BP ANN Research Articles

APPLICATION OF MACHINE LEARNING MODELS AND GSA METHOD FOR DESIGNING STUD CONNECTORS

The design of stud connectors is aided by determining the relationship between shear strength and the input variables (number, diameter, height, tensile strength and elastic modulus of the studs, and compressive strength and elastic modulus of the concrete) that influence strength. Since strength is nonlinearly related to the influencing variables, which makes the predictions of the relevant empirical equations unreliable, the use of machine learning (ML) models is preferred. The prediction results of eight machine learning models were evaluated, including linear regression (LR1), ridge regression (RR), lasso regression (LR2), back-propagation artificial neural network (BP ANN), genetic algorithm optimized BP ANN (GA-BP ANN), extreme learning machines (ELM), random forests (RF), and support vector machines (SVM). The results show that the GA-BP ANN model is the most accurate model for prediction with a mean absolute percentage error (MAPE) of 6.17% and an R2 of 0.9599. Based on the GA-BP ANN model and the global sensitivity analysis (GSA) method, a new parameter importance analysis method was developed to compare the magnitude of the effect of different input variables on strength. It was found that stud diameter had the greatest effect on shear strength.

Identification and verification of PCDD/Fs indicators from four typical large-scale municipal solid waste incinerations with large sample size in China

Monitoring PCDD/Fs emissions from municipal solid waste incinerations (MSWIs) is of paramount importance, yet it can be time-consuming and labor-intensive. Predictive models offer an alternative approach for estimating their levels. However, robust models specific to PCDD/Fs were lacking. In this study, we collected 190 PCDD/Fs samples from 4 large-scale MSWIs in China, with the average PCDD/Fs levels and TEQ levels of 0.987 ng/m3 and 0.030 ng TEQ/m3, respectively. We developed and evaluated predictive models, including traditional statistical methods, e.g., linear regression (LR) as well as machine learning models such as back propagation-artificial neural networks (BP ANN) and random forest (RF). Correlation analysis identified 2,3,4,7,8-PeCDF, 1,2,3,6,7,8-HxCDF, 2,3,4,6,7,8-HxCDF were better indicator congeners for PCDD/Fs estimation (R2 > 0.9, p < 0.001). The predictive results favored the RF model, exhibiting a high R2 value and low root mean square error (RMSE) and mean absolute error (MAE). Additionally, the RF model showed excellent prediction ability during external validation, with low absolute relative error (ARE) of 10.9 %-12.6 % for the three indicator congeners in the normal PCDD/F TEQ levels group (<0.1 ng TEQ/m3) and slightly higher ARE values (13.8 %-17.9 %) for the high PCDD/F TEQ levels group (>0.1 ng TEQ/m3). In conclusion, our findings strongly support the RF model’s effectiveness in predicting PCDD/Fs TEQ emission from MSWIs.

Modelling crash severity outcomes for low speed urban roads using back propagation – Artificial neural network (BP – ANN) – A case study in Indian context

This work analyses influence of road, weather and crash-specific factors on crash severity outcomes for low-speed urban midblock sections and intersections, for day and night time, using Backpropagation–Artificial Neural Network (BP–ANN). Five-year crash data (2015–2019) from 82Km urban road network of Patna, India was used for the study. The road factors include pavement width, distress condition, marking; shoulder type, condition; road section type as mid-block, intersection and intersection control. Weather factors include season of crash, fog or rain at crash time. Crash factor include collision partner, type and crash time. The most appropriate BP–ANN model architecture was estimated using Misclassification-Rate. It was observed that midblock segments witness higher severities during daytime, whereas intersections witness higher severities during night. Controlled intersections are safer compared to un-controlled intersections. Pavement distress greatly increase the chance of higher severities. Narrow roads record greater severities during day due to lack of surveillance.

