Search

Catastrophic forgetting is a major problem that affects neural networks during progressive learning. In it, the previously learned representation vanishes as the network learns new information. The extreme learning machine is one of the variants of the neural network. It is used in many domains due to fast training and good generalization ability. However, like other neural networks, it suffers from catastrophic forgetting and negative forward and backward transfer during the progression of neurons in incremental learning. The study hypothesizes that it is due to overlapping in hidden neurons and output weights. The global representation by an activation function further supports this hypothesis. To address this, the study proposes a neuron clustering approach to mitigate it in an adaptive class incremental extreme learning machine. The neuron clustering method activates k nearest neurons during learning and testing. It helps to partition the network to select overlapping subnetwork. Experimental results on four food datasets show that the proposed approach reduces negative forward and backward transfer when neurons are added incrementally during progressive learning.

Hypertension is an epidemic restricted not only to the developing but also to the developed nations. It is triggered by various lifestyle choices that depend on each individual based on their personal physiology and lifestyle. Early diagnosis is possible, but it requires continuous blood pressure monitoring. Various machine learning methods have been proposed for early diagnosis of hypertension by predicting blood pressure and detecting high spikes in the values. However, these solutions are built upon the generic guidelines which may not be applicable for every patient. Most of these solutions incorporate batch learning and require all data to be present before prediction and do not support any online learning mechanism. This leads to potentially outdated models. Furthermore, there is also a lack of an intelligent approach to handling incomplete time series while training the model. This paper presents a personalized approach to estimate blood pressure that eliminates the need for continuous monitoring based on the Online recurrent extreme learning machine (OR-ELM). The missing values are imputed using Gaussian mixture models. The prediction model learns from the historical data and learns online as more data becomes available. The proposed scheme is developed and deployed on a mobile application for secured prediction results. The method is used to predict blood pressure in Malaysian population and compared with existing batch-learning and online learning methods. The results show that OR-ELM based model outperforms the existing online techniques such as the Online sequential extreme learning machine and batch learning technique such as Extreme learning machine.

Typically, it is desired to operate the wind turbine at the maximum power point. However, in small wind turbines which have a storage system integrated with them, harvesting as much energy as possible is more crucial. This may be achieved by reducing the cut-in speed while maximizing the mechanical power. These two goals may be achieved by optimizing the turbine blades. In this article, the turbine blades are optimized using improved blade element momentum theory including Viterna-Corrigan stall model with the objective to yield low cut-in speed and high power level. Using the blade element momentum analysis, the power coefficient curves as functions of tip-speed ratio at various range of wind speeds are obtained for the optimized turbine. Using MATLAB/Simulink tool, a wind energy system, which consists of a wind turbine, permanent magnet synchronous generator and a resistive load, is simulated. The curves obtained from the blade element momentum analysis are used to emulate the wind turbine. The results obtained from the simulation are compared to experimental results. It is noticed that the wind turbine may be optimized to harvest more energy.

In order to provide the reactive power demand of a self-excited induction generator which is required to achieve a voltage build-up, a three-phase capacitor bank is connected between the generator terminals. As loading increases, the operating point on the magnetizing curve moves toward the linear region which may lead to the collapse of the generated voltage. In this article, genetic algorithms are used to evaluate the value of the excitation capacitance that makes the machine operate in the saturation region which ensures a stable generated voltage. To verify the effectiveness of this method, a laboratory machine which has a relatively high stator and rotor resistances and leakage reactance is considered. The values of the excitation capacitances predicted by the genetic algorithms are applied to the machine Simulink-based model. The results obtained by the simulation are compared with experimental results which show a good agreement.

Due to the fluctuation of the wind speed, wind energy systems are typically connected to the utility grid through back-to-back converters and a DC coupling. Although this topology provides a stable voltage on the grid side, it makes the control system relatively complicated. In this paper, a model for the machine-side converter and the controller is derived in the rotor reference frame. Using αβ0 transformation to detect the voltage phase angle, a decoupled control of the active and the reactive power is achieved. As a result, a simple PI controller is used to control the active and reactive power flow of the machine side converter. The performance of the controller is tested using the MatLab/Simulink simulation. The results show that the controller has provided the desired performance.

This paper discusses the integration of wind energy system in Derna, Libya to the main grid of General Electricity company of Libya (GECOL) through a back-to-back converter. The machine side converter is used to regulate the AC and DC voltages while the grid side converter is used to control the active and reactive power supplied by the wind energy system. To achieve a relatively simple and decoupled control, the controllers are implemented in the QD reference frame. The proposed system is simulated using MatLab/Simulink environment. The simulation results show the rapid response of the control system in following the commanded active and reactive power commands and the robustness of the proposed system to the changes in operating conditions.

In order to meet the continuous growth in the global energy demand and to reduce the negative impact which is associated with the global warming, adopting renewable energy sources by policy makers need to be considered. Although Libya is very rich in solar energy resource, the country has been recently struggling to satisfy its energy demand. Consequently, centralized PV power plants may be considered as renewable and clean source of energy. This paper, presents the technical and economic feasibility of a 14MW photovoltaic (PV) power plant near the Houn city which is located in the central-south of Libya in the Jufra Region. In addition, In order to prove the design validity of the proposed system, models and simulations in NEPLAN software is carried out for a practical distribution grid. The result show that PV power plant is economically feasible and sustainable.

News articles provide a large data set which can be used to classify information such as author's point of view, affiliation and biasedness. Classification on the basis of positive and negative content can help readers to understand the author's sentiments. mydNews application is a proposed a solution to achieve this objective of presenting author's opinion in an article. We performed sentiment analysis using CoreNLP API presented by Stanford and through experimentation we have reached to a conclusion that using 60% sentiment value of text and 40% of headline produces results nearest to human evaluation. We have proposed a new algorithm Recursive Cosine for similarity matching of news. Our algorithm uses classical cosine in a different way. We compared latest news articles with the one's published in same day by different authors and we have achieved 5% better results than classical cosine measure by using the newly constructed algorithm.

Due to their robustness, low cost, and low maintenance, induction generators are good candidates for wind energy systems. In addition, when the wind turbine is isolated from the grid, the induction generator can operate within a wide speed range which utilize maximum power extraction. To ensure that voltage buildup will take place between the machine terminals, the reactive power required by the machine must be supplied. To achieve that, a three phase capacitor bank is connected to the generator terminals. Evaluating the adequate value of the capacitance is very crucial since it affects the value of the generated voltage. In this paper, the desired operating voltage and the corresponding value of the magnetizing inductance are defined from the magnetizing curve. By using genetic algorithms (GAs), the corresponding values of the excitation capacitance and the rotor slip are evaluated. Matlab/Simulink model is then used for validation. It is noticed that there is a good agreement between the values predicted by the GAs and the ones obtained from the simulation.