Artificial Network Model Research Articles

PENGGUNAAN MODEL JARINGAN SARAF TIRUAN (ARTIFICIAL NEURON NETWORK) UNTUK MEMPREDIKSI HASIL TANDAN BUAH SEGAR (TBS) KELAPA SAWIT BERDASAR CURAH HUJAN DAN HASIL TBS SEBELUMNYA

Abstrak. Untuk memprediksi hasil kelapa sawit 2018 di 4 kebun percobaan Pusat Penelitian Kelapa Sawit (Padang Mandarsah, Dalu-dalu, Bukit Sentang, dan Aek Pancur), maka disusun model jaringan saraf tiruan (JST). Data yang digunakan untuk menyusun model ini adalah hasil dan curah hujan bulanan pada 2013-2017. Keluaran model diperolah dari hubungan non-linear Autoregressive dengan masukkan eksternal (External/Exogenous input, NARX) yaitu curah hujan bulanan. Proses pemodelan meliputi training model menggunakan data 2013-2015, validasi model menggunakan data 2016, dan pengujian model menggunakan data 2017. Dari hasil pengujian model diperolah arsitektur model JST n-d-h-o (peubah input,n ; d-tapped delayed , d, node tersembunyi, h; lapisan keluaran, o) yang sesuai untuk kebun Padang Mandarsah , Dalu-dalu, Bukit Sentang, dan Aek Pancur berturut-turut adalah 2-3-4-1, 2-24-5-1, 2-24-10-1, dan 2-3-5-1 dengan hubungan keeratan (r) antara keluaran model dengan data aktual berturut-turut 0,84; 0,74; 0,84; dan 0,86.

Multiplex networks of musical artists: The effect of heterogeneous inter-layer links

The way the topological structure goes from a decoupled state into a coupled one in multiplex networks has been widely studied by means of analytical and numerical studies, involving models of artificial networks. In general, these experiments assume uniform interconnections between layers offering, on the one hand, an analytical treatment of the structural properties of multiplex networks but, on the other hand, losing applicability to real networks where heterogeneity of the links’ weights is an intrinsic feature. In this paper, we study 2-layer multiplex networks of musicians whose layers correspond to empirical datasets containing, and linking the information of: (i) collaboration between them and (ii) musical similarities. In our model, connections between the collaboration and similarity layers exist, but they are not ubiquitous for all nodes. Specifically, inter-layer links are created (and weighted) based on structural resemblances between the neighborhood of an artist, taking into account the level of interaction at each layer. Next, we evaluate the effect that the heterogeneity of the weights of the inter-layer links has on the structural properties of the whole network, namely the second smallest eigenvalue of the Laplacian matrix (algebraic connectivity). Our results show a transition in the value of the algebraic connectivity that is far from classical theoretical predictions where the weight of the inter-layer links is considered to be homogeneous.

Realization of Direction Selective Motor Learning in the Artificial Cerebellum: Simulation on the Vestibuloocular Reflex Adaptation.

The vestibule-ocular reflex (VOR) has been one of the most popular model systems to investigate the role of the cerebellum in adaptive motor control. VOR motor learning can be experimentally induced by continuous application of combination of head rotating stimulus and optokinetic stimulus. For instance, in phase application of those stimuli decreases VOR gain defined by eye velocity of VOR in the dark divided by head velocity, while out of phase of those increases VOR gain. It has been known that VOR gain is modifiable context-dependently. Namely, VOR gains for leftward and rightward head rotations can be respectively increased and decreased simultaneously. The cerebellar signal processing underlying the context dependent VOR motor learning, however, is not fully uncovered. In the present study, we simulated direction selective VOR motor learning, using the artificial cerebellar neuronal network model that we developed to understand the origin of the cerebellar motor learning.

Predicting energy Ccnsumption using artificial neural networks: a case study of the UAE

Predicting energy consumption is very important for improving resource planning and for more efficient production. This study uses artificial neural network (ANN) models to predict energy consumption in the United Arab Emirates (UAE). The multilayer perceptron model (MLP) and Radial Basis Function (RBF) were used for this purpose. Historical input and output data related to the long-term energy consumption in the UAE were used for training, validation, and testing. The developed neural network models were compared to find the most suitable model with high accuracy.