Detection of apple moldy cores using transmittance spectroscopy combined with supervised classifier comparison and parameter optimization

It is difficult to determine which apples have moldy cores just by looking at the outside of the apple. In the present study, we investigated identifying moldy cores using near-infrared transmittance spectra. First, input spectral features selected by noise adjusted principal component analysis (NAPCA) for back propagation artificial neural network (BP ANN) was used to reduce the dimensions of the original data. Then, four factors and five levels uniform design of the input nodes, training functions, transfer layer functions and output layer functions for NAPCA-BP ANN optimization is proposed. And the original data were input into NAPCA-BP ANN to obtain the recognition accuracy and NAPCA-support vector machine (SVM) was as a comparative recognition model. The results showed that through the uniform design-based NAPCA-BP ANN optimization, the NAPCA method had higher identification accuracy, precision, recall and F1 score, than either full spectrum or principal component analysis. Being assessed by different ratio of model test, functions in the hidden layer and output layer of NAPCA-BP ANN, the proposed method achieved the best accuracy to 98.03%. The accuracy, precision, recall and F1 score based on NAPCA-BP ANN were 3.92%, 2.86%, 2.78% and 2.82% higher than those based on NAPCA-SVM, respectively. This method provides a theoretical basis for the development of on-line monitoring of the internal quality of apples.

Research and practice of blaze new trails and start an undertaking cultivate model based on big data technology in computer special field of study

Abstract With the advent of the era of big data and cloud computing, big data is more and more widely used in various industries, and the research and practice of using big data technology to realize blaze new trails and start an undertaking cultivate mode in computer special field of study will become an important focus of the blaze new trails and start an undertaking cultivate system nowadays. In this paper, we construct the blaze new trails and start an undertaking cultivate model based on BP ANN, construct the blaze new trails and start an undertaking cultivate evaluation index system through Delphi method, and finally design the model of blaze new trails and start an undertaking cultivate evaluation in order to analyze the results of blaze new trails and start an undertaking cultivate. The scoring results correspond to the data of 21 evaluation indexes of a student, and the various scores can map to the level of individual students’ blaze new trails and start an undertaking ability, and provide effective countermeasures for computer special field of study students’ blaze new trails and start an undertaking cultivate. This paper effectively evaluates the blaze new trails and start an undertaking ability of students, broadens the application value and research field of BP ANN, and actively explores the blaze new trails and start an undertaking cultivate of computer special field of study students, which helps the operation mode of blaze new trails and start an undertaking cultivate of computer special field of study special field of study, provides step foreward ideas, and is of great significance to promote the step forward of blaze new trails and start an undertaking cultivate mode.

Improved flavonoid content in mulberry leaves by solid-state fermentation: Metabolic profile, activity, and mechanism

This study was aimed at investigating the mechanisms of release and transformation of total flavonoids (TFs) from mulberry leaves (MLs) by solid-state fermentation (SSF) via Aspergillus cristatus. An intelligent model based on a back-propagation artificial neural network (BP–ANN) was established to predict the TFs in fungi co-fermented MLs. After prediction and experimental validation, the TFs released by SSF (72.55 mg rutin equivalents/g dry weight) were significantly higher than that of unfermented (24.42 mg rutin equivalents/g dry weight). In addition, the biotransformation mechanism of flavonoids during SSF was proposed by analyzing the untargeted metabolome, including the flavonoid and phenylpropanoid biosynthetic pathways. Pearson's correlation coefficient analysis indicated that the 29 up-regulated catabolites (especially loureirin D and arachidoside) after SSF contributed to increased ABTS (2,2′-azino-bis (3-ethylbenzothiazoline-6-sulfonic acid)) radical scavenging activities and α-glucosidase inhibitory activities. Furthermore, structural characteristics (by scanning electron microscopy, thermogravimetric analysis, and FTIR spectroscopy) and analyses of dynamic changes in carbohydrate-hydrolyzing enzymes revealed that the destruction of hemicellulose was essential to releasing TFs by SFF. Industrial relevanceIn this study, a back-propagation artificial neural network (BP–ANN) was successfully used to establish a co-fermentation with Aspergillus cristatus for releasing and transforming flavonoids of mulberry leaves (MLs) through solid-state fermentation (SSF). The present study has immeasurable importance toward making maximum use of MLs. Moreover, the findings of this study also provide new insights into understanding the release and biotransformation mechanism of flavonoids from MLs by SSF.

