Input Data For Artificial Neural Network Research Articles

Evaluation of Hybrid Soft Computing Model’s Performance in Estimating Wave Height

In coastal and port engineering, wind-generated waves have always been a crucial, fundamental, and important topic. As a result, various methods for estimating wave parameters, including field measurement and numerical methods, have been proposed over time. This study evaluates the wave height at Sri-Lanka Hambantota Port using soft computing models such as Artificial Neural Networks (ANNs) and the M5 model tree (M5MT). In order to overcome its nonstationarity, the primary wave height time series were divided into subtime series using the wavelet transform. The collected subtime series were then utilized as input data for ANN and M5MT in order to determine the wave height. For the sake of the model performance, the daily wind and wave data from the Acoustic Wave and Current (AWAC) sensor for Hambantota Port in 2020 and Sanmen Bay in 2017 were used in this study. The training state utilizes 80% of the available data, while the test state uses 20%. The Root Mean Square Error (RMSE) of the ANN, M5, WANN, and Wavelet-M5 models in the Hambantota Port for the test stage are 0.12, 0.11, 0.04, and 0.06, respectively. While in Sanmen Bay, the RMSE of the ANN, M5, WANN, and Wavelet-M5 models for the test stage are 0.14, 0.16, 0.06, and 0.08, respectively. According to the findings of this study, the accuracy of WANN and Wavelet-M5 hybrid models in evaluating wave height is superior to that of classic ANN and M5MT, and it is recommended that WANN and Wavelet-M5 hybrid models be used to estimate wave height.

Identification of Power Quality Disturbances Based on Fast Fourier Transform and Artificial Neural Network

This paper presents the proposed algorithms for the identification of Short Duration RMS Variations and Long Duration RMS Variations combined with harmonic. The proposed algorithms are Fast Fourier Transform (FFT) and Artificial Neural Network (ANN). The Algorithms identify nine types of Power Quality (PQ) disturbances such as normal signal, voltage sag, voltage swell, under voltage, over voltage, voltage sag combined harmonic, voltage swell combined harmonic, undervoltage combined harmonic, and over voltage combined harmonic. FFT is used to obtain the frequency spectrum of each PQ disturbance with frequency sampling of 1000 Hz, data length of 200. Output FFT is used to input data for ANN. Output ANN is a type of nine PQ disturbances. The result shows that proposed algorithms (FFT combined ANN) are effective for identification, which ANN with 20 neurons in the hidden layer has an accuracy of approximately 99.95 %

Quality Prediction in Directed Energy Deposition Using Artificial Neural Networks Based on Process Signals

The Directed Energy Deposition process is used in a wide range of applications including the repair, coating or modification of existing structures and the additive manufacturing of individual parts. As the process is frequently applied in the aerospace industry, the requirements for quality assurance are extremely high. Therefore, more and more sensor systems are being implemented for process monitoring. To evaluate the generated data, suitable methods must be developed. A solution, in this context, was the application of artificial neural networks (ANNs). This article demonstrates how measurement data can be used as input data for ANNs. The measurement data were generated using a pyrometer, an emission spectrometer, a camera (Charge-Coupled Device) and a laser scanner. First, a concept for the extraction of relevant features from dynamic measurement data series was presented. The developed method was then applied to generate a data set for the quality prediction of various geometries, including weld beads, coatings and cubes. The results were compared to ANNs trained with process parameters such as laser power, scan speed and powder mass flow. It was shown that the use of measurement data provides additional value. Neural networks trained with measurement data achieve significantly higher prediction accuracy, especially for more complex geometries.

P–135 Artificial neural networks (ANNs) for live birth prediction in frozen embryo transfers: the strength of post-warmed blastocyst dynamics

