Multilayer Feed-forward Artificial Neural Network Research Articles

The Prediction of the Compressibility Factor of Sour and Natural Gas by an Artificial Neural Network System

In this study, a multilayer normal feed-forward artificial neural network with three layers has been developed for prediction of compressibility factors of gases with different compositions. This model was developed using 977 experimental data of compressibility factors obtained from the literature. In this model, some statistical criteria such as R2, root mean square error (RMSE), and average absolute deviation (AAD) are obtained 0.991, 0.024, and 0.965, respectively. Model validity was obtained by comparison between the experimental results and the results of various equations of state, such as van der Waals (1910), Redlich-Kwong (1949), Lawal-Lake-Silberberg (Lawal, 1999), Peng-Robinson (1976), and pseudo-experimental correlations such as Dranchuk-Abu-Kassem (1975), Dranchuk-Purvis-Robinson (1974), Hall-Yarborough (1973), Brill-Beggs (1974), Shell Oil Company (2003), Gopal (1977) obtained result of this model. It can be inferred that the results of this model are compatible with the experimental data and the obtained result of this model is more accurate than other correlations.

Modeling of some properties of the crushed tile concretes exposed to elevated temperatures

In this study, artificial neural network (ANN) and fuzzy logic (FL) models have been developed for predicting the compressive strength ( f c ) and dynamic modulus of elasticity ( E d) of the crushed tile concretes (CTC) exposed to elevated temperatures. Some relationships are established between chosen inputs and outputs by developing and testing a multi-layered feed forward ANN and FL trained with the back-propagation algorithm. First of these relationships is established between the outputs as f c of CTC after being exposed to elevated temperatures and the inputs as exposed temperature ( T), crushed tile aggregate (CT) and crushed stone II (CSII) contents of concrete. The second one is the relationship between E d of concretes and the same inputs. In this aim, concrete specimens are produced by CT replacing 16–31.5 mm coarse aggregate at the ratios of 0%, 10%, 25%, 50%, 75% and 100%. Concrete specimens are exposed to 20, 150, 300, 400, 600, 900 and 1200 °C high temperatures corresponding TS EN 1363-1 after an initial 28 day curing period. After heating, the specimens are slowly air-cooled to the room temperature and then E d and f c of concretes were determined. Experimental results are also predicted by constructing models in ANN and FL methods. In the models, the training and testing results have shown that ANN and FL methods have strong potential for predicting the f c and E d of crushed tile concretes exposed to elevated temperatures.

Illuminance-based slat angle selection model for automated control of split blinds

Venetian blinds play an important role in controlling daylight in buildings. Automated blinds overcome some limitations of manual blinds; however, the existing automated systems mainly control the direct solar radiation and glare and cannot be used for controlling innovative blind systems such as split blinds. This research developed an Illuminance-based Slat Angle Selection (ISAS) model that predicts the optimum slat angles of split blinds to achieve the designed indoor illuminance. The model was constructed based on a series of multi-layer feed-forward artificial neural networks (ANNs). The illuminance values at the sensor points used to develop the ANNs were obtained by the software EnergyPlus™. The weather determinants (such as horizontal illuminance and sun angles) were used as the input variables for the ANNs. The illuminance level at a sensor point was the output variable for the ANNs. The ISAS model was validated by evaluating the errors in the calculation of the: 1) illuminance and 2) optimum slat angles. The validation results showed that the power of the ISAS model to predict illuminance was 94.7% while its power to calculate the optimum slat angles was 98.5%. For about 90% of time in the year, the illuminance percentage errors were less than 10%, and the percentage errors in calculating the optimum slat angles were less than 5%. This research offers a new approach for the automated control of split blinds and a guide for future research to utilize the adaptive nature of ANNs to develop a more practical and applicable blind control system.

ARTIFICIAL NEURAL NETWORK APPROACH: AN APPLICATION TO HARMONIC LOAD FLOW FOR RADIAL SYSTEMS

Radial Distribution Systems (RDS) require special load flow methods to solve power flow equations owing to their high R/X ratio. Increasing use of power electronic devices and effect of magnetic saturation cause harmonics in RDS. This paper reports a multi-layer feed forward ANN with error back propagation learning algorithm for the calculation of bus voltages and power loss for different harmonic components. The proposed method is tested upon a 33-bus RDS and the results are reported for various harmonics. Extensive testing of the proposed ANN based approach indicates its viability for harmonic load flow assessment for radial systems.