Neural networks for predicting air gap membrane distillation performance

The hollow fiber air gap membrane distillation (AGMD) has recently attracted tremendous attention for desalination and wastewater treatment due to its high packing density, low conductive heat loss, and latent heat recovery capability. Utilizing fast and accurate modeling tools to predict MD performance can result in the further development of desalination technologies. However, simple and time-saving prediction models to assess the AGMD performance were not abundant. Herein, AGMD performance, including permeate flux (J) and gained output ratio (GOR) was predicted through multiple linear regression (MLR) model, back propagation neural network (BP ANN) and radial basis function neural network (RBF ANN) under different hot temperatures (Th), coolant temperatures (Tc), feed flow rates (F), and feed concentration (c). A total of 30 sets of data were used to train the proposed models, the other 10 external validation datasets not used for training the models were applied to validate the prediction accuracy. The results depicted that RBF ANN (SPREAD = 30, N = 30) showed greatest prediction performance (R2 = 0.99–1) compared with BP ANN and MLR models (R2 = 0.98–0.99; R2 = 0.89–0.97). The computing time consumption of RBF ANN was higher than BP ANN. According to the Mean impact value (MIV) analysis, Th had the strongest effect on J and GOR. Increasing Th and decreasing c both had positive impacts on J and GOR, but increasing Tc or F resulted in a trade-off influence. A genetic algorithm (GA) was employed to optimize J and GOR simultaneously, the optimum J and GOR could reach 6.00 kg/m2·h and 7.70 respectively. In this study, the three prediction models proved their abilities to predict AGMD performance and further provide guidance in the actual membrane distillation water treatment process.

Research on Establishment and Application of Digital Twin for a Phase-Shift Full-Bridge Current Doubling Rectifier Converter

In order to improve the safety and reliability of airborne power supply and distribution system, a digital twin of a phase-shift full-bridge current doubling rectifier converter based on gallium nitride metal-oxide-semiconductor field-effect transistor (GaN MOSFET) is built. The digital twin can be regarded as virtual symmetry of the real converter. The particle swarm optimization (PSO) algorithm is applied to compare the voltage and current waveforms of the real converter with relative waveforms of its digital model established in simulation software, and the interior parameters of the model are constantly updated until the waveforms of the digital twin coincide with that of the converter. Applications of this digital twin are discussed through experimental verification, and the results show that the digital twin can deduce the parameter value of key components of the converter in the case of soft faults. In the case of hard faults, a back propagating artificial neural net (BP ANN) is trained by data collected from the digital twin, and the BP ANN is able to identify different running states of the converter. The outcomes of this paper present a non-intrusive and high real-time method of condition monitoring and fault diagnosis, which is beneficial to improve the reliability of airborne HVDC power supply system.

Mechanical Dispatch Reliability Prediction for Civil Aircraft Considering Operational Parameters

To effectively predict the mechanical dispatch reliability (MDR), the artificial neural networks method combined with aircraft operation health status parameters is proposed, which introduces the real civil aircraft operation data for verification, to improve the modeling precision and computing efficiency. Grey relational analysis can identify the degree of correlation between aircraft system health status (such as the unscheduled maintenance event, unit report event, and services number) and dispatch release and screen out the most closely related systems to determine the set of input parameters required for the prediction model. The artificial neural network using radial basis function (RBF) as a kernel function, has the best applicability in the prediction of multidimensional, small sample problems. Health status parameters of related systems are used as the input to predict the changing trend of MDR, under the artificial neural network modeling framework. The case study collects real operation data for a certain civil aircraft over the past five years to validate the performance of the model which meets the requirements of the application. The results show that the prediction quadratic error Ep of the model reaches 6.9 × 10−8. That is to say, in the existing operating environment, the prediction of the number of delay ＆ cancel events per month can be less than once. The accuracy of RBF ANN, BP ANN and GA-BP ANN are compared further, and the results show that RBF ANN has better adaptability to such multidimensional small sample problems. The efforts of this paper provide a highly efficient method for the MDR prediction through aircraft system health state parameters, which is a promising model to enhance the prediction and controllability of the dispatch release, providing support for the construction of the civil aircraft operation system.