Abstract Study question Does the post-warmed blastocyst dynamics have an impact over the likelihood of achieving a live birth? Summary answer Variables related to dynamics of vitrified/warmed blastocysts have shown a greater effect on the live birth prediction than only embryo morphological quality through artificial intelligence. What is known already Morphological dynamics of vitrified/warmed blastocysts were described by Coello et al., in 2017. The investigated markers were the thickness of zona pellucida (µm) and blastocysts area (µm2) after warming and before transfer, the area of the inner cell mass (µm2), time of initiation of reexpansion (in minutes), and presence of collapse or contraction. They found a correlation between blastocyst reexpansion and implantation rate and developed a hierarchical model for implantation prediction. In our study, we evaluated the post-warmed blastocyst dynamics for live birth prediction by using novel artificial intelligence techniques. Study design, size, duration This retrospective analysis included 415 vitrified/warmed blastocysts with known live birth data. Blastocysts after warming were placed in EmbryoScope (Vitrolife) immediately until embryo transfer. Embryo evaluation and selection were performed by senior embryologists according to fresh blastocyst morphology (before vitrification). Then, parameters related to post-warmed blastocyst dynamics were calculated. Finally, these variables and the embryo morphological grade before the vitrification were used as input data for ANNs optimized with genetic algorithm for live birth prediction. Participants/materials, setting, methods Blastocysts were vitrified and warmed by the Cryotop method (Kitazato,Biopharma). During the period between the warming procedure and the embryo transfer, the following variables were measured with the drawing tools provided by the EmbryoViewer workstation: zona pellucida thinning (µm), blastocyst expansion (um) and the speed of these two events (µm/h). Finally, multilayer perceptron neural networks were trained with data of 331 embryos by using the backpropagation learning algorithm and tested with data of 84 embryos. Main results and the role of chance We trained and tested three architectures of ANNs with different input variables as follows: post-warmed variables (thinning of the zona pellucida, blastocyst expansion, thinning speed and expansion speed) and morphological grade (A, B or C) for ANN1, only post-warmed variables for ANN2 and only morphological grade for ANN3. The highest success rate when ANNs classified embryos as positive and negative live birth (LB+ and LB-) was achieved by combining post-warmed variables and morphological grade before embryo vitrification. The general accuracies for the blind tests were: 73.8% for ANN1, 66.7% for ANN2 and 71.4% for ANN3. Likewise, this combination achieved the highest AUC on test dataset to predict LB- (0.76 for ANN1, 0.74 for ANN2 and 0.67 for ANN3). However, the ANN2 trained with only post-warmed variables showed the best capacity to predict LB+ with an AUC of 0.73 (versus 0.46 for ANN1 and 0.5 for ANN3). Limitations, reasons for caution The main limitation is the subjectivity of manual annotations, although only one embryologist participated in this task. Wider implications of the findings: The dynamics of vitrified/warmed blastocysts prior to embryo transfer could be more relevant variables than the morphological quality on day 5 before the cryopreservation. The analysis of embryo behavior after warming could improve clinical outcomes in frozen embryo transfers. Trial registration number none

Software implemented fault diagnosis of natural gas pumping unit based on feedforward neural network

In recent years, more and more attention has been paid to the use of artificial neural networks (ANN) for the diagnostics of gas pumping units (GPU). Usually, ANN training is carried out on GPU workflow models, and generated sets of diagnostic data are used to simulate defect conditions. At the same time, the results obtained do not allow assessing the real state of the GPU. It is proposed to use the characteristics of the acoustic and vibration processes of the GPU as the input data of the ANN. A descriptive statistical analysis of real vibration and acoustic processes generated by the operation of the GPU type GTK-25-i (Nuovo Pignone, Italy) was carried out. The formation of batches of diagnostic features arriving at the input of the ANN was carried out. Diagnostic features are the five maximum amplitude components of the acoustic and vibration signals, as well as the value of the standard deviation for each sample. Diagnostic features are calculated directly in the ANN input data pipeline in real time for three technical states of the GPU. Using the frameworks TensorFlow, Keras, NumPy, pandas, in the Python 3 programming language, an architecture was developed for a deep fully connected feedforward ANN, trained on the backpropagation algorithm. The results of training and testing the developed ANN are presented. During testing, it was found that the signal classification precision for the “nominal” state of all 1,475 signal samples is 1.0000, for the “current” state, precision equals 0.9853, and for the “defective” state, precision is 0.9091. The use of the developed ANN makes it possible to classify the technical states of the GPU with an accuracy sufficient for practical use, which will prevent the occurrence of GPU failures. ANN can be used to diagnose GPU of any type and power

Prediction of instantaneous real-world emissions from diesel light-duty vehicles based on an integrated artificial neural network and vehicle dynamics model

This paper presents a road vehicle emission model that integrates an artificial neural network (ANN) model with a vehicle dynamics model to predict the instantaneous carbon dioxide (CO2), nitrogen oxides (NOx) and total hydrocarbon (THC) emissions of diesel light-duty vehicles. Real-world measurement data were used to train a multi-layer feed-forward ANN model. The optimal combination of the various experimental variables was selected as the ANN input through a parametric study considering both practicality and accuracy. For CO2 prediction, two variables (engine speed and engine torque) are enough to develop an accurate ANN model. In order to achieve satisfactory accuracy for CO and NOx prediction, more variables were used for ANN training. The trained ANN model was used to predict road vehicle emissions by integrating the vehicle dynamics model, which was used as a supplementary tool to produce ANN input data. The integrated model is practical because it requires relatively simple data for input such as vehicle specifications, velocity, and road gradient. In the accuracy validation, the proposed model showed satisfactory prediction accuracy for road vehicle emissions.