PREDICTING ENGINEERING FACTORS RELATED TO SOIL AMENDMENTS USING NEURAL NETWORKS

Artificial Neural Network model is used to predict engineering factors [bulk density, hydraulic conductivity, infiltration rate, soil penetration resistance and available water], under three different soil amendments [Bitumen Emulsion, Polyacrylamide and Organic Manure]. Multilayer feedforward ANN with 11 input and 5 output neurons was trained using a backpropagation learning algorithm. The data needed to train and test the ANN model was obtained from previous literatures. The inputs were soil amendments [Bitumen Emulsion (BE), Polyacrylamide (PAM) and Organic Manure (OM)], soil texture [Sand (S), Silt (Si) and Clay (C)], engineering factors [Initial bulk density (IBd), Initial hydraulic conductivity (IKa), Initial infiltration rate (IIr), Initial Soil penetration resistance (ISp) and Initial available water (IAW)]. The outputs were final Bd, Ka, Ir, Sp and AW.The result of Artificial Neural Network model showed that the variations between measured and predicted engineering factors were very small. A field experiment was carried out in the Agricultural Experiment Station of the Desert Research Center at Ras Sidr (رأس سدر), South Sinai Governorate. The experiment studied the effect of three soil amendments on some engineering factors, productivity of sorghum yield and water use efficiency. The results showed that the soil amendments improved the engineering factors in general. Optimum values for productivity and water use efficiency were obtained by applying Organic Manure 23128 Mg/fed and 26.6 kg/m3 respectively. Least values were obtained by applying Polyacrylamide 15559 Mg/fed and 13.14 kg/m3 respectively.

Artificial intelligence applied to floating boom behavior under waves and currents

Floating containment booms are an essential device in the fight against coastal pollution, allowing to contain the pollutant prior to its recovery. An Artificial Intelligence model is developed with the focus on the effective draft, or draft available for containment that a floating boom will provide in open waters. The dataset is obtained through an extensive laboratory campaign in which seven model booms are subjected to numerous wave and current combinations. This dataset is randomly divided into two subsets, one for training, the other for testing or validating the model. Input and output variables are selected based on dimensional analysis and laboratory results. The AI technique chosen for the model is multilayer feedforward artificial neural networks trained with the back-propagation algorithm, for their capability to apprehend higher-order patterns from the training examples and subsequently generalize them to other (validation) cases. In order to find an efficient network architecture, a comparative study involving 640 neural networks is carried out. Having selected the best performing architecture, the model is successfully validated; it is a virtual laboratory allowing to determine the effective draft that a certain boom design will provide under given wave and current conditions.

A multifactorial analysis of obesity as CVD risk factor: Use of neural network based methods in a nutrigenetics context

BackgroundObesity is a multifactorial trait, which comprises an independent risk factor for cardiovascular disease (CVD). The aim of the current work is to study the complex etiology beneath obesity and identify genetic variations and/or factors related to nutrition that contribute to its variability. To this end, a set of more than 2300 white subjects who participated in a nutrigenetics study was used. For each subject a total of 63 factors describing genetic variants related to CVD (24 in total), gender, and nutrition (38 in total), e.g. average daily intake in calories and cholesterol, were measured. Each subject was categorized according to body mass index (BMI) as normal (BMI ≤ 25) or overweight (BMI > 25). Two artificial neural network (ANN) based methods were designed and used towards the analysis of the available data. These corresponded to i) a multi-layer feed-forward ANN combined with a parameter decreasing method (PDM-ANN), and ii) a multi-layer feed-forward ANN trained by a hybrid method (GA-ANN) which combines genetic algorithms and the popular back-propagation training algorithm.ResultsPDM-ANN and GA-ANN were comparatively assessed in terms of their ability to identify the most important factors among the initial 63 variables describing genetic variations, nutrition and gender, able to classify a subject into one of the BMI related classes: normal and overweight. The methods were designed and evaluated using appropriate training and testing sets provided by 3-fold Cross Validation (3-CV) resampling. Classification accuracy, sensitivity, specificity and area under receiver operating characteristics curve were utilized to evaluate the resulted predictive ANN models. The most parsimonious set of factors was obtained by the GA-ANN method and included gender, six genetic variations and 18 nutrition-related variables. The corresponding predictive model was characterized by a mean accuracy equal of 61.46% in the 3-CV testing sets.ConclusionsThe ANN based methods revealed factors that interactively contribute to obesity trait and provided predictive models with a promising generalization ability. In general, results showed that ANNs and their hybrids can provide useful tools for the study of complex traits in the context of nutrigenetics.

