Inputs Of Artificial Neural Network Research Articles

Predicting the dry bulk density of soils across Denmark: Comparison of single-parameter, multi-parameter, and vis–NIR based models

Dry bulk density (BD) is a key soil physical property, characterizing the level of soil compaction and controlling flow and transport of fluids and solutes. In this study, predictive BD models were developed from 2462 BD measurements of mineral soils representing both top- and subsoil horizons across Denmark. Three types of BD models were compared: (i) single-parameter based on soil organic carbon (SOC), (ii) multi-parameters based on SOC, soil texture and depth, and using multiple linear regression (MLR), and (iii) visible–near infrared spectroscopy (vis–NIRS) based, using the spectral information and partial least squares regression (PLSR). Also, three machine learning techniques (random forest (RF), regression rules (RR), and artificial neural networks (ANN)) were applied for multi-parameter and vis–NIRS based models to potentially improve predictions. BD models were calibrated on 70% of dataset and validated on the remaining 30%. Single-parameter models with SOC had the lowest predictive potential and were not able to predict BD above 1.62 g cm−3. Machine learning using RF, RR and ANN on soil properties resulted in significantly higher prediction accuracies (RMSE ≤ 0.11 g cm−3 and R2 ≥ 0.60) than MLR (RMSE = 0.12 g cm−3 and R2 = 0.41) and SOC-based models. The relative importance of variables was obtained from the multi-parameter models with SOC and clay as the most important variables. The prediction accuracies MLR and vis–NIRS did not differ but were better than SOC-based models, suggesting that vis–NIRS could be a cost- and time-efficient alternative to pedotransfer functions that generally require determination of one or more soil properties for developing the calibration models. Decomposing the spectral data into principal components and using selected components as inputs for RF, RR and ANN improved the model performances only for RF models. Instead and in perspective, combining techniques by supplementing spectral data with readily available soil information and using machine learning techniques may further improve BD predictions.

Solar Irradiance Fluctuation Prediction Methodology Using Artificial Neural Networks

Abstract Renewable energy is an attractive alternative source of energy to fossil fuels, as it can help prevent global warming and air pollution. Solar energy, one of the most promising renewable energy sources, can be converted into electricity using photovoltaic power generation systems. Anywhere on the Earth, solar irradiance generally fluctuates during the day but depends on atmospheric conditions. Thus, when a photovoltaic power generation system is connected to a conventional electricity network, predicting near-future global solar irradiance, especially its drastic increases and decreases, is critical to stabilize the network. In this research, a simple method utilizing artificial neural networks to predict large increases and decreases in global solar irradiance is developed. The red–blue ratio (RBR) values, which are extracted from a set of sampling points in images of the sky, as well as the corresponding global solar irradiance values, are used as the artificial neural network inputs. The direction of the movement of clouds is predicted using RBR data at the sampling points. Then, solar irradiance is predicted using the RBR values along the axis closest to the predicted cloud movement direction and the corresponding solar irradiance measurements. The proposed methodology is able to predict both large increases and decreases in solar irradiance greater than 50 through 100 W/m2 1 min in advance with a 40% prediction error. A significant reduction in computational effort is achieved compared to existing sky image-based methodologies using limited sky image data.

Harvest chronological planning using a method based on satellite-derived vegetation indices and artificial neural networks

Aim of study: Wheat appropriate harvest date (WAHD) is an important factor in farm monitoring and harvest campaign schedule. Satellite remote sensing provides the possibility of continuous monitoring of large areas. In this study, we aimed to investigate the strength of vegetation indices (VIs) derived from Landsat-8 for generating the harvest schedule regional (HSR) map using Artificial Neural Network (ANN), a robust prediction tool in the agriculture sector.Area of study: Qorveh plain, Iran.Material and methods: During 2015 and 2016, a total of 100 plots was selected. WAHD was determined by sampling of plots and specifying wheat maximum yield for each plot. The strength of eight Landsat-8 derived spectral VIs (NDVI, SAVI, GreenNDVI, NDWI, EVI, EVI2, CVI and CIgreen) was investigated during wheat growth stages using correlation coefficients between these VIs and observed WAHD. The derived VIs from the required images were used as inputs of ANNs and WAHD was considered as output. Several ANN models were designed by combining various VIs data.Main results: The temporal stage in agreement with dough development stage had the highest correlation with WAHD. The optimum model for predicting WAHD was a Multi-Layer Perceptron model including one hidden layer with ten neurons in it when the inputs were NDVI, NDWI, and EVI2. To evaluate the difference between measured and predicted values of ANNs, MAE, RMSE, and R2 were calculated. For the 3-10-1 topology, the value of R2 was estimated 0.925. A HSR map was generated with RMSE of 0.86 days.Research highlights: Integrated satellite-derived VIs and ANNs is a novel and remarkable methodology to predict WAHD, optimize harvest campaign scheduling and farm management.

