Contents of Volume 29 (2019)

1/2019 2/2019 3/2019 4/2019 5/2019

5/2019

  • [1] Jha S.K., Bilalovikj J. (Vietnam, Macedonia) ,
    A comparative approach of neural network and regression analysis in very short-term wind speed prediction , pp. 285-300

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.018

    Abstract: The accurate estimation of very short-term wind speed is essential for planning, management, and distribution of wind power produced by any installed wind turbine at a power plant. This study is based on very short-term wind characteristics and meteorological data measured from the wind farm at Bogdanci, in the Former Yugoslav Republic of Macedonia (FYROM) in between May-September 2015. Moreover, the study focuses on the comparative analysis of conventional polynomial based regression analysis and artificial neural network (ANN) methods for very short-term wind speed prediction at the interval of 10 min using four types of wind directions, and three atmospheric parameters. Polynomial regression analysis results in the maximum accuracy (R2 = 0.71) in the prediction of wind speed rotation mean (WSRM) using the wind direction base mean (WDBM) and temperature. The ANN method achieves the best efficiency (R2 = 0.97) in the prediction of WSRM using four types of wind directions and three atmospheric parameters. The ANN performs better than the conventional regression analysis in the prediction of each of the target wind speeds.

  • [2] Raja Basha A., Yaashuwanth C. (India)
    An optimal data aggregation scheme for wireless sensor network using QOS parameters with efficient failure detection and loss recovery technique, pp. 301-324

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.019

    Abstract: WSN: Wireless Sensor Networks play a significant part in its modern era but its limited power supply acts as a blocking stone in it growth. In order to save energy in WSN the concept of aggregator node is introduced, where the aggregator node would act as a mid-point between the source and destination node during the data transmission. The data aggregation process creates major problems like excess energy expenditure, and delay. In the process of eliminating or reducing the delay and energy expenditure, the researchers have been handled in different ways. Applications like environment monitoring, target tracking, military surveillance and health care require reliable and accurate information. Many researchers have proposed data aggregation techniques to enhance the latency, average energy consumption and average network lifetime. However, these techniques are not sufficient to address situations like node failure and loss recovery. This paper proposes to build a solid wireless sensor system which concentrate on efficient optimal data aggregation along with additional QoS metrics such as failure detection and loss recovery. The first contribution of this paper is to propose an Improved Wolf Optimization (IWO) algorithm for clustering. The clustering process includes an efficient cluster formation like, Cluster Head (CH), and Sub Head (SH) selection. The second contribution of this paper is inclusion of failure detection and loss recovery. The former is developed based on Multi-criteria Moths-Flame Decisionmaking (MMFD) model and the latter is achieved through SH. SH node will act as the backup node for cluster head when failure instances are detection. CH recovers the lost data through SH, which minimize the additional delay of backup node selection process and save much more energy. The results are simulated using network simulator 2 tool and it is compared with existing techniques. The Network Simulator - 2 results disclose that the findings are better than the available existing methodologies.

  • [3] Henclová K. (CZ)
    Using CMA-ES for tuning coupled PID controllers within models of combustion engines, pp. 325-344

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.020

    Abstract: Proportional integral derivative (PID) controllers are important and widely used tools of system control. Tuning their gains is a laborious task, especially for complex systems such as combustion engines. To minimize the time an engineer spends tuning the gains in a simulation software, we propose to formulate a part of the problem as a black-box optimization task. In this paper, we summarize the properties and practical limitations of gain tuning in this particular application. We investigate the latest methods of black-box optimization and conclude that the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) with bi-population restart strategy, elitist parent selection and active covariance matrix adaptation is best suited for this task. Details of the algorithm's experiment-based calibration are explained as well as derivation of a suitable objective function. The method's performance is compared with that of PSO and SHADE. Finally, its usability is verified on six models of real engines.

  • [4] Su T., Min S., Shi A., Cao Z., Dong M. (China)
    A CNN-LSVM model for imbalanced images identification of wheat leaf, pp. 345-361

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.021

    Abstract: In order to improve the accuracy of convolutional neural networks (CNN) in imbalanced dataset classification, a novel hierarchical CNN-LSVM is proposed. Considering the imbalance in the number and spatial distribution of wheat leaf disease images, the improved local support vector machine (LSVM) re- places Softmax as the classifier of the model, and meanwhile a cost sensitive matrix is designed to assign the value for penalty factors in the optimized objective function of LSVM. It effectively improves the sensitivity of misclassification caused by imbalanced data. To verify the validity and practicability of CNN-LSVM, 6028 wheat leaf disease images containing 8 species are collected from planting bases in Shandong Agricultural University. Then the imbalanced and balanced standard image sets are generated by data augmentation and Borderline-Synthetic Minority Oversampling (Borderline-SMOTE). They have 36168 and 46176 images, respec- tively. The experimental results show that the average identification accuracies of the CNN-LSVM obtained on imbalanced and balanced standard datasets are 90.32% and 93.68%, respectively. And it starts to converge when the iteration times are close to 13000. CNN-LSVM has higher classification accuracy and lower iteration times, compared with CNN-Softmax, CNN-SVM, LSVM and support vector machine (SVM).

