Multilayer Feedforward Neural Network Model Research Articles

A neural network model for predicting weighted mean temperature

Water vapor is an important element of the Earth’s atmosphere, and most of it concentrates at the bottom of the troposphere. Knowledge of the water vapor measured by Global Navigation Satellite Systems (GNSS) is an important direction of GNSS research. In particular, when the zenith wet delay is converted to precipitable water vapor, the weighted mean temperature $$T_\mathrm{m}$$ is a variable parameter to be determined in this conversion. The purpose of the study is getting a more accurate $$T_\mathrm{m}$$ model for global users by a combination of two different characteristics of $$T_\mathrm{m}$$ (i.e., the $$T_\mathrm{m}$$ seasonal variations and the relationships between $$T_\mathrm{m}$$ and surface meteorological elements). The modeling process was carried out by using the neural network technology. A multilayer feedforward neural network model (the NN) was established. The NN model is used with measurements of only surface temperature $$T_\mathrm{S}$$ . The NN was validated and compared with four other published global $$T_\mathrm{m}$$ models. The results show that the NN performed better than any of the four compared models on the global scale.

Prediction of hourly solar radiation in Abu Musa Island using machine learning algorithms

Accurate forecasting of renewable energy sources plays a key role in their integration into the grid. This study proposes machine learning algorithms to predict the hourly solar irradiance. Forecasting models were developed based two types of the input data. The first one uses local time, temperature, pressure, wind speed, and relative humidity as input variables of the models (N1); the second one is the time-series prediction of solar irradiance (N2) (forecasting models only use from past time-series solar radiation values to estimate the future values). For this purpose, multilayer feed-forward neural network (MLFFNN), radial basis function neural network (RBFNN), support vector regression (SVR), fuzzy inference system (FIS) and adaptive neuro-fuzzy inference system (ANFIS) are developed. The results demonstrated that for the N1, SVR and MLFFNN models have the maximum performance to predict the solar irradiance with R = 0.9999 and 0.9795, respectively. For the N2, SVR, MLFFNN and ANFIS models have reported the correlation coefficient more than 0.95 for the testing dataset.

PREDICTING THE RHEOLOGICAL PROPERTIES OF BITUMEN-FILLER MASTIC USING ARTIFICIAL NEURAL NETWORK METHODS

This study was conducted to develop two types of artificial neural network (ANN) model to predict the rheological properties of bitumen-filler mastic in terms of the complex modulus and phase angle. Two types of ANN models were developed namely; (i) a multilayer feed-forward neural network model and (ii) a radial basis function network model. This study was also conducted to evaluate the accuracy of both types of models in predicting the rheological properties of bitumen-filler mastics by means of statistical parameters such as the coefficient of determination (R2), mean absolute error (MAE), mean squared error (MSE) and root mean squared error (RMSE) for every developed model. A set of dynamic shear rheometer (DSR) test data was used on a range of the bitumen-filler mastics with three filler types (limestone, cement and grit stone) and two filler concentrations (35 and 65% by mass). Based on the analysis performed, it was found that both models were able to predict the complex modulus and phase angle of bitumen-filler mastics with the average R2 value exceeding 0.98. A comparison between the two types of models showed that the radial basis function network model has a higher accuracy than multilayer feed-forward neural network model with a higher value of R2 and lower value of MAE, MSE and RMSE. It can be concluded that the ANN model can be used as an alternative method to predict the rheological properties of bitumen-filler mastic.

Retrieval of fractional snow covered area from MODIS data by multivariate adaptive regression splines

In this paper, a novel approach to estimate fractional snow cover (FSC) from MODIS data in a complex and heterogeneous Alpine terrain is represented by using a state-of-the-art nonparametric spline regression method, namely, multivariate adaptive regression splines (MARS). For this purpose, twenty MODIS - Landsat 8 image pairs acquired between April 2013 and December 2016 over European Alps are used. Fifteen of the image pairs are employed during model training and five images are reserved as an independent test dataset. MARS models are trained by using MODIS top-of-atmosphere reflectance values of bands 1–7, normalized difference snow index, normalized difference vegetation index and land cover class as predictor variables. Reference FSC maps are generated from higher spatial resolution Landsat 8 binary snow cover maps. Multilayer feedforward artificial neural network (ANN) models are also trained by using the same input data. During the training and the testing, the effects of the training data size and the sampling type on the predictive performance of ANN and MARS models are investigated. An additional search is also conducted to reveal whether the choice of the transfer function used in the output layer of ANN has a significant contribution to the network's FSC mapping performance. The final ANN and MARS FSC products are at 500m spatial resolution. The results on the independent test scenes indicate that the developed ANN models with linear and hyperbolic tangent transfer functions in the output layer and the MARS models are in good agreement with reference FSC data with the same average values of R=0.93. In contrast, the standard MODIS snow fraction product, namely, MOD10 FSC, exhibits slightly poorer performance with average R=0.88. The proposed MARS approach is statistically proven to have the same performance with ANN, yet it is computationally more efficient in model building.

