Fourier convolution-parallel neural network framework with library matching for multi-tool processing decision-making in optical fabrication.

This thesis deals with recurrent neural networks, a particular class of artificial neural networks which can learn a generative model of input sequences. The input is mapped, through a feedback loop and a non-linear activation function, into a hidden state, which is then projected into the output space, obtaining either a probability distribution or the new input for the next time-step. This work consists mainly of two parts: a theoretical study for helping the understanding of recurrent neural networks framework, which is not yet deeply investigated, and their application to non-linear prediction problems, since recurrent neural networks are really powerful models suitable for solving several practical tasks in different fields. For what concerns the theoretical part, we analyse the weaknesses of state-of-the-art models and tackle them in order to improve the performance of a recurrent neural network. Firstly, we contribute in the understanding of the dynamical properties of a recurrent neural network, highlighting the close relation between the definition of stable limit cycles and the echo state property of an echo state network. We provide sufficient conditions for the convergence of the hidden state to a trajectory, which is uniquely determined by the input signal, independently of the initial states. This may help extend the memory of the network and increase the design options for the network. Moreover, we develop a novel approach to address the main problem in training recurrent neural networks, the so-called vanishing gradient problem. Our new method allows us to train a very simple recurrent neural network, making the gradient not to vanish even after many time-steps. Exploiting the singular value decomposition of the vanishing factors in the gradient and random matrices theory, we find that the singular values have to be confined in a narrow interval and derive conditions about their root mean square value. Then, we also improve the efficiency of the training of a recurrent neural network, defining a new method for speeding up this process. Thanks to a least square regularization, we can initialize the parameters of the network, in order to set them closer to the minimum and running fewer epochs of classical training algorithms. Moreover, it is also possible to completely train the network with our initialization method, running more iterations of it without losing in performance with respect to classical training algorithms. Finally, it is also possible to use it as a real-time learning algorithm, adjusting the parameters to the new data through one iteration of our initialization. In the last part of this thesis, we apply recurrent neural networks to non-linear prediction problems. We consider prediction of numerical sequences, estimating the following input choosing it from a probability distribution. We study an automatic text generation problem, where we need to predict the following character in order to compose words and sentences, and a path prediction of walking mobile users in the central area of a city, as a sequence of crossroads. Then, we analyse the prediction of video frames, discovering a wide range of applications related to the prediction of movements. We study the collision problem of bouncing balls, taking into account only the sequence of video frames without any knowledge about the physical characteristics of the problem, and the distribution over days of mobile user in a city and in a whole region. Finally, we address the state-of-the-art problem of missing data imputation, analysing the incomplete spectrogram of audio signals. We restore audio signals with missing time-frequency data, demonstrating via numerical experiments that a performance improvement can be achieved involving recurrent neural networks.

Induction machines have become the standard for highly demanding industrial applications. This has led to the utilization of modern discrete-time control techniques (such as model predictive control) that require the estimation of internal variables that are not subject to measurement (such as the rotational velocity in sensorless applications). From this point of view, it is fundamental to have accurate discrete-time models of induction machines, particularly given their nonlinear nature, so that control techniques perform according to design requirements. In spite of the above, the modeling of induction machines has not received much attention in the literature, even though more powerful machines and faster microcontrollers are currently being used. To better understand induction machine models for control, in this paper, we develop and compare various discrete-time models of induction machines based on Euler, Taylor, and Runge–Kutta methods. In addition, we compare the Extended Kalman Filter and Unscented Kalman Filter for state estimation in terms of accuracy and computational burden. The models are derived and compared through extensive Monte Carlo simulations and the state estimation techniques are compared in terms of root mean squared error, execution time, and maximum absolute error. Our simulations show that, in general, the Taylor method yields more accurate models than both the Runge–Kutta and Euler methods. In particular, the Taylor method results in a root mean square error that is one order of magnitude smaller than the Euler method for stator current and rotor flux linkages. For rotor angular speed, the Runge–Kutta methods are more accurate than both the Taylor and Euler methods, resulting in a root mean square error that is two orders of magnitude smaller than the Euler method. On the other hand, the Extended Kalman Filter results in smaller execution time than the Unscented Kalman Filter, up to two orders of magnitude. In terms of root mean squared error and maximum absolute error, both filtering algorithms perform similarly.

