Artificial Neural Network Training Phase Research Articles

On the Training Algorithms for Artificial Neural Network in Predicting the Shear Strength of Deep Beams

This study aims to predict the shear strength of reinforced concrete (RC) deep beams based on artificial neural network (ANN) using four training algorithms, namely, Levenberg–Marquardt (ANN-LM), quasi-Newton method (ANN-QN), conjugate gradient (ANN-CG), and gradient descent (ANN-GD). A database containing 106 results of RC deep beam shear strength tests is collected and used to investigate the performance of the four proposed algorithms. The ANN training phase uses 70% of data, randomly taken from the collected dataset, whereas the remaining 30% of data are used for the algorithms’ evaluation process. The ANN structure consists of an input layer with 9 neurons corresponding to 9 input parameters, a hidden layer of 10 neurons, and an output layer with 1 neuron representing the shear strength of RC deep beams. The performance evaluation of the models is performed using statistical criteria, including the correlation coefficient (R), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The results show that the ANN-CG model has the best prediction performance with R = 0.992, RMSE = 14.02, MAE = 14.24, and MAPE = 6.84. The results of this study show that the ANN-CG model can accurately predict the shear strength of RC deep beams, representing a promising and useful alternative design solution for structural engineers.

Predicting the compressive strength of normal and High-Performance Concretes using ANN and ANFIS hybridized with Grey Wolf Optimizer

Achieving a reliable model for predicting the compressive strength (CS) of concrete can save in time, energy, and cost and also provide information about scheduling for construction and framework removal. In this study, Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS) techniques were hybridized by Grey Wolf Optimizer (GWO) to develop the predictive models for predicting the CS of Normal Concrete (NC) and High-Performance Concrete (HPC). The classical optimization algorithms (COAs) served in training of ANN and ANFIS have a high capability in the exploitation phase. In this study, GWO was used in the training phase of ANN and ANFIS to eliminate this weakness. In this regard, a comprehensive dataset containing 2817 distinctive data records was collected to develop six ANN and three ANFIS models. In case of ANN models, three models were developed using three different COAs and the others were constructed using hybridization of these COAs and GWO. With regard to ANFIS models, one model was developed using the original version of ANFIS and two models were hybridized with GWO. The results indicate that the hybridization of the models with GWO improves the training and generalization capability of both ANN and ANFIS models. It is also deduced that ANN models trained with Levenberg-Marquardt algorithm outperformed other ANN-based models as well as all ANFIS-based models.

A particle-based optimization of artificial neural network for earthquake-induced landslide assessment in Ludian county, China

The focal point of this study is to assess the efficacy of a state-of-the-art optimization technique namely, particle swarm optimization (PSO) for enhancing the performance of the artificial neural network (ANN) in modeling the seismic landslides at Ludian districts, China. Twelve geological and hydrological landslide-conditioning factors namely, elevation, lithology, slope degree, slope aspect, stream power index, peak ground acceleration, topographic wetness index, distance to river, distance to road, distance to fault, normalized difference vegetation index and plan curvature were considered within a geographic information system (GIS). After achieving the optimal structure of the multilayer perceptron neural network, the PSO algorithm was applied to improve its efficiency. The landslide susceptibility maps were generated in a GIS environment and area under the curve (AUC) criterion was used to assess the integrity of employed predictive models. The results showed that after applying the PSO algorithm, AUC experiences a significant increase from 0.765 to 0.825 in the validation phase. Moreover, respective AUCs of 0.812 and 0.828 obtained for the training phase of ANN and PSO-ANN reveal the efficiency of the proposed algorithm in improving the ANN accuracy.