Non-probabilistic method to consider uncertainties in frequency response function for vibration-based damage detection using Artificial Neural Network

Artificial neural network (ANN) has become a popular computational approach in the field of vibration-based damage detection, based on its ability to relate the nonlinear relationship between structural vibration characteristics and damage information. In the meantime, frequency response function (FRF) estimation has been proven effective as a dynamic parameter for damage detection due to its prevention of information leakage. In this regard, FRF is chosen as the input variable for ANN to detect structural damage in this study. However, the main concern in damage detection using FRF with ANN is the size of FRF data. A full-size FRF data will result in a wide composition range of the ANN input layer, thus affecting the iteration divergence in the network training process and resulting in the computational inefficiency. In most applications, principal component analysis (PCA) has been used to reduce the size of the FRF data before being fed to an ANN model. However, as the structures become more complex, the FRF data size also increases. The large size of FRF data may result in the PCA not being effective in selecting important information from the actual FRF data, leading to false damage detection. Moreover, the existence of uncertainties from modelling error and measurement error may also amplify the error in damage detection. Hence, this study proposes a combination of a non-probabilistic method with PCA to consider the problem of the existing uncertainties and the inefficiency of using FRF data in ANN-based damage detection. In this study, ANN is used to relate the FRF data to a damage feature. The input data for the network are the compressed real FRFs and the outputs are the elemental stiffness parameter (ESP). The compressed FRF data obtained from PCA provide a new damage index (DI) that is used as the input layer of the ANN. Based on the interval analysis method, the uncertainties in the new DI are considered to bind together to obtain the interval bound (lower and upper bounds) of the DI changes. The possibility of damage existence (PoDE) is designed to ascertain the relationship between the input and output parameters in the form of undamaged and damaged conditions. The verification conducted on a numerical model and a laboratory tested steel truss bridge model demonstrated that the proposed method is efficient in dealing with uncertainties using FRF for damage detection.

Detection of Moldy Cores in Apples with Near-Infrared Transmission Spectroscopy Based on Wavelet and BP Network

Moldy cores in apples are not initially obvious from the outside of the fruit, so developing methods to detect moldy cores is an important area of research in the apple industry. The objective of this study was to improve the ability of near-infrared spectrometry to detect moldy cores in apples. Transmission spectra were recorded for 200 apple samples in the range of 200–1100[Formula: see text]nm, and 140 and 60 samples were randomly selected as training and test sets, respectively. Signal de-noising was performed by wavelet thresholding based on the results of orthogonal experiments. The best wavelengths for discriminating between healthy and diseased apples were selected by a successive projection algorithm (SPA). The extracted wavelengths were used as the input in a back propagation artificial neural network (BP-ANN). Through these experiments, this study compared the correct recognition rates using different ratios of training to test numbers in the model, and functions in the hidden and output layers of the BP-ANN. The proposed method achieved the highest accuracies of 95.00% and 95.71% for the test and training sets, respectively. This method could be used to develop a portable instrument for detecting moldy cores in apples.

Detection of focal epilepsy in brain maps through a novel pattern recognition technique

Focal epilepsy is a common neurological disorder of the brain. It is symbolized by recurring seizure activities in particular regions of the brain. Electroencephalogram (EEG) signals contain information about the brain activity that can be used to identify areas affected by seizures. For this, however, neurologists have the challenging task to analyze the EEG signal to imply where the source of epilepsy is located. A novel approach to detect regions of focal epilepsy is proposed in this paper. By using an adaptive mixture of independent component analysis as a decomposition algorithm and information about the position of the measuring electrodes, EEG signals are transformed into two-dimensional brain maps that are further altered by a symmetric-weighted scale-invariant local ternary pattern technique. Features from this altered brain map are used as input in a three-layered artificial neural network for classification between epileptic and artifact brain maps. The brain maps classified as artifact are eliminated, and the epileptic components are given as output for doctors to investigate the regions of the brain affected by focal epilepsy. With the proposed methodology, an accuracy of 99.53% to detect epileptic components in EEG signals is reached.

