Minimum Square Error Research Articles

Enhanced oceanic fog nowcasting through satellite-based recurrent neural networks

ABSTRACT The presence of fog in offshore regions poses significant hazards to navigation and aviation, making fog nowcasting indispensable for various industries, including oil and gas. This study presented a novel approach utilizing Recurrent Neural Networks (RNN) within a deep learning framework to address this need. Leveraging geostationary GOES-16 satellite data from the summers of 2018 and 2019, fog maps were generated as input. The model incorporated Convolutional Long Short-Term Memory (ConvLSTM) layers and was trained with a unique loss function combining Minimum Squared Error (MSE) and structural DISSIMilarity (DSSIM) metrics. Validation results demonstrated an approximate 60% accuracy for both two-hour and three-hour nowcasting. Furthermore, evaluation against in-situ data from an offshore platform revealed a Probability of Detection (PoD) of 0.75 and False Alarm Rate (FAR) of 0.14 for two-hour nowcasting, PoD of 0.75 and FAR of 0.20 for three-hour nowcasting, and PoD of 0.70 and FAR of 0.20 for six-hour nowcasting. These findings suggested the operational viability of the proposed method for short-term fog forecasting in offshore environments.

Minimum Squared Error Method based on Modal Analysis for Acoustic Emission Source Location using Microfiber Coupler Sensor

This study focuses on an improved modal analysis method to the source location using a microfiber coupler sensor (MFCS). This method uses the Minimum Squared Error(MSE) to reduce the dependence of locating results on frequency selection and automate the locating process of contour threshold modal analysis. First, the secondary enclosed package method of MFCS was studied to ensure that the signal detected by this sensor meets the requirements of the contour threshold modal analysis and can extend the MFCS working life as well. The MSE method was then presented detailed. Second, the directionality experiment of the MFCS was performed to determine the best layout method for sensors in planar localization. The planar source localization method with both two sensors and three sensors are then introduced. Last, both linear and planar location experiments were performed on an aluminium thin plate. The results indicate that the sensor packaged by cuboid glass tube is more suitable for practical ultrasonic testing; the location process is automated by valid contour extraction and initial frequency selection, and the linear error using this method is effectively reduced as well; the best arrangement of the sensor is to place the sensor at the corner of the planar, and the angle between the sensor and the horizontal line is 45°; compared with two sensors, the planar localization with three sensors provides a more accurate and high repeatability results. This study proved that the combination of the MFCS and the MSE method is quite feasible and can be applied to the planar localization in plate-like structures effectively.

Parameter identification algorithm for dynamic modeling of industrial-scale vanadium redox flow batteries

In this work, we propose a two-stage parameter identification algorithm for the zero-dimensional model of vanadium redox flow battery (VRFB) based on the fitting of the available experimental data. Specifically, in the first stage, the parameters of overpotentials are obtained from the discharge polarization curve at the state of charge of 50%. In the second stage, the parameters of capacity loss and crossover are identified by the minimization of square error between the predicted and experimental charge-discharge curves at a certain value of load current. The algorithm allows to assess the degree of degradation of the battery components, and capture the effects associated with crossover and mass transfer limitations at high load currents and low flow rates. We verify this algorithm with a reference simulation considering the most interesting practical cases of battery use and validate it with three experimental VRFB systems with power and capacity ratings in the range of 5–10 kW and 5–100 kWh, correspondingly. The results show a good agreement with experimental data having an average error value of less than 1%. The results of the study are an important step in further developments of advanced control algorithms for large-scale energy storage systems.

