Data Handling Type Neural Network Research Articles

Numerical examination of exergy performance of a hybrid solar system equipped with a sheet-and-sinusoidal tube collector: Developing a predictive function using artificial neural network

Integrating cooling systems with photovoltaic-thermal (PVT) collectors has the potential to mitigate the exergy consumption in the building sector due to their capability for simultaneous power and thermal energy generation. The simultaneous utilization of nanofluid and geometry modification resulted in a synergetic enhancement in the performance of PVTs and thereby reducing their sizes and costs. In addition, there is still a lack of high accurate predictive model for the estimation of the performance of PVTs at a given Re number and nanofluid concentration ratio to be used in engineering design for the further product commercialization. To this end, the current numerical study investigates the exergy electricity, thermal, and overall exergies of a building-integrated photovoltaic thermal (BIPVT) solar collector with Al2O3/water coolant. The increase in nanoparticle concentration (ω) from 0 % to 1 % increased the useful thermal exergy and overall exergy efficiency (Exu,t/ Υov) by 0.3999 %/0.0497 %, 1.3959 %/0.2598 %, and 0.7489 %/0.1771 % at Re numbers of 500, 1000, and 1500, respectively, while Exu,t/ Υov exhibited a reducing trend at Re = 2000; 0.3928 %/0.1056 % decrease. In addition, the increase in ω from 0 % to 1 % caused the useful electricity and electrical exergy (Exu,e/ Υe) to be diminished by 0.0060 %/0.0025 % at Res 500 and 1000, and to be escalated by 0.0113 %/0.0055 % at Res of 1500 and 2000. Meanwhile, the Re augmentation, from 500 to 2000, improved the Exu,t, Exe, Υe, and Υov by 60 %, 1.26 %, 1.26 %, and 17.50 %, respectively, at different ω s. In addition, two functions were developed and proposed by applying a group method of data handling-type neural network (GMDH-ANN) to forecast the value of Υov based on two input values (Re and ω). The results showed high accuracy of the proposed model with MSE, EMSE, and R2 of 0.0138, 0.1143, and 0.99785, respectively.

MATHEMATICAL MODELLING OF THE CENTRIFUGAL COMPRESSOR

The paper is aimed to create the mathematical model of the centrifugal compressor based on the group method of data handling-type neural networks to determine the compressor volumetric flow rate as the dependence on the centrifugal compressor’s technological parameters (the rotor’s angular velocity, the compressor’s inlet and outlet temperatures, the compressor’s inlet and outlet pressures, the atmospheric pressure). It is the important scientific task, because most centrifugal compressors used in the process industry don’t have equipment needed to measure the volumetric flow rate. It does not allow to estimate the compressor’s technical state during its operation. Verification of the developed model has been performed, based on the 336 data points (collected from the field measurements) and with using the centrifugal compressor of natural gas (16ГЦ2-395/53-76C) of Dolyna linear production administration of gas transmittal pipelines. The test results have been showed the adequate efficiency of the mathematical model. Keywords: volume flow, centrifugal supercharger, mathematical model, method of group consideration of arguments, neural networks, technological parameters, correlation coefficient.

Performance of Radial Basis Function and Group Method of Data Handling-type Neural-Networks in Flammability Characteristics Prediction of PMMA

Performance of Radial Basis Function and Group Method of Data Handling-type Neural-Networks in Flammability Characteristics Prediction of PMMA

Formulating a novel combined equation for coal calorific value estimation by group method data handling type neural network

ABSTRACT The accurate estimation of the calorific value of coal with the help of models is very important for the design and operation of coal-based technologies. This study proposes a new combined estimation equation for the calorific value of coal. The modeling process in this study consists of two stages. In Stage-1, a high-performance gross calorific value (GCV) prediction equation is developed using the Group Method Data Handling type Neural Network (GMDH-type NN) method based on proximate analysis components. In addition, by comparing the performance of the developed GCV equation with the performance of the model equations in the literature, three model equations providing the best estimation performance are determined. In Stage-2, a combined equation for GCV prediction is formed based on GMDH-type NN. This combined equation uses the estimation results of the best three equations as model inputs, determined in Stage-1. A large dataset consisting of 8501 coal samples (as-received basis) is used in the development and testing of GCV estimation equations. According to the results obtained from estimation equations, the estimation equation developed in Stage-1 provides better estimation performance than the equations in the literature. It is found that the combined equation developed in Stage-2 improves considerably the prediction performance values.

