Neural Network Sensitivity Analysis Research Articles

Artificial neural network modeling and sensitivity analysis for soiling effects on photovoltaic panels in Morocco

In the present work, an Artificial Neural Network (ANN) methodology for studying and modeling the soiling effect on solar photovoltaic (PV) glass is presented. To perform the study, a solar PV glazing was exposed outdoor at the home solar energy platform of Physic of Semi-conductors and Solar Energy research structure (PSES) at Mohammed V University in Rabat, Morocco. Regular measurements from April 20, to December 31, 2016, were carried out to monitor the soiling rate changes over time. Meteorological data were used as input variables for ANN modeling. The model performance was evaluated using a statistical comparison between experimental and simulated values. Results show that the implementation of Levenberg-Marquardt backpropagation algorithm, and the active functions Tansig, and Purline achieve the best estimations (R2 = 0.928) in an ANN architecture 6-35-1. Additionally, a sensitivity analysis approach was employed to determine the effect of input parameters on model output and the behavior of the model with the variation of each input parameter. Sensitivity analysis results indicate that the most influential parameter for PV soiling rate was the relative humidity, followed by wind direction. The ANN model coupled with sensitivity analysis show be a promising framework for its application in smart sensors on cleaning systems for PV modules to improve their operational efficiency.

A Sensitivity Analysis Method for Finite Element Simulation of CFRP Drilling Process

This paper developed a sensitivity analysis method for FE simulation of CFRP drilling process to rank the parameters in order of importance. Firstly, a FE model of CFRP is established using ABAQUS/Explicit. A standard twist drill is imported from stp-formatted files, then it is meshed using C3D4 element and treated as rigid body. The CFRP is modeled as a shell and 2D Hashin Criteria is used to represented the damage. Secondly, a back-propagation artificial neural network is established and trained with the data generated by the FE models using different parameters, which is designed by uniform design method. Lastly, the sensitivity analysis of the back-propagation artificial neural network is carried out using analytical method. The sensitivity of the parameters in the FE model is calculated using a MATLAB program.

Application and Sensitivity Analysis of Artificial Neural Network for Prediction of Chemical Oxygen Demand

Integrating water quality forecasting model with river restoration techniques makes river restoration more effective and efficient. This research investigates how to use the Artificial Neural Network (ANN) to predict the Chemical Oxygen Demand (COD) during river restoration in Wuxi city, China. Specifically, we applied a Multi-Layer Perceptron (MLP) using ten neurons in a single hidden layer and seven input variables (Temperature, Dissolved Oxygen, Total Nitrogen, Total Phosphorus, Suspended Sediment, Transparency, and NH3-N) to simulate COD. The modeled results have a correlation coefficient of 0.966, 0.949, and 0.890 with the observations for the raining, validation, and testing phases, respectively. When presenting the trained network to an independent data set, the ANN model still shows a good predictive capability, indicating by a correlation coefficient of 0.978, a root mean square error (RMSE) of 0.628 mg/L, and a mean square error (MSE) of 0.394 mg2/L2. A sensitivity analysis was further implemented to analyze the effect of each of the input variables on prediction of COD. DO, TO, and Transparency have relatively low influences on the estimate of COD, and can be removed from the input variables. The results from this study indicate that ANN models can provide satisfactory estimates of COD during the process of bacterial treatment and is a useful supportive tool for river restoration.

Integrating artificial neural network and imperialist competitive algorithm (ICA), to predict the energy consumption for land leveling

PurposeThis work aims to determine the best linear model using an artificial neural network (ANN) with the imperialist competitive algorithm (ICA-ANN) and ANN to predict the energy consumption for land leveling.Design/methodology/approachUsing ANN, integrating artificial neural network and imperialist competitive algorithm (ICA-ANN) and sensitivity analysis (SA) can lead to a noticeable improvement in the environment. In this research, effects of various soil properties such as embankment volume, soil compressibility factor, specific gravity, moisture content, slope, sand per cent and soil swelling index on energy consumption were investigated.FindingsAccording to the results, 10-8-3-1, 10-8-2-5-1, 10-5-8-10-1 and 10-6-4-1 multilayer perceptron network structures were chosen as the best arrangements and were trained using the Levenberg–Marquardt method as the network training function. Sensitivity analysis revealed that only three variables, namely, density, soil compressibility factor and cut-fill volume (V), had the highest sensitivity on the output parameters, including labor energy, fuel energy, total machinery cost and total machinery energy. Based on the results, ICA-ANN had a better performance in the prediction of output parameters in comparison with conventional methods such as ANN or particle swarm optimization (PSO)-ANN. Statistical factors of root mean square error (RMSE) and correlation coefficient (R2) illustrate the superiority of ICA-ANN over other methods by values of about 0.02 and 0.99, respectively.Originality/valueA limited number of research studies related to energy consumption in land leveling have been done on energy as a function of volume of excavation and embankment. However, in this research, energy and cost of land leveling are shown to be functions of all the properties of the land, including the slope, coefficient of swelling, density of the soil, soil moisture and special weight dirt. Therefore, the authors believe that this paper contains new and significant information adequate for justifying publication in an international journal.