Novel Planning Methodology for Spatially Optimized RES Development Which Minimizes Flexibility Requirements for Their Integration into the Power System

An optimization model which determines optimal spatial allocation of wind (WPPs) and PV power plants (PVPPs) for an energy independent power system is developed in this paper. Complementarity of the natural generation profiles of WPPs and PVPPs, as well as differences between generation profiles of WPPs and PVPPs located in different regions, gives us opportunity to optimize the generation capacity structure and spatial allocation of renewable energy sources (RES) in order to satisfy the energy needs while alleviating the total flexibility requirements in the power system. The optimization model is based on least squared error minimization under constraints where the error represents the difference between total wind and solar generation and the referent consumption profile. This model leverages between total energy and total power requirements that flexibility resources in the considered power system need to provide in the sense that the total balancing energy minimization implicitly bounds the power imbalances over the considered time period. Bounding the power imbalances is important for minimizing investment costs for additional flexibility resources. The optimization constraints bound the installed power plant capacity in each region according to the estimated technically available area and force the total energy production to equal the targeted energy needs. The proposed methodology is demonstrated through the example of long-term RES planning development for complete decarbonization of electric energy generation in Serbia. These results could be used as a foundation for the development of the national energy strategy by serving as a guidance for defining capacity targets for regional capacity auctions in order to direct the investments in wind and solar power plants and achieve transition to dominantly renewable electricity production.

Optimized modeling of energy and environmental metrics of mixed flow turbofan engine used regional aircraft.

As a kind of gas turbine engines, turbofan engines have powered a number of aero-vehicles in aviation sector. The necessity of turbofan with higher energy efficiency has been greatly drawn attention since these are operating dependent to fossil fuels. In this study, energy and emission metrics of fifty-one mixed flow turbofan engines (MFTE) with different bypass ratio, overall pressure ratio and fuel flow are modeled with multi-regression (MR) method. The obtained models are subjected to metaheuristic approaches involving genetic algorithm (GA) and simulated annealing (SA) so as to decrease error of the models. According to MR findings, rated thrust of MFTEs is estimated with 1.4877 of minimum square error (MSE) whereas GA and SA make it lower as 1.3404 and 1.2524, respectively. On the other hand, NOx emission index of MFTEs is predicted with relatively low coefficient of determination (R2) as 0.8620. However, its accuracy is enhanced to 0.8633 (with GA) and 0.8655 (with SA). Finally, exergy efficiency of MFTEs is estimated the highest model correctness with GA. Namely, R2 of the model is computed as 0.9280 with GA and 0.9277 with SA. Without applying these methods, its R2 is obtained as 0.9263 with MR. When considering these outcomes, thanks to modeling and optimization methods, prediction of performance and emission indexes of mixed flow turbofan engines could be performed with lower error values. It is thought that the study helps in prediction of environmental effect regarding turbofan engines that are utilized at busy airports.

ScSTAR reveals hidden heterogeneity with a real-virtual cell pair structure across conditions in single-cell RNA sequencing data.

Cell-state transition can reveal additional information from single-cell ribonucleic acid (RNA)-sequencing data in time-resolved biological phenomena. However, most of the current methods are based on the time derivative of the gene expression state, which restricts them to the short-term evolution of cell states. Here, we present single-cell State Transition Across-samples of RNA-seq data (scSTAR), which overcomes this limitation by constructing a paired-cell projection between biological conditions with an arbitrary time span by maximizing the covariance between two feature spaces using partial least square and minimum squared error methods. In mouse ageing data, the response to stress in CD4+ memory T cell subtypes was found to be associated with ageing. A novel Treg subtype characterized by mTORC activation was identified to be associated with antitumour immune suppression, which was confirmed by immunofluorescence microscopy and survival analysis in 11 cancers from The Cancer Genome Atlas Program. On melanoma data, scSTAR improved immunotherapy-response prediction accuracy from 0.8 to 0.96.

Prediction of viscosity of MWCNT-Al[formula omitted]O[formula omitted] (20:80)/SAE40 nano-lubricant using multi-layer artificial neural network (MLP-ANN) modeling

This study, a comprehensive study on dynamic viscosity (μnf) of MWCNT-Al2O3(20:80)/SAE40 nano-lubricant is done with an artificial neural network (ANN). In this research, Multilayer Perceptron-ANN (MLP-ANN) modeling with Levenberg–Marquardt algorithm (MLA) has been used for model of ANN. The optimal method is consists of four hundred different ANN structures on MWCNT-Al2O3/SAE40 nano-lubricant. It is contain of two hidden layers with ten neurons optimal structure in first layer and ten neurons in second layer. The volume fraction of nanoparticles (φ), rate of shear (γ̇) and temperature have been considered as input parameters and μnf has been considered an output parameter in the ANN model. The optimal ANN with eight neurons per layer has the minimum square error (MSE) and the maximum regression coefficient R near to one to predict the μnf. The margin of deviation range (MOD) is −2% < MOD < 2% in the predicted values. A comparison of three groups of lab data, computational and prediction of ANN shows the predicting data predicted by the ANN is much more powerful than a new relation and approximates lab data accurately.