Mathematical Modelling for Predicting Thermal Properties of Selected Limestone

Due to a lack of geotechnical and geothermal studies on Jordanian limestone, this paper aims to provide the thermal properties, including thermal conductivity, thermal diffusivity, and specific heat, using the Hot Disk Transient Plane Source (TPS) 2200 method. It also aims to provide a set of mathematical models through which the thermal properties can be indirectly predicted from the rocks’ physical and engineering properties. One hundred cylindrical rock specimens with a height of 20 cm and a diameter of 10 cm were extracted and prepared. The results showed that the thermal conductivity values ranged between (1.931–3.468) (W/(m*k)), thermal diffusivity (1.032–1.81) (mm2/s), and specific heat (1.57–2.563) ((MJ)/(m3*K)). The results also suggest a direct relationship between conductivity and diffusivity and an inverse relationship between conductivity and specific heat. On the other hand, the results indicate the direct relationship between the conductivity and diffusivity, and the inverse relationship between the specific heat and density, hardness, sound velocity, and rock strength; the opposite happens when the rock’s porosity is considered. Simple regression, multivariate regression, and the backpropagation–artificial neural network (BP–ANN) approach were utilized to predict the thermal properties of limestone. Results indicated that the ANN model provided superior prediction performance compared to other models.

An improved model for predicting dynamic total pressure distortion

In order to accurately predict the dynamic total pressure distortion produced by crescent baffle distortion simulator in its practical application, an improved turbulence correlation model is proposed in this paper. Several turbulence parameters from CFD solution are correlated with the measured turbulence via the improved model. Back-propagation artificial neural network (BP ANN) is used to find the correlation and establish a turbulence prediction model. Dynamic distortion intensity is represented by the predicted turbulence. The results show that dynamic distortion indexes predicted by the improved model are more accurate, with a maximum error of 7.72%. Finally, the dynamic distortion flow field of crescent baffle distortion simulator is predicted by applying the established prediction model. Analysis of dynamic distortion characteristics provides important guidance for the application of crescent baffle distortion simulator.

QSPR models to predict quantum chemical properties of imidazole derivatives using genetic algorithm–multiple linear regression and back‐propagation–artificial neural network

AbstractImidazole derivatives are the foundation of different types of drugs with a wide range of biological activities. In this study, the genetic algorithm–multiple linear regression (GA–MLR), and backpropagation–artificial neural network (BP–ANN) were applied to design QSPR models to predict the quantum chemical properties like the entropy (S) and enthalpy of formation (∆Hf) of imidazole derivatives. In order to draw molecular structure of 84 derivative compounds Gauss View 05 program was used. These structures were optimized at DFT‐B3LYP/6‐311G* level with Gaussian09W. The Dragon software was used to calculate a set of different molecular descriptors, and the genetic algorithm procedure and backward stepwise regression were applied for the selection of descriptors. The resulting quantitative GA–MLR model of ∆Hf, showed that there is good linear correlation between the selected descriptors and ∆Hf of compounds. Also the results show that the BP–ANN model appeared to be superior to GA–MLR model for prediction of entropy. Different internal and external validation metrics were adopted to verify the predictive performance of QSPR models. The predictive powers of the models were found to be acceptable. Thus, these QSPR models may be useful for designing new series of imidazole derivatives and prediction of their properties.