A neural network sensitivity analysis in the presence of random fluctuations

In this work we investigate the sensitivity analysis of the noise applied to a neural network. The goal of this study is to try to understand the network's behavior and the outputs sensitivity from the perturbation of weights during the training process. From an engineering point of view, noise is perceived as detrimental to the system and the quality of the output, but in biological neural systems, we can observe noise fluctuations, which possess certain abilities to improve information processing. By means of sensitivity analysis tools we show quantitatively the acceptable level of noise which provides optimal solution to the random fluctuations without sacrificing the behavior of the network. The three different indicators utilized in this endeavor allow us to observe whether the noise variance is detrimental or beneficial, and whether it acts as a source of fluctuation.

LS-SVM data mining analysis: how does biochar influence soil net nitrogen mineralization in the field?

Biochar has been considered as a stable-carbon source for improving soil quality and long-term sequestration of carbon. However, in view of ecological environmental feedback and the tightly coupled system of carbon-nitrogen cycling, further attention has shifted to the effect of biochar on soil net nitrogen mineralization (SNNM). Recently, ecological evaluations of biochar were mostly based on laboratory incubation or pot experiments, ignoring external and uncontrollable natural factors. Therefore, the essential characteristics of local environments were not accurately described. In this paper, a nonlinear stochastic model of SNNM based on least squares support vector machine (LS-SVM) was set up to study the effect of biochar on nitrogen cycling in a field experiment. In order to explore this effect in natural conditions, partial derivative (PaD) sensitivity analysis of LS-SVM was firstly proposed, evaluated by the data from a known equation, and then applied to open the “black-box” stochastic model of SNNM. Comparing with the sensitivity analysis of artificial neural networks (ANNs), the RD values of LS-SVM PaD1 algorithm were almost the same as those of ANNs PaD1 algorithm. However, the RSD values of LS-SVM PaD2 algorithm were closer to the given equation. In the SNNM model, RD values of LS-SVM PaD1 algorithm of initial nitrogen, time, and precipitation were 21, 15, and 14 %, and the biochar RD value was only 0.51 %, implying that biochar did not influence SNNM directly. However, the cumulative RSD of the PaD2 algorithm of biochar with the other factors was 15.05 %, the maximum of the interactions, implying that it could greatly enhance the tendency for SNNM by interacting with other factors. PaD sensitivity analysis of LS-SVM was a stable and reliable data mining method. In the SNNM model, initial nitrogen, time, and precipitation were the main controlling factors of the SNNM model. Biochar did not directly influence SNNM; however, it could greatly enhance the tendency for SNNM by interactions with other factors by decreasing the inhibitory effect of initial nitrogen on SNNM and modifying soil condition to change the effect of other factors on SNNM.

Improving management of windrow composting systems by modeling runoff water quality dynamics using recurrent neural network

The recurrent neural network is a tool that can provide valuable insights when forecasting future likelihood of events using dynamic time series. One of the challenging research problems is to extend the black-box modeling into white-box modeling in order to gain insights into the physical processes. Sensitivity analysis has shown a great contribution in overcoming this challenge. The main objective of this study was to perform a detailed sensitivity analysis of recurrent neural network in order to identify parameters that are important for predicting water quality constituents.We used a windrow composting pad located at UGA Bioconversion center, Athens, GA, USA as our study site. Runoff from windrow composting pad was collected in order to prevent the discharge of organic pollutants. We used time series data from nine years of precipitation, temperature, pond volume, material volume on the pad, total suspended solids (TSS), biological oxygen demand (BOD) and nitrate (NO3) concentration levels of stored runoff in the collection pond.Previously, we applied recurrent neural network for predicting TSS, BOD and NO3 as well as performed auto-correlation and cross-correlation analysis (Shim and Tollner, 2014). We used first eight years of data (from January 2001 to December 2008) to build the model and last year of data (from January 2009 to December 2009) to evaluate the model. Within this paper, we showed that the detailed sensitivity analysis of recurrent neural network can allow a better understanding of water quality dynamics of collected runoff and assist in identifying strategies for better management of windrow composting systems.