Neural network model of 100 W portable PEM fuel cell and experimental verification

The inherent properties of artificial neural networks (ANNs) such as low sensitivity to noise and incomplete information make the ANN a promising candidate to model the fuel cell system. In this paper, an ANN-based model of 100 W portable direct hydrogen fed proton exchange membrane fuel cell (PEMFC) is presented. The model is built based on experimentally collected data from a portable 100 W direct hydrogen fed PEMFC in the authors’ laboratory. A multilayer feedforward ANN with back-propagation training algorithm is used to model the portable PEMFC. The ANN consists of fully connected four layers network with two hidden layers. The PEMFC ANN model is trained using extracted data from experimentally measured and calculated parameters. To validate the model, the outputs of the PEMFC ANN are compared against experimental data and results from a dynamic model of portable direct hydrogen fed PEMFC. In addition, three statistical indices to measure variations, unbiasedness (precision), and accuracy in voltage, power, and hydrogen flow are used to evaluate the PEMFC ANN model performance. The indices indicate that the maximum variations, unbiasedness, and accuracy of the voltage, power, and hydrogen flow are 1.45%, 2.04%, and 1.90%, respectively, which shows a close agreement between the outputs of the PEMFC ANN and the experimental results.

Performance comparison of conventional and wiper ceramic inserts in hard turning through artificial neural network modeling

Hard turning with ceramic tools provides an alternative to grinding operation in machining high precision and hardened components. But, the main concerns are the cost of expensive tool materials and the effect of the process on machinability. The poor selection of cutting conditions may lead to excessive tool wear and increased surface roughness of workpiece. Hence, there is a need to investigate the effects of process parameters on machinability characteristics in hard turning. In this work, the influence of cutting speed, feed rate, and machining time on machinability aspects such as specific cutting force, surface roughness, and tool wear in AISI D2 cold work tool steel hard turning with three different ceramic inserts, namely, CC650, CC650WG, and GC6050WH has been studied. A multilayer feed-forward artificial neural network (ANN), trained using error back-propagation training algorithm has been employed for predicting the machinability. The input–output patterns required for ANN training and testing are obtained from the turning experiments planned through full factorial design. The simulation results demonstrate the effectiveness of ANN models to analyze the effects of cutting conditions as well as to study the performance of conventional and wiper ceramic inserts on machinability.

A low cost INS/GPS navigation system integrated with a multiplayer feed forward neural network

This article investigates the use of a multilayer feedforward artificial neural network into a GPS integrated low cost inertial navigation system based on MEMS sensors. The neural network is applied as an alternative of integration technique, with the purpose of providing better navigation solutions, during the lack of information in GPS outages portions of time. An input-output neural network signals model is proposed, based on a set of simplified terrestrial vehicle navigation equations. Also an adaptive Kalman filter training methodology is tested with real navigation data. Preliminary simulated numerical results are presented, based on urban vehicular positioning application data trials, acquired from low cost Crossbow CD400-200 IMU and an Ashtech Z12 GPS receiver.

A B-spline network based neural controller for power electronic applications

Conventional multi-layer feedforward ANN controllers with back-propagation training are too complex to be implemented in fast-dynamic power electric systems. This paper proposes a controller for power electric systems based on a type of on-line trained neural network called the B-spline network (BSN). Due to its linear nature and local weight updating, the BSN controller is more suitable for real-time implementation than conventional multi-layer feedforward neural controllers. Based on a frequency domain stability analysis, a design methodology for determining the two main parameters of the BSN is presented. The design procedure of the proposed BSN controller is straightforward and simple. Experimental results of UPS inverters with the proposed controller under various conditions show that the proposed controller can achieve excellent performance.

