has been used to estimate the risks of drought in the Tagus river basin. This is the longest river in the Iberian Peninsula. In the methodology applied, an essential phase is the generation of multiple future hydrological scenarios spanning several months in the future (between 24 and 60, depending on the basin); these are conditioned to the hydrological situation at the moment of the inquiry to the DSS. These multiple future scenarios are used to simulate the management of the water resource in the basin; the results are used in turn to estimate the statistics of future deficits (i.e. means, probabilities and distribution functions) and statistics of state variables (e.g. reservoir storage). Most of the rivers in the Iberian Peninsula experience droughts that last for several months. Hence, it is crucial that the synthetically generated future scenarios reproduce closely the statistics related to drought and storage. Presently, the DSS allows for synthetic data generation by means of classical stochastic models, such as autoregressive moving average (ARMA) models. The present work explores the possibilities of usingneuralnetworks (NN) as generators of future scenarios, with emphasis on the ability to reproduce the statistics related to drought and storage. This application relates to a study on a sub-basin of the Tagus river basin involving two reservoirs, the Entrepeñas with a capacity of 1639 hm³ that regulates the river Guadiela, a tributary of the Tagus, and the Buendía, in the main course, with a capacity of 803 hm³. The location of the reservoirs is indicated in Fig. 1. Both reservoirs are used for irrigation, production of hydroelectricity, and urban water supply (CHT, 1999). If the results of this study show that NN are useful, the work can be extended to the other sub-basins, where multivariate modelling with more than two sites will be necessary.
RegnANN is a novel method for reverse engineering gene networksbasedon an ensemble of multilayer perceptrons. The algorithm builds a regressor for each gene in the network, estimating its neighborhood independently. The overall network is obtained by joining all the neighborhoods. RegnANN makes no assumptions about the nature of the relationships between the variables, potentially capturing high-order and non linear dependencies between expression patterns. The evaluation focuses onsynthetic data mimicking plausible submodules of larger networks and on biological data consisting of submodules of Escherichia coli. We consider Barabasi and Erdo¨s-Re´nyi topologies together with two methods for data generation. We verify the effect of factors such as network size and amount of data to the accuracy of the inference algorithm. The accuracy scores obtained with RegnANN is methodically compared with the performance of three reference algorithms: ARACNE, CLR and KELLER. Our evaluation indicates that RegnANN compares favorably with the inference methods tested. The robustness of RegnANN, its ability to discover second order correlations and the agreement between results obtained with this new methods on both synthetic and biological data are promising and they stimulate its application to a wider range of problems.
Many authors showed the applicability of ANN in different nontraditional machining processes such as, Kuo Tsai et la., showed the ability of different neuralnetworks models to predict the surface finish basedon the effect of changing the electrode polarity in the EDM process. . Mohan Sen et la.,presented a hybridneural network and genetic algorithm (NNGA) approach for the multi-response optimization of the electro jet drilling (EJD) process, where ANN model was used to predict the response parameters of the process and then, a genetic algorithm was applied to the trained neural network model to obtain the optimal process parameters .whereas, Soleimanimehr h. et la, Developed an artificialneural network (ANN) for prediction of aluminum workpieces' surface roughness in ultrasonicvibration assisted turning (UAT) and also, investigated the effect of tool wear as the main cause of surface roughness .
All the engineering issues will have plenty of solutions. The greatest task is to select the best from the available solutions. ArtificialNeural Network (ANN) is a model of biological neuron system. ANN can be trained by known results and the knowledge acquired from training can be used to foretell or compute the unknown output. The most popular and broadly acclimated ANN is Multi Layer Back Propagation Network (MLBPN). In BPN the weights of input-hidden layers and hidden-output layers are computed using gradient search method. This will lead the network to local optimum solutions. Moreover, the BPN is unable to work with new occurrence far from training. This research work aims to abbreviate the drawbacks of BPN by affiliation of Genetic Algorithm (GA) with a back propagation network. Although GA is not an assuring global solution it is found that GA is able of bearing adequate acceptable solutions. GA may even be accomplishing the results with less number of iterations. GA was visualized by Holland in 1975 and applied auspiciously in structural engineering. Later it was extended to all fields due to its distinctive features like random search basedon natural genetics, population of points at time, etc. Anatomy of proposed hybrid genetic algorithm —neuralnetworks system is shown in Fig. 5. The program generates preliminary population randomly. The fitness of each chromosome in population will be evaluated by weights of genes. If convergence is not reached, genetic operations reproduction, crossover and mutation will be carried out to decide the new population. Again, the fitness of new population is checked and this system will continue until the function is converged. If the convergence is attained, the program stops and gives the result. The various steps of proposed genetic algorithm basedneural network system to predict the tool wear are discussed as following. In this study, an artificialneural network with input, hidden and output layers was thought about. The number of neurons and parameters used in each layer is shown in Table 5.
