نشریه محاسبات نرم
سال دوم شماره 2 (پیاپی 4، پاییز و زمستان 1392)

Optimum operation of power networks has attracted a great deal of attention considering the latest technological advancement taken place in these networks. Considering the huge costs associated with installation and development of such networks، employing any mean to increase efficiency، reduce loss and operation problems (such as reducing energy losses، improving voltage profile and current reduction of lines) seem to be inevitable. Shunt compensators such as DSTATCOM could be employed for these purposes. In this article، optimal allocation and sizing of this element to mitigate such problems both in the presence of distributed generation (DG) and without it are studied. Hence، immune algorithm (inspired originally by immune system of human body) is employed to simulate such optimization using the IEEE 33-bus standard test system. Results show that proposed method is capable of optimizing the distribution system by presenting optimal allocation and sizing of DSTATCOM.

The extent and diversity of systems provided by IP networks have made various technologies to approach integrating various types of access networks and converting to next generation network. The Session Initiation Protocol (SIP) with respect to facilities such as being in text form، end-to-end connection، independence from the type of transmitted data، and supporting various forms of transmission، is an appropriate choice for signaling protocol in order to make connection between two IP network users. These advantages have made SIP to be considered as a signalling protocol in IMS، a proposed signalling platform for next generation networks. Despite having all these advantages، SIP protocol is in lack of appropriate mechanism for addressing overload. Overload can cause serious problems for SIP servers. SIP overload occurs when a SIP server does not have enough resources to process messages. It is a proven fact that the performance of SIP servers is largely degraded during overload periods because of the built in message retransmission mechanism of SIP. In this Thesis، we propose an advanced method which is an improved method of the Windows Based Overload Control. In the Windows based Overload Control method، the window is used to limit the amount of messages that are generated at once by SIP proxy server. An adaptive window based overload control algorithm، which does not use explicit feedback from the downstream server، is proposed. The number of confirmation messages is used as a measure of the amount of load on the downstream server by upstream ones. Thus، the proposed algorithm does not impose any additional complexity or processing on the downstream server which is overloaded، making it a very robust approach. Our proposed algorithm is developed and implemented on Asterisk open source proxy and evaluated. The results indicate that the related server maintains its throughput at the maximum capacity in under-load conditions. Also it is concluded that the suggested method can change the maximum windows size dynamically through considering the number of confirmation message.

With advances in computer networks، special attention has been paid to Wireless Sensor Networks (WSN). Because of their wireless nature، the probability of the fault occurrence in such networks is high. There has been a lot of research conducted on the field of fault detection in WSN. Among the fault detection methods، the disturbed ones are more important because they outperform other methods with respect to the scalability and the network life time. In this paper we propose a new distributed trust-based fault detection algorithm that resolves the drift problem. Experimental results show that the proposed algorithm gains a higher correct detection rate and a lower false alarm rate compared with similar fault detection algorithms.

Intelligent age estimation via face images has become an important research topic in machine vision and pattern recognition fields because it has a key role in many applications such as customer behavior analysis in a business intelligence system. Age estimation is a process which analyses an individual face image and estimates his/her age based on the year measure. The age estimation process includes two parts: 1) face representation and 2) learning the age estimation function. So far several methods have been presented for improvement of one or both parts of the age estimation process. In this paper، a new method is presented for improvement of learning age estimation part. In the proposed method faces are represented by BIF (Bio-Inspired Features) and PLS (Partial Least Squares) methods، and then individual’s age is estimated by a new two-step method which is implemented based on MLP (Multi-Layer Perceptron) neural network. The proposed method is evaluated using two benchmark databases FG-NET and MORPH2. The results show that the proposed method improves the age estimation process (The proposed method gains 54% of improvement on FG-NET and also 78% of improvement on MORPH2).

The uniformity of yarn is one of the major quality parameters which significantly influences on yarn characteristics، warping، weaving، and ultimately fabric production. This parameter depends on fiber properties and spinning process directly. In this study، yarn non-uniformity in a worsted spinning system was predicted by using a hybrid technique involving Kohonen''s self-organized and perceptron neural network. A series of 2490 experiments involving parameters of raw materials and production quality performed from a worsted spinning factory was collected and analyzed. In step one، Kohonen neural network was used to cluster data. Each cluster was fed into the perceptron network، and the predicting error of the hybrid model was calculated for a new series of data. As the second step، the whole data were fed into the perceptron network and the error percentage was computed. Comparing the results obtained from the hybrid technique and the perceptron network showed that the rate of predicting error can be reduced by 3. 34% when the hybrid approach of Kohonen and perceptron neural network was used.