Optimal Wavelength Selection in Ultraviolet Spectroscopy for the Estimation of Toxin Reduction Ratio during Hemodialysis

Introduction The concentration of substances, including urea, creatinine, and uric acid, can be used as an index to measure toxic uremic solutes in the blood during dialysis and interdialytic intervals. The on-line monitoring of toxin concentration allows for the clearance measurement of some low-molecular-weight solutes at any time during hemodialysis.The aim of this study was to determine the optimal wavelength for estimating the changes in urea, creatinine, and uric acid in dialysate, using ultraviolet (UV) spectroscopy. Materials and Methods In this study, nine uremic patients were investigated, using on-line spectrophotometry. The on-line absorption measurements (UV radiation) were performed with a spectrophotometer module, connected to the fluid outlet of the dialysis machine. Dialysate samples were obtained and analyzed, using standard biochemical methods. Optimal wavelengths for both creatinine and uric acid were selected by using a combination of genetic algorithms (GAs), i.e., GA-partial least squares (GA-PLS) and interval partial least squares (iPLS). Results The Artifitial Neural Network (ANN) sensitivity analysis determined the wavelengths of the UV band most suitable for estimating the concentration of creatinine and uric acid. The two optimal wavelengths were 242 and 252 nm for creatinine and 295 and 298 nm for uric acid. Conclusion It can be concluded that the reduction ratio of creatinine and uric acid (dialysis efficiency) could be continuously monitored during hemodialysis by UV spectroscopy.Compared to the conventional method, which is particularly sensitive to the sampling technique and involves post-dialysis blood sampling, iterative measurements throughout the dialysis session can yield more reliable data.

The identification of the relationship between chemical and electrical parameters of honeys using artificial neural networks

A number of significant scientific studies have confirmed the health benefits of honey. Due to the high price of natural honey, it is a common target for adulteration which reduces its medicinal value. Adulteration detection methods require specific laboratory equipment and are very expensive. The development of measurement techniques enables the measurement of electrical characteristics of strained honey. Honey electrical parameters can possibly be used for its quality assessment. The identification of the relationship between chemical and electrical parameters of honeys and analysis to determine if there are frequency-dependent changes, can help in developing of that group of methods. The aim of this research was to determine how the chemical parameters of certain honeys influence the dielectric loss factor and the permittivity of strained honey measured in various frequencies. Another aim was to determine whether the percentage influence of certain chemical parameters of honeys on electrical characteristics significantly depends on frequency value. The research was based on neural network models and sensitivity analysis. The percentage influence of certain chemical parameters on electrical characteristics significantly depends on frequency value.

Efficiency of Neural Networks for Estimating the Patch Load Resistance of Plate Girders with a Focus on Uncertainties in Material and Geometrical Properties

In this paper, a sensitivity analysis of artificial neural networks (NNs) is presented and employed for estimating the patch load resistance of plate girders subjected to patch loading. To evaluate the accuracy of the proposed NN model, the results are compared with the previously proposed empirical models, so that we can estimate the resistance of plate girders subjected to patch loading. The empirical models are calibrated, for improving the formulae, with experimental data set which was collected from the corresponding literature. NNs models are later trained and validated through using the existing experimental data. In this process several NNs architectures are taken into account. A set of good NNs models are selected and then analyzed regarding their robustness when confronted with the test data set and regarding their ability to reproduce the effect of uncertainty on the data. A sensitivity analysis is conducted herein in order to investigate the effect of variability in material and geometrical properties of plate girders. Thereafter, several estimates measuring the efficiency and the quality of the NN model and the calibrated models are obtained and discussed.