Force and vibration generated in apical direction by three endodontic files of different kinematics during simulated canal preparation: An in vitro analytical study.

Three-Dimensional Quantitative Comparative Analysis of Trapezial-Metacarpal Joint Surface Curvatures in Human Populations

The wide-field telescope will be an important tool in the discovery and birth of new theories of astronomy in future. The detectors in this kind of telescope mostly adopt a mosaic focal plane array. The focal depth of a large-F-number optical system is very small, and the flatness of the mosaic detector should be less than half the focal depth. In this paper, a new flatness measurement and data processing method is developed. A flatness measurement platform composed of a high-precision gantry platform and a coaxial multispectral displacement meter was built. The flatness of a single detector was measured under uncooled and cooled working conditions, and data processing was conducted, in which the root mean square (RMS) and the peak-valley (PV) values of the cooled detector were 0.0017 mm and 0.0112 mm, respectively. Next, a 2 × 6 mosaic model of the focal plane with a dimension of 148.5 mm × 168.5 mm was built using a metal detector model. The measurement platform was used to measure the flatness of the mosaic focal plane. According to the measurement results, the preliminary installation and mosaic adjustment were completed. The final RMS and PV values of the mosaic focal plane are 0.009 mm and 0.0808 mm, respectively. The experimental results show that the measurement and the data processing method can accurately reflect surface information on the detectors. This suggests that the method has great potential for use in ensuring the accuracy of wide-field telescope equipment in all applicable research areas.

The aim of the present research is to calculate the Relative Accuracy (R.A) and Root Mean Square Error (RMSE) for the location of Technical Institute, Mosul in Nineveh Governorate, Iraq. at the northern part of Mosul city by using surveying Instruments (GPS 1230 , Total Station 06 & Level NAK2), to calculate the coordinates and elevations of (16) Ground Control Points which covered the study area. The field data, remote sensing techniques and GIS software (Arc Map GIS10.3) are used to check the digital map of study area that helped the administrator to take the correct decision, determining positions of the future expansionist for buildings, roads, measuring lengths, areas and there coordinates, directions as well as the total costs estimation needed for any development from the final map. The gradient percentage between the highest and lowest points is calculated to be (9.954%). Relative accuracy for the based measured points (R.A) = 1:60000), and Root Mean Square error in Easting coordinates (RMSE) = ± 0.102m, in Northing coordinates (RMSN) = ± 0.096m, the Root Mean Square for the resultant (RMSR) = ± 0.099m, and Root Mean Square for Elevations ( RMSRL)= ± 0.005m. We Matched and correlated (10) measured points from study area with their graphic data positions on the aerial photo, determining the offsets for position and elevations as follows: (RMSE = ± 0.076m), (RMSN = ± 0.125m), (RMSR = ± 0.088m), and (Root Mean Square for Elevations = RMSRL= ± 0.045m). This research enable us after checking the results to produce new maps from Aerial photograph or Digital Image to use it for land uses and future extensions in whole study area.

Radar antennas in the open air are affected by dynamic wind loading during azimuthal scanning. In this state, the acquisition and prediction of the aerodynamic load is the focus of the structure design for the antenna. We performed wind tunnel force tests on flat plate antennas with different aspect ratios. The changes in mean, maximum, and root mean square values of the aerodynamic coefficients relative to the antenna aspect ratio and reduced frequency are given for the first time. The effects of the antenna aspect ratio and reduced frequency on aerodynamic characteristics are analyzed. The experimental results indicate that the magnitude of the azimuthal moment is related to the antenna aspect ratio and the reduced frequency. It is worth noting that the antenna aspect ratio has a significant impact on the mean, maximum, and root mean square values of dynamic aerodynamic coefficients. The mean, maximum, and root mean square values of the rolling moment coefficient, azimuth moment coefficient, and pitching moment coefficient monotonously decrease with the increase in aspect ratio. In addition, the influence of reduced frequency on the dynamic aerodynamic coefficient is closely related to the antenna aspect ratio. Under the same antenna aspect ratio, increasing the reduced frequency will increase the mean, maximum, and root mean square values of the aerodynamic coefficient. For flat plate antennas with different aspect ratios, at the condition of increasing the aspect ratios, the mean, maximum, and root mean square values of aerodynamic coefficients do not change significantly with the change of reduced frequency.