ESTIMATING CANOLA’S BIOPHYSICAL PARAMETERS FROM TEMPORAL, SPECTRAL, AND POLARIMETRIC IMAGERY USING MACHINE LEARNING APPROACHES

Abstract. The objective of this study was to investigate the application of multi-temporal optical and polarimetric synthetic aperture radar (PolSAR) Earth observations for crop characterization. Crop dry biomass, Leaf Area Index (LAI), and Plant Water Content (PWC) were estimated and assessed using Machin learning approaches. An accurate estimation of crop parameters provides essential information to increased food production and plays a crucial role in the management of agricultural lands. Multispectral and PolSAR data provide valuable observations of spectral and structural properties which are essential for crops parameter modelling. The Earth observations used in this paper were collected by RapidEye satellites and Unmanned Aerial Vehicle Synthetic Aperture Radar (UAVSAR) system in the summer of 2012, over an agriculture area in Winnipeg, Manitoba, Canada. The RapidEye vegetation indices (VIs) and UAVSAR polarimetric parameters were used as inputs in artificial neural network (ANN) and support vector regression (SVR) models for canola biophysical parameters estimation. The best models were provided by SVR for canola. Also combining optical VIs and polarimetric features appeared as a powerful tool for crop parameters estimation in agricultural lands.

Using atmospheric inputs for Artificial Neural Networks to improve wind turbine power prediction

A robust machine learning methodology is used to generate a site-specific power-curve of a full-scale isolated wind turbine operating in an atmospheric boundary layer to drastically improve the power predictions, and, thus, the forecasting of the monthly energy production estimates. The study has important implication in measuring the financial feasibility of wind farms by improving the accuracy of monthly energy estimates. The significance of the study is that atmospheric stability and air-density are accounted in the power predictions of the wind turbine. Artificial Neural Networks (ANN) machine learning approach is used to generate multi-parameter input models to estimate the power produced by the wind turbine. The ANN model in this study uses Feed Forward Back Propagation (FFBP) algorithm. The power- and wind-data is obtained from a 2.5 MW wind turbine that has a Meteorological tower located 900 m Southwest of the wind turbine in Kirkwood, Iowa, USA. The study investigates the role of atmospheric boundary-layer metrics – Wind Speed, Density (a measure of stratification), Richardson Number, turbulence intensity, and wind shear as input parameters into the ANN model. The study investigates the influence of FFBP ANN hyper-parameters on the power prediction accuracy. Comparison of the FFBP ANN model to other power curve correction techniques demonstrated an improvement in the Mean Absolute Error (MAE) of 40% when compared to the density correction (the next closest). The five-parameter 4-layer FFBP ANN has an average energy production error of 0.4% for the nine months while the IEC this error is −3.7% and for the air density correction the error is −1.9%, respectively. Finally, the study determines the performance of the FFBP ANN model for different atmospheric stability regimes (Unstable, Stable, Strongly Stable, Strongly Unstable and Neutral) classified using two criterions - Richardson number and Turbulence intensity. The largest MAE occurs during the strongly stable regime of the atmospheric boundary layer for both criteria.

A Novel Relaying Approach for Performance Enhancement in a STATCOM Integrated Wind-Fed Transmission Line Using Single-Terminal Measurement

This paper presents the novel relaying algorithm for fault detection in static synchronous compensator integrated transmission line including wind firm. The scheme is used for single-terminal data measurement. The faulty current signals are retrieved from single end terminal of the transmission line through the current transformer placed at sending end and process through DWT to obtain the spectral energy of the current signals. The spectral energy of the current signal obtained from the sending end using DWT is fed to the input of artificial neural network (ANN)-based fault detector for evaluating fault detection. The combined approach of DWT-ANN is used to detect and classify the fault. The proposed scheme uses the faulty current signal from single-terminal data. So, the requirement of communication link is not needed and it involves less cost. Another merit of the proposed scheme is that the fault detection time is reduced when DWT-ANN approach is used as compared to other existing schemes. In addition to this, the scheme works fine to detect the fault with the parameter variation of the line such as source impedance, fault inception angle, wind speed, fault location, fault type, fault resistance and high-resistance fault. It senses the fault in an accurate manner and classifies the fault at any parameter variation of the line. The fault detection time (delay time) remains within a half-cycle period of time (10 ms). The overall performance of the scheme is found to be nearly 99.98%. Therefore, the scheme is very effective, accurate and simple to understand the fault detection and classification of the line.