A deep learning based solution for performing measurements of knee range of motion using convolutional neuralnetworks is presented, supported by the generation of a synthetic dataset. A 3D human-body model was used to generate realistic and varied images, perpendicular to the plane defined by the lower limb, simulating a patient lying pronate on a treatment table with the leg flexed at arbitrary angles. Such images were labeled, being registered the coordinates of three key points for the calculation of the flexion angle: the centers of the thigh, the knee and the lower leg. This data was used to train convolutional neuralnetworks for performing regression and predicting these six coordinates. Transfer learning was used with the VGG16 and InceptionV3 models pre-trained on the ImageNet dataset. An additional custom model was trained from scratch. These networks were tested under various conditions using different combinations of image augmentation techniques on the training sets. Out of the three architectures, Eva achieved the best results, closely followed by InceptionV3. VGG16’s results were unsatisfactory. Posterior testing was performed using a few real images to test how well the network generalized, and the results were also favorable.
The architectural diagram of a typical voice and speaker recognition system is shown in Figure 1. The system is trained to recognize the voice of individual speakers with each speaker providing specific sets of utterances through a microphone terminal or telephone. The captured analog voice waveform has three components: speech segment, silence or non-voiced segment, and background noise signals. To extract the relevant speech signals, the voice waveform is digitized and signal processing is carried out to remove the noise signals and the silence or non-voiced components. Any relevant information that is ignored during this processing is completely considered as lost and conversely, any irrelevant information such as fundamental frequency of the speaker and the characteristics of the microphone that is allowed to pass is treated as useful with implications on the speech feature classification performance. The extracted speech signals are then converted into streams of template feature vectors of the voice pattern for classification and training. In the event that irrelevant information is allowed, then the speech features that may be generated from the corrupted speech signals may no longer be similar to the class distributions that are learned from the training data. The system recognizes the voice of individual speakers by comparing the extracted speech features of their utterances with the respective template features invoked from the training systems. The GMM recognizer computes scores that are used for the matching of the most distinctive speech features of speakers. The decision criteria for the voice recognition of speakers were basedon correlation analysis of the speech features of speakers from the ANN and GMM.
The present study will be done to determine land use level change of Siah Mazgi basin forests of shaft city and explain the factors influencing these changes using remote sensing and GIS. Basedon this, Landsat satellite images of 2000 and 2013 will be used. Research methodology will be done usingneural network classification method. Neuralnetworks method is considered as a proper method for classification of land use and land cover, because it can be used for all kinds of data in various statistical scales (Chavez 1988). One of the most important features of neural network is its independence to statistical distribution of the input data. This important feature of neuralnetworks
ABSTRACT: Digital soil mapping is an alternative for the recognition of soil classes in areas where pedological surveys are not available. The main aim of this study was to obtain a digital soil map usingartificialneuralnetworks (ANN) and environmental variables that express soil- landscape relationships. This study was carried out in an area of 11,072 ha located in the Barra Bonita municipality, state of São Paulo, Brazil. A soil survey was obtained from a reference area of approximately 500 ha located in the center of the area studied. With the mapping units identified together with the environmental variables elevation, slope, slope plan, slope profile, convergence index, geology and geomorphic surfaces, a supervised classification by ANN was implemented. The neural network simulator used was the Java NNS with the learning algorithm "back propagation." Reference points were collected for evaluating the performance of the digital map produced. The occurrence of soils in the landscape obtained in the reference area was observed in the following digital classification: medium-textured soils at the highest positions of the landscape, originating from sandstone, and clayey loam soils in the end thirds of the hillsides due to the greater presence of basalt. The variables elevation and slope were the most impor- tant factors for discriminating soil class through the ANN. An accuracy level of 82% between the reference points and the digital classification was observed. The methodology proposed allowed for a preliminary soil classification of an area not previously mapped using mapping units obtained in a reference area.