Abstract. Hydrological models are usually systems of nonlinear differential equations for which no analytical solutions exist and thus rely on numerical solutions. While some studies have investigated the relationship between numerical method choice and model error, the extent to which extreme precipitation such as that observed during hurricanes Harvey and Katrina impacts numerical error of hydrological models is still unknown. This knowledge is relevant in light of climate change, where many regions will likely experience more intense precipitation. In this experiment, a large number of hydrographs are generated with the modular modeling framework FUSE (Framework for Understanding Structural Errors), using eight numerical techniques across a variety of forcing data sets. All constructed models are conceptual and lumped. Multiple model structures, parameter sets, and initial conditions are incorporated for generality. The computational cost and numerical error associated with each hydrograph were recorded. Numerical error is assessed via root mean square error and normalized root mean square error. It was found that the root mean square error usually increases with precipitation intensity and decreases with event duration. Some numerical methods constrain errors much more effectively than others, sometimes by many orders of magnitude. Of the tested numerical methods, a second-order adaptive explicit method is found to be the most efficient because it has both a small numerical error and a low computational cost. A small literature review indicates that many popular modeling codes use numerical techniques that were suggested by this experiment to be suboptimal. We conclude that relatively large numerical errors may be common in current models, highlighting the need for robust numerical techniques, in particular in the face of increasing precipitation extremes.

The aim of a personalized heating system is to provide a desirable microclimate for each individual when heating is needed. In this paper, we present a method based on machine learning algorithms for generation of predictive models for use in control of personalized heating systems. Data was collected from two individual test subjects in an experiment that consisted of 14 sessions per test subject with each session lasting 4 h. A dynamic recurrent nonlinear autoregressive neural network with exogenous inputs (NARX) was used for developing the models for the prediction of personalized heating settings. The models for subjects A and B were tested with the data that was not used in creating the neural network (unseen data) to evaluate the accuracy of the prediction. Trained NARX showed good performance when tested with the unseen data, with no sign of overfitting. For model A, the optimal network was with 12 hidden neurons with root mean square error equal to 0.043 and Pearson correlation coefficient equal to 0.994. The best result for model B was obtained with a neural network with 16 hidden neurons with root mean square error equal to 0.049 and Pearson correlation coefficient equal to 0.966. In addition to the neural network models, several other machine learning algorithms were tested. Furthermore, the models were on-line tested and the results showed that the test subjects were satisfied with the heating settings that were automatically controlled using the models. Tests with automatic control showed that both test subjects felt comfortable throughout the tests and test subjects expressed their satisfaction with the automatic control.

Soil Roughness Measuring System Combined With Image Processing

Uniqueness in economies and stock markets has given rise to an interesting domain of exploring data mining techniques across global indices. Previously, very few studies have attempted to compare the performance of data mining techniques in diverse markets. The current study adds to the understanding regarding the variations in performance of data mining techniques across the global stock indices. We compared the performance of Neural Networks and Support Vector Machines using accuracy measures Mean Absolute Error (MAE) and R­­­­oot Mean Square Error (RMSE) across seven major stock markets. For prediction purpose, technical analysis has been employed on selected indicators based on daily values of indices spanning a period of 12 years. We created 196 data sets spanning different time periods for model building such as 1 year, 2 years, 3 years, 4 years, 6 years and 12 years for selected seven stock indices. Based on prediction models built using Neural Networks and Support Vector Machines, the findings of the study indicate there is a significant difference, both for MAE and RMSE, across the selected global indices. Also, Mean Absolute Error and Root Mean Square Error of models built using NN were greater than Mean Absolute Error and Root Mean Square Error of models built using SVM.

Accurate electrical load forecasting is of great significance to help power companies in better scheduling and efficient management. Since high levels of uncertainties exist in the load time series, it is a challenging task to make accurate short-term load forecast (STLF). In recent years, deep learning approaches provide better performance to predict electrical load in real world cases. The convolutional neural network (CNN) can extract the local trend and capture the same pattern, and the long short-term memory (LSTM) is proposed to learn the relationship in time steps. In this paper, a new deep neural network framework that integrates the hidden feature of the CNN model and the LSTM model is proposed to improve the forecasting accuracy. The proposed model was tested in a real-world case, and detailed experiments were conducted to validate its practicality and stability. The forecasting performance of the proposed model was compared with the LSTM model and the CNN model. The Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) were used as the evaluation indexes. The experimental results demonstrate that the proposed model can achieve better and stable performance in STLF.