A Scheme for Quickly Simulating Extraterrestrial Solar Radiation over Complex Terrain on a Large Spatial-Temporal Span—A Case Study over the Entirety of China

Extraterrestrial solar radiation (ESR) is the essential basic background for solar radiation, which determines the occurrence of the weather and atmospheric phenomena. Since the influence of ESR variation on actual rugged terrain is a diverse, complex, and dynamic process, simulating ESR over a large spatial-temporal span, especially with a high-resolution digital elevation model (DEM), is a significant challenge. In this paper, we developed a new scheme for simulating ESR over the entirety of China using a DEM with a resolution of 30 m. To fully consider regional terrain status, the feature variables used were elevation, slope, and aspects of the located grid and the surrounding four grids to reveal the topography. In addition, latitude was used as a feature variable to consider the geographical location, and the month number was used to consider the duration. On the basis of different geographical locations, the training dataset was established from 20,000 grids. With the feature variable composition and training dataset, a backpropagation artificial neural network (BP ANN) was found to have the best performance compared with the other three machine learning methods in simulating ESR for a DEM. In terms of the proposed scheme and BP ANN, we drew an ESR map of China with a resolution of 30 m. The determination coefficient of the simulation result achieved 0.99 and the root-mean-square error was less than 50 MJ/m2 in all sample areas, confirming its remarkable accuracy. In terms of efficiency, the time consumption of ESR simulated using the proposed scheme shrinks over 150 times in all sample areas compared to that simulated via the theoretical model. Simultaneously, the developed scheme was also used to simulate an ESR for a DEM with a resolution of 90 m to verify the universality and robustness of the developed scheme. In addition, we used the proposed scheme to derive the direct solar radiation and global solar radiation, thereby further proving the reliability and applicability of our study. Overall, our work convincingly proved that the proposed scheme is a potential and effective approach for quickly simulating ESR with high accuracy. This study provides the basis for different solar radiation inversions of long time series and large spatial scales, offering additional insights for simulating ESR on a large spatial-temporal span.

Land use optimization of rural production–living–ecological space at different scales based on the BP–ANN and CLUE–S models

Rural production–living–ecological (PLE) space is the essential carrier of China's rural land resource planning and management. However, the pattern identification, optimal prediction, and multi–scale integration of rural PLE space lack sufficient scientific evidence and reliable quantitative analysis. To fill this gap, this paper proposes a multi-scale land use optimization method based on benefit coupling evaluation, BP–ANN and CLUE–S models. The typical hilly area of the upper reaches of the Yangtze River in southwest China were used as a case for empirical study. The results showed that there was a high correlation between land use patterns and benefits. The key to land use optimization in hilly areas is to increase the area of ecological land and improve the capacity of regional ecosystem services. Through the evaluation of the degree of comprehensive benefit coupling, 44 sample towns with an optimal land use pattern were effectively identified to construct the optimization model. At the regional scale, the BP-ANN model predicted the optimal ratio of land use structure for each township unit based on natural, social, and economic influencing factors. The optimization results reduced production land by 8.94% on average, increased ecological land by 9.2% on average, and kept living land relatively stable. The proportions of production, living, and ecological land were adjusted to 59.85%, 8.34%, and 31.81%, respectively, which can better meet the land space demand for food security and ecological protection in the future. Regarding the smaller scale, the CLUE-S model took the regional-scale optimization results as the goal to simulate the spatial distribution of land use. The proportion of production, living, and ecological land in the town of Taiping was optimized from the previous 88.17%, 6.25%, and 5.58% to 62.92%, 7.83%, and 29.25%, respectively. The optimization results not only ensure the stability of high-quality cultivated land, but also effectively improve ecological functions such as soil conservation and water purification. This novel method was proven to be effective for quantitative optimization of land use and multi-scale functional space cohesion and integration, providing scientific support for the sustainable use of rural land resources in China.