Decodingβ-decay systematics: A global statistical model forβ−half-lives

Statistical modeling of nuclear data provides a novel approach to nuclear systematics complementary to established theoretical and phenomenological approaches based on quantum theory. Continuing previous studies in which global statistical modeling is pursued within the general framework of machine learning theory, we implement advances in training algorithms designed to improve generalization, in application to the problem of reproducing and predicting the half-lives of nuclear ground states that decay $100%$ by the ${\ensuremath{\beta}}^{\ensuremath{-}}$ mode. More specifically, fully connected, multilayer feed-forward artificial neural network models are developed using the Levenberg-Marquardt optimization algorithm together with Bayesian regularization and cross-validation. The predictive performance of models emerging from extensive computer experiments is compared with that of traditional microscopic and phenomenological models as well as with the performance of other learning systems, including earlier neural network models as well as the support vector machines recently applied to the same problem. In discussing the results, emphasis is placed on predictions for nuclei that are far from the stability line, and especially those involved in $r$-process nucleosynthesis. It is found that the new statistical models can match or even surpass the predictive performance of conventional models for $\ensuremath{\beta}$-decay systematics and accordingly should provide a valuable additional tool for exploring the expanding nuclear landscape.

A Novel Algorithm for Translation, Rotation and Scale Invariant Character Recognition

This paper presents a novel algorithm, called Radial Sector Coding (RSC), for Translation, Rotation and Scale invariant character recognition. Translation invariance is obtained using Center of Mass (CoM). Scaling invariance is achieved by normalizing the features of characters. To obtain most challenging rotation invariance, RSC searches a rotation invariant Line of Reference (LoR) by exploiting the symmetry property for symmetric characters and Axis of Reference (AoR) for nonsymmetric characters. RSC uses the LoR to generate invariant topological features for different characters. The topological features are then used as inputs for a multilayer feed-forward artificial neural network (ANN). We test the proposed approach on two widely used English fonts Arial and Tahoma and got 98.6% recognition performance on average.

Modeling and Analysis of Machinability Characteristics in PA6 and PA66 GF30 Polyamides through Artificial Neural Network

The traditional metallic materials are replaced by some applications for turning process in PA6 and PA66 GF30 polyamides due to excellent properties such as high specific strength and stiffness, wear resistance, dimensional stability, low weight and directional properties. The addition of short fibers to the polyamides improves the properties over the unreinforced polyamides. As a result of these improved properties and potential applications in several fields of engineering, there is a need to understand the machining of unreinforced and reinforced polyamides. Selection of cutting tool and process parameters is important in machining of these composites. This article presents the application of artificial neural network (ANN) modeling to assess the machinability characteristics of unreinforced polyamide (PA6) and reinforced polyamide with 30% of glass fibers (PA66 GF30). The effects of process parameters such as work material, tool material, cutting speed, and feed rate on three aspects of machinability, namely, machining force, power, and specific cutting force have been analyzed through a multilayer feed forward ANN. The input—output patterns required for training are obtained through turning experiments planned as per full factorial design. The model analysis revealed that the minimum machining force results at low feed rate and independent of cutting speed, whereas the power is minimal when both the cutting speed and feed rate are at low levels for PA6 and PA66 GF30 polyamides machining irrespective of the cutting tool. On the other hand, the specific cutting force is minimal at low cutting speed and high feed rate in case of PA6 material, whereas high values of cutting speed and feed rate are essential for minimizing the specific cutting force for PA66 GF30 polyamide machining.

Computational intelligence-based congestion prediction for a dynamic urban street network

This paper develops a hybrid model for single point short term traffic flow forecasting in an urban traffic network. The hybrid model consists of two main modules: a fuzzy input fuzzy output filter (FIFO-filter) and a multi-layer feed-forward artificial neural network architecture optimized using evolution strategies (MLFN-ES). The FIFO-filter performs the data clustering operation and provides a rough forecasted prediction value based on the input data to the MLFN-ES associated with each cluster for modeling the input–output relation to provide accurate short term forecast value. The performance of the proposed model is demonstrated by predicting the traffic flow for an intersection in the central business district (CBD) area of Singapore. The hybrid model proposed in this paper gave a mean absolute percentage error (MAPE) of 8.35% on weekdays and 9.73% on weekends for the test data. A comparison analysis shows improved performance of the proposed hybrid method in short term traffic prediction over popular approaches like ARIMA and artificial neural network based systems.