Nowadays , nonlinear time series and artificial neural networks (ANN) models are used for forecasting in the field of business, agriculture and soon. Recent studies have shown, ANN have been successfully used for forecasting of financial and agriculture data series The classical methods used for time series prediction like Box-Jenkins or ARIMA assumes that there is a linear relationship between inputs and outputs. ANN have more advantages that can approximate to model both linear and nonlinear structures in time series, they are not able to handling both structures equally well. The autoregressive integrated moving average (ARIMA) model and two ANN models namely, Radial basis function neural networks (RBFNN), and Elman recurrent neural networks (ERNN) methods were applied to Hyderabad airport traffic data. The data obtained for 15 years from 2002–2003 to 2016–2017 about domestic and international passenger of International Airport of Hyderabad, India. In this research paper, we compared the performances of ARIMA, RBFNN and ERNN were based on three measures: mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE). The results showed that RBFNN obtained the smallest MAE, MAPE and RMSE in both the modeling and forecasting processes. The performances of the three models ranked in ascending order were: ARIMA, ERNN and the RBFNN model. Keywords: T ime series, forecasting, artificial neural networks, ARIMA models, radial basis function neural networks, and Elman recurrent neural networks Cite this Article R. Ramakrishna, Berhe Aregay, Tewodros Gebregergs. The Comparison in Time Series Forecasting of Air Traffic Data by Autoregressive Integrated Moving Average Model, Radial Basis Function and Elman Recurrent Neural Networks. Research & Reviews: Journal of Statistics . 2018; 7(3): 75–90p.

Volatility forecasting of financial market returns is vital for risk management and portfolio optimization. In the quest for accurate volatility modeling, this article introduces an Artificial Neural Network -Generalized Autoregressive Conditional Heteroscedasticity model (ANN-GARCH) that fuses the strengths of neural networks with the well-known GARCH framework to deliver superior volatility predictions. The approach integrates volatility captured by conventional GARCH models into a neural network framework, combining return and realized volatility series. Then, the ANN architecture is designed to learn and process patterns, ultimately generating enhanced volatility predictions. The proposed model consistently outperforms traditional volatility forecasting methods when applied to Dow Jones Index returns during the COVID-19 period and its aftermath. Across various error distributions, the ANN-GARCH model demonstrates superior forecasting accuracy, evidenced by significantly lower root mean square error (RMSE) and mean absolute error (MAE) compared to competing models. This underscores the model's reliability and effectiveness in volatility forecasting. This innovative model will help academics and finance practitioners make informed decisions when the market conditions are stressed.

Fossil fuels have traditionally played a crucial role in global energy production, but they cause significant environmental challenges. In response, the scientific community has shifted its focus toward renewable alternative fuels. In this context, biodiesel and alcohols have emerged as promising options for diesel engines. This study is centered on predicting the cylinder pressure of a single-cylinder four-stroke diesel engine fuelled with a diesel fuel-biodiesel (methyl ester)-isopropanol ternary blend using three machine learning algorithms: Gaussian Process Regression (GPR), Artificial Neural Networks (ANN), and Ensembles of Trees (ET). The cylinder pressure data is collected under the full throttle condition and different engine speeds. GPR, ANN, and ET algorithms are trained and compared using root mean square error and regression analysis. GPR exhibits outstanding prediction performance during the validation, with a lower root mean square error of 0.12686, and r2 of 1.00. ANN also exhibits strong prediction performance, with a validation of root mean square error of 0.47081, and a r2 of 1.00. ET, while showing a slightly higher validation root mean square error of 1.73370, maintains strong predictive capability with an r2 of 0.99. However, a comparison between the measured cylinder pressure data and the predicted values reveals a qualitatively and quantitatively closer agreement, particularly for ANN. These findings can suggest the practicality and reliability of these algorithms for predicting cylinder pressure in internal combustion engine studies. In conclusion, this study can contribute to the expanding body of research on alternative fuels and machine learning applications in internal combustion engines.