Sensitive Input Variables Research Articles

An integrated approach of artificial neural networks and polynomial chaos expansion for prediction and analysis of yield and environmental impact of oil shale retorting process under uncertainty

Shale oil reserves exploration is getting huge investments due to the depletion of conventional reserves. Computational methods have been used to realize optimum design and operation of shale oil and gas reserves exploration and processing. The uncertainty associated with the composition of shale reserves and operating conditions during processing put a challenge to the realization of high yield and mitigation of environmental impact. In the current work, machine learning (ML) based models are proposed for the estimation of the yield and environmental impact of the oil shale retorting process under uncertainty. An artificial uncertainty of 1% was inserted in feed composition and process conditions of an Aspen model of the process to generate data for the development of the ML models. Artificial Neural Network (ANN), Least Square Boosting (LSB), and Bagging techniques were compared to find the best ML model. ANN models, with the highest correlation coefficient of 0.995 and 0.999 for the oil yield and CO2 content respectively, are used as surrogates in a Polynomial Chaos Expansion (PCE) framework for the uncertainty analysis of the process. For 1% uncertainty in feed composition and process conditions, a mean absolute deviation of 0.319 and 0.580 was obtained for the oil yield and carbon dioxide emissions respectively. To find the hierarchy in the process inputs in terms of their effect on the oil yield and carbon dioxide emissions, the ANN model is used as a surrogate in sensitivity analysis through Sobol and Fourier Amplitude Sensitivity Test (FAST) indices. The most sensitive input variables were feed temperature and air molar flow rate. The proposed modeling framework will provide a base for future real-time monitoring and analysis of the oil shale retorting processes.

Feature selection and Gaussian process prediction of rougher copper recovery

The emergence of advanced computational techniques provides opportunity in better modelling complex, nonlinear relationship of industrial processes. In this work, the performance of Gaussian Process Regression (GPR) models in predicting rougher copper recovery, following development with and without model-selected, relevant rougher flotation variables has been investigated. Randomly selected training and validation data sets as well as an independent testing data set (later than the original training and validation sets) were used to demonstrate the model predictive behaviour. The results showed that exponential GPR covariance function, modelled with selected relevant rougher flotation variables (throughput, feed particle size, xanthate dosage, frother dosage and froth depth), was the best model yielding correlation coefficient (r), root mean square error (RMSE), normalised root mean square error (NRMSE), variance accounted for (VAF) and performance index (PI) values of 0.99, 0.10, 0.01%, 99.99% and 1.97, respectively for training set. In the same manner, the validation set yielded 0.95, 0.40, 0.06%, 91.0%, and 1.75 whilst the testing set gave 0.96, 0.39, 0.07%, 92.10% and 1.78, confirming that exponential GPR algorithm can make good rougher copper recovery prediction given a set of useful rougher flotation variables. In all cases, the Regularised Neighbourhood Component Analysis algorithm performed best better in selecting useful flotation variables for rougher copper recovery prediction. Sobol’s global sensitivity analysis indicated feed particle size as the most sensitive input variable recording first and total sobol indices values of ~63% and ~67%, respectively. Partial dependence plots revealed the functional relationship between the top two sensitive input variables (feed particle size and froth depth of tank cell 1) and predicted rougher copper recovery, suggesting their critical operating regime.

Multi-fidelity approach for uncertainty quantification of buried pipeline response undergoing fault rupture displacements in sand

A new efficient approach for uncertainty quantification of pipeline structural response undergoing reverse-slip fault rupture displacements in sand is presented using multi-fidelity Gaussian processes. Uncertainty quantification problems generally take large numbers of scenarios to be analysed considering variations in the influencing parameters. Analysing large numbers of computationally expensive detailed geo-technical numerical models is practically not feasible. Hence, a multi-fidelity approach employing Gaussian process is proposed to tackle this problem which combines the accuracy of a relatively few number of detailed geo-technical numerical models and the efficiency of large numbers of simplified numerical models to track the uncertainty response. The detailed model utilizes a previously validated pipe-soil finite element (FE) model including a non-linear sand constitutive model implemented in finite element software ABAQUS. The simplified model utilizes beam element pipe and bi-linear soil springs. The multi-fidelity model is first trained using data from high-fidelity and low-fidelity model analyses, thereafter cross-validated and subsequently used to quantify uncertainty in the peak compressive strains generated and to identify the most sensitive input variables. Finally, fragility curves are derived for a site specific pipe-soil fault rupture problem.