Salloum & Kuo  proposed an aggregation of RNN architectures for analysis and classification after selecting QRS segments without further feature extraction methods. For the identification problem, it was reported nearly 100% classification accuracy on the ECG-ID database. Values of EER between 3.5% and 0% in an authentication task using the MIT-BIH database. This proved the efficacy of RNNs for this kind of problem. Zhang et al.  used a multiresolution one-dimensional CNN, where the feature extraction step comprises discrete wavelet transform, autocorrelation and component selection, to obtain 93.5% identification rate as the average result on 8 different datasets. Labati et al.  also applied a 1D CNN to identify patients using the IDEAL database as well as the PTB diagnostic ECG database, achieving 100% accuracy on the latter. Zhao et al.  obtained promising results in a recently published ECG dataset  by feeding a 2D CNN with ECG trajectories obtained by applying generalized S transforms to the signals. Wu et al.  reached state of the art performance on MIT-BIH Arrhythmia and ECG-ID by applying a two-stage approach, consisting of a 1D CNN, followed by an attention-based bidirectional LSTM.
UPFC is one of the famous FACTs devices that is used to improve power system stability. Figure 1 shows a single machine infinite bus (SMIB) system (Heffron-Philips model of a power system installed with UPFC) with UPFC. It is assumed that the UPFC performance is basedon pulse width modulation (PWM) converters. In figure 1 m e , m b and δ e , δ b are the amplitude modulation ratio and phase angle of control signal of
The relationship between the basic density and wood moisture after saturation showed a Pearson correlation coefficient of 0.995, which can be explained by the fact that materials with lower basic density have higher void volume, such as lumen fibers, vessel elements, and other cavities filled with water during the saturation period, which increases the wood moisture content (ENGELUND et al., 2013; SIAU, 1971). Moreover, these spaces facilitate the release of water (KOLLMANN; CÔTÉ, 1968). After 1.5 days of drying, the relationship between the basic density and moisture was inversely proportional, but with a lower ratio between these parameters. At seven days of drying, a Pearson correlation coefficient of 0.737 showed that materials with lower basic density had lower moisture due to their higher drying rate, and this trend continued until the end of drying. After 15 days of drying, the wood moisture stabilized, indicating that the samples reached the equilibrium moisture content, but its value differed with the wood density. High Pearson correlation coefficients between basic density and moisture during drying are essential to produce artificialneural network models with high accuracy.
The transformations that medical practice has undergone in recent years – especially with the incor- poration of new information technologies – point to the need to broaden discussions on the teaching- -learning process in medical education. The use of new computer technologies in medical education has shown many advantages in the process of acquiring skills in problem solving, which encourages creativity, critical thinking, curiosity and scientific spirit. In this context, it is important to highlight artificialneuralnetworks (ANN) – computer systems with a mathematical structure inspired by the human brain – which proved to be useful in the evaluation process and the acquisition of knowledge among medical students. The purpose of this communication is to review aspects of the application of ANN in medical education.
Abstract— In this paper an artificialneural network (ANN) model has been developed for design of circular microstrip patch antenna. The major advantage of the proposed approach is that, after proper training, proposed neuralmodel completely bypasses the repeated use of complex iterative process for design of such types of antennas, thus resulting in an extremely fast solution with high accuracy. The difficulty in designing microstrip patch antennas is due to the involvement of a large number of physical parameters and hence their associated optimal values. It is indeed very difficult to formulate an exact numerical solution through empirical studies basedon practical observations. In order to circumvent this problem, an alternative solution is achieved usingartificialneuralnetworks. The proposed technique used feed-forward back-propagation artificialneural network (FFBP-ANN) with one hidden layer& trained by Levenberg-Marquardt back propagation learning algorithm as an approximate model for design of circular micro strip patch antennas with reasonable accuracy. IE3D software has been used to generate data dictionary for training and validation set of ANN. Also Results of proposed ANN models are compared with simulated and theoretical values, these results are found in agreement with same.
Road extraction and vehicle detection using aerial imagery involves information and data related to GIS and these data need to be updated every certain period of time . Road extraction and vehicle detection for incident detection in aerial imagery is a new, controversial issue in computer vision, which also influences many other projects and operations, such as traffic control and incident detection on highways. Some other real-time applications using sensors  or ground cameras have been implemented for traffic control, such as [16- 20]. Since the development of new satellite systems such as Quick Bird, IKONOS and Geoeye-2, remote sensing imagery is available with a 0.25 meter resolution. Vehicles can be observed clearly in this type of satellite images. Some vehicle detection methods have been studied using aerial imagery [21,22,23]. The study most closely related to this paper is . In this study the authors have developed a system for traffic incident recognition using aerial imagery. Their approach focused on road extraction and incident detection using the Radon transform method. The main difference between our system and  is a new input feature for incident and bottleneck detection. This feature is traffic flow, which helps us improve the detection rate.