  • [5] Svítek M. (CZ)
    Quantum multidimensional models of complex systems, pp. 363-371

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.022

    Abstract: The paper presents a new methodology how to extend the well-known quantum model [2] with (2N-1) free parameters (moduli and phases) of wave probabilistic functions ψ(Ai) assigned into events Ai, i∈(1,2,...,N) to (N(N+1))/2 free parameters necessary for full N-dimensional representation of complex system. Our approach generally enables to include additional functions applied on events Ai, i∈(1,2,...,N). In the paper, we will demonstrate this mathematical instrument on additional wave probabilistic functions ψ(Ak ∩ Am ∩ ... ∩ An) connected with macroscopic events' intersections Ak ∩ Am ∩ ... ∩ An where k,m,...,n ∈(1,2,...,N)


4/2019

  • [1] Tekerek A., Bay O.F. (Turkey)
    Design and implementation of an artificial intelligence-based web application firewall model, pp. 189-206

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.013

    Abstract: Attacks on web applications and web-based services were conducted using Hyper-Text Transfer Protocol (HTTP), which is also used as the communication protocol of web-based applications. Due to the dynamic structure of web applications and the fact that they have many variables, detection and prevention of web-based attacks are made more difficult. In this study, a hybrid learning-based web application firewall (WAF) model is proposed to prevent web-based attacks, by using signature-based detection (SBD) and anomaly-based detection (ABD). Detection of known web-based attacks is done by using SBD, while detection of anomaly HTTP requests is done by using ABD. Learning-based ABD is implemented by using Artificial Neural Networks (ANN). Thus, an adaptation of the model against zero-day attacks is ensured by learning-based ABD by using ANN. The proposed model is tested by using WAF2015, CSIC 2010 and ECML-PKDD datasets which are open source datasets. According to the test results, a high mean achievement percentage (96,59%) was obtained. Detection results are also compared to previous studies. After comparison, the proposed model promises higher performance than what the existing studies until now have to offer.

  • [2] Alquran H., Alqudah A.M., Abu-Qasmieh I., Al-Badarneh A., Almashaqbeh S. (Jordan)
    ECG classification using higher order spectral estimation and deep learning techniques, pp. 207-219

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.014

    Abstract: Electrocardiogram (ECG) is one of the most important and effective tools in clinical routine to assess the cardiac arrhythmias. In this research higher order spectral estimations, bi-spectrum and third-order cumulants, are evaluated, saved, and pre-trained using convolutional neural networks (CNN) algorithm. CNN is transferred in this study to carry out automatic ECG arrhythmia diagnostics after employing the higher-order spectral algorithms. Transfer learning strategies are applied on pre-trained convolutional neural network, namely AlexNet and GoogleNet, to carry out the final classification. Five different arrhythmias of ECG waveform are chosen from the MIT-BIH arrhythmia database to evaluate the proposed approach. The main contribution of this study is to utilize the pre-trained convolutional neural networks with a combination of higher-order spectral estimations of arrhythmias ECG signal to implement a reliable and applicable deep learning classification technique. The Highest average accuracy obtained is 97.8% when using third cumulants and GoogleNet. As is evident from these results, the proposed approach is an efficient automatic cardiac arrhythmia classification method and provides a reliable recognition system based on well-established CNN architectures instead of training a deep CNN from scratch.

  • [3] Miloševic N., Rackovic, M. (Serbia)
    Classification based on missing features in deep Convolutional Neural Networks, pp. 221-234

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.015

    Abstract: Artificial Neural Networks, notably Convolutional Neural Networks (CNN) are widely used for classification purposes in different fields such as image classification, text classification and others. It is not uncommon therefore that these models are used in critical systems (e.g. self-driving cars), where robustness is a very important attribute. All Convolutional Neural Networks used for classification, classify based on the extracted features found in the input sample. In this paper, we present a novel approach of doing the opposite - classification based on features not present in the input sample. Obtained results show not only that this way of learning is indeed possible but also that the trained models become more robust in certain scenarios. The presented approach can be applied to any existing Convolutional Neural Network model and does not require any additional training data.