Mapping of pore pressure, in-situ stress and brittleness in unconventional shale reservoir of Krishna-Godavari basin

A methodology is demonstrated to estimate brittleness coefficient and in-situ stress from well log data linking post-stack seismic data in the unconventional Raghavapuram Shale reservoir of Krishna-Godavari basin, India. Post-stack seismic data has been used to estimate pore pressure, in-situ stress from well logs through multilayered feedforward neural network (MLFN) model. Shear impedance and density sections from post-stack seismic data are generated from conversion of acoustic impedance to shear impedance linking with well log data. Inverted acoustic impedance, shear impedance, density and compressional to shear wave velocity ratio (Vp/Vs) have been used to train the MLFN model for mapping pore pressure and vertical stress obtained from well logs. Minimum horizontal stress is computed for anisotropic shale medium. Differential stress ratio (DHSR) section is generated from maximum and minimum horizontal stress magnitudes using MLFN models. Brittleness coefficient estimated from static Young's modulus and Poisson's ratio is mapped into post-stack section. Fracture index is computed from dipole shear sonic log data. The brittleness coefficient of 20–45% and DHSR of 7–16% with 20–46% fracture index are noticed in the Raghavapuram Shale. This zones exhibit relatively high values of brittleness coefficient and relatively high DHSR. Fractures will be aligned in high DHSR zones and oriented to the direction of maximum horizontal stress. This information may be useful for fracture type prediction and designing hydraulic fracture stimulation.

Prediction of Compressional Wave Velocity Using Regression and Neural Network Modeling and Estimation of Stress Orientation in Bokaro Coalfield, India

Velocity of compressional wave (V P) of coal and non-coal lithology is predicted from five wells from the Bokaro coalfield (CF), India. Shear sonic travel time logs are not recorded for all wells under the study area. Shear wave velocity (Vs) is available only for two wells: one from east and other from west Bokaro CF. The major lithologies of this CF are dominated by coal, shaly coal of Barakar formation. This paper focuses on the (a) relationship between Vp and Vs, (b) prediction of Vp using regression and neural network modeling and (c) estimation of maximum horizontal stress from image log. Coal characterizes with low acoustic impedance (AI) as compared to the overlying and underlying strata. The cross-plot between AI and Vp/Vs is able to identify coal, shaly coal, shale and sandstone from wells in Bokaro CF. The relationship between Vp and Vs is obtained with excellent goodness of fit (R 2) ranging from 0.90 to 0.93. Linear multiple regression and multi-layered feed-forward neural network (MLFN) models are developed for prediction Vp from two wells using four input log parameters: gamma ray, resistivity, bulk density and neutron porosity. Regression model predicted Vp shows poor fit (from R 2 = 0.28) to good fit (R 2 = 0.79) with the observed velocity. MLFN model predicted Vp indicates satisfactory to good R2 values varying from 0.62 to 0.92 with the observed velocity. Maximum horizontal stress orientation from a well at west Bokaro CF is studied from Formation Micro-Imager (FMI) log. Breakouts and drilling-induced fractures (DIFs) are identified from the FMI log. Breakout length of 4.5 m is oriented towards N60°W whereas the orientation of DIFs for a cumulative length of 26.5 m is varying from N15°E to N35°E. The mean maximum horizontal stress in this CF is towards N28°E.