Prediction of Solid Conversion Process in Direct Reduction Iron Oxide Using Machine Learning

The direct reduction process has been developed and investigated in recent years due to less pollution than other methods. In this work, the first direct reduction iron oxide (DRI) modeling has been developed using artificial neural networks (ANN) algorithms such as the multilayer perceptron (MLP) and radial basis function (RBF) models. A DRI operation takes place inside the shaft furnace. A shaft furnace reactor is a gas-solid reactor that transforms iron oxide particles into sponge iron. Because of its low environmental pollution, the MIDREX process, one of the DRI procedures, has received much attention in recent years. The main purpose of the shaft furnace is to achieve the desired percentage of solid conversion output from the furnace. The network parameters were optimized, and an algorithm was developed to achieve an optimum NN model. The results showed that the MLP network has a minimum squared error (MSE) of 8.95 × 10−6, which is the lowest error compared to the RBF network model. The purpose of the study was to identify the shaft furnace solid conversion using machine learning methods without solving nonlinear equations. Another advantage of this research is that the running speed is 3.5 times the speed of mathematical modeling.

AUC optimization for deep learning-based voice activity detection

Voice activity detection (VAD) based on deep neural networks (DNN) have demonstrated good performance in adverse acoustic environments. Current DNN-based VAD optimizes a surrogate function, e.g., minimum cross-entropy or minimum squared error, at a given decision threshold. However, VAD usually works on-the-fly with a dynamic decision threshold, and the receiver operating characteristic (ROC) curve is a global evaluation metric for VAD at all possible decision thresholds. In this paper, we propose to maximize the area under the ROC curve (MaxAUC) by DNN, which can maximize the performance of VAD in terms of the entire ROC curve. However, the objective of the AUC maximization is nondifferentiable. To overcome this difficulty, we relax the nondifferentiable loss function to two differentiable approximation functions—sigmoid loss and hinge loss. To study the effectiveness of the proposed MaxAUC-DNN VAD, we take either a standard feedforward neural network or a bidirectional long short-term memory network as the DNN model with either the state-of-the-art multi-resolution cochleagram or short-term Fourier transform as the acoustic feature. We conducted noise-independent training to all comparison methods. Experimental results show that taking AUC as the optimization objective results in higher performance than the common objectives of the minimum squared error and minimum cross-entropy. The experimental conclusion is consistent across different DNN structures, acoustic features, noise scenarios, training sets, and languages.

Lead ferrite-activated carbon magnetic composite for efficient removal of phenol from aqueous solutions: synthesis, characterization, and adsorption studies

A novel lead ferrite-magnetic activated carbon (lead ferrite-MAC) composite was developed using the chemical co-precipitation method. Instrumental analyses such as X-ray diffraction (XRD), scanning electron microscopy (SEM), Fourier-transform infrared spectroscopy (FTIR), and Brunauer–Emmett–Teller (BET) analysis were performed to characterize adsorbent. The uptake of phenol from aqueous solutions using the developed adsorbent was compared to that of pristine activated carbon. The maximum adsorption capacity of lead ferrite-MAC composite (145.708 mg/g) was more than that of pristine activated carbon (116.606 mg/g) due to the metal hydroxides coated on activated carbon since they improve the retention of phenol on the available active sites of adsorbent and create an additional electrostatic interaction with the phenol adsorbate. Regarding the high value of the coefficient of determination (R2) and adjusted determination coefficient (R2adj), coupled with the lower values of average relative error (ARE) and minimum squared error (MSE), it can be found that the isothermal data for the lead ferrite-MAC adsorbent were in agreement with the isotherm models of Redlich-Peterson and Langmuir. From the kinetic viewpoint, pseudo-second-order and linear driving force models explained the phenol adsorption data for both adsorbents. The reusability tests for lead ferrite-MAC composite revealed that after six cycles, 85% of the initial adsorption capacity was maintained. The developed adsorbent can be successfully applied to uptake phenol from aqueous solutions.