  • [4] Kilic H., Yuzgec U., Karakuzu C. (Turkey)
    Improved Antlion Optimizer Algorithm and its performance on neuro fuzzy inference system, pp. 235-254

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.016

    Abstract: Antlion optimizer algorithm (ALO) is inspired by hunting strategy of antlions. In this study, an improved antlion optimization algorithm is proposed for training parameters of adaptive neuro fuzzy inference system (ANFIS). In the standard ALO algorithm, the greatest deficiency is its long running time during optimization process. The random walking model of ants, the selection procedure and boundary checking mechanism have been developed to speed up standard ALO algorithm. To evaluate the performance of the improved antlion optimization algorithm (IALO), it has been tested on dynamic system modelling problems. ANFIS's parameters has been optimized by IALO algorithm to model five dynamic systems. ANFIS training procedure has been performed with 30 independent runs. Each training has been started with the random initial parameters of ANFIS and performance metrics have been obtained at the end of training. The results show that the IALO algorithm is able to provide competitive results in terms of mean, best, worst, standard deviation, training time metrics. According to the training time result, the proposed IALO algorithm has better performance than standard ALO algorithm and the average training time has been reduced to approximately 80 %.

  • [5] M. Křížek (CZ)
    Do Einstein’s equations describe reality well?, pp. 255-283

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.017

    Abstract: The standard cosmological model that is based on Einstein's equations possesses many paradoxes. Therefore, we take a closer look at the equations themselves, and not only on cosmological scales. In this survey paper, we present 10 significant problems and drawbacks of Einstein's equations investigated by the author. They include their extremely large complexity, non-differentiability of the metric, difficulties with initial and boundary conditions, multiple divisions by zero, excessive extrapolations to cosmological distances leading to mysterious dark matter and dark energy entities, unconvincing relativistic tests, the absence of aberration effects, and scale non-invariance. We also discuss a slight violation of the laws of conservation of energy and of momentum.


3/2019

  • [1] Sharif S.M.A., Mahboob M. (Bangladesh, US) ,
    Deep HOG: A hybrid model to classify Bangla isolated alpha-numerical symbols, pp. 111-133

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.009

    Abstract: Bangla is known to be the second most widely used script in the South Asian region. Despite its wide usage, a complete study with all available Bangla handwritten image classes is still due. This work proposes a hybrid model to classify all available handwritten image classes and unifying the existing benchmark datasets. The feasibility of the di erent handcrafted features in the hybrid model also has been demonstrated. Moreover, the proposed hybrid model obtain a maximum accuracy of 89.91% in validation phase with a total of 259 Bangla alpha-numerical image classes. With the same number of image classes, the proposed hybrid model shows a testing accuracy of 89.28% on 15,175 testing samples. The comparison results demonstrate that the proposed hybrid-HOG model can outperform the existing state-of-the-art classification models in Bangla handwritten alpha-numerical image classification. The code will be available on https://github.com/sharif-apu/hybrid-259.

  • [2] Min Xia, Chong Zhang, Yin Wang, Jia Liu, Chunzheng Li (China)
    Memory based decision making: A spiking neural circuit model, pp. 135-149

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.010

    Abstract: Conscious decision making is one of the important functions of human behavior. Episodic memory is the source of knowledge for conscious decision making. The mechanism of how episodic memory a ects conscious decision-making is unclear. To investigate the brain mechanism of conscious decision making, we investigated a biologically-based network model of spiking neurons for competition between automatic response and conscious decision making. The proposed model integrates episodic memory modular and brain decision-making modular, and uses episodic memory output as the top-down input of decision making. In the decision making, the network realizes the competition between decision patterns through mutual inhibition, finally reaches the conscious decision making. The simulations show that the proposed model can well implement multimodal coherent decision making under sequential memory control. The proposed model can effectively explain the transmission mechanism of conscious decision information.

  • [3] Fister D., Mun J.C., Jagric V., Jagric T. (Slovenia, US) ,
    Deep learning for stock market trading: a superior trading strategy?, pp. 151-171

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.011

    Abstract: Deep-learning initiatives have vastly changed the analysis of data. Complex networks became accessible to anyone in any research area. In this paper we are proposing a deep-learning long short-term memory network (LSTM) for automated stock trading. A mechanical trading system is used to evaluate its performance. The proposed solution is compared to traditional trading strategies, i.e., passive and rule-based trading strategies, as well as machine learning classifiers. We have discovered that the deep-learning long short-term memory network has outperformed other trading strategies for the German blue-chip stock, BMW, during the 2010–2018 period.