Incorporating Forecasts of Rainfall in Two Hydrologic Models Used for Medium-Range Streamflow Forecasting

This study reports on the performance of two medium-range streamflow forecast models: (1) a multilayer feed-forward artificial neural network; and (2) a distributed hydrologic model. Quantitative precipitation forecasts were used as input to both models. The Furnas Reservoir on the Rio Grande River was selected as a case study, primarily because of the availability of quantitative precipitation forecasts from the Brazilian Center for Weather Prediction and Climate Studies and due to its importance in the Brazilian hydropower generating system. Streamflow forecasts were calculated for a drainage area of about 51,900 km(2), with lead times up to 12 days, at daily intervals. The Nash-Sutcliffe efficiency index, the root-mean-square error, the mean absolute error, and the mean relative error were used to assess the relative performance of the models. Results showed that the performance of streamflow forecasts was strongly dependent on the quality of quantitative precipitation forecasts used. The artificial neural network (ANN) method seemed to be less sensitive to precipitation forecast error relative to the distributed hydrological model. Hence, the latter presented a better skill in flow forecasting when using the more accurate perfect precipitation forecast. The conceptual hydrological model also demonstrates better forecast skill than ANN models for longer lead times, when the representation of the rainfall-runoff process and of the water storage in the watershed becomes more important than the flow routing along the drainage network. In addition, results obtained by incorporating a quantitative precipitation forecast in both models performed better than the current streamflow obtained by the Brazilian national electric system operator using statistical models which do not utilize information on precipitation, whether observed or forecast. (Less)

Analysis of drought and storage for Mula project using ANN and stochastic generation models

A hybrid model for streamflow generation is presented to explore the possibilities of using the multilayer feedforward artificial neural networks (ANNs) as generators of future scenarios, with emphasis on the ability to reproduce the statistics of flows related to drought and storage. The artificial neural network model has two components: deterministic and random. The second part of the model incorporates the uncertainty associated with the hydrological processes. The model is applied to the monthly inflows of Mula irrigation project in Maharashtra, India. A comparison of drought and storage among other statistics was made between the performance of the ANN-based model results and the results of the Thomas–Fiering models. The results show that ANN is a promising alternative modelling approach for flow simulation purposes, with interesting potential in the context of water resources systems management and optimization.

Hybrid Neural Network-Based Fault Diagnosis and Fault-Tolerance Design with Application in Electro-Hydraulic Servovalve

In this study, to cope with the needs of the predictive maintenance for complex systems, a hybrid dynamic Arti- ficial Neural Network (ANN) based fault and degradation diagnosis and tolerance method is designed. The multi-layer feed forward ANN and recurrent ANN are combined, so as to be able to form a dynamic identification model for the non- linear time-varying system. It has three work modes, and can perform the fault and degradation diagnosis and tolerance by using these modes alternately. The result of its application in an Electro-Hydraulic Servovalve of a Hydroelectric Genera- tion Unit shows that it is effective and feasible, has the advantages of the simple and fast algorithms, working online, and no disturbance signals importing to the system.

ANN based reactive power control of isolated wind-diesel-micro-hydro hybrid power systems

This paper develops a mathematical modelling of isolated wind-diesel-micro-hydro hybrid power system and presents an artificial neural network (ANN) based approach to tune the parameters of static var compensator (SVC) to meet the reactive power requirement of hybrid system. In the hybrid system considered, a synchronous generator is connected to a diesel-generator (DG) and induction generators connected to the wind and micro-hydro system. The system also has a SVC to provide the required reactive power in addition to the reactive power generated by the synchronous generator. The multi-layer feed-forward ANN with the error back-propagation training is employed to tune the gain of SVC. The dynamic responses presented show that SVC tuned by the ANN can provide optimum dynamic performance of the hybrid power system over a wide range of typical load models.

Development and testing of a prototype reflex measurement system employing artificial neural networks

This paper presents the development, testing and performance evaluation of a patellar tendon reflex measurement system to provide a quantitative reflex evaluation for use by medical practitioners and in a telemedicine or E-medicine environment. A prototype was developed that makes use of XSens MTx orientation sensors, force-sensitive resistors and an electromyogram to measure the reflex response. Suitable parameters from the sensors were identified for analysis, and clinical testing was performed on 20 subjects to collect data to evaluate the system’s performance. Subjective reflex evaluations were conducted by three medical doctors according to a standard reflex grading scale using video recordings of the tests. Multi-layer feed-forward (MLFF) artificial neural networks (ANNs) were used to analyze the collected data with the aim of pattern identification and reflex grading prediction. It was found that the MLFF network delivered the corresponding reflex grading with an accuracy of 85%, which was of the same order as the rate of differences between the subjective reflex evaluations performed by the doctors (80%). The use of ANNs to analyze a reflex measurement offers a repeatable and concise representation of the reflex that is familiar to doctors and can be developed for use in a general clinical setting or for telemedicine purposes.