Mooring Damage Identification of Floating Wind Turbine Using a Non-Probabilistic Approach Under Different Environmental Conditions

This paper discusses the damage identification in the mooring line system of a floating wind turbine (FWT) exposed to various environmental loads. The proposed method incorporates a non-probabilistic method into artificial neural networks (ANNs). The non-probabilistic method is used to overcome the problem of uncertainties. For this purpose, the interval analysis method is used to calculate the lower and upper bounds of ANNs input data. This data contains some of the natural frequencies utilized to train two different ANNs and predict the output data which is the interval bounds of mooring line stiffness. Additionally, in order to reduce computational time and more importantly, identify damage in various conditions, the proposed method is trained using constant loads (CL) case (deterministic loads, including constant wind speed and airy wave model) and is tested using random loads (RL) case (including Kaimal wind model and JONSWAP wave theory). The superiority of this method is assessed by applying the deterministic method for damage identification. The results demonstrate that the proposed non-probabilistic method identifies the location and severity of damage more accurately compared to a deterministic one. This superiority is getting more remarkable as the difference in uncertainty levels between training and testing data is increasing.

Optimization of a flow regime identification system and prediction of volume fractions in three-phase systems using gamma-rays and artificial neural network

This study presents a method based on gamma-ray densitometry using only one multilayer perceptron artificial neural network (ANN) to identify flow regime and predict volume fraction of gas, water, and oil in multiphase flow, simultaneously, making the prediction independent of the flow regime. Two NaI(Tl) detectors to record the transmission and scattering beams and a source with two gamma-ray energies comprise the detection geometry. The spectra of gamma-ray recorded by both detectors were chosen as ANN input data. Stratified, homogeneous, and annular flow regimes with (5 to 95%) various volume fractions were simulated by the MCNP6 code, in order to obtain an adequate data set for training and assessing the generalization capacity of ANN. All three regimes were correctly distinguished for 98% of the investigated patterns and the volume fraction in multiphase systems was predicted with a relative error of less than 5% for the gas and water phases.

Efficient Structural Design of a Prefab Concrete Connection by Using Artificial Neural Networks

In the built environment, one of the main concerns during the design stage is the selection of adequate structural materials and elements. A rational and sensible design of both materials and elements results not only in economic benefits and computing time reduction, but also in minimizing the environmental impact. Nowadays, Artificial Neural Networks (ANNs) are showing their potential as design tools. In this research, ANNs are used in order to foster the implementation of efficient tools to be used during the early stages of structural design. The proposed networks are applied to a dry precast concrete connection, which has been modelled by means of the Finite Element Method (FEM). The parameters are: strength of concrete and screws, diameter of screws, plate thickness, and the posttensioning load. The ANN input data are the parameters and nodal stresses obtained from the FEM models. A multilayer perceptron combined with a backpropagation algorithm is used in the ANN architecture, and a hyperbolic tangent function is applied as an activation function. Comparing the obtained predicted stresses to those of the FEM analyses, the difference is less than 9.16%. Those results validate their use as an efficient structural design tool. The main advantage of the proposed ANNs is that they can be easily and effectively adapted to different connection parameters. In addition, their use could be applied both in precast or cast in situ concrete connection design.