Meta-Transfer Learning Using Wavelet Decomposition for Multi-Horizon Time Series Forecasting

Multi-horizon time series forecasting is a very challenging task in many fields of research. In the field of machine learning, artificial neural networks have been used to carry out these tasks. However, there are still problems that are of general interest to researchers such as: Loss of data in data acquisition and long-term forecast. In this paper, we propose a hybrid Meta-Transfer Learning technique based on transfer-learning, meta-learning and signal detection by means of the discrete wavelet transform to solve the aforementioned problems in multi-horizon time series forecasting. Input-to-state stability analysis and the strong and weak convergence analysis for the proposed method are included. To validate the effectiveness of the method, the long-term prediction of earthquakes magnitude (M>4.5) in Italy is taken as a case of study, using information from Italy and Mexico. Simulations of classic methods for forecasting time series based on neural models are performed. The forecasting performance obtained is the minimum square error (MSE) is 0.091, while for the meta-transfer learning, the MSE is 0.032.

A Novel Technique of Noise Cancellation based on Stationary Bionic Wavelet Transform and WATV: Application for ECG Denoising

In this paper, is proposed a novel technique of Electrocardiogram (ECG) denoising. It is based on the application of Wavelet/Total-Variation (WATV) denoising approach in the domain of the Stationary Bionic Wavelet Transform (SBWT). It consists firstly in applying the SBWT to the noisy ECG signal for obtaining two noisy coefficients namedwtb1 and wtb2 which are respectively details and approximation coefficients. For estimating the level of noise altering the signal, namedσ, we use wtb1. This noise is an additive Gaussian white noise. The thresholding ofwtb1 is secondly performed employing the soft thresholding and a denoised coefficient wtd1 is obtained. This thresholding requires the use of a certain threshold, thr which is computed using σ. Thedenoising ofwtb2 is performed using WATV denoising method and we obtain a denoised coefficient, wtd2. This WATV denoising method also uses σ. The denoised ECG signal is finally obtained by applying the inverse of SBWT (SBWT-1) to wtd1 and wtd2. The proposed technique performance is justified by the results obtained from the computations of Signal to Noise Ratio (SNR), Minimum Square Error (MSE), Mean Absolute Error (MAE), Peak-SNR (PSNR) and Cross-Correlation (CC).

材料パラメータ同定問題におけるガウスニュートン法の適用

In this study, the application of the Gauss-Newton method to several material parameter identification problems was investigated. The parameters of materials with complicated constitutive laws are determined to reproduce known physical quantities such as stiffness and natural frequency. The identification problem is usually formulated as a squared error minimization problem for the physical quantity. Gauss-Newton method is one of the important solution methods. In this method, the Hessian matrix is approximated using the first order differentiation of the physical quantity with respect to the parameters. Although this method obtains good convergence in the initial stage, it is observed that it becomes unstable in the final stage. We introduced a line search technique based on the Marquardt method to avoid the instability. We compared this method with the general gradient method and confirmed good convergence of this method.

Towards Grading Chest X-rays of COVID-19 Patients Using a Dynamic Radial Basis Function Network Classifier

The high volume of COVID-19 Chest X-rays and less number of radiologists to interpret those is a challenge for the highly populous developing nations. Moreover, correct grading of the COVID-19 stage by interpreting the Chest X-rays manually is time-taking and could be biased. It often delays the treatment. Given the scenario, the purpose of this study is to develop a deep learning classifier for multiple classifications (e.g., mild, moderate, and severe grade of involvement) of COVID-19 Chest X-rays for faster and accurate diagnosis. To accomplish the goal, the raw images are denoised with a Gaussian filter during pre-processing followed by the Regions of Interest, and Edge Features are identified using Canny's edge detector algorithm. Standardized Edge Features become the training inputs to a Dynamic Radial Basis Function Network classifier, developed from scratch. Results show that the developed classifier is 88% precise and 86% accurate in classifying the grade of illness with a much faster processing speed. The contribution lies in the dynamic allocation of the (i) number of Input and Hidden nodes as per the shape and size of the image, (ii) Learning rate, (iii) Centroid, (iv) Spread, and (v) Weight values during squared error minimization; (vi) image size reduction (37% on average) by standardization, instead of dimensionality reduction to prevent data loss; and (vii) reducing the time complexity of the classifier by 26% on average. Such a classifier could be a reliable assistive tool to human doctors in screening and grading COVID-19 patients and in turn, would help faster management of the patients as per the stages of COVID-19.

