فصلنامه هوش محاسباتی در مهندسی برق
سال ششم شماره 2 (تابستان 1394)

In this paper, a bidirectional pre-training method for initializing weights of hetero-associative deep neural network was presented. Training of deep neural networks, because of confrontation with a large number of local minima, is not often converged. This is while through proper initializing weights instead of random values at the beginning of the training; it is possible to avoid many local minima. The bidirectional layer-by-layer pre-training method pre-train weights in forward and backward manners in parallel. Afterwards, the weight values resulted from their training are applied in the deep neural network. The bidirectional layer-by-layer pre-training was applied for pre-training of the classifier deep neural network weights, and revealed that both the training speed and the recognition rate were improved in Bosphorus and CK+ databases.

This paper proposes a new method for designing fractional order PID to control boost converter. FOPID is a PID controller where the integration and derivation orders are of fractional order rather than integer. In this paper FOPID controller is used for controlling the boost converter, and Imperialist Competitive Algorithm is employed to determine FOPID parameters because of its good performance and high accuracy. To illustrate the performance of the proposed controller, some simulations have been carried out in MATLAB and the results have been compared with Genetic algorithm. Moreover, FOPID controller has been compared with PID and PI controllers optimized by ICA. The Simulation results illustrate the good performance of the proposed controller.

Multi-FPGA systems are an alternative to the reconfiguration limitation of a single FPGA, when a number of real-time tasks arrive together and must be scheduled and executed before a specified deadline. Energy consumption, consisting of static and dynamic, is an important factor in such systems, especially when used as battery powered applications. In this paper, heterogeneous multi-FPGA system is proposed to reduce the static energy consumption of a multi-FPGA system. Ant colony optimization (ACO) is used to schedule the real-time tasks that periodically enter the system. The number of these tasks, as the case of many applications, supposed to be different from one period to another. The consumption energy is estimated for both the homogeneous and heterogeneous systems. Results show that the heterogeneous system saves 6.44 percent of energy, compared with the homogeneous system. If the number of tasks, and so the required number of blocks are small, this amount could be much higher.

This paper proposes an algorithm for the optimal location and control of Flexible AC Transmission System (FACTS) devices for enhancing the loadability in transmission system using particle swarm optimization (PSO) and Imperialist Competitve Algorithm (ICA) for pool and hybrid model in deregulated electricity market. This approach uses AC load flow equations with the constraints on power generation, transmission line flow, magnitude of bus voltages and FACTS device settings. The bilateral transactions are modeled using secured bilateral transaction matrix utilizing AC distribution factor with the slack bus contribution. In this proposed approach, Thyristor Controlled Series Compensator (TCSC) is used. To validate the proposed approach, simulations are performed on 39-bus New England test system and IEEE 118-bus system. Comparisons are made in terms of maximum loadability, computation time. The simulation results obtained indicate that by optimal location and control of TCSC, using ICA, the loadability in transmission system is enhanced and a less computation time is achieved. The comparative study concludes that by the optimal location and control of TCSC using ICA method the secured loadability enhancement is obtained in the transmission system in deregulated electricity market.

The convergence of learning rate in neural networks identifier and controller is one of challenging issues which attracts great interest from researchers. This paper suggests the adaptive gradient descent algorithm with learning laws which assures the convergence of multi-layer perceptron neural network based on Taylor series expansion of output error. In the proposed method the learning rate can be calculated online. To increase the accuracy and the speed of convergence, the second and higher order terms of the Taylor series expansion are not considered constant and are updated during the algorithm. Simulating the suggested algorithm on two examples reveals that with considering the bounds in the proposed method, the aims for learning rate, convergence of learning algorithm are guaranteed and the speed of convergence of training algorithm is increased.

In this paper, a method is presented to online Farsi handwritten isolated characters. In the proposed method, the information of main body and tiny movements are simultaneously used to improve the validation of output class recognition. Farsi handwritten isolated characters are categorized in 18 groups based on similarity in main body and also 11 groups based on tiny movement. According to the proposed method in this paper, the main body and tiny movements are recognized to identify unknown input characters. If detected groups from main body and tiny movements are corresponded, the unknown character is recognized; otherwise this mistake will be corrected by correction algorithm, as much as possible. In this paper, point features and global features are extracted from main body. Principle Component Analysis (PCA) and Linear Discriminate Analysis (LDA) are applied to reduce computational burden and to increase the quality of features. Using PCA and LDA, feature dimension is reduced from 102 to 17 for main body. One Versus One (OVO) approach of Support Vector Machine (SVM) classifier is used to classify the main body of characters and also tiny movements. The obtained results show that by using the proposed method; about 98 percent of online Farsi handwritten isolated characters are correctly recognized.