S. M. Shafiul Alam

Detection of Seizure and Epilepsy Using Higher Order Statistics in the EMD Domain

In this paper, a method using higher order statistical moments of EEG signals calculated in the empirical mode decomposition (EMD) domain is proposed for detecting seizure and epilepsy. The appropriateness of these moments in distinguishing the EEG signals is investigated through an extensive analysis in the EMD domain. An artificial neural network is employed as the classifier of the EEG signals wherein these moments are used as features. The performance of the proposed method is studied using a publicly available benchmark database for various classification cases that include healthy, interictal (seizure-free interval) and ictal (seizure), healthy and seizure, nonseizure and seizure, and interictal and ictal, and compared with that of several recent methods based on time-frequency analysis and statistical moments. It is shown that the proposed method can provide, in almost all the cases, 100% accuracy, sensitivity, and specificity, especially in the case of discriminating seizure activities from the nonseizure ones for patients with epilepsy while being much faster as compared to the time-frequency analysis-based techniques.

Distribution Grid State Estimation from Compressed Measurements

Real-time control of a smart distribution grid with renewable energy based generators requires accurate state estimates, that are typically based on measurements aggregated from smart meters. However, the amount of data/measurements increases with the scale of the physical grid, posing a significant stress on both the communication infrastructure as well as data processing control centers. This paper first investigates the effect of geographical footprint of distributed generation (DG) on the voltage states of a smart distribution system. We demonstrate that the strong coupling in the physical power system results in estimated voltage phasors exhibiting a correlation structure that allows for compression of measurements. Specifically, by exploiting principles of 1D and 2D compressed sensing, we develop two approaches, an indirect and direct method for state estimation starting from compressed power measurements. We illustrate the effectiveness of voltage estimation with significantly low number of random spatial, temporal as well as spatio-temporal power measurements using the IEEE 34 node distribution test feeder and a larger 100 node radial distribution system. Results show similar performance for both methods at all levels of compression. It is observed that, even with only 50% compressed power measurements, both methods estimate the states of the test feeder with high level of accuracy.

Design and construction of an automatic solar tracking system

Energy crisis is the most important issue in todays world. Conventional energy resources are not only limited but also the prime culprit for environmental pollution. Renewable energy resources are getting priorities in the whole world to lessen the dependency on conventional resources. Solar energy is rapidly gaining the focus as an important means of expanding renewable energy uses. Solar cells those convert suns energy into electrical energy are costly and inefficient. Different mechanisms are applied to increase the efficiency of the solar cell to reduce the cost. Solar tracking system is the most appropriate technology to enhance the efficiency of the solar cells by tracking the sun. A microcontroller based design methodology of an automatic solar tracker is presented in this paper. Light dependent resistors are used as the sensors of the solar tracker. The designed tracker has precise control mechanism which will provide three ways of controlling system. A small prototype of solar tracking system is also constructed to implement the design methodology presented here.

Goal-Based Holonic Multiagent System for Operation of Power Distribution Systems

Large-scale integration of rooftop solar power generation is transforming traditionally passive power distribution systems into active ones. High penetration of such devices creates new dynamics for which the current power distribution systems are inadequate. The changing paradigm of power distribution system requires it to be operated as cyber-physical system. A goal-based holonic multiagent system (HMAS) is presented in this paper to achieve this objective. This paper provides details on design of the HMAS for operation of power distribution systems. Various operating modes and associated goals are discussed. Finally, the role of HMAS is demonstrated for two applications in distribution systems. The first one is associated with control of reactive power at solar photovoltaic installations at individual homes for optimal operation of the system. The second deals with the state estimation of the system leveraging different measurements available from smart meters at homes.

Detection of epileptic seizures using chaotic and statistical features in the EMD domain

An artificial neural network (ANN)-based method, using a combination of statistical and chaotic features, is proposed to discriminate electroencephalogram (EEG) signals for seizure detection. The EEG signals are subjected to empirical mode decomposition, generating intrinsic mode functions. Statistical and chaotic features such as skewness, kurtosis, variance, and largest Lyapunov exponent, correlation dimension and approximate entropy are extracted from these modes and fed to the ANN to classify the EEG signals. It is shown that the proposed method can achieve up to 100% accuracy as compared to several state-of-the-art techniques in discriminating the seizure signals from the non-seizure ones.