A risk assessment framework for construction project using artificial neural network

The current trend of increasing construction project size and complexity results in higher level of project risk. As a result, risk management is a crucial determinant of the success of a project. It seems necessary for construction companies to integrate a risk management system into their organizational structure. The main aim of this paper is to propose a risk assessment framework using Artificial Neural Network (ANN) technique. Three main phases of the proposed framework are risk management phase, ANN training phase and framework application phase. Thereby, Risk Factors are identified and analysed using Failure Mode and Effect Analysis (FMEA) technique. ANN model is created and trained to evaluate the impact of Risk Factors on Project Risk which is represented through the ratio of contractor’s profit to project costs. As a result, the framework with successful model is used as a tool to support the construction company in assessing risk and evaluate their impact on the project’s profit for new projects.
 Keywords: risk management; risk assessment; Artificial Neural Network (ANN); Failure Mode and Effect Analysis (FMEA); construction project.

Spatially continuous probabilistic prediction of sparsely measured ground properties constrained by ill-posed tomographic imaging considering data uncertainty and resolution

Probabilistic prediction of 2D or 3D distributions of sparsely measured borehole or direct-push logging data can contribute to solving hydrological, petroleum, or engineering exploration tasks. We use and improve a recently developed workflow constrained by ill-posed geophysical tomography to achieve 2D probabilistic predictions of geotechnical exploration target parameters that could only be measured by 1D borehole or direct-push logging. We use artificial neural networks (ANNs) to find the optimal prediction models between ensembles of equivalent geophysical tomograms and sparsely measured logging data. During the training phase of ANNs, we consider four different training strategies taking into account the logging data uncertainty and geophysical tomographic ambiguity to avoid data overfitting of the ANNs. Thus, we successfully transform the logging data uncertainty and geophysical tomographic reconstruction ambiguity as well as differences in spatial resolution of logging and tomographic models into the probabilistic 2D prediction of our target parameters in a data-driven manner, which allows application of our methodology to any combination of geophysical tomograms and hydrologic, petroleum, or engineering target parameters solely measured in boreholes. To illustrate our workflow, we use an available field data set collected at a field site south of Berlin, Germany, to characterize near-subsurface sedimentary deposits. In this example, we employ cross-borehole tomographic radar-wave velocity, P-wave velocity, and S-wave velocity models to constrain the prediction of tip resistance, sleeve friction, and dielectric permittivity as target parameters.

Density prediction for petroleum and derivatives by gamma-ray attenuation and artificial neural networks

This work presents a new methodology for density prediction of petroleum and derivatives for products' monitoring application. The approach is based on pulse height distribution pattern recognition by means of an artificial neural network (ANN). The detection system uses appropriate broad beam geometry, comprised of a 137Cs gamma-ray source and a NaI(Tl) detector diametrically positioned on the other side of the pipe in order measure the transmitted beam. Theoretical models for different materials have been developed using MCNP-X code, which was also used to provide training, test and validation data for the ANN. 88 simulations have been carried out, with density ranging from 0.55 to 1.26gcm−3 in order to cover the most practical situations. Validation tests have included different patterns from those used in the ANN training phase. The results show that the proposed approach may be successfully applied for prediction of density for these types of materials. The density can be automatically predicted without a prior knowledge of the actual material composition.

Multilayer Perceptron Neural Network in a Data Assimilation Scenario

Multilayer Perceptron Neural Network (MLP-NN) have been successfully applied to solve nonlinear problems in meteorology and oceanography. In this work, MLP-NN is applied to completely emulate an Extended Kalman Filter (EKF) in a data assimilation scenario. Data assimilation is a process for producing a good combination of data from observations and data from a mathematical model. This is a fundamental issue in an operational prediction system. The one-dimensional shallow water equation DYNAMO-1D is employed here for testing the assimilation schemes. The DYNAMO model is derived from depth-integrating the Navier-Stokes equations, in the case where the horizontal length scale is much greater than the vertical length scale, where the Coriolis force is also considered in atmospheric flows. Techniques, such as Extend Kalman Filter, are available to track non-linear dynamical models under certain conditions. Under strong non-linearity, the fourth-order moment EKF works well when applied to high-dimensional state space for data assimilation, but the computational burden is a barrier in this kind of application. Artificial Neural Network (ANN) is an alternative solution for this computational complexity problem, once the ANN is trained offline with a high order Kalman filter, even though this Kalman filter has high computational cost (which is not a problem during ANN training phase). The results achieved in this research encourage us to apply this technique on operational models. However, it is not yet possible to assure convergence in high dimensional problems.