A Multi-model Approach in Developing an Intelligent Assistant for Diagnosis Recommendation in Clinical Health Systems

Clinical health information systems capture massive amounts of unstructured data from various health and medical facilities. This study utilizes unstructured patient clinical text data to develop an intelligent assistant that can identify possible related diagnoses based on a given text input. The approach applies a one-vs-rest binary classification technique wherein given an input text data, it is identified whether it can be positively or negatively classified for a given diagnosis. Multi-layer Feed-Forward Neural Network models were developed for each individual diagnosis case. The task of the intelligent assistant is to iterate over all the different models and return those that output a positive diagnosis. To validate the performance of the models, the performance metrics were compared against Naive Bayes, Decision Trees, and K-Nearest Neighbor. The results show that the neural network learner provided better performance scores in both accuracy and area under the curve metric scores. Further, testing on multiple diagnoses also shows that the methodology for developing the diagnosis models can be replicated for development of models for other diseases as well.

Improvement of Envelope Design Through Multilayer Feed-Forward Neural Networks

The performance of the building envelope of a large-scale public building significantly influences the energy consumption of such a building. This study aims to determine the best strategy for the envelope by examining the engineering design of the building in Nanchang University. The building shape coefficient, sun-shading strategies, window–wall ratio, roof, and walls were studied through a method involving multilayer feed-forward neural network model simulations. Results show that the optimum shape coefficient value is 0.32. The combination of interior and exterior blinds and electrochromic glass is the ideal option to reduce the increase in the energy consumption of the architecture caused by solar radiation. Maintaining the window–wall ratio at 0.4 is ideal. A green roof exerts a minimal effect on building energy consumption decrease (only 0.4%). Applying the strategy of vertical greening to the external wall can reduce cooling energy consumption by as much as 5.4%. Adopting the best envelope strategy combination can further decrease energy consumption by 20.8%. This strategy is also applicable to the middle and lower reaches of Yangtze River in China, which flow through Nanchang and have a climate similar to that of the said area. Future research should be directed toward applying artificial neural networks to quantitatively evaluate the effects of a design strategy and produce the best design strategy combination.

Prediction of Coal Proximate Parameters and Useful Heat Value of Coal from Well Logs of the Bishrampur Coalfield, India, Using Regression and Artificial Neural Network Modeling

The energy of coal is expressed by its useful heat value (UHV), and it is the major key player in coal pricing. The objectives of this paper are to obtain (a) the regression relationship between coal proximate parameters and UHVs, and (b) multilayered feed forward neural network (MLFN) models between geophysical log responses and UHVs. Six wells are used for training the networks, and three wells are used for validating the obtained results in the Bishrampur coalfield. The mean square error (MSE) of MLFN models along with correlation (R2) values at their training, validation, and testing stages are the criteria for selecting the best model for estimation of coal proximate parameters and UHVs using geophysical log responses. The final model is selected based on low MSE (≤0.07) and high R2 values (≥0.80) at training, validation, and testing stages. The predicted UHV obtained from the best MLFN model has excellent correlation (R2 = 0.98) with the laboratory determined UHVs of three major coal seams. The pred...

Weld quality prediction in friction stir welding using wavelet analysis

The paper aims to present the application of wavelet packet transformation for feature extraction from the signals acquired during friction stir welding of aluminum alloy. One of the challenges encountered while implementing wavelet packet transformation is the selection of an appropriate mother wavelet function. In this study, a new method is proposed for the selection of an appropriate mother wavelet function based on the ratio of energy of the signal to the entropy of the decomposed wavelet packets. Main spindle motor and feed motor current signals are acquired during 65 welding experiments designed through full factorial method by varying three process parameters in four levels. Features obtained from wavelet packet transformation along with process parameters are fed to two artificial neural network models: multi-layer feed-forward neural network model trained with back propagation algorithm and radial basis function neural network model for the prediction of ultimate tensile strength and yield strength of the welds. The prediction performance of the former model is found to be superior to the later model for both ultimate tensile strength and yield strength.

Optimization of friction stir welding process parameters using soft computing techniques

In welding processes, desired weld quality is highly dependent on the selection of optimal process conditions. In this work, the influence of input parameters of friction stir welding process is studied using Taguchi method and full factorial design of experiment. The experimental data set is used to develop multilayer feed-forward artificial neural network (ANN) models using back-propagation training algorithm. These models are used to predict weld qualities as a function of eight process parameters. The weld qualities of the welded joint, such as ultimate tensile strength, yield stress, percentage elongation, bending angle and hardness, are considered. In order to offline optimize these quality characteristics, four evolutionary algorithms, namely binary-coded genetic algorithm, real-coded genetic algorithm, differential evolution and particle swarm optimization, are coupled with the developed ANN models. The optimized quality characteristics obtained from these proposed techniques are compared and verified with experimental results.