A disjunctive program formulation to generate regular public transit timetables adhering to prioritized planning requirements

Timetable regularity, that is, equability of headways, is an important measure for service quality in high frequency public transit systems, assuring an evenly distributed passenger load as well as improving product attractiveness. However, to be feasible during daily operation a timetable may also have to adhere to other planning requirements, such as departure time coordination with other service providers or deliberately short headways to reduce the passenger load of follow‐up vehicles. In this article, a disjunctive program formulation combining aspects of two previous optimization models is proposed, to generate regular public transit timetables adhering to planning requirements. The modeled requirements not only allow for the consideration of feasibility constraints from daily operations, but also for the consideration of simultaneous departures for transfer connections, an objective traditionally opposed to regularity. To show its applicability the approach is applied to two models of artificial transit networks as well as to models of the public transit network of Cologne, Germany. The results show that the proposed formulation can be used to generate timetables for network instances of realistic size in acceptable time. For networks consisting of multiple connected components it is shown that a decomposition approach can significantly reduce run times.

Application of Artificial Neural Networks to the prediction of out-of-plane response of infill walls subjected to shake table

The main purpose of this paper is to predict missing absolute out-of-plane displacements and failure limits of infill walls by artificial neural network (ANN) models. For this purpose, two shake table experiments are performed. These experiments are conducted on a 1:1 scale one-bay one-story reinforced concrete frame (RCF) with an infill wall. One of the experimental models is composed of unreinforced brick model (URB) enclosures with an RCF and other is composed of an infill wall with bed joint reinforcement (BJR) enclosures with an RCF. An artificial earthquake load is applied with four acceleration levels to the URB model and with five acceleration levels to the BJR model. After a certain acceleration level, the accelerometers are detached from the wall to prevent damage to them. The removal of these instruments results in missing data. The missing absolute maximum out-of-plane displacements are predicted with ANN models. Failure of the infill wall in the out-of-plane direction is also predicted at the 0.79 g acceleration level. An accuracy of 99% is obtained for the available data. In addition, a benchmark analysis with multiple regression is performed. This study validates that the ANN-based procedure estimates missing experimental data more accurately than multiple regression models.

Compressive strength estimation of concrete containing zeolite and diatomite: An expert system implementation

In this study, we analyze the behavior of concrete which contains zeolite and diatomite. In order to achieve the goal, we utilize expert system methods. The utilized methods are artificial neural network and adaptive network-based fuzzy inference systems. In this respect, we exploit seven different mixes of concrete. The concrete mixes contain zeolite, diatomite, mixture of zeolite and diatomite. All seven concrete mixes are exposed to 28, 56 and 90 days\' compressive strength experiments with 63 specimens. The results of the compressive strength experiments are used as input data during the training and testing of expert system methods. In terms of artificial neural network and adaptive network-based fuzzy models, data format comprises seven input parameters, which are; the age of samples (days), amount of Portland cement, zeolite, diatomite, aggregate, water and hyper plasticizer. On the other hand, the output parameter is defined as the compressive strength of concrete. In the models, training and testing results have concluded that both expert system model yield thrilling medium to predict the compressive strength of concrete containing zeolite and diatomite.

An automated data-driven tool to build artificial neural networks for predictive decision-making

We propose the development of an automated data-driven tool to assist data analysts in building an optimal artificial neural network (ANN) model to solve their domain-specific problems for predictive decision making. The proposed approach combines the strengths of both sequential training methods and multi-hidden-layer learning algorithms to dynamically learn the best-fitted parameters, including learning rate (LR), momentum rate (MR), number of hidden layers (NHL), and number of neurons in each hidden layer (NNHL), for the given set of key input attributes and multiple output nodes. Specifically, the contributions of this work are three-fold: 1) develop the new extended algorithm, i.e., multidimensional parameter learning (MPL), to learn the optimal ANN parameters; 2) provide the user-friendly GUI tool for data analysts to maintain the data manipulations and the tool operations; 3) conduct the experimental case study, i.e., determining the severity level of Alzheimer's patients, to present the superior result (i.e., 95.33%) in terms of prediction accuracy and model complexity by using the learned parameters (i.e., LR = 0.6, MR = 0.8, NHL = 2, NNHL at the 1st layer = 28, and NNHL at the 2nd layer = 24) from the MPL algorithm.

Hybrid fuzzy model to predict strength and optimum compositions of natural Alumina-Silica-based geopolymers

This study introduces the supervised committee fuzzy model as a hybrid fuzzy model to predict compressive strength (CS) of geopolymers prepared from alumina-silica products. For this purpose, more than 50 experimental data that evaluated the effect of Al2O3/SiO2, Na2O/Al2O3, Na2O/H2O and Na/[Na+K] on (CS) of geopolymers were collected from the literature. Then, three different Fuzzy Logic (FL) models (Sugeno fuzzy logic (SFL), Mamdani fuzzy logic (MFL), and Larsen fuzzy logic (LFL)) were adopted to overcome the inherent uncertainty of geochemical parameters and to predict CS. After validating the model, it was found that the SFL model is superior to MFL and LFL models, but each of the FL models has advantages to predict CS. Therefore, to achieve the optimal performance, the supervised committee fuzzy logic (SCFL) model was developed as a hybrid method to combine the benefits of individual FL models. The SCFL employs an artificial neural network (ANN) model to re-predict the CS of three FL model predictions. The results also show significant fitting improvement in comparison with individual FL models.