A General Multisegment Artificial Neural Network Architecture for the Efficient Evaluation of Electromagnetic Plane-Wave Wedge Diffraction

A multisegment artificial neural network (ANN) is proposed as an interpolation technique for the evaluation of the electromagnetic field diffracted at the edge of anisotropic impedance wedges under plane wave illumination at oblique incidence. Multisegmentation is needed as the high-frequency wedge diffracted field is characterized by a number of discontinuities at the shadow boundaries of the geometrical optics and surface wave fields. The proposed approach is applied, as a test case, to the problem of an anisotropic impedance right-angled wedge illuminated by a skewly incident plane wave. Some exact analytical solutions valid for specific surface impedance tensors are used to obtain numerical data for the ANN training phase as well as to show the interpolation capabilities of the implemented ANN. Nevertheless, the proposed ANN structure is general and can be trained with data obtained from other available solutions (analytical, perturbative, numerical) valid for more general wedge configurations, eventually leading to a single software tool encompassing all of them and providing accurate approximations of the wedge diffracted field in a relatively short time, comparable to that of a closed form analytical solution.

New approach to applying neural network in nonlinear dynamic model

In this work, radial basis function neural network (RBF-NN) is applied to emulate an extended Kalman filter (EKF) in a data assimilation scenario. The dynamical model studied here is based on the one-dimensional shallow water equation DYNAMO-1D. This code is simple when compared with an operational primitive equation models for numerical weather prediction. Although simple, the DYNAMO-1D is rich for representing some atmospheric motions, such as Rossby and gravity waves. It has been shown in the literature that the ability of the EKF to track nonlinear models depends on the frequency and accuracy of the observations and model errors. In some cases, just fourth-order moment EKF works well, but will be unwieldy when applied to high-dimensional state space. Artificial Neural Network (ANN) is an alternative solution for this computational complexity problem, once the ANN is trained offline with a high order Kalman filter, even though this Kalman filter has high computational cost (which is not a problem during ANN training phase). The results achieved in this work encourage us to apply this technique on operational model. However, it is not yet possible to assure convergence in high dimensional problems.

A SELECTIVE IMPROVEMENT TECHNIQUE FOR FASTENING NEURO-DYNAMIC PROGRAMMING IN WATER RESOURCE NETWORK MANAGEMENT

An approach to the integrated water resources management based on Neuro-Dynamic Programming (NDP) with an improved technique for fastening its Artificial Neural Network (ANN) training phase will be presented. When dealing with networks of water resources, Stochastic Dynamic Programming provides an effective solution methodology but suffers from the so-called “curse of dimensionality”, that rapidly leads to the problem intractability. NDP can sensibly mitigate this drawback by approximating the solution with ANNs. However in the real world applications NDP shows to be considerably slowed just by this ANN training phase. To overcome this limit a new training architecture (SIEVE: Selective Improvement by Evolutionary Variance Extinction) has been developed. In this paper this new approach is theoretically introduced and some preliminary results obtained on a real world case study are presented.

Neural net based determination of generator-shedding requirements in electric power systems

This paper presents an application of artificial neural networks (ANN) in support of a decision-making process by power system operators directed towards the fast stabilisation of multi-machine systems. The proposed approach considers generator shedding as the most effective discrete supplementary control for improving the dynamic performance of faulted power systems and preventing instabilities. The sensitivity of the transient energy function (TEF) with respect to changes in the amount of dropped generation is used during the training phase of ANNs to assess the critical amount of generator shedding required to prevent the loss of synchronism. The learning capabilities of neural nets are used to establish complex mappings between fault information and the amount of generation to be shed, suggesting it as the control signal to the power system operator.