A Monte Carlo based approach for exergo-economic modeling of solar water heater

ABSTRACT In life-cycle costing of thermal energy systems, the basis of costing could be mass or exergy and the approach followed could be deterministic or stochastics. In thermal energy systems with end products/services such as hot air, hot water, steam etc. the value addition is due to higher exergy content; therefore, exergy is a logical basis of costing and stochastics is a practical approach capturing uncertainties of input variables. This paper proposes a novel framework named as stochastic Monte Carlo-based exergy costing (SMXC) for assessment of solar hot water systems. The annual hours of operation, maintenance cost, service life, and capital cost have been identified as highly sensitive input variables. The costs based on mass and exergy content of hot water in deterministic life-cycle costing method are estimated at 0.296 and 0.304 US cent/kg, respectively. The mean values of mass and exergy costs of hot water using Monte Carlo-based stochastics life-cycle costing method are 0.302 and 0.310 US cent/kg. A very low value (i.e. 2.4%) of the exergo-economic factor (f) for the solar water heater indicates the poor exergetic efficiency; therefore, capital investment to improve its efficiency is justified. The methodological approach can be extended to examine the probabilistic exergo-economic cost of array of thermal energy products when the parametric uncertainties play a key role.

Artificial neural network model to predict transport parameters of reactive solutes from basic soil properties

Measurement of solute-transport parameters through soils for a wide range of solute- and soil-types is time-consuming, laborious, expensive and practically impossible. So, indirect methods for estimating the transport parameters by pedo-transfer functions are now advancing. This study developed and evaluated an Artificial Neural Network (ANN) model for estimating the transport velocity (V), dispersion coefficient (D) and retardation factor (R) of NaAsO2, Pb(NO3)2, Cd(NO3)2, C9H9N3O2 and CaCl2 from the basic soil properties. Breakthrough data of the solutes were measured in 14 agricultural soils of Bangladesh by time-domain reflectometry (TDR) in repacked soil columns under unsaturated steady-state water-flow conditions. The transport parameters of the chemicals were determined by analyzing the solute breakthrough data. Bulk density (γ), organic carbon (OC), clay (C) content, pH, median grain diameter (D50) and uniformity coefficient (Cu) of the soils were determined. An ANN model for V, D and R was developed by using data of eight soils, validated/tested with the data of five soils and verified with the data of one soil. Clay content and bulk density of the soils were the most sensitive input variables to the ANN model followed by other soil properties (OC, C, pH, D50 and Cu). The model reliably predicted V, D and R with relative root-mean-square error (RRMSE) of 0.028–0.363, mean error (ME) of – 0.00004 to 0.0005, bias error (BOE%) of 0–0.003 and modeling efficiency (EF) of >0.99. Thus, the ANN model can significantly enhance prediction of pollution transport through soils in terms of cost and effort.

A newly developed method to extract the optimal hyperspectral feature for monitoring leaf biomass in wheat

Hyperspectral image provides a plethora of information, some of which may be redundant. Therefore, reducing information’s dimensionality using feature selection methods becomes essential. Here, we proposed a new technique, named synergy interval partial least squares (SIPLS) with successive projections algorithm (SPA) (SIPLS-SPA), combining SIPLS and SPA, to efficiently extract optimal spectral features of wheat biomass from hyperspectral image data. In this study, hyperspectral images and leaf biomass were acquired from two-year wheat field experiments with varied nitrogen rates, planting densities, and cultivars. The results showed that eight wavelengths (706, 724, 734, 806, 808, 810, 812, and 816 nm) were selected as the sensitive input variables to establish partial least squares regression (PLSR) model for wheat leaf biomass. The SIPLS-SPA biomass model performed better with higher Rc2 (0.79) in calibration, lower RMSEv (0.059 kg/m2) and RRMSEv (38.55%) in validation. Compared with other state-of-the-art feature selection techniques, the SIPLS-SPA method provided significantly fewer unrelated, collinear spectral variables, and showed a promising application in terms of lower overall complexity, reduced computational complexity, and shorter running time. Furthermore, this work demonstrates the potential of SIPLS-SPA method for extracting other plant traits from hyperspectral data in future agricultural applications.