Probabilistic and Deterministic Assessment of Liquefaction Potential Using Piezocone Data

This research provides two probabilistic and deterministic methods to assess the liquefaction potential based on the piezocone (CPTu), and then a procedure is proposed to develop the seismic fragility curves of soil liquefaction. For this purpose, a reliable liquefaction case history dataset based on CPT and CPTu tests is compiled. Next, a deterministic correlation of the liquefaction potential is proposed by the use of the Group Method of Data Handling type neural network optimized with genetic algorithms. Data mining techniques are then used to compare the results of the various models, including the proposed deterministic model. The results demonstrate that an improvement in statistical measures of performance of the proposed model has been achieved. Through this study, a sensitivity analysis is carried out to clarify the more properties of the proposed method. Finally, a new probabilistic function is expanded based on the results of the deterministic approach. The developed probabilistic function is utilized as a proper tool to evaluate the fragility curves of liquefaction occurrence.

A novel framework for approximation of magneto-resistance curves of a superconducting film using GMDH-type neural networks

Vortex behavior, in particular magneto-resistance curves, has been vastly studied and discussed in the literature, often for the purpose of observing the effect of varying period of antidots on superconducting films. It has been shown that by decreasing the period of samples, the number of matching fields increases, and energy losses in nano-engineered thin films may be minimized. While the importance of studying magneto-resistance curves is well researched and understood, means to ease the procedure of obtaining these measurements has somewhat been overlooked. In this work, we motivate to use approximation techniques to extrapolate − instead of incessantly measuring − magneto-resistance characteristics, and propose an entire framework for this purpose. The latter exploits a machine learning method, called the Group Method of Data Handling type neural networks, which is known to be capable of solving complex, nonlinear problems. Our simulation results show that the proposed technique yields mean-squared error in the range of 10−8 when compared to the measured curves.

Mathematical modeling of ethylene polymerization over advanced multisite catalysts: an artificial intelligence approach

Recent developments in multisite catalysts based on metallocenes and post-metallocenes attracted the attention of researchers and industrial petrochemical companies due to the production of high-performance polymeric materials which generally are not achievable based on Ziegler–Natta catalysts. In this study, with the aim of predicting the average molecular weight of produced polyethylene and activity of ethylene polymerization using multisite catalysts, robust precise models based on artificial neural networks are developed. The average error for the prediction of the average molecular weight and activity are 3.76% and 5.89%, respectively. The Leverage method was used to check the reliability of the proposed model and the quality of experimental data which have been used for model development. The results showed that just a few data points are outside of the applicability domain of the developed models, confirming that both developed models and their predictions are statistically correct. Comparison of the artificial neural network models with other artificial intelligence approaches including support vector machine and group method of data handling type neural networks illustrates the better performance and robustness of the proposed models. The results of this study promise that neural networks can be used as reliable models with reasonable accuracy to estimate the performance of ethylene polymerization over this type of new metallocene/post-metallocene multisite catalysts.

Data-driven modeling of interfacial tension in impure CO2-brine systems with implications for geological carbon storage

Global climate change is a pressing problem, and given the current state of development of renewable energies, Carbon Capture and Storage (CCS) appears a promising solution. Deep saline aquifers are the most accessible geological locations for safe CO2 sequestration, where CO2 can be securely trapped mostly by structural and capillary trapping mechanisms. The efficiency of both mechanisms is governed by multiphase flow characteristics that are heavily dependent on interfacial tension (IFT) between the gaseous and aqueous phases. In this study, we conduct a comprehensive data-driven study on the interfacial tension of pure and impure gas-brine mixtures within saline aquifers. We carefully collect 2517 experimental data, upon which we develop seven machine-learning based models for predicting the IFT. These include, namely, four differently optimized multilayer perceptron models, a radial basis function neural network, a least squares support vector machine, and a group method of data handling-type neural network. The three most accurate intelligent models are subsequently incorporated into a state-of-the-art committee machine intelligent system. We use a pseudo (average) critical temperature variable to capture the gas phase impurities while developing a general-in-purpose predictive tool for any gas mixture. We comprehensively evaluate the accuracy and statistical validity of our models, and demonstrate their robustness relative to major empirical correlations currently in use. The analyses reveal the non-trivial behavior of IFT against various parameters. Finally, we elucidate the impact of gas and brine impurities on the storage capacity of saline aquifers at various in-situ conditions.

Effect of the magnetic field on the heat transfer coefficient of a Fe3O4-water ferrofluid using artificial intelligence and CFD simulation

A ferrofluid is a magnetic fluid which