Implementing Any Nonlinear Quantum Neuron.

The ability of artificial neural networks (ANNs) to adapt to input data and perform generalizations is intimately connected to the use of nonlinear activation and propagation functions. Quantum versions of ANN have been proposed to take advantage of the possible supremacy of quantum over classical computing. To date, all proposals faced the difficulty of implementing nonlinear activation functions since quantum operators are linear. This brief presents an architecture to simulate the computation of an arbitrary nonlinear function as a quantum circuit. This computation is performed on the phase of an adequately designed quantum state, and quantum phase estimation recovers the result, given a fixed precision, in a circuit with linear complexity in function of ANN input size.

A novel system for effective speech recognition based on artificial neural network and opposition artificial bee colony algorithm

The problem related to speech recognition system becomes challenging if vocabularies are having too many similar-sounding words. To overcome these types of challenges, an effective speech recognition system using artificial neural network (ANN) with optimization technique is proposed. In this system, distinct words spoken by different people are considered as input speech signal. The features of these input speech signals are extracted using amplitude modulation spectrogram. The extracted features are then the input to the ANN for training. The trained ANN inputs are used for predicting the isolated words during testing. In this work, the default structure of ANN is redesigned using Levenberg–Marquardt algorithm, to retrieve optimal prediction rate with accuracy. The hidden layers and neurons of the hidden layers are further optimized using the opposition artificial bee colony optimization technique. The outcome of the system demonstrates that the sensitivity, specificity, and accuracy of the proposed technique is 90.41%, 99.66%, and 99.36%, respectively, which is better than all the existing methods.

Prediction of displacement in the equine third metacarpal bone using a neural network prediction algorithm

Bone is a nonlinear, inhomogeneous and anisotropic material. To predict the behavior of bones expert systems are employed to reduce the computational cost and to enhance the accuracy of simulations. In this study, an artificial neural network (ANN) was used for the prediction of displacement in long bones followed by ex-vivo experiments. Three hydrated third metacarpal bones (MC3) from 3 thoroughbred horses were used in the experiments. A set of strain gauges were distributed around the midshaft of the bones. These bones were then loaded in compression in an MTS machine. The recordings of strains, load, load exposure time, and displacement were used as ANN input parameters. The ANN which was trained using 3,250 experimental data points from two bones predicted the displacement of the third bone (R2 ≥ 0.98). It was suggested that the ANN should be trained using noisy data points. The proposed modification in the training algorithm makes the ANN very robust against noisy inputs measurements. The performance of the ANN was evaluated in response to changes in the number of input data points and then by assuming a lack of strain data. A finite element analysis (FEA) was conducted to replicate one cycle of force-displacement experimental data (to gain the same accuracy produced by the ANN). The comparison of FEA and ANN displacement predictions indicates that the ANN produced a satisfactory outcome within a couple of seconds, while FEA required more than 160 times as long to solve the same model (CPU time: 5 h and 30 min).

An electronic expansion valve modeling framework development using artificial neural network: A case study on VRF systems

Electronic expansion valves (EEV) are widely used in variable refrigerant flow systems (VRF) to control the mass flow rate of each indoor unit. EEV model is used to predict the mass flow rate through an EEV. While the power-law correlation method has been used to build the EEV model so far, Artificial Neural Network (ANN) methods have been adapted to model the EEV with a fixed speed compressor thanks to its higher accuracy. However, the EEV is typically working with the variable speed compressor in a VRF system. In addition, the parameters used in ANN modeling could be further optimized. The objective of this study is to develop an EEV modeling framework and optimize the input parameter number and hidden neuron number. We presented the framework through an EEV model development for a VRF system. For these, we used the field test data, applied a principal components analysis approach in optimizing the ANN input parameter number, and investigated the proper number of hidden neurons and appropriate transfer function pairs. We found the performance of the ANN model would not improve much as the number of input parameters and the number of hidden neurons reached a threshold. Only three transfer function pairs are suitable for this case in total nine groups we studied. Our ANN model has the average absolute deviation of 2.2%. The framework can be easily applied to EEV modeling in other VRF and air condition systems with necessary data support.