A Comparison Study of MIMO Water Wall Model with Linear, MFNN and ESN Models

A water wall system is one of the most important components of a boiler in a thermal power plant, and it is a nonlinear Multi-Input and Multi-Output (MIMO) system, with 6 inputs and 3 outputs. Three models are developed and comp for the controller design, including a linear model, a multilayer feed-forward neural network (MFNN) model and an Echo State Network (ESN) model. First, the linear model is developed by linearizing a given nonlinear model and is analyzed as a function of the operating point. Second, the MFNN and the ESN are developed by using training data from the nonlinear model. The three models are validated using Matlab with nonlinear input-output data that was not used during training.

Artificial Neural Network Model for Estimating Temporal and Spatial Freeway Work Zone Delay Using Probe-Vehicle Data

Highway lane closures due to road reconstruction and the resulting work zones have been a major source of nonrecurring congestion on freeways. It is extremely important to calculate the safety and cost impacts of work zones: the use of new technologies that track drivers and vehicles make that possible. A multilayer feed-forward artificial neural network (ANN) model is developed in this paper to estimate work zone delay by using the probe-vehicle data. The probe data include the travel speeds under normal and work zone conditions. Unlike previous models, the proposed model estimates temporal and spatial delays, which are applied to a real world case study in New Jersey. The work zone data (i.e., starting time, duration, length, and number of closed lanes) were collected on New Jersey freeways in 2014 together with actual probe-vehicle speeds. A comparative analysis was conducted; the results indicate that the ANN model outperforms the traditional deterministic queuing model in terms of the accuracy in estimating travel delays. The ANN model can be used to calculate contractor penalty in terms of cost overruns as well as incentivize a reward schedule in case of early work competition. The model can assist work zone planners in designing optimal start and end time of work zone as function of time of day. In assessing the performance of work zones, the model can assist transportation engineers to better develop and evaluate traffic mitigation and management plans.

Feedforward Neural Network Modeling of Biomass Pyrolysis Process for Biochar Production

Growing energy needs and increasing environmental issues are creating awareness for alternative energy which substitutes the non-renewable and polluting fossil fuels. Biomass is a good feedstock for biochar production through the pyrolysis process. There is potential to generate solid fuel from biomass, as there are large quantities of agricultural wastes available in Malaysia. This paper outlines the experimental study on the pyrolysis of durian rinds in Thermogravimetric Analyzer (TGA). The effects of temperatures on the yield of biochar from the durian rinds were investigated. Increasing temperature resulted in increasing weight loss of the biomass sample. The total weight loss at the end of 920 °C was 86.9 %. This corresponds to the loss of water and volatile matter from the durian rinds. A multilayer feed-forward neural network (FANN) model was trained with an error back-propagation algorithm. Reaction time, temperature were used as the input parameters and weight loss was the output for the study. A FANN model with modeling performance of 2-20-1 was obtained for the study.

Predicting customer profitability over time based on RFM time series

Predicting consumer profitability dynamically over time plays a vital role in today's customer-centric business. In this paper, we adopt a dynamic systems approach to address the dynamic prediction problem of customer profitability. Based on customer transaction records, RFM score-based time series are generated using cluster analysis. These time series are used to measure and describe customer profitability. Furthermore, multilayer feed-forward neural network models are trained to capture the dynamics of the evolving customer profitability. A set of real transactions from a UK-based online retailer is used in this study. Relevant experimental results have shown good performance of the proposed approach.

