N-K contingency scenarios، the power system security is determined and the ANN inputs data base will be established. In each scenario، the necessary active and reactive load shedding value for preserving power system stability is determined by solving an offline optimization problem by using intelligent hybrid CPCE algorithm. The values of the active and reactive load shedding in each load shedding step in each contingency scenario are considered as the ANN outputs. Genetic algorithm is employed for optimizing the ANN training process. The trained ANN will be used for online application in power system by using real time operation information that is collected by wide area monitoring system (WAMS) and phasor measurement units (PMU). Simulation results for IEEE 118-bus test system shows the effectiveness of the proposed method.

The price signal in a competitive electricity market has a major importance in all planning and commissioning activities. Also، the electricity price has a non-deterministic nature and is affected by various parameters in short and long terms. Active players in electricity market need accurate and effective price forecasting for risk management. With the increased use of renewable energies، especially wind energy، the electricity price is being affected by this new parameter، as the intermittent nature of wind generation has further complicated the process of instantaneous balancing of power system demand against power generation. In this paper، using the Nord Pool electricity market data، the effect of wind units'' generation on price forecasting is studied. The main idea is based on presenting an intelligent model for forecasting the Market Clearing Price through the use of a multilayer perceptron neural network based on hybrid genetic model and Imperialist Competitive algorithm. This hybrid model has a better accuracy، compared to the conventional neural networks (based on gradient-based optimization algorithms)، and has the ability of converging towards the absolute optimum. The results verify the high accuracy of this model in short term electricity price forecasting.