Classification of drug-induced hERG potassium-channel block from electrocardiographic T-wave features using artificial neural networks.

Human ether-à-go-go-related gene (hERG) potassium-channel block represents a harmful side effect of drug therapy that may cause torsade de pointes (TdP). Analysis of ventricular repolarization through electrocardiographic T-wave features represents a noninvasive way to accurately evaluate the TdP risk in drug-safety studies. This study proposes an artificial neural network (ANN) for noninvasive electrocardiography-based classification of the hERG potassium-channel block. The data were taken from the "ECG Effects of Ranolazine, Dofetilide, Verapamil, and Quinidine in Healthy Subjects" Physionet database; they consisted of median vector magnitude (VM) beats of 22 healthy subjects receiving a single 500μg dose of dofetilide. Fourteen VM beats were considered for each subject, relative to time-points ranging from 0.5hr before to 14.0hr after dofetilide administration. For each VM, changes in two indexes accounting for the early and the late phases of repolarization, ΔERD30% and ΔTS/A , respectively, were computed as difference between values at each postdose time-point and the predose time-point. Thus, the dataset contained 286 ΔERD30% -ΔTS/A pairs, partitioned into training, validation, and test sets (114, 29, and 143 pairs, respectively) and used as inputs of a two-layer feedforward ANN with two target classes: high block (HB) and low block (LB). Optimal ANN (OANN) was identified using the training and validation sets and tested on the test set. Test set area under the receiver operating characteristic was 0.91; sensitivity, specificity, accuracy, and precision were 0.93, 0.83, 0.92, and 0.96, respectively. OANN represents a reliable tool for noninvasive assessment of the hERG potassium-channel block.

Quantitative CT texture analysis for diagnosing systemic sclerosis: Effect of iterative reconstructions and radiation doses.

To test whether texture analysis (TA) can discriminate between Systemic Sclerosis (SSc) and non-SSc patients in computed tomography (CT) with different radiation doses and reconstruction algorithms.In this IRB-approved retrospective study, 85 CT scans at different radiation doses [49 standard dose CT (SDCT) with a volume CT dose index (CTDIvol) of 4.86 ± 2.1 mGy and 36 low-dose (LDCT) with a CTDIvol of 2.5 ± 1.5 mGy] were selected; 61 patients had Ssc ("cases"), and 24 patients had no SSc ("controls"). CT scans were reconstructed with filtered-back projection (FBP) and with sinogram-affirmed iterative reconstruction (SAFIRE) algorithms. 304 TA features were extracted from each manually drawn region-of-interest at 6 pre-defined levels: at the midpoint between lung apices and tracheal carina, at the level of the tracheal carina, and 4 between the carina and pleural recesses. Each TA feature was averaged between these 6 pre-defined levels and was used as input in the machine learning algorithm artificial neural network (ANN) with backpropagation (MultilayerPerceptron) for differentiating between SSc and non-SSc patients.Results were compared regarding correctly/incorrectly classified instances and ROC-AUCs.ANN correctly classified individuals in 93.8% (AUC = 0.981) of FBP-LDCT, in 78.5% (AUC = 0.859) of FBP-SDCT, in 91.1% (AUC = 0.922) of SAFIRE3-LDCT and 75.7% (AUC = 0.815) of SAFIRE3-SDCT, in 88.1% (AUC = 0.929) of SAFIRE5-LDCT and 74% (AUC = 0.815) of SAFIRE5-SDCT.Quantitative TA-based discrimination of CT of SSc patients is possible showing highest discriminatory power in FBP-LDCT images.