Prediction of Carbon Dioxide Solubility in Polymers Based on Adaptive Particle Swarm Optimization and Least Squares Support Vector Machine

AbstractSolubility is a significant physical and chemical property. The solubility of carbon dioxide(CO2) in polymers is an important application of green chemistry. Aimed at the problem of insufficient precision of the existing prediction model, a solubility prediction model based on the adaptive particle swarm optimization algorithm and the least‐squares support vector machine(APSO‐LSSVM) is proposed. Different from the traditional particle swarm algorithm, APSO algorithm improves the problem of easily falling into local optimal solution. The regularization parameters and kernel function tuning parameters of the LSSVM were optimized by APSO and then use this model to predict the solubility of CO2 in eight polymers within a wide range of temperatures and pressures. APSO‐LSSVM model has great prediction ability, with a good correlation between prediction data and experimental data, high prediction accuracy, short calculation time and strong stability. Compared with back propagation artificial neural network(BP ANN), back propagation‐particle swarm optimization algorithm artificial neural network(BP‐PSO ANN) and LSSVM, APSO‐LSSVM has better comprehensive performance. The average absolute relative deviation(AARD), root mean square error(RMSE), determination coefficient(R2) were respectively 0.2130, 0.0120, 0.9853. In addition, the model has good expansibility and can be applied to other fields such as chemistry and medicine.

Deformation Behavior and Constitutive Equation of 42CrMo Steel at High Temperature

High-temperature reduction pretreatment (HTRP) is a process that can significantly improve the core quality of a billet. The existing flow stress data cannot meet the needs of simulation due to lack of high temperature data. To obtain the hot forming process parameters for the high-temperature reduction pretreatment process of 42CrMo steel, a hot compression experiment of 42CrMo steel was conducted on Gleeble-3500 thermal-mechanical at 1200–1350 °C with the rates of deformation 0.001–10 s−1 and the deformation of 60%, and its deformation behavior at elevated temperature was studied. In this study, the effects of flow stress temperature and strain rate on austenite grain were investigated. Moreover, two typical constitutive models were employed to describe the flow stress, namely the Arrhenius constitutive model of strain compensation and back propagation artificial neural network (BP ANN) model. The performance evaluation shows that BP ANN model has high accuracy and stability to predict the curve. The thermal processing maps under strains of 0.1, 0.2, 0.3, and 0.4 were established. Based on the analysis of the thermal processing map, the optimal high reduction process parameter range of 42CrMo is obtained: the temperature range is 1250–1350 °C, and the strain rate range is 0.01–1 s−1.

Evaluation and Development of Pedo-Transfer Functions for Predicting Soil Saturated Hydraulic Conductivity in the Alpine Frigid Hilly Region of Qinghai Province

In recent years, Pedo-Transfer Functions (PTFs) have become a commonly used tool to predict the hydraulic properties of soil. As an important index to evaluate the function of forest water conservation, the prediction of saturated hydraulic conductivity (KS) on the regional scale is of great significance to guide the vegetation construction of returning farmland to forest area. However, if the published PTFs are directly applied to areas where the soil conditions are different from those where the PTFs are established, their predictive performance will be greatly reduced. In this study, 10 basic soil properties were measured as input variables for PTFs to predict KS in the three watersheds of Taergou, Anmentan, and Yangjiazhai in the alpine frigid hilly region of Qinghai Province, China. The parameters of the eight published PTFs were modified by the least-squares method and new PTFs were also constructed, and their prediction performance was evaluated. The results showed that the KS of coniferous and broad-leaved mixed forests and broad-leaved pure forests in the study area were significantly higher than those of pure coniferous forests, and grassland and farmland were the lowest (p &gt; 0.05). Soil Organic Matter plays an important role in predicting KS and should be used as an input variable when establishing PTFs. The Analysis-Back Propagation Artificial Neural Network (BP ANN) PTF that was established, with input variables that were, Si·SOM, BD·Si, ln2Cl, SOM2, and SOM·lnCl had a better predictive performance than published PTFs and MLR PTFs.