An Inversion-free Predictive Controller for Piezoelectric Actuators Based on A Dynamic Linearized Neural Network Model

Piezoelectric actuators (PEAs) are widely used in high-precision positioning applications. However, the inherent hysteresis nonlinearity seriously deteriorates the tracking performance of PEAs. To deal with it, the compensation of the hysteresis by using its inverse model (called inversion-based) is the popular method in the literature. One major disadvantage of this method is that the tracking performance of PEAs highly relies on its inverse model. Meanwhile, the computational burden of obtaining the inverse model is overwhelming. In addition, the physical constraints of the input voltage of PEAs is hardly handled by the inversion-based method. This paper proposes an inversion-free predictive controller, which is based on a dynamic linearized multilayer feedforward neural network (MFNN) model. By the proposed method, the inverse model of the inherent hysteresis is not required, and the control law can be obtained in an explicit form. By using the technique of constrained quadratic programming, the proposed method still works well when dealing with the physical constraints of PEAs. Moreover, an error compensation term is introduced to reduce the steady-state error if the dynamic linearized MFNN cannot approximate the PEA's dynamical model satisfactorily. To verify the effectiveness of the proposed method, experiments are conducted on a commercial PEA. The experiment results show that the proposed method has a satisfactory tracking performance even with high-frequency references. Comparisons demonstrate that the proposed method outperforms some existing results.

Neural networks for predicting the duration of new software projects

The duration of software development projects has become a competitive issue: only 39% of them are finished on time relative to the duration planned originally. The techniques for predicting project duration are most often based on expert judgment and mathematical models, such as statistical regression or machine learning. The contribution of this study is to investigate whether or not the duration prediction accuracy obtained with a multilayer feedforward neural network model, also called a multilayer perceptron (MLP), and with a radial basis function neural network (RBFNN) model is statistically better than that obtained by a multiple linear regression (MLR) model when functional size and the maximum size of the team of developers are used as the independent variables. The three models mentioned above are trained and tested by predicting the duration of new software development projects with a set of projects from the International Software Benchmarking Standards Group (ISBSG) release 11. Results based on absolute residuals, Pred(l) and a Friedman statistical test show that prediction accuracy with the MLP and the RBFNN is statistically better than with the MLR model.

Comparison of classification algorithms with wrapper-based feature selection for predicting osteoporosis outcome based on genetic factors in a taiwanese women population.

An essential task in a genomic analysis of a human disease is limiting the number of strongly associated genes when studying susceptibility to the disease. The goal of this study was to compare computational tools with and without feature selection for predicting osteoporosis outcome in Taiwanese women based on genetic factors such as single nucleotide polymorphisms (SNPs). To elucidate relationships between osteoporosis and SNPs in this population, three classification algorithms were applied: multilayer feedforward neural network (MFNN), naive Bayes, and logistic regression. A wrapper-based feature selection method was also used to identify a subset of major SNPs. Experimental results showed that the MFNN model with the wrapper-based approach was the best predictive model for inferring disease susceptibility based on the complex relationship between osteoporosis and SNPs in Taiwanese women. The findings suggest that patients and doctors can use the proposed tool to enhance decision making based on clinical factors such as SNP genotyping data.

Applying acoustic emission and neural network to classify wheat seeds from weed seeds.

In the present study, an expert weed seeds recognition system combining acoustic emissions analysis, Multilayer Feedforward Neural Network (MFNN) classifier was developed and tested for classifying wheat seeds. This experiment was performed for classifying two major important wheat varieties from five species of weed seeds. In order to produce sound signals, a 60o inclined glass plate was used. Fast Fourier Transform (FFT), Phase and Power Spectral Density (PSD) of impact signals were calculated. All features of sound signals are computed via a 1024-point FFT. After feature generation, 60% of data sets were used for training, 20% for validation, and remaining samples were selected for testing. The optimized MFNN model was found to have 500-12-2 and 500-10-2 architectures for

Predicting the performance measures of a message-passing multiprocessor architecture using artificial neural networks

In this paper, we develop multi-layer feed-forward artificial neural network (MFANN) models for predicting the performance measures of a message-passing multiprocessor architecture interconnected by the simultaneous optical multiprocessor exchange bus (SOME-Bus), which is a fiber-optic interconnection network. OPNET Modeler is used to simulate the SOME-Bus multiprocessor architecture and to create the training and testing datasets. The performance of the MFANN prediction models is evaluated using standard error of estimate (SEE) and multiple correlation coefficient (R). Also, the results of the MFANN models are compared with the ones obtained by generalized regression neural network (GRNN), support vector regression (SVR), and multiple linear regression (MLR). It is shown that MFANN models perform better (i.e., lower SEE and higher R) than GRNN-based, SVR-based, and MLR-based models for predicting the performance measures of a message-passing multiprocessor architecture.