Suman Ahmmed

An adaptive merging and growing algorithm for designing artificial neural networks

This paper presents a new algorithm, called adaptive merging and growing algorithm (AMGA), for designing artificial neural networks (ANNs). The new algorithm merges and adds hidden neuron during training. The merging operation introduced here is a kind mixed mode operation that is equivalent to pruning two neurons and adding one neuron. Unlike most previous studies on designing ANNs, AMGA puts emphasis on adaptive functioning in designing ANNs. This is the main reason why AMGA merges and adds hidden neurons repeatedly (or alternatively) based on the learning ability of hidden neurons and training progress of ANNs, respectively. AMGA has been tested on five benchmark problems including the Australian credit card, cancer, diabetes, glass and thyroid problems. The experimental results show that AMGA can produce ANNs with good generalization ability compared to other algorithms.

Constrained non-linear optimization by modified particle swarm optimization

This paper presents a modified particle swarm optimization (MPSO) for constrained non-linear optimization problems. Optimization problems are very complex in real life applications. The proposed modified PSO consists of problem (complexity) dependent variable number of promising values (in velocity vector), error-iteration dependent step length, unlocking the dead look of idle particles and so on. It reliably and accurately tracks a continuously changing solution of the complex function and no extra concentration/effort is needed for more complex higher order functions. Constraint management is incorporated in the modified PSO by penalty function. The modified PSO has balance between local and global searching abilities, and an appropriate fitness function helps to converge it quickly. To avoid the method to be frozen, stagnated/idle particles are reset. Finally, benchmark data and methods are used to show the effectiveness of the proposed method.

STLF using Neural Networks and Fuzzy for anomalous load scenarios - A case study for Hajj

Load Forecasting has an important role in load generation, scheduling, planning etc. in power system. Different computational intelligent techniques are used in Short Term Load Forecasting (STLF) to make it more effective. Neural Networks (NN) is an effective mapping algorithm that can map complex input-output relationships, which is an important technique to do STLF having existing dataset. Usually a proper NN is sufficient to achieve accepted level of performance. But different load dataset may bear some irregular nature of load demand scenario due to having special events, where accuracy of NN suffers significantly. To enhance the performance for those situations, the authors propose a hybrid STLF approach-Neural Networks and Fuzzy (NNF) method. The authors first try to select the best possible trained NN and do STLF. Considering historical data trend and of existing errors of NN solution, for those special days, NNF determine the Load Change trend. Fuzzy Inference Rules (FIR) have been developed to further improvement by fuzzy method. In fuzzy part the NNF apply FIR on two inputs: STLF of NN and Load Change trend, to enhance the performance of STLF for special events. To evaluate the proposed method it is applied on the dataset of Regional Power Control Center of Saudi Electricity Company, Western Operation Area (SEC-WOA). Since the authors deal with the daily load dataset of Saudia Arabia of Hijri years (Arabian years), Hajj has been chosen as one of the anomalous load scenario. Empirical results show that for Hajj event of Hijri 1428 year, the accuracy of STLF by NN is approximately 6.37%, whereas proposed NNF can decline the error at only 1.92%.

Computational intelligence approach to load forecasting - a practical application for the desert of Saudi Arabia

This paper presents the development of an Artificial Neural Networks and Particle Swarm Optimization (ANN-PSO) based short-term load forecasting model with improved generalization technique for the Regional Power Control Center of Saudi Electricity Company, Western Operation Area (SEC-WOA). Weather, load demand, wind speed, wind direction, heat, sunlight, etc. are quite different in a desert land than other places. Thus this model is different from a typical forecasting model considering inputs and outputs. In this research paper two steps have been introduced, first load forecasting made by mapping mechanism and then optimization technique applied to improve its accuracy. This paper includes ANN and PSO models for 24-hours ahead load forecasting. ANN is an effective mathematical tool for mapping complex relationships. It is also successful for doing forecasting, categorization, classification, and so forth. On the other hand, PSO is the most promising optimization tool. It has better information sharing and conveying mechanism; it has better balance of local and global searching abilities; and can handle huge multi-dimensional optimization problems efficiently with hundreds of thousands of constraints. Thus PSO is chosen as the optimization model of the weight matrix of ANN. Results show that the proposed ANN-PSO performs much better than ANN for the load forecasting in a desert like Saudi Arabia.

Designing ANN using sensitivity & hypothesis correlation testing

Now a day artificial neural network (ANN) has become one of the most prominent concepts in the field of artificial intelligence. ANN has already been applied in the thousands of real life applications. In the arena of classification problem ANN is used massively. But the key issue is in almost all situations the performance of it depends on the architecture of the ANN. As a result designing a proper ANN is always a vital issue in the field of neural networks. The determination of an appropriate ANN architecture is always a challenging task for the ANN designers. This paper proposes a pruning algorithm for designing a three layered ANN architectures. It is well known that a three layered ANN can solve any kind of linear and nonlinear problems. The proposed algorithm uses some major mathematical concepts: correlation coefficients, standard deviations, and statistical hypothesis testing scheme for designing the ANNs. For that reason the authors propose the new pruning algorithm, ANN designing by sensitivity and hypothesis correlations testing (SHCT), to determine ANN architectures automatically. The salient features of SHCT are that it uses statistical hypothesis testing scheme, standard deviations, correlation coefficients, merging with proper replacements to design the ANNs. To justify the performances of SHCT it has been tested on a number of benchmark problem datasets such as Australian credit cards, breast cancer, diabetes, heart disease, and thyroid.

Information theoretic SOP expression minimization technique

The efficient design of multiple Boolean functions is becoming important and necessary during computer aided design for circuit and systems (CADCS), especially the manufacture of chips have reached a density of several ten thousands transistors per chip, so called very large scale integration (VLSI). To simplify the Boolean expressions by conventional approaches like graphical observation-based K-MAP or other simplification procedures become tedious or even impossible when the number of variables in a truth table exceeds certain limit. In this paper we propose an information theory based circuit designing approach for deriving minimal sum of product (SOP) expressions for unlimited number of variable. We have verified our approach on a number of cases with the conclusion that the proposed approach is a better alternative to conventional approaches particularly when the numbers of variables restrict the use of conventional approaches. The key feature of proposed method is that it performs a hill-climbing search through the state space of Boolean variables using information theoretic heuristic to find the minimal SOP expression.

A new CAC protocol for optimizing revenue and ensuring QoS

Call admission control (CAC) protocol will try to be fair, fast, reducing inconveniences, and maintaining quality of service (QoS) at the time of deciding whether a call request will be accepted or rejected into a network or communication system. A good CAC should have the ability to select the calls among the requested calls in such a way so that maximum fairness is justified by its performances at the sometime it should achieve some targets such as maximizing the revenue considering the constraints of the network system such as limited bandwidth. In the todaypsilas communication world the demand of bandwidth increases sharply due to the different users different types of service requirements such as audio, video, voice etc. from the same communication system. Besides need to deal with different tariff. So considering all these issues a CAC needs to decide which call requests should be admitted or not, which is obviously a critical and difficult task. In this paper a new CAC protocol is proposed that can select the best calls for the communication system, it can optimize the generated revenue from the network system, it can handle the different call admission situations in a fairness way, it can consider the different priorities based on class, can maximize the revenue of the communication system and also maintains the overall QoS for any network or communication system. Empirical results have shown its excellent performances for any network and communication platform.