Lipid profile prediction based on artificial neural networks

Lipid profile usually includes levels of total cholesterol (TCH), low density lipoprotein (LDL), high density lipoprotein (HDL) and triglycerides (TG), all of which require a blood test. Using advances in machine learning and a relationship between lipid profile and obesity, a model that predicts lipid profile without using any laboratory results can be developed and used in clinical diagnosis. The causal relationship between lipid profile and obesity is well known—TCH, LDL and TG show an increase, while HDL is decreased in obese persons. In this paper we are using artificial neural networks (ANN) to estimate the lipid profile values using non-lab electronic health record data and some measures of obesity. The ANN inputs are gender, age, systolic and diastolic blood pressures, and a single or a combination of multiple obesity parameters, which include body mass index, saggital abdominal diameter to height ratio, waist to height ratio and body fat percentage. Study shows that the presented solution is suitable for prediction of TCH (with accuracy 81.89%), LDL (with accuracy 79.29%) and HDL (with accuracy 81.23%), while not suitable for TG prediction (with accuracy 44.48%).

Real-time simulation for long paths in laser-based additive manufacturing: a machine learning approach

This study predicts in real time the evolution of temperature and density for arbitrary long tracks in laser-based additive manufacturing using artificial neural networks (ANNs). First, a random laser path is transformed appropriately to serve as an ANN input via a custom trajectory decomposition method providing a local description of each trajectory point relative to its surroundings, which is calculated and load-optimized using K-d trees. For each trajectory point, a “Master” ANN calculates the present temperature, using the trajectory descriptors and previous step temperature as inputs. This is made into a parallel procedure by exploiting a “Pilot” ANN without temperature feedback, making a first pass over the whole trajectory to estimate temperature responses. The Master ANN uses these results as a feedback that is iteratively refined. A “Rider” ANN uses the final temperatures as input and calculates local density evolution. Four hundred fifty finite element model experiments of short random walk trajectories were used for ANN training. Results are presented for various types of laser paths, including random, hatch, spiral, fractal, and spline. ANN simulation execution time consistently taking under 50% of the real process time in all validated cases for track length in excess of 100 m proved the ability of the platform to operate at the full layer-scale of LBAM.

Diagnosis and Classifications of Bearing Faults Using Artificial Neural Network and Support Vector Machine

The research paper presents a comparative study of artificial neural network (ANN) and support vector machine (SVM) using continuous wavelet transforms and energy entropy approaches for fault diagnosis and classification of rolling element bearings. An experimental test rig is used to acquire the vibration signals of healthy and faulty bearings. Four real-valued base wavelets are considered. Out of these wavelets, mother base wavelet is selected on behalf of maximum energy and minimum entropy criterions and extracts the statistical features from wavelet coefficient of raw vibration signals. These statistical features are used as input of ANN and SVM for classifying the faults of bearings. Finally, Morlet wavelet is selected on the basis of energy and entropy criterions. The test results show that SVM gives the better fault diagnosis and classification accuracy than ANN.

Artificial neural networks modeling of kinetic curves of celeriac (Apium graveolens L.) in vacuum drying

Abstract The objective of this study was to predict celeriac drying curves using artificial neural networks (ANNs). The experimental data for vacuum drying kinetics of celeriac slices reported by other researcher in the previously published article was used. The air temperature, chamber pressure and time values were used as ANN inputs. To predict the moisture content, the multilayer feed forward back propagation neural network, as a well-known network, was used. The network with Levenberg-Marquardt learning algorithm, hyperbolic tangent sigmoid transfer function, and 3-6-9-1 topology provided the superior results.

Rapid discrimination of multiple myeloma patients by artificial neural networks coupled with mass spectrometry of peripheral blood plasma

Multiple myeloma (MM) is a highly heterogeneous disease of malignant plasma cells. Diagnosis and monitoring of MM patients is based on bone marrow biopsies and detection of abnormal immunoglobulin in serum and/or urine. However, biopsies have a single-site bias; thus, new diagnostic tests and early detection strategies are needed. Matrix-Assisted Laser Desorption/Ionization Time-of Flight Mass Spectrometry (MALDI-TOF MS) is a powerful method that found its applications in clinical diagnostics. Artificial intelligence approaches, such as Artificial Neural Networks (ANNs), can handle non-linear data and provide prediction and classification of variables in multidimensional datasets. In this study, we used MALDI-TOF MS to acquire low mass profiles of peripheral blood plasma obtained from MM patients and healthy donors. Informative patterns in mass spectra served as inputs for ANN that specifically predicted MM samples with high sensitivity (100%), specificity (95%) and accuracy (98%). Thus, mass spectrometry coupled with ANN can provide a minimally invasive approach for MM diagnostics.