Model identification and parametric adaptive control of hydraulic manipulator with neighborhood field optimization

The model identification and uncertain parameters estimation are common control problems in multi-DOF manipulator system, since both model parametric uncertainties and unknown external disturbance often degrade the output performance and the close-loop system stability. In this study, by using the joint angle and the torque information, a neighborhood field optimization is adopted to identify the Lagrange model parameters of two-DOF hydraulic manipulator. The fitness function of the neighborhood field optimization is designed to optimize the minimal squared error of the torque estimation and to obtain the lumped estimated parameters with high accuracy. Then based on the identification model, a parametric adaptive controller is designed to address uncertain electro-hydraulic parameters and external disturbances. Furthermore, a new stable control variable is redesigned to avoid the redundant input saturation of electro-hydraulic actuator. The effectiveness of the proposed controller is verified by the comparative experimental results with the general backstepping controller.

Vibration direction sensitivity of the cochlea with bone conduction stimulation in guinea pigs

Sound and vibrations that cause the skull bone to vibrate can be heard as ordinary sounds and this is termed hearing by bone conduction (BC). Not all mechanisms that causes a skull vibration to result in BC hearing are known, and one such unknown is how the direction of the vibration influences BC hearing. This direction sensitivity was investigated by providing BC stimulation in five different directions at the vertex of the guinea pig skull. The hearing thresholds for BC stimulation was obtained in the frequency range of 2 to 20 kHz by measurements of compound action potential. During the stimulation by BC, the vibration of the cochlear promontory was measured with a three-dimensional laser Doppler vibrometer resulting in a set of unique three-dimensional velocity magnitude combinations for each threshold estimation. The sets of three-dimensional velocity magnitude at threshold were used to investigate nine different predictors of BC hearing based on cochlear promontory velocity magnitudes, six single direction (x, y and z directions in isolation, the normal to the stapes footplate, the oval to round window direction, and the cochlear base to apex direction), one linear combination of the three dimension velocity magnitudes, one square-rooted sum of the squared velocity magnitudes, and one sum of the weighted three dimensional velocity magnitudes based on a restricted minimum square error (MSE) estimation. The MSE gave the best predictions of the hearing threshold based on the cochlear promontory velocity magnitudes while using only a single direction gave the worst predictions of the hearing thresholds overall. According to the MSE estimation, at frequencies up to 8 kHz the vibration direction between the right and left side gave the greatest contribution to BC hearing in the guinea pig while at the highest frequencies measured, 16 and 20 kHz, the anteroposterior direction of the guinea pig head gave the greatest contribution.

Transmitter and Receiver Window Designs for Orthogonal Time-Frequency Space Modulation

In this paper, we investigate the impacts of transmitter and receiver windows on the performance of orthogonal time-frequency space (OTFS) modulation and propose window designs to improve the OTFS channel estimation and data detection performance. In particular, assuming ideal pulse shaping filters at the transceiver, we derive the impacts of windowing on the effective channel and its estimation performance in the delay-Doppler (DD) domain, the total average transmit power, and the effective noise covariance matrix. When the channel state information (CSI) is available at the transceiver, we analyze the minimum squared error (MSE) of data detection and propose an optimal transmitter window to minimize the detection MSE. The proposed optimal transmitter window can be interpreted as a mercury/water-filling power allocation scheme, where the mercury is firstly filled before pouring water to pre-equalize the time-frequency (TF) domain channels. When the CSI is not available at the transmitter but can be estimated at the receiver, we propose to apply a Dolph-Chebyshev (DC) window at either the transmitter or the receiver, which can effectively enhance the sparsity of the effective channel in the DD domain. Thanks to the enhanced DD domain channel sparsity, the channel spread due to the fractional Doppler is significantly reduced, which leads to a lower error floor in both channel estimation and data detection compared with that of rectangular window. Simulation results verify the accuracy of the obtained analytical results and confirm the superiority of the proposed window designs in improving the channel estimation and data detection performance over the conventional rectangular window design.