  • [4] Rybicková A., Mocková D., Teichmann D. (CZ)
    Genetic algorithm for the continuous location-routing problem, pp. 173-187

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.012

    Abstract: This paper focuses on the continuous location-routing problem that comprises of the location of multiple depots from a given region and determining the routes of vehicles assigned to these depots. The objective of the problem is to design the delivery system of depots and routes so that the total cost is minimal. The standard location-routing problem considers a finite number of possible locations. The continuous location-routing problem allows location to infinite number of locations in a given region and makes the problem much more complex. We present a genetic algorithm that tackles both location and routing subproblems simultaneously.


2/2019

  • [1] Kratochvíl R., Jánešová M. (CZ)
    Use of clustering for creating economic-mathematical model of a web portal, pp. 61-70

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.005

    Abstract: This article describes the mathematical economic model of a communication web portal. To create the model, we use Cluster analysis as one of the areas of artifficial intelligence. Based on real data obtained from the operation of the communication web portal and the subsequent identification of the individual data clusters, a model is created that mathematically describes the dependence of economic variables (income from sales of services and selling price of services) on other variables (time of sales of services and field of offered services). Using this analysis, the model clusters together all data to parameterized sets of given properties. The main purpose of creating a model is the suitable classification of data. Consequently, it is possible to streamline the sale of the services and maximize the profits of web portals offering this type of service.

  • [2] Ling Y.,Chai C., Hou W., Hei D., Qing S.,Jia W. (China)
    A new method for nuclear accident source term inversion based on ga-bpnn algorithm, pp. 71-82

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.006

    Abstract: Rapid and accurate prediction and evaluation of accident consequences can provide scientific basis for decision-making of nuclear emergency measures. Accident source term estimation under reactor accident conditions is an important part of nuclear accident consequence evaluation. In order to accurately estimate the information of radioactive source terms released from nuclear power plants to the environment, an inversion model of accident source terms based on BP neural network algorithm (BPNN) was constructed. And to resolve the defect that BPNN is easy to fall into local minimum during training process, genetic algorithm (GA) was used to optimize the weights and thresholds of BPNN. In this paper, referring to the release rates of radioactive source term from the Fukushima nuclear accident. The release rates of 131I and 137Cs diffused into the environment in stable atmosphere were taken as the two target outputs of the GA-BPNN, and the meteorological data for one hour at fixed monitoring points were taken as the target inputs. And the simulation results showed that for the release rate of 131I and 137Cs, the mean relative errors of the training and the testing sample sets were both below 2% which indicates that the GA-BPNN model not only improves the shortcoming of BPNN, but also increases the speed and accuracy of source term inversion.

  • [3] Cheng Y., Ye Z., Wang M., Zhang Q. (China)
    Document classification based on convolutional neural network and hierarchical attention network, pp. 83-98

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.007

    Abstract: Numerous studies have demonstrated that the neural network model can achieve satisfactory performance in various natural language processing (NLP) tasks. In recent years, document classification is one of the NLP tasks that has gain considerable attention from researchers. For NLP tasks, convolutional neural network (CNN), recurrent neural network (RNN) and attention mechanism can be used. In this work, it is assumed that a document can be divided into two levels, word level and sentence level. In this paper, an effective and novel model called C-HAN (Convolutional Neural Network-based and Hierarchical Attention Network with RNN as basic units-based model) is proposed for document classification by combining the advantages of CNN, RNN and attention model. The CNN is used to extract the abstract relations between different words that are then fed into an attention based bidirectional long short-term memory recurrent neural network (Bi-LSTM) to obtain the high-level abstract representation of sentences. The representation of a document consists of sentences is obtained by using another attention based Bi-LSTM. Lastly, the classification ability of the proposed C-HAN model is evaluated on two datasets. The experimental results demonstrate that the C-HAN model outperforms previous deep learning methods and achieves the state-of-art performance.

  • [4] Ata A., Khan M.A., Abbas S., Ahmad G., Fatima A. (Pakistan)
    Modelling smart road traffic congestion control system using machine learning techniques, pp. 99-110

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.008

    Abstract: By the dramatic growth of the population in cities requires the traffic systems to be designed eciently and sustainably by taking full advantage of modern-day technology. Dynamic traffic flow is a significant issue which brings about a block of traffic movement. Thus, for tackling this issue, this paper aims to provide a mechanism to predict the traffic congestion with the help of Artificial Neural Networks (ANN) which shall control or minimize the blockage and result in the smoothening of road traffic. Proposed Modeling Smart Road Traffic Congestion Control using Artificial Back Propagation Neural Networks (MSR2C-ABPNN) for road traffic increase transparency, availability and efficiency in services offered to the citizens. In this paper, the prediction of congestion is operationalized by using the algorithm of backpropagation to train the neural network. The proposed system aims to provide a solution that will increase the comfort level of travellers to make intelligent and better transportation decision, and the neural network is a plausible approach to find traffic situations. Proposed MSR2C-ABPNN with Time series gives attractive results concerning MSE as compared to the fitting approach.