STATIC AIR-GAP ECCENTRICITY FAULT DETECTION OF INDUCTION MOTOR USING ARTIFICIAL NEURAL NETWORK (ANN)

Induction motor plays an important role in the industrial, commercial and residential industries, owing to its immense advantages over the opposite types of motors. Such motors have to operate under different operating conditions that cause engine degradation leading to fault occurrences. There are numerous fault detection techniques available. There are numerous fault detection techniques available. The technique used in this paper to prove the effect of static air gap eccentricity on behaving or performing of the three-phase induction motor is the artificial neural network (ANN) as ANN depends on detecting the fault on the amplitude of positive and negative harmonics of frequencies. In this paper, we used two motors to achieve real malfunctions and to get the required data and for three different load tests. In this paper, we adopted MCSA to detect the fault based on the stator current. The ANN training algorithm used in this paper is back propagation and feed forward. The inputs of ANN are the speed and the amplitudes of the positive and the negative harmonics, and the type of fault is the output. To distinguish between healthy and faulty motor, the input data of ANN are well-trained via experiments test. The methodology applied in this paper was MATLAB and present how we can distinguish between healthy and faulty motor.

Performance comparison of artificial intelligence techniques in short term current forecasting for photovoltaic system

<span>This paper presents artificial intelligence approach of artificial neural network (ANN) and random forest (RF) that used to perform short-term photovoltaic (PV) output current forecasting (STPCF) for the next 24-hours. The input data for ANN and RF is consists of multiple time lags of hourly solar irradiance, temperature, hour, power and current to determine the movement pattern of data that have been denoised by using wavelet decomposition. The Levenberg-Marquardt optimization technique is used as a back-propagation algorithm for ANN and the bagging based bootstrapping technique is used in the RF to improve the results of forecasting. The information of PV output current is obtained from Green Energy Research (GERC) University Technology Mara Shah Alam, Malaysia and is used as the case study in estimation of PV output current for the next 24-hours. The results have shown that both proposed techniques are able to perform forecasting of future hourly PV output current with less error.</span>

ULTRASHORT IMPULSE RADAR FOR DETECTION AND CLASSIFICATION OF OBJECTS IN LAYERED MEDIUM BY ARTIFICIAL NEURAL NETWORK

The application of artificial neural networks (ANN) for object classification based on data received by impulse radar is analyzed. Instead of spectrum analysis or threshold criterion application, the amplitudes of received impulse electromagnetic field in a definite time and spatial points are processed by the neural network directly to make an object recognition procedure. Several electromagnetic problems and structures of ANN are considered as examples to obtain the layered medium parameter, to find some object and its coordinates in a medium. The stability of the determination of object placement and medium geometrical characteristics are studied in the presence of experimental errors and noises. The normal incidence of short impulse plane wave of Gaussian time form on the complex medium with losses is considered. Finite Difference in Time Domain (FDTD) method is used for obtaining the reflected electromagnetic field above a medium. The amplitudes of the electrical component of the field are the input data for ANN. The thicknesses of one of the layers or distances to an object are the ANN output parameters that are sought for.

A regional suspended load yield estimation model for ungauged watersheds

Developing regional models using physiographic, climatic, and hydrologic variables is an approach to estimating suspended load yield (SLY) in ungauged watersheds. However, using all the variables might reduce the applicability of these models. Therefore, data reduction techniques (DRTs), e.g., principal component analysis (PCA), Gamma test (GT), and stepwise regression (SR), have been used to select the most effective variables. The artificial neural network (ANN) and multiple linear regression (MLR) are also common tools for SLY modeling. We conducted this study (1) to obtain the most effective variables influencing SLY through DRTs including PCA, GT, and SR, and then, to use them as input data for ANN and MLR; and (2) to provide the best SLY models. Accordingly, we used 14 physiographic, climatic, and hydrologic parameters from 42 watersheds in the Hyrcanian forest region (in northern Iran). The most effective variables as determined through DRTs as well as the original data sets were used as the input data for ANN and MLR in order to provide an SLY model. The results indicated that the SLY models provided by ANN performed much better than the MLR models, and the GT-ANN model was the best. The determination of coefficient, relative error, root mean square error, and bias were 99.9%, 26%, 323 t/year, and 6 t/year in the calibration period, and 70%, 43%, 456 t/year, and 407 t/year in the validation period, respectively. Overall, selecting the main factors that influence SLY and using artificial intelligence tools can be useful for water resources managers to quickly determine the behavior of SLY in ungauged watersheds.