On attracting sets in artificial networks: cross activation

Mathematical models of artificial networks can be formulated in terms of dynamical systems describing the behaviour of a network over time. The interrelation between nodes (elements) of a network is encoded in the regulatory matrix. We consider a system of ordinary differential equations that describes in particular also genomic regulatory networks (GRN) and contains a sigmoidal function. The results are presented on attractors of such systems for a particular case of cross activation. The regulatory matrix is then of particular form consisting of unit entries everywhere except the main diagonal. We show that such a system can have not more than three critical points. At least n–1 eigenvalues corresponding to any of the critical points are negative. An example for a particular choice of sigmoidal function is considered.

Development of Heavy Rain Damage Prediction Function Using Artificial Neural Network and Multiple Regression Model

Development of Heavy Rain Damage Prediction Function Using Artificial Neural Network and Multiple Regression Model

Scriptaid cause histone deacetylase inhibition and cell cycle arrest in HeLa cancer cells: A study on structural and functional aspects

Scriptaid (SCR), a well-known histone deacetylase inhibitor, cause various cellular effects such as cell growth inhibition and apoptosis. In this study, we have evaluated the anti-cancer effects of Scriptaid in HeLa cells, IMR-32 and HepG2 cells. Scriptaid inhibited the growth of HeLa cells with IC50 of 2μM at 48h in a dose-dependent manner. Flow-cytometric analysis indicated that SCR induced apoptosis. Scriptaid was found to inhibit HDAC-8 effectively than other HDAC inhibitor such as TSA as observed by HDAC-8 assay, Western blotting and modelling study. This observation was further strengthened by an artificial neuronal network (ANN) model.

Development and formulation of floating tablet formulation containing rosiglitazone maleate using Artificial Neural Network

The aim of this study was to develop a new gastro-retentive formulation that has the capability to maintain the release of rosiglitazone in the gastric medium. For this purpose, we have used formulation by design approach and predicted the formulation characteristics of the final formulation by using Artificial Neural Network. Tablet formulations were prepared by using HPMC K100LV and HPMC K4M for the maintenance of sustained release and sodium bicarbonate and citric acid were used as the gas forming agents for floating. Possible interactions between the rosiglitazone maleate and the excipients were investigated by FT-IR and DCS analyses. Dissolution studies were performed in order to characterize the zero order release pattern of the formulations in three different media as pH:1 hydrochloric acid-potassium chloride, pH 2.5:phthalate and pH:4 acetate buffer solutions, under sink conditions. In the end, Artificial Network Model predicted and experimentally quantified results were compared with each other with f2 test. As a result, these two profiles are found to be similar with f2 values of 68.98, 59.98 and 62.99 for pH:1 hydrochloric acid-potassium chloride, pH: 2.5 phthalate and pH:4 acetate buffer solutions, respectively.

Comprehensive heat transfer correlation for water/ethylene glycol-based graphene (nitrogen-doped graphene) nanofluids derived by artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS)

Herein, artificial neural network and adaptive neuro-fuzzy inference system are employed for modeling the effects of important parameters on heat transfer and fluid flow characteristics of a car radiator and followed by comparing with those of the experimental results for testing data. To this end, two novel nanofluids (water/ethylene glycol-based graphene and nitrogen-doped graphene nanofluids) were experimentally synthesized. Then, Nusselt number was modeled with respect to the variation of inlet temperature, Reynolds number, Prandtl number and concentration, which were defined as the input (design) variables. To reach reliable results, we divided these data into train and test sections to accomplish modeling. Artificial networks were instructed by a major part of experimental data. The other part of primary data which had been considered for testing the appropriateness of the models was entered into artificial network models. Finally, predictad results were compared to the experimental data to evaluate validity. Confronted with high-level of validity confirmed that the proposed modeling procedure by BPNN with one hidden layer and five neurons is efficient and it can be expanded for all water/ethylene glycol-based carbon nanostructures nanofluids. Finally, we expanded our data collection from model and could present a fundamental correlation for calculating Nusselt number of the water/ethylene glycol-based nanofluids including graphene or nitrogen-doped graphene.

Moment-Based Spectral Analysis of Large-Scale Generalized Random Graphs

This paper investigates the spectra of the adjacency matrix and Laplacian matrix for an artificial complex network model—the generalized random graph. We deduce explicit expressions for the first four asymptotic spectral moments of the adjacency matrix and the Laplacian matrix associated with a generalized random graph with independent and identically distributed vertex weights. An estimate for the upper bound of the spectral radius is obtained on the basis of the fourth asymptotic spectral moment of the adjacency matrix, as well as the Laplacian matrix. These expressions are applied to study the behavior of a viral infection in a generalized random graph. On the basis of our results, we can design generalized random graphs with good antivirus ability when facing an initial virus infection. Numerical simulations agree with our analytical predictions.