Evaluation of Winter Maintenance Chemicals and Crashes with an Artificial Neural Network

The objective of this study was to investigate and evaluate the effects of winter maintenance chemicals on road safety. To this end, a winter chemical usage model was developed. A methodology combining artificial neural network (ANN) methods and sensitivity analysis is presented. In this method, the ANN system was used to establish the relationship between road safety and associated factors, and sensitivity analysis was used to identify significant variables and quantify their effects. The chemical usage model and the ANN system were then applied to case studies in Idaho. The results from the case study showed that the use of winter maintenance chemicals played an important role in road safety. Furthermore, benefit–cost analysis of the case studies indicated that the use of winter maintenance chemicals produced more benefits than costs.

An artificial neural network approach and sensitivity analysis in predicting skeletal muscle forces.

This paper presents the use of an artificial neural network (NN) approach for predicting the muscle forces around the elbow joint. The main goal was to create an artificial NN which could predict the musculotendon forces for any general muscle without significant errors. The input parameters for the network were morphological and anatomical musculotendon parameters, plus an activation level experimentally measured during a flexion/extension movement in the elbow. The muscle forces calculated by the 'Virtual Muscle System' provide the output. The cross-correlation coefficient expressing the ability of an artificial NN to predict the "true" force was in the range 0.97-0.98. A sensitivity analysis was used to eliminate the less sensitive inputs, and the final number of inputs for a sufficient prediction was nine. A variant of an artificial NN for a single specific muscle was also studied. The artificial NN for one specific muscle gives better results than a network for general muscles. This method is a good alternative to other approaches to calculation of muscle force.

Sensitivity Analysis of Training Neural Networks by Orthogonal Weight Functions and its Application in Intrusion Detection

The analysis of statistical sensitivity for training neural networks using a new kind of orthogonal weight functions (OWFs) is discussed in this paper. The weights obtained after training are orthogonal functions defined on the sets of input variables (input patterns). We design a kind of classifier for intrusion detection. By extracting some parameters using the sensitivity formula of OWFs neural networks given in this paper, the test data for intrusion detection are optimized. We show that the classifier of OWFs neural networks has the advantages of optimized architecture and high detection rate.

Design of an Ideal Grain-Refiner Alloy for Al-7Si Alloy Using Artificial Neural Networks

An ideal grain refiner has been designed for Al-7Si alloy by performing sensitivity analysis of trained artificial neural network (ANN) model. An ANNs model has been developed for solving these complex grain refinement phenomena in Al-7Si alloy. The model predictions and the analysis are well in agreement with the experimental results and existing metallurgical facts. Uncertainty in predictions helped in finding a new phenomenon at lower addition levels of grain refiner.

On reliability of neural network sensitivity analysis applied for sensor array optimization

Sensor arrays are nowadays first choice solution for low-cost and portable gas mixtures analysis systems. The key issue in construction of such systems is selection of sensors. The authors try to apply neural network sensitivity analysis for this task. The algorithm starts from huge set of sensors, which provides satisfying operation of the system, and then detects the most redundant elements, which may be removed without significant decrease of system accuracy. Eventually the small but efficient array of sensors is obtained. Authors present the method, propose some modifications and discuss problems with its application. Results of sensitivity analysis approach are compared with exhaustive search for the best set of sensors. The case study is quantitative analysis of volatile organic compounds mixtures by means of commercial, tin dioxide, TGS 800 series sensors characterized in in-house developed gas chamber.

Modeling of the carbon dioxide capture process system using machine intelligence approaches

Improving the efficiency of the carbon dioxide (CO 2) capture process requires a good understanding of the intricate relationships among parameters involved in the process. The objective of this paper is to study the relationships among the significant parameters impacting CO 2 production. An enhanced understanding of the intricate relationships among the process parameters supports prediction and optimization, thereby improving efficiency of the CO 2 capture process. Our modeling study used the 3-year operational data collected from the amine-based post combustion CO 2 capture process system at the International Test Centre (ITC) of CO 2 Capture located in Regina, Saskatchewan of Canada. This paper describes the data modeling process using the approaches of (1) neural network modeling combined with sensitivity analysis and (2) neuro-fuzzy modeling technique. The results from the two modeling processes were compared from the perspectives of predictive accuracy, inclusion of parameters, and support for explication of problem space. We conclude from the study that the neuro-fuzzy modeling technique was able to achieve higher accuracy in predicting the CO 2 production rate than the combined approach of neural network modeling and sensitivity analysis.