Corrosion Control by Impressed Cathodic Protection Using Intelligent Systems

The aim of this study is to adopt the Artificial Neural Network (ANN) to Model the Cathodic Protection system (CPS) and evaluation the potential required to protect the coated and bared pipeline as well as to the prediction of corrosion rate. On the other hand, the experimental work was carried out to collect the required data to be used for training and testing the neural network. The objective of this research paper is to corrosion control in the pipeline with different potential values. The proposed structure of ANN for potential and corrosion is an input layer, two hidden layers and one output layer and this structure is arbitrarily selected. The transfer function that has been used in the first hidden layer for each network is the Tan-Sigmoid function and for the second layer is the pure line. The back propagation training algorithm with one variable learning rate is used to train these neural networks. For the potential assessment; the ANN input data includes the distance between anodes and cathodes (D), Current Density (CD), length of pipe from end to the drain point (L), resistivity of solution (ρ) and the voltage of power stations, while the potential is the network output. For the corrosion rate prediction, the network input information is only time, surface area and resistivity of the soil (solution) (ρ), while the corrosion rate is the network output. Many networks are constructed by changing the number of neurons for the hidden layers. This has been simulated by using the MATLAB R2009a software. The optimum network for coated pipe was (13) neurons in the first hidden layer and (8) neurons in the second hidden layer which is tested and trained by using (120 data sample). This network has proved to be reliable and can be used to assess the potential required for CPS. Concerning the bared pipe-lines, the collected experimental data is not stable and the fluctuation of the data occurs due to the interference between the corroded part of the pipe and the protected parts, which causes the un-stability of potential. The optimum network for coated pipe was (15) neurons in the first hidden layer and (4) neurons in the second hidden layer) which is tested and trained by using (250 data sample). This network demonstrates to be reliable and capable of predicting the corrosion rate.

A new collection of compressed damage indices for multi-damage detection of cold formed steel shear walls based on neural network ensembles

This study presents a methodology that utilizes a new combination of two compressed damage indices as input data of an artificial neural network (ANN) ensemble to detect multi-damages in the braces of cold formed steel shear walls. To identify an efficient input data for ANN, first, three main groups of damage indices are considered: modal parameter-based damage indices; frequency response functions (FRFs)-based damage indices and time series-based damage indices. Furthermore, principal component analysis (PCA) technique is applied to reduce the dimensions of FRFs and time series-based input pattern. By a sensitivity study, two suitable damage indices of PCA-compressed time series data and PCA-compressed FRFs are identified and then combined to produce a new efficient input data for a hierarchy of ANN ensembles. The numerical results show that the ANN ensemble-based damage detection approach with the proposed collection of two damage indices is effective and reliable.

Artificial intelligence approach based on near-infrared spectral data for monitoring of solid-state fermentation

This work aimed to establish a chemometric technique for quantifying amylase and protease activities as well as protein concentration in aqueous extracts of Rhizopus microsporus var. oligosporus obtained via solid-state fermentation (SSF). The kinetics of four agro-industrial wastes (wheat bran, soybean meal, type II wheat flour and sugarcane bagasse) were studied for 144h, along with two different sets of their ternary mixtures, at a constant fermentation time of 120h, to obtain primary data (biochemical parameters as well as near-infrared (NIR) spectral data). Then, models such as artificial neural network (ANN) and partial least squares (PLS) were calibrated to predict biochemical parameters using the spectral data. Primary data and three methods of preprocessing data – first, second and third derivatives – were assessed as inputs for both chemometric tools. The third derivative, that is, spectral pre-processing plus an optimized ANN, showed the least relative errors (<8.3%±10.5%). The third-derivative spectrum was found to be suitable as the ANN input data for monitoring amylase and protease activities and protein concentration in the SSF under study. The proposed methodology can serve as a foundation for at-line sensor development and decrease the time and cost of bioprocess development using Rhizopus microsporus var. oligosporus.

Neutron spectrum unfolding using artificial neural network and modified least square method