Medical images segmentation based on improved three-dimensional pulse coupled neural network

Pulse coupled neural network (PCNN) is the third-generation model of artificial networks, which is based on the construction of the cat visual principle. When processing the image, it has a unique advantage, and PCNN is widely used in various fields, especially in the aspect of image segmentation, image fusion, and so on. However, the traditional PCNN model has a lot of problems, such as multi-parameters, parameter setting. Moreover, exponential decay mechanism will sometimes bring certain difficulty for image segmentation, etc. To solve these problems, a simplified and improved 3D-PCNN model is proposed in this paper, through which the whole 3D brain image segmentation is achieved. The experimental results show that, the 3D-PCNN algorithm reduced the segmentation time and improved the efficiency of segmentation when compared with the traditional 2D-PCNN model, the traditional 3D-PCNN algorithm and the 3D Otsu algorithm.

RETRACTED ARTICLE: A comparative study on parameters of leaf-shaped patch antenna using hybrid artificial intelligence network models

This study proposes a very compact coaxial-fed planar antenna for X band applications. The antenna design includes a tulip-shaped radiator on the FR4 dielectric substrate. The antenna parameters, such as return losses, bandwidth and operating frequency, have close relationships with patch geometry. In order to obtain desired antenna parameters for X band application, patch dimension is necessary to be optimized. In this article, four different hybrid artificial intelligence network models are suggested for optimization. These are particle swarm optimization, differential evolution, grey wolf optimizer and vortex search algorithm. Also, they are combined with artificial neural network for the purpose of estimating dimension of patch. Therefore, the comparison of different proposed algorithms is analyzed to obtain higher characteristics for antenna design. Their results are compared with each other in HFSS 13.0 software. The antenna with the most suitable return loss, bandwidth and operating frequency is selected to be used in antenna design.

Reliability and Survivability Analysis of Artificial Cobweb Network Model Used in the Low-Voltage Power-Line Communication System

In order to verify the feasibility of the artificial cobweb model used in low-voltage power-line communication (LVPLC) routing to improve the reliability of LVPLC, the all-terminal reliability of the artificial cobweb model has been calculated based on the factorization method in this paper. Analysis shows that the proposed artificial cobweb shows much better survivability than existing tree and ring network topologies. Simulation results of delay, throughput, and bit-error ratio are further presented to verify the correctness of the theoretical analysis about the reliability and survivability of LVPLC networks based on the artificial cobweb model, which further proves the feasibility of the artificial cobweb model for LVPLC routing. These results potentially provide a guideline to practical LVPLC system design.

Illumination of parameter contributions on uneven break phenomenon in underground stoping mines

One of the most serious conundrum facing the stope production in underground metalliferous mining is uneven break (UB: unplanned dilution and ore-loss). Although the UB has a huge economic fallout to the entire mining process, it is practically unavoidable due to the complex causing mechanism. In this study, the contribution of ten major UB causative parameters has been scrutinised based on a published UB predicting artificial neuron network (ANN) model to put UB under the engineering management. Two typical ANN sensitivity analysis methods, i.e., connection weight algorithm (CWA) and profile method (PM) have been applied. As a result of CWA and PM applications, adjusted Q rate (AQ) revealed as the most influential parameter to UB with contribution of 22.40% in CWA and 20.48% in PM respectively. The findings of this study can be used as an important reference in stope design, production, and reconciliation stages on underground stoping mine.

Computing the pull-in voltage of fixed–fixed micro-actuators by artificial neural network

In this study, a feed-forward backpropagation (FFBP) artificial neural network (ANN) model based on multilayer perceptron (MLP) for computing the pull-in voltage value of fixed–fixed micro-actuators is presented. The proposed ANN model is trained using the numerous pull-in voltage value results of clamped–clamped actuators, which have various physical properties, simulated by a software that employs the finite element method (FEM). As the presented ANN model has been trained for the simulation data, it implicitly takes into account the fringing field, mid-plane stretching and size effects. The accuracy and robustness of the results obtained by the proposed ANN model are verified by comparing with those of the theoretical ones through the simulation and experimental studies previously presented in the literature. The average percentage errors for training and testing data are found to be 0.30 and 0.40 %, respectively, while the maximum percentage error for literature data is calculated as 1.96 %. These results show that the main advantage of the presented ANN model gives satisfying pull-in voltage results over solution space of the simulations with effortless way.