1/2019

  • [1] Vaitová M., Štemberk P., Rosseel T.M. (CZ)
    Fuzzy logic model of irradiated aggregates , pp. 1-18

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.001

    Abstract: The worldwide need for nuclear power plant (NPP) lifetime extension to meet future national energy requirements while reducing greenhouse gases raises the question of the condition of concrete structures exposed to ionizing radiation. Although research into the effects of radiation has a long history and the phenomenon of deterioration of concrete due to irradiation is not yet completely understood, the main assumed degradation mode is radiation-induced volumetric expansion of aggregates. There are experimental data on irradiated concrete obtained over decades under different conditions; however, the collection of data exhibits considerable scatter. Fuzzy logic modeling offers an effective tool that can interconnect various data sets obtained by different teams of experts under different conditions. The main goal of this work is to utilize available data on irradiated concrete components such as minerals and aggregates that expand upon irradiation. Furthermore, aggregate radiation-induced volumetric expansion gives an estimate of the change in mechanical properties of aggregate after years of reactor operation. The mechanical properties of irradiated aggregate can then be used for modeling irradiated concrete in the actual NPP structure based on the composition of concrete, the average temperature on the surface of the biological shield structure, and the neutron dose received by biological shield.

  • [2] Snor J., Kukal J., Van Tran Q. (CZ)
    SOM in Hilbert space , pp. 19-31

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.002

    Abstract: The self organization can be performed in an Euclidean space as usually defined or in any metric space which is generalization of previous one. Both approaches have advantages and disadvantages. A novel method of batch SOM learning is designed to yield from the properties of the Hilbert space. This method is able to operate with finite or infinite dimensional patterns from vector space using only their scalar product. The paper is focused on the formulation of objective function and algorithm for its local minimization in a discrete space of partitions. General methodology is demonstrated on pattern sets from a space of functions.

  • [3] Fu X.Y., Luo H., Zhang G.Y., Zhong S.S. (China)
    A lazy support vector regression model for prediction problems with small sample size, pp. 33-44

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.003

    Abstract: Prediction problems with small sample size are problems which widely exist in engineering application. Because lazy prediction algorithms can utilize the information of predicted individual, it is often possible for them to achieve better predictive effect. Traditional lazy prediction algorithms generally use sample information directly, and therefore the predictive effect still has room for improvement. In this paper, we combine support vector regression (SVR) with lazy prediction algorithm, and propose a lazy support vector regression (LSVR) model. The insensitive loss function in LSVR depends on the distance between the individual in training sample set and the predicted individual. The smaller the distance, the smaller the lossless interval of the individual in training sample set, which means that the individual in training sample set has a great impact on the predicted individual. To solve the LSVR model, a generalized Lagrangian function is introduced to obtain the dual problem of the primal problem, and the solution to the primal problem is obtained by solving the dual problem. Finally, three numerical experiments are conducted to validate the predictive effect of LSVR. The experimental results show that the predictive effect of LSVR is better than those of e-SVR, neural network (NN) and random forest (RF), and it is also better than that of k-nearest neighbor (k-NN) algorithm when the sample size is not too small and the distance between the predicted individual and the individual in training sample set is not too large. Therefore, LSVR not only has the advantage of good generalization ability of traditional SVR, but also has the advantage of good local accuracy of lazy prediction algorithm.

  • [4] Yildirim O., Baloglu U.B. (Turkey, UK) ,
    RegP: A new pooling algorithm for deep convolutional neural networks, pp. 45-60

      Full text     DOI: http://dx.doi.org/10.14311/NNW.2019.29.004

    Abstract: In this paper, we propose a new pooling method for deep convolutional neural networks. Previously introduced pooling methods either have very simple assumptions or they depend on stochastic events. Different from those methods, RegP pooling intensely investigates the input data. The main idea of this approach is finding the most distinguishing parts in regions of the input by investigating neighborhood regions to construct the pooled representation. RegP pooling improves the efficiency of the learning process, which is clearly visible in the experimental results. Further, the proposed pooling method outperformed other widely used hand-crafted pooling methods on several benchmark datasets.