A statistical method for automatic detection of seizure and epilepsy in the dual tree complex wavelet transform domain

In this paper, a statistical method for automatic detection of seizure and epilepsy in the dual-tree complex wavelet transform(DT-CWT) domain is proposed. Variances calculated from the EEG signals and their DT-CWT sub-bands are utilized as features in the classifiers such as artificial neural network(ANN) and support vector machine(SVM). Studies are conducted using EEG signals from a publicly available benchmark EEG database to assess the ability of the proposed method for a number of clinically relevant classification scenario which include healthy vs seizure, healthy and non-seizure(inter-ictal) vs seizure(ictal), and finally, ictal vs inter-ictal records. It is shown that the proposed method using SVM performs better than employing ANN. It gives 100% accuracy, sensitivity and specificity; at least the same or better than those corresponding to several existing techniques. In addition, the proposed method is computationally faster than the time-frequency and EMD-based techniques.

Statistical parameters in the dual tree complex wavelet transform domain for the detection of epilepsy and seizure

In this paper, a comprehensive statistical analysis of electroencephalogram (EEG) signals is carried out in the dual tree complex wavelet transform domain using a publicly available EEG database. It is shown that variance and kurtosis can be effective in distinguishing EEG signals at sub-band levels. It is further shown that the parameters of a normal inverse Gaussian probability density function can equally discriminate the EEG signals at sub-band levels. Thus, these statistical quantities may be used to characterize EEG signals and help the researchers in developing improved classifiers for the detection of epilepsy and seizure and building a better understanding of the diverse process of EEG signals.

Stability of Agent Based Distributed Model Predictive Control Over a Lossy Network

In this paper, an agent based control formulation of a large-scale cyber-physical system is proposed. Each agent can partially observe a part of the global dynamical process and estimate the associated local states through a combination of traditional Kalman filtering algorithm and consensus. The local estimates are then used for state feedback control. The optimal feedback gain of individual agents is obtained through dynamically solving a moving horizon linear quadratic optimization problem. The agents also exchange information among neighbors to coordinate the agent-wise state-feedback controls. Finally, a control decision incurring the least cost among all agents is applied to the global system. A Lyapunov function-based stability analysis is performed to obtain a bound over the degree of agent negotiation in designing the control decision. Besides, the effect of lossy communication network in control design and henceforth in global system stability is also investigated and the corresponding bound in control consensus is obtained. The theoretical results are verified via simulation of a 10-agent and a 50-agent dynamical process within a radial topology.

Bounds on decentralized concave optimization in energy harvesting wireless sensor networks

Wireless sensor networks (WSNs) have increasingly become the viable means of distributed sensing and control for a wide array of applications. The energy-sensitive sensor nodes in these systems are often augmented by an energy harvesting device, allowing for continuous operation. The drawback, however, is that the availability of communication resources is uncertain. Decentralized optimization is a common technique implemented to coordinate such a disparate collection of devices. Most decomposition methods involve iterative updates where public information about joint constraints or objectives must be shared. Recent work in distributed optimization has provided some new insights on the performance of optimization in such a distributed network. For perfect and unlimited communication, the convergence of the optimization performs as good as a centralized controller. However, limited communication introduces delays and quantization errors which affect solution convergence, especially for algorithms utilizing multi-hop updates. In this paper, we analyze the effect of deterministic delays and quantization errors on the convergence of decentralized optimization in an energy harvesting wireless sensor network. The corresponding utility maximization problem is being solved through a combination of dual decomposition and alternating direction method of multipliers (ADMM). The convergence bound on the associated dual function update exhibits a square law uncertainty with respect to the maximum allowable communication delay and quantization noise variance.

Distribution of decentralized optimization convergence bounds in energy harvesting wireless sensor networks

In this paper, we attempt to uncover the fundamental limitations of implementing decentralized optimization in an energy harvesting sensor network by quantifying the impact of stochastic energy availability on convergence. Specifically, the discrete energy quanta being harvested by a network of wireless sensors are modelled via a marked Poisson process. The wireless sensors are involved in updating the Lagrange multipliers associated with the constraints for a decentralized concave optimization problem. The convergence of the corresponding Lagrange dual function is monotonically dependent upon the maximum delay in communication. The probabilistic nature of the harvested energy quanta results in a random crossover with respect to the minimum energy threshold required for successful communication. As a consequence, we investigate the probabilistic behaviour of the Lagrange dual function convergence bound under the effect of random delay considering single/multi-hop communication among neighbouring sensors. The maximum delay distribution is derived for both deterministic and stochastic models for the energy harvested at sensor nodes. Simulation shows the efficacy of the theoretical distributions in modelling the delay dependent Lagrange dual function convergence bound behaviour. Copyright