In the present paper, neutron spectrum is reconstructed using the Artificial Neural Network (ANN) and Modified Least Square (MLSQR) methods. The detector's response (pulse height distribution) as a required data for unfolding of energy spectrum is calculated using the developed MCNPX-ESUT computational code (MCNPX-Energy engineering of Sharif University of Technology). Unlike the usual methods that apply inversion procedures to unfold the energy spectrum from the Fredholm integral equation, the MLSQR method uses the direct procedure. Since liquid organic scintillators like NE-213 are well suited and routinely used for spectrometry of neutron sources, the neutron pulse height distribution is simulated/measured in the NE-213 detector. The response matrix is calculated using the MCNPX-ESUT computational code through the simulation of NE-213 detector's response to monoenergetic neutron sources. For known neutron pulse height distribution, the energy spectrum of the neutron source is unfolded using the MLSQR method. In the developed multilayer perception neural network for reconstruction of the energy spectrum of the neutron source, there is no need for formation of the response matrix. The multilayer perception neural network is developed based on logsig, tansig and purelin transfer functions. The developed artificial neural network consists of two hidden layers of type hyperbolic tangent sigmoid transfer function and a linear transfer function in the output layer. The motivation of applying the ANN method may be explained by the fact that no matrix inversion is needed for energy spectrum unfolding. The simulated neutron pulse height distributions in each light bin due to randomly generated neutron spectrum are considered as the input data of ANN. Also, the randomly generated energy spectra are considered as the output data of the ANN. Energy spectrum of the neutron source is identified with high accuracy using both MLSQR and ANN methods. The results obtained from MLSQR and ANN methods for 252Cf and 241Am-9Be source are validated against the ISO spectrum. The unfolded neutron energy spectra from both MLSQR and ANN methods show a good agreement with the actual spectrum of 252Cf and 241Am-9Be source.

Digital Soil Mapping Using Artificial Neural Networks and Terrain-Related Attributes

Detailed soil surveys involve costly and time-consuming work and require expert knowledge. Since soil surveys provide information to meet a wide range of needs, new methods are necessary to map soils quickly and accurately. In this study, multilayer perceptron artificial neural networks (ANNs) were developed to map soil units using digital elevation model (DEM) attributes. Several optimal ANNs were produced based on a number of input data and hidden units. The approach used test and validation areas to calculate the accuracy of interpolated and extrapolated data. The results showed that the system and level of soil classification employed had a direct effect on the accuracy of the results. At the lowest level, smaller errors were observed with the World Reference Base (WRB) classification criteria than the Soil Taxonomy (ST) system, but more soil classes could be predicted when using ST (7 soils in the case of ST vs. 5 with WRB). Training errors were below 11% for all the ANN models applied, while the test error (interpolation error) and validation error (extrapolation error) were as high as 50% and 70%, respectively. As expected, soil prediction using a higher level of classification presented a better overall level of accuracy. To obtain better predictions, in addition to DEM attributes, data related to landforms and/or lithology as soil-forming factors, should be used as ANN input data.

Infrared and Microwave Image Fusion for Rainfall Detection over Northern Algeria

Rain areas delineation proposed in this paper is based on the image fusion from geostationary Meteosat Second Generation (MSG) satellite, with the low-earth orbiting passive Tropical Rainfall Measuring Mission (TRMM) satellite. The fusion technique described in this work used an artificial neural network (ANN). It's has been developed to detect instantaneous rainfall by using information from the IR images of MSG satellite and from TRMM Microwave Imager (TMI). The study is carried out over north of Algeria. Seven spectral parameters are used as input data of ANN to identify raining or non - raining pixels. Corresponding data of raining /non-raining pixels are taken from a PR (precipitation radar) issued from TRMM. Results from the developed scheme are compared with the results of SI method (Scattering Index) taken as reference method. The results show that the developed model performs very well and overcomes the deficiencies of use a single satellite.

Root‐zone soil moisture estimation using data‐driven methods

AbstractThe soil moisture state partitions both mass and energy fluxes and is important for many hydro‐geochemical cycles, but is often only measured within the surface layer. Estimating the amount of soil moisture in the root‐zone from this information is difficult due to the nonlinear and heterogeneous nature of the various processes which alter the soil moisture state. Data‐driven methods, such as artificial neural networks (ANN), mine data for nonlinear interdependencies and have potential for estimating root‐zone soil moisture from surface soil moisture observations. To create an ANN root‐zone model that was nonsite‐specific and physically constrained, a training set was generated by forcing HYDRUS‐1D with meteorological observations for different soil profiles from the unsaturated soil hydraulic database. Ensemble ANNs were trained to provide soil moisture at depths of 10, 20, and 50 cm below the surface using surface soil moisture observations and local meteorological information. Insights into the processes represented by the ANNs were derived from a clamping sensitivity analysis and by changing the ANNs input data. Further model testing based on synthetic soil moisture profiles from three McMaster Mesonet and three USDA soil climate analysis network sites suggests that ANNs are a flexible tool capable of predicting root‐zone soil moisture with good accuracy. It was found that ANNs could well represent soil moisture as estimated by HYDRUS‐1D, but performance was reduced in comparison to in situ soil moisture observations outside the training conditions. The transferability of the model appears limited to the same geographic region.