Utilization of the sustainable and renewable sea wave energy has recently received special attention by the virtue of being a free, clean and zero-carbon footprint power source. This paper presents a novel approach to model, design, analyze and control a sea wave electric power generating system using an artificial intelligent nonlinear auto regressive with external input neural networks (NARX-NN). Modeling design, and analysis of an electro-mechanical power-generating system using linear permanent magnet generator attached to a dual spring-mass-damper platforms is introduced. The purpose of this proposed generator is to convert sea and ocean wave kinetic energy into a useful electrical power generated as a result of the linear motion core through an electromagnetic stator. One of the direct applications of the sea wave generator is to install one or more units on shipboard to contribute to its power utility needs whether it is moving or floating. The dynamical stability and compensator control of the spring-mass damper generator platform is analyzed along with its associated electric power. Faraday’s law based results show that the output induced voltage ranges from -60 to 60 volts (120 volts p-p). Moreover, artificial intelligent nonlinear auto-regressive neural networks are used to train, validate, and test the sea wave electric generator output. Two-layer NN are used to train the dynamical input-output relationship of the proposed system using one hidden layer that contains of 10 neurons. Two delays are used, one for motion input and one for voltage output. The NARX-NN training demonstrates that the network is being trained efficiently and tracks the actual sea wave electric generator output with a very low mean-square-error performance response without the need to measure the variables.

Due to freedom to express views, opinions, news, etc and easier method to disseminate the information to large population worldwide, social media platforms are inundated with big streaming data characterized by both short text and long normal text. Getting the glimpse of ongoing events happening over social media is quintessential from the viewpoint of understanding the trends, and for this, topic modeling is the most important step. With reference to increase in proliferation of big data streaming from social media platforms, it is crucial to perform large scale topic modeling to extract the topics dynamically in an online manner. This paper proposes an adaptive framework for dynamic topic modeling from big data using deep learning approach. Approach based on approximation of online latent semantic indexing constrained by regularization has been put forth. The model is designed using deep network of feed forward layers. The framework works in an adaptive manner in the sense that model is extracts incrementally according to streaming data and retrieves dynamic topics. In order to get the trends and evolution of topics, the framework supports temporal topic modeling, and enables to detect implicit and explicit aspects from sentences also.

The aim of this research is the study of pathogenic signs, prognostically significant for the outcome of the disease and restoration of impaired functions at various stages of recovery after a stroke. This work describes a new method of applying a group of artificial neural network algorithms for each of the criteria and for each period of rehabilitation, and it is aimed at analyzing the structural and functional support of motor and higher cognitive functions, including speech and language as well as brain plasticity after ischemic stroke. The functional magnetic resonance imaging (fMRI, DTI) and clinical data machine learning algorithms were used. Self-organizing Kohonen and probabilistic neural network-based models with different structures and parameters were developed and applied for each criterion for periods of 3, 6, and 12 months of rehabilitation. For correlation analyses and modeling additional classifiers, we used: Decision Tree (DT), Support Vector Machine (SUM), k-Nearest Neighbor (KNN) clustering, and Logistic Regression (LR). In the performance evaluation, sensitivity, specificity, accuracy, error rate, and f-measure were used. The using of clinical parameters and mathematical modeling for analysis of brain plasticity mechanisms in stroke patients allowed in some cases to predict cognitive functions within the accuracy of 85-97%. Moreover, it is shown that the functional systems is represented by various brain structures, its synchronous activity and structural connectivity ensures the rapid and most complete restoration of motor and higher cognitive functions, including speech and language (effective post-stroke plasticity of the brain) after a course of neurorehabilitation.

Speed control for an I.M is a few what complex strategies; the complexity is regularly increasing in line with the required system achievement. The main forms of control strategies are scalar, direct torque, adaptive, sensorless, and vector or Field Oriented Control (FOC). The FOC method is the most efficient technique in which machine parameters: Rotor flux, unit vector, and electromagnetic torque, usually are estimated by means of using Digital Signal Processing (DSP). The Artificial Neural Network (ANN) becomes an effective tool for controlling nonlinear device in present time. This paper proposes the using of ANN instead of DSP to estimate the machine parameters in order to reduce the hardware complexity and the Electromagnetic Interference (EMI) impact. Also, it presents the PI-NN controller which is based totally on ANN. The systems simulations for both DSP and ANN are depicted. The performance of the ANN-based system gives excellent results: overshot less than 0.5%, rise time 0.514 s, steady state error less than 0.2%, settling time 0.7 s. in conjunction with that of DSP-based performance: overshot about 2%, rise time 0.64 s, steady state error less than 0.4%, settling time 0.75 s.

In most of the clustering algorithms, the assignment of initial centroids is performed randomly, which affects both the final outcome and the number of iterations required. Another aspect of the approaches in clustering algorithms is the use of Euclidean distance as the measure of similarity between data points, which is handicapped by linear separability of input data. The purpose of this paper is to combine suitable techniques so that both the above problems can be handled suitably leading to efficient algorithms. For the initial assignment of centroids we use Firefly and Fuzzy Firefly algorithms. We replace the Euclidean distance by Kernels (Gaussian and Hyper-tangent) leading to hybridized versions. For experimental analysis we use five different images from different domains as input. Two efficiency measures; Davis Bouldin index (DB) and Dunn index (D) are used for comparison. The tabular values, their graphical representations and output images are generated to support the claims. The analysis proves the superiority of the optimized algorithms over their existing counterparts. We also find that Hyper-tangent kernel with Rough Intuitionistic Fuzzy C-Means algorithm using Fuzzy Firefly algorithm produces the best results and has a much faster convergence rate. The analysis of medical, satellite or geographical images can be done more efficiently using the proposed optimized algorithms. It is supposed to play an important role in image segmentation and analysis.

Big Data is unstructured data that overcome the processing complexity of conventional database systems. The dimensionality reduction approach, which is a fundamental technique for the large-scale data-processing, try to maintain the performance of the classifier while reduce the number of required features. The pedestrian data includes a number of features compare to the other data, so pedestrian detection is the complex task. The accuracy of detection and location directly affect the performance of the entire system. Moreover, the pedestrian based approaches mainly suffer from huge training samples and increase the computation complexity. In this paper, an efficient dimensionality reduction model and pedestrian data classification approach has been proposed. The proposed model has three steps Histogram of Oriented Gradients (HOG) descriptor used for feature extraction, Orthogonal Locality Preserving Projection (OLPP) approach for feature dimensionality reduction. Finally, the relevant features are forwarded to the Support Vector Machine (SVM) to classify the pedestrian data and non-pedestrian data. The proposed HOG+OLPP+SVM model performance was measured using evaluation metrics such as precision, accuracy, recall and f-measure. The proposed model used the Penn-Fudan Database and compare to the existing research the proposed model improved approximately 6% of pedestrian data classification accuracy.