Pedo-transfer functions with multiple linear regressions to predict solute-transport parameters

Transport parameters of soluble chemicals through soils are needed to assess the pollution risks of soil and groundwater resources. But, it is time consuming, laborious, expensive and, practically, impossible to experimentally measure such parameters for a wide range of solutes and soil types. So, indirect estimate of the parameters by pedo-transfer function is becoming popular. The aim of this study was to develop and evaluate pedo-transfer functions (PTFs) for solute-transport parameters by multiple linear regression (MLR) analysis. For this, transport parameters of three heavy metal /metalloid compounds (NaAsO 2 , Pb(NO 3 ) 2 , Cd(NO 3 ) 2 ), a pesticide (carbendazim) and an inert salt (CaCl 2 ) through 14 agricultural soils of Bangladesh were determined. The transport experiments were done in repacked soil columns under unsaturated steady-state water flow conditions. Breakthrough data of the solutes were measured with time-domain reflectometry (TDR), and velocity ( V ), dispersion coefficient ( D ) and retardation factor ( R ) of the solutes were determined by analyzing the data by a transfer-function method. Bulk density ( g ), organic carbon ( OC ) content, clay ( C ) content, pH, median grain diameter ( D 50 ) and uniformity coefficient ( C u ) of the soils were determined. Regression models for V , D and R were developed with g , OC , C , pH, D 50 and C u as the input variables. Bulk density and clay content were found the most sensitive input variables to the MLR models. The MLR models fairly predicted V , D and R , and thus provide a way of significantly enhancing prediction of reactive solute transport through agricultural soils.

Eco-efficiency of centralized wastewater treatment plants in industrial parks: A slack-based data envelopment analysis

China has more than 2500 industrial parks that are above the provincial level and contribute more than half of the GDP of China. Centralized wastewater treatment plants (CWWTPs) are one of the most important environmental infrastructures of these industrial parks, and their performance and efficiency usually directly influences the pollutants discharged and the surrounding water environment. Data envelopment analysis (DEA) is a method that has been increasingly popular for assessing the efficiency of wastewater and wastewater treatment plants (WWTPs) in recent decades. In this study, a slack-based DEA (SBM-DEA) model, which includes five input variables and four output variables, is applied to assess the eco-efficiency of 281 CWWTPs in 126 national-level industrial parks (NIPs). Sensitivity analysis is used to identify the most sensitive input and output variables. Next, a Pearson correlation analysis is used to evaluate the correlations between the eco-efficiency factors and some implicit factors, which are not selected as input and output variables but may have an underlying impact on the eco-efficiency of CWWTPs. Then, the uncertainty and limitations of the DEA method, in terms of its ability to assess the efficiency of CWWTPs, are discussed. Finally, the conclusion is presented and some policy suggestions are proposed to improve the eco-efficiency of CWWTPs in NIPs. The key findings will have referential significance of improving eco-efficiencies of industrial parks around the world.

Automated Advanced Calibration and Optimization of Thermochemical Models Applied to Biomass Gasification and Pyrolysis

This paper presents a methodology that combines physicochemical modeling with advanced statistical analysis algorithms as an efficient workflow, which is then applied to the optimization and design of biomass pyrolysis and gasification processes. The goal was to develop an automated flexible approach for the analyses and optimization of such processes. The approach presented here can also be directly applied to other biomass conversion processes and, in general, to all those processes for which a parametrized model is available. A flexible physicochemical model of the process is initially formulated. Within this model, a hierarchy of sensitive model parameters and input variables (process conditions) is identified, which are then automatically adjusted to calibrate the model and to optimize the process. Through the numerical solution of the underlying mathematical model of the process, we can understand how species concentrations and the thermodynamic conditions within the reactor evolve for the two proce...