Measurement and correlation of 1, 4-diamino-2-methoxyanthraquinone and 1- amino-2-methoxy-4-hydroxyanthraquinone in supercritical CO2

ABSTRACT The solubility of 1, 4-diamino-2-methoxyanthraquinone and 1-amino−2-methoxy−4-hydroxyanthraquinone in supercritical carbon dioxide (SC-CO2) was measured by a dynamic method under the temperature ranges of 343.15 to 383.15 K, and pressures of 16 to 24 MPa. The results of 1, 4-diamino-2-methoxyanthraquinone were correlated by the Mende-Santiago-Teja (MST) equation and the backpropagation artificial neural network (BP ANN). The MST equation is a three-parameter linear equation, and the model could correlate satisfactorily the experimental results between the temperature ranges of 353.15 K and 383.15 K with the correlation coefficient of 0.82. In BP ANN built-in MATLAB software, temperature and pressure were used as the two input layers and solubility as the output layer. A non-linear function was selected as a hidden layer function. The calculated results showed that the solubility data at the temperature ranges of 353.15 to 383.15 K, and pressures of 16 to 24 MPa could be estimated accurately by BP ANN with fitting level of 0.99.

Dynamic behavior and modified artificial neural network model for predicting flow stress during hot deformation of Alloy 925

In this work, the hot deformation behavior of Alloy 925 with intermediate strain rates of 0.01s−1-10s−1 ranging from 900 °C to 1150 °C has been investigated through the isothermal hot compression tests. Some features of the flow curves corresponding to negative strain rate sensitivity have been observed and discussed associated with the microstructural evolution, showing the nucleation mechanism of dynamic recrystallization and the occurrence of dynamic strain ageing. Besides, we have employed two typical constitutive models, namely, Arrhenius model and backpropagation artificial neural network (BP ANN) model to describe the flow behavior, and also developed a modified BP ANN model based on genetic algorithm (GA-BP ANN). The results show that the GA-BP ANN model has the highest accuracy and stability for predicting the flow stress. The correlation coefficient between the predicted and experimental values is 99.99 %, and the average absolute relative error is only 0.54 %. The comparative investigation on the predicted values of different ANN models reflects that GA can reduce the randomness of initial weights and thresholds of BP ANN and also can further increase the accuracy and stability of ANN model. Moreover, the increasing number of input training data can improve the prediction performance of neural network. For a single-layer neural network, 12 hidden layers can effectively ensure the reliability of the constitutive model.

Impact factor analysis, prediction, and mapping of soil corrosion of carbon steel across China based on MIV-BP artificial neural network and GIS

Q235 carbon steel is one of the most widely used carbon steels, and soil corrosion and failures of it caused accidents, casualties, and great financial losses. Corrosion of Q235 carbon steel differed in spatial because of spatial variation in soil environmental factors. However, the national scale spatial pattern of soil corrosion of Q235 carbon steel across China has not been explored. The values of 12 impact factors, corrosion rate, and pitting corrosion rate at 25 sites covered all soil types in China were collected. Mean impact value (MIV) algorithm and back propagation artificial neural network (BP ANN) were combined and applied in the impact factor analysis. Prediction models for corrosion rate and pitting corrosion were developed based on BP ANN. The proposed prediction models and information about soil properties with high resolution (1 km × 1 km) were used in the prediction of corrosion rate. Based on geographical information system (GIS), the national scale spatial pattern of soil corrosion of Q235 carbon steel across China were analyzed. The water content and pH were of the largest influence (|MIV| > 0.522) on both corrosion rate and pitting corrosion rate. For prediction models of corrosion rate and pitting corrosion rate, the predicted values were close to the measured values. Corrosion rates were of higher spatial differences, ranged from 0.632 to 5.181 g/(dm2·a). Pitting corrosion rates in the northeast were higher than other areas, which might be caused by higher values of total salt content, organic matter, and total voidage. Average corrosion rate in soil type 1 (1.161 mm/a) was nearly two times of that in soil type 2 (0.610 mm/a) and over three times of that in other types, indicating that corrosion rate varied largely among different soil types. The importance of 12 impact factors of corrosion of Q235 carbon steel were evaluated, and the national scale corrosion rate and pitting corrosion rate in China were predicted and mapped for the first time. Both pitting corrosion rates and corrosion rates were of strong spatial variation.