Spatial Wideband Channel Estimation for mmWave Massive MIMO Systems With Hybrid Architectures and Low-Resolution ADCs

In this article, a channel estimator for wideband millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) systems with hybrid architectures and low-resolution analog-to-digital converters (ADCs) is proposed. To account for the propagation delay across the antenna array, which cannot be neglected in wideband mmWave massive MIMO systems, the discrete time channel that models the spatial wideband effect is developed. Also, the training signal design that addresses inter-frame, inter-user, and inter-symbol interferences is investigated when the spatial wideband effect is not negligible. To estimate the channel parameters over the continuum based on the maximum a posteriori (MAP) criterion, the Newtonized fully corrective forward greedy selection-cross validation-based (NFCFGS-CV-based) channel estimator is proposed. NFCFGS-CV is a gridless compressed sensing (CS) algorithm, whose termination condition is determined by the CV technique. The CV-based termination condition is proved to achieve the minimum squared error (SE). The simulation results show that NFCFGS-CV outperforms state-of-the-art on-grid CS-based channel estimators.

Optimization of Normalized Least Mean Square Algorithm of Smart Antenna Beamforming for Interference Mitigation

A smart antenna is an array of antennas with signal processing capabilities to transmit/receive information in an adaptive manner. This capability continues to be investigated to find the best adaptive algorithm for the desired beamforming capability. This paper aims to provide a study and analysis of the effect of the normalized least mean square (NLMS) algorithm on step size parameter settings which have an impact on the desired minimum square error (MSE) value and on the nulling beam radiation pattern arrangement of smart antennas in its role in mitigating interference. The simulation for beamformer performance for 1000 iterations was carried out using the Matlab tool on the additional white noise gaussian (AWGN) channel and the simulation parameters were changed to some step size values (μ) with the NLMS algorithm for 16 antenna elements. The effect of the value of the step size μ is seen in the iterations number that takes place before the minimum error noise is obtained, where the increase in the value of this step size reduces the iterations number required; at the same time further improving MSE levels. From the pattern of amplitude response after the beamforming process, the result is an escalation in the number of nulling to increase the distance between the spaced elements taken. The interference source is eliminated/closed by placing the nulls in the path of the interference source where the worst interference level is obtained around -90 dB when the smallest step size is taken.

Artificial Neural Network Intelligent System on the Early Warning System of Landslide

Landslide is a natural sloping ground movement disaster that can occur due to several factors such as high rainfall, soil moisture in the depth of the soil of an area, vibrations experienced in the region, and the slope of the ground structure. A system that can deliver these factor values into the levels of vulnerability of landslide disasters is needed. The system uses Arduino Mega 2560 to process the level of vulnerability. It can predict the moment and the probability of the disaster occurring as an early warning system. The artificial neural network (ANN) intelligent system can expect an event of a disaster. The designed ANN used five parameters causing landslide as input data: rainfall, slope, soil moisture on the surface, soil moisture in the ground’s depth, and soil vibration. The ANN system output delivered three-level conditions: the safe, the standby, and the hazardous. The feed-forward backpropagation (FFBP) and the cascade forward backpropagation (CFBP) methods were analyzed. The performance of both methods was compared in terms of minimum square error (MSE). The MSE results of FFBP and CFBP in the safe, the standby, and the hazardous conditions were 0.017076 and 0.034952; 0.049597 and 0.046764; 0.062105 and 0.060355; respectively. The results point to the supremacy of CFBP to FFBP in standby and hazardous conditions. Therefore, the CFBP is implemented into the hardware of the early warning system.