Unsupervised load shape clustering for urban building performance assessment

This paper presents a method to automatically cluster typical days of energy consumption in one or several buildings. The method is based on an optimized version of the Symbolic Aggregate approXimation (SAX) method. SAX is a data mining technique for clustering time series with recent applications in building fault detection and building performance assessment. The number of clusters and accuracy of SAX highly depends on two highly sensitive input variables, i.e., the word size and the alphabet size. We propose the use of the genetic algorithm NSGA-II to optimize the number of words and alphabet size of SAX subjected to three fitness objectives, i.e., maximize data accuracy and compression and minimize complexity. In addition, we propose the use of MAVT as selection method of the optimal solution. The methodology is applied to measured energy consumption data of three representative buildings on a university campus in Singapore. Potential future uses of the approach include advanced studies in fault detection and calibration of urban building performance models.

An experimental design-based evaluation of sensitivities of MEPDG prediction: investigating main and interaction effects

The mechanistic–empirical pavement design guide (MEPDG) uses mechanistic–empirical models by analysing the impacts of traffic, climate, materials and pavement structure to predict performances of pavement. The MEPDG software uses a three-level hierarchical input to predict performance in terms of terminal International Roughness Index, permanent deformation, total cracking (reflective and alligator), asphalt concrete (AC) thermal fracture, AC bottom-up fatigue cracking and AC top-down fatigue cracking. However, these inputs with different levels of accuracy may have significant impact on performance prediction. This study focuses on the sensitivity of the inputs of MEPDG distresses to identify the effect of the accuracy level of inputs based on experimental design. A local sensitivity analysis is carried out to identify the main effect of inputs considering them as independent variables. Interaction effects are also analysed based on random combination of the inputs. Sensitive input variables and their combinations are evaluated through a multiple regression analysis for respective distresses.

A hybrid evolutionary data driven model for river water quality early warning

China's fast pace industrialization and growing population has led to several accidental surface water pollution events in the last decades. The government of China, after the 2005 Songhua River incident, has pushed for the development of early warning systems (EWS) for drinking water source protection. However, there are still many weaknesses in EWS in China such as the lack of pollution monitoring and advanced water quality prediction models. The application of Data Driven Models (DDM) such as Artificial Neural Networks (ANN) has acquired recent attention as an alternative to physical models. For a case study in a south industrial city in China, a DDM based on genetic algorithm (GA) and ANN was tested to increase the response time of the city's EWS. The GA-ANN model was used to predict NH3–N, CODmn and TOC variables at station B 2 h ahead of time while showing the most sensitive input variables available at station A, 12 km upstream. For NH3–N, the most sensitive input variables were TOC, CODmn, TP, NH3–N and Turbidity with model performance giving a mean square error (MSE) of 0.0033, mean percent error (MPE) of 6% and regression (R) of 92%. For COD, the most sensitive input variables were Turbidity and CODmn with model performance giving a MSE of 0.201, MPE of 5% and R of 0.87. For TOC, the most sensitive input variables were Turbidity and CODmn with model performance giving a MSE of 0.101, MPE of 2% and R of 0.94. In addition, the GA-ANN model performed better for 8 h ahead of time. For future studies, the use of a GA-ANN modelling technique can be very useful for water quality prediction in Chinese monitoring stations which already measure and have immediately available water quality data.

Calibration of HDM-4 Emission Models for Indian Conditions

The Highway Development and Management-4 (HDM-4) tool has been widely used for the pavement management activities across the world. This tool has inbuilt pavement performance prediction models, vehicular performance models and economic analysis tools. Exhaust emissions are one of the important outputs of vehicular performance models that are helpful in assessing viability of investment options and environment impact assessment activities. There are seven exhaust emission models (for different components like hydro carbon, carbon monoxide, particulate emissions etc.) available within HDM-4. These models are required to be calibrated so that the predictions made by calibrated HDM-4 models represent the specific local ground conditions. The work presented here is an attempt to calibrate the HDM-4 emission models to Indian conditions. Initially sensitivity analysis of emission models was conducted to find sensitive input variables in emission model that affect model output significantly. It was found that operating weight, pavement gradient and vehicle life are very sensitive inputs into HDM-4 emission models. Based on the sensitivity analysis and data obtained from a previous study were used in calibration of emission models for Indian conditions. Further these calibrated emission models were used to predict emissions for urban conditions prevailing in India. Comparison of predicted and measured values indicate that all emission models for two lane road and Carbon Monoxide emission model for four lane road over-predicts for two wheelers, car, light commercial vehicles and busses, while under-predicting for trucks.