The three first steps in the data mining process regards the collection, understanding and preparation of data, which has partially been done outside of SAS Miner. These three steps are essential because real-world data has issues that influence (highly) the models, hence why missing or inconsistent records needed to be removed from consideration (Hand, 2007). As for data exploration, the software has several nodes that output the characteristics of the data set, such as StatExplore and MultiPlot. These will give a better idea on what variables are worth the most to the classification process.
2004). They have been used to solve complex problems that are difficult to be solved if not impossible by the conventional approaches, such as control, optimization, pattern recognition, classification, and so on, Specially it is desired to have the minimum difference between the predicted and observed (actual) outputs (Richon and Laugier, 2003). Artificialneuralnetworks are biological inspirations basedon the various brain functionality characteristics. They are composed of many simple elements called neurons that are interconnected by links and act like axons to determine an empirical relationship between the inputs and outputs of a given system. Multiple layers arrangement of a typical interconnected neural network is shown in Figure (1). It consists of an input layer, an output layer, and one hidden layer with different roles. Each connecting line has an associated weight. Artificialneuralnetworks are trained by adjusting these input weights (connection weights), so that the calculated outputs may be approximated by the desired values. The output from a given neuron is calculated by applying a transfer function to a weighted summation of its input to give an output, which can serve as input to other neurons, as follows (Gharbi, 1997).
Benefits of genetic algorithms include robustness and ease of implementation, but suffer from the number of runs to produce the final solutions as it may stick into local optimum points. 30 different runs are an acceptable minimum . Back propagation of neural network is a "typical delegate". Neural Network may suffer from local minimum and also from the slowness of convergence. Hence, a combination of GA and ANN may overcome these issues. ANN can be used for training and to build prediction models while GA can be used to speed up the process of ANN by tuning up the design parameters of the ANN . It must be noted that a good design of the model with tuning some parameters can enhance the speed of convergence . Combining neuralnetworks and GA in one model has been investigated and built by different researchers -.
When using WT an important decision has to be made: the selection of a suitable mother wavelet, in the particular STLF case, the choice of Daubechies family wavelets is almost consensual, with special emphasis on Daubechies of order 2 and 4 (db2 and db4) , , , . However, larger temporal data sets, specifically a load time series, can present considerable variability over time. Thus, the best mother (base) wavelet for a certain time interval may not be the most appropriate for others, for example in  the authors found that different orders Daubechies were best suited for each season of the year. The same reason led the authors in  to implement a wavelet-based ensemble scheme composed by eight mother wavelets, sectioning the load data series and finding the mother wavelet that best fits each section. Yet when selecting the eight best candidates, the authors also relied on the consensus about Daubechies in selecting half the base wavelets, while the remaining four where picked from the Coiflets family, basedon experimental results.
The RT+ANN approach was applied to this power system, in order to derive security structures to be used for security assessment purposes related with a critical pre-selected disturbance. The following disturbance was considered: short- circuit in the eastern side of the island, causing the disconnection of all wind (Pw1 to Pw6), and Pc7 to Pc8 power plants. This regards to situations where the dynamic security of the system is reduced in case of short-circuits that take place near to power production facilities, leading to these facilities disconnection (due to under-voltage conditions). As wind parks sites are more expose to adverse climatic conditions, short-circuits usually take place near these facilities. This disturbance was selected by the utility as one of the most important to be included as a security restriction within the system operating policies, and therefore to be considered in a security assessment process. In fact, this is a particular severe disturbance that may provoke large frequency drops, leading to load shedding activation, or to system instability. Basedon the set-point values of the installed load shedding relays, the system was considered to lose security if the negative frequency deviations ('f) go bellow –2 Hz.
The formulation of products with low levels of saturated and trans fatty acids is a new challenge for industries, and alternative raw materials have been studied. Artificialneuralnetworks (ANNs) have been used for this process. The objective of the present study was to formulate blends, with the help of an ANN, using soybean-based interesterified fats for the production of a zero trans fat margarine similar to a margarine produced using a specific commercial fat. The software was trained with three raw materials to generate formulations with a solid fat content (SFC) and a melting point (MP) similar to specific commercial fats. The SFC, MP, fatty acid and triacylglycerol composition were determined for all ANN blends and commercial fats. Margarines were produced in a pilot plant and evaluated for consistency and stability under temperature cyclization. The ANN showed efficiency in to predict SFC and MP of the suggested formulations, although there were differences at low temperatures for the desired SFC. Differences in the consistency of the commercial fats and ANN blends were observed; however, the margarines produced in the pilot plant had a similar consistency. The margarine prepared with ANN formulation had a higher emulsion stability. Overall, the margarine produced with ANN formulation had characteristics very similar to margarine produced with the commercial fat, and the margarine with soybean-based fat contained reduced saturated and trans fat levels.