From neural network to neuro-fuzzy modeling: Applications to the carbon dioxide capture process

Abstract Research on improving efficiency of the amine-based post combustion carbon dioxide (CO 2 ) capture process has been ongoing during the past decade. A good understanding of the intricate relationships among parameters involved in the CO 2 capture process is important for process optimization. The objective of this study is to uncover relationships among the significant parameters impacting CO 2 production by modeling the historical real-time process data. The data were collected from the amine-based post combustion CO 2 capture process at the International Test Centre of CO 2 Capture (ITC) located in Regina, Saskatchewan of Canada. Relevant literature review and opinions from the experienced engineers of the ITC CO 2 capture plant suggested that the four parameters of reboiler heat duty, lean loading, CO 2 absorption efficiency and CO 2 production rate are the key parameters for assessing efficiency of the process. The eight process parameters that influence these four consequent or output parameters were identified as the conditional or input parameters. In this study, two artificial intelligence techniques were applied for modeling the relationships among the conditional and consequent parameters: (1) artificial neural network combined with sensitivity analysis and (2) neuro-fuzzy modeling. The results from the two modeling processes were compared, and it was observed that the neuro-fuzzy modeling technique was able to achieve on average higher accuracies than the combined approach of neural network modeling and sensitivity analysis.

Sensitivity analysis of Hopfield neural network in classifying natural RGB color space

This paper presents a study of the sensitivity analysis of the artificial Hopfield Neural Network (HNN) when segmenting natural color images. The color distinction or vision system relies on two st...

Modeling the impact of landscape types on the distribution of stream fish species

Modifications of the landscape adjoining streams perturb their local habitat and their biological diversity, but little quantitative information is available on land cover classes that influence the fish species individually. Data collected from 191 sites in the Adour–Garonne Basin (France) were analyzed to assess the effects of land cover on the distribution of fish species. A multimodel approach was carried out to predict fish species using land cover classes and to define the most important classes applying a hierarchical filtering based on artificial neural network method and sensitivity analysis. Firstly, using three single-class models, a selection of the land cover subclasses contributing the most was carried out for each fish species and each class. Secondly, multiclass models were built with all the previously selected subclasses to predict each species (n-selected subclass model). Finally, the percentages of contribution for artificial, agricultural, and forest areas obtained for the different model architectures (three class, n-selected subclass, and global multiclass models) were compared. The majority of the distribution of fish species was correctly predicted by the single-class models, and different land cover subclasses have been selected depending on the species. Using the n-selected subclass models, the predictive performances were globally better than those obtained with other multiclass models.

Sensitivity analysis of neural network parameters to improve the performance of electricity price forecasting

This paper presents a sensitivity analysis of neural network (NN) parameters to improve the performance of electricity price forecasting. The presented work is an extended version of previous works done by authors to integrate NN and similar days (SD) method for predicting electricity prices. Focus here is on sensitivity analysis of NN parameters while keeping the parameters same for SD to forecast day-ahead electricity prices in the PJM market. Sensitivity analysis of NN parameters include back-propagation learning set (BP-set), learning rate (η), momentum (α) and NN learning days (dNN). The SD parameters, i.e. time framework of SD (d=45 days) and number of selected similar price days (N=5) are kept constant for all the simulated cases. Forecasting performance is carried out by choosing two different days from each season of the year 2006 and for which, the NN parameters for the base case are considered as BP-set=500, η=0.8, α=0.1 and dNN=45 days. Sensitivity analysis has been carried out by changing the value of BP-set (500, 1000, 1500); η (0.6, 0.8, 1.0, 1.2), α (0.1, 0.2, 0.3) and dNN (15, 30, 45 and 60 days). The most favorable value of BP-set is first found out from the sensitivity analysis followed by that of η and α, and based on which the best value of dNN is determined. Sensitivity analysis results demonstrate that the best value of mean absolute percentage error (MAPE) is obtained when BP-set=500, η=0.8, α=0.1 and dNN=60 days for winter season. For spring, summer and autumn, these values are 500, 0.6, 0.1 and 45 days, respectively. MAPE, forecast mean square error and mean absolute error of reasonably small value are obtained for the PJM data, which has correlation coefficient of determination (R2) of 0.7758 between load and electricity price. Numerical results show that forecasts generated by developed NN model based on the most favorable case are accurate and efficient. Copyright © 2008 John Wiley & Sons, Ltd.