Sensitivity Analysis of a Standalone Photovoltaic System Model Parameters

The values of input variables of any model are cause to undergo changes due to influence of environmental conditions. These changes can be investigated by conducting the sensitivity analysis of input variables with respect to output variables. Sensitivity analysis increases the validity, credibility and assurance of model estimates. The purpose of this study was to identify the most important and sensitive input variables and to prioritize the parameters based on their influence on the model outputs of a standalone photovoltaic (SAPV) system. For that, a normalized local sensitivity analysis and sensitivity index of seven input variables of a SAPV system model with reference to three output parameters namely amount of absorbed solar radiation, maximum photovoltaic (PV) module power output and optimum PV array area has been carried out. It was revealed from the analysis that the most important and sensitive input variable was the amount of total solar radiation and the least important variable was solar azimuth angle and the lowest sensitive variable was wind speed.

Identification of bottlenecks to improve equipment availability: a case study

In this highly competitive market where the demand for the products is increasing gradually, it is essential for a manufacturing company to reduce the production cycle time and cost accordingly. One of the methods for this is to increase the availability of process. This paper attempts to identify and quantify the causes of machine downtime in a manufacturing company. For this a four-prong approach is taken: Firstly, the major causes for the low availability of process are determined, where bearing failure is found to be the major causes of low availability. Second, combining desirability function approach and data envelopment analysis are performed to identify the sensitive input variables that affect the bearing failure. Thirdly, ordinary linear regression is used to validate the results of DEA. Lastly, some managerial implications to reduce the downtime were suggested. We have considered an electrolytic tinplating line process in a tin sheet manufacturing company as a case study.

Prediction of moisture content in pre-osmosed and ultrasounded dried banana using genetic algorithm and neural network

In this study, application of a versatile approach for estimation moisture content of dried banana using neural network and genetic algorithm has been presented. The banana samples were dehydrated using two non-thermal processes namely osmotic and ultrasound pretreatments, at different solution concentrations and dehydration times and were then subjected to air drying at 60 and 80 °C for 4, 5 and 6 h. The processing conditions were considered as inputs of neural network to predict final moisture content of banana. Network structure and learning parameters were optimized using genetic algorithm. It was found that the designed networks containing 7 and 10 neurons in first and second hidden layers, respectively, give the best fitting to experimental data. This configuration could predict moisture content of dried banana with correlation coefficient of 0.94. In addition, sensitivity analysis showed that the two most sensitive input variables towards such prediction were drying time and temperature.

Predicting project performance through neural networks

Successful project delivery of construction projects depends on many factors. With regard to the construction of a facility, selecting a competent contractor for the job is paramount. As such, various approaches have been advanced to facilitate tender award decisions. Essentially, this type of decision involves the prediction of a bidder’s performance based on information available at the tender stage. A neural network based prediction model was developed and presented in this paper. Project data for the study were obtained from the Hong Kong Housing Department. Information from the tender reports was used as input variables and performance records of the successful bidder during construction were used as output variables. It was found that the networks for the prediction of performance scores for Works gave the highest hit rate. In addition, the two most sensitive input variables toward such prediction are “Difference between Estimate” and “Difference between the next closest bid”. Both input variables are price related, thus suggesting the importance of tender sufficiency for the assurance of quality production.

Wavelet-Based Regularization of Dynamic Data Reconciliation

Dynamic data reconciliation can supply more accurate data for dynamic optimization, dynamic fault diagnosis, and control by means of incorporating process information in some mathematical model. It will be an ill-posed inverse problem if the sensitive input variables are unmeasured; here, the sensitive input variable is defined as the variable that, if it is unmeasured, can only be estimated through the differentiation of other measured variables. In such a case, existing methods cannot obtain correct and usable data effectively. To address the problem, based on the principle of regularization, the wavelets are adopted to construct regular operators. And, a new approach is proposed to determine the optimal scale level corresponding to the optimal approximate operator in which the prior statistical information of the signal is utilized. The algorithm can deal with the estimation of unknown sensitive input variable effectively. The results show that more accurate estimation of the sensitive input variable can be obtained by using the proposed method as compared with the one obtained by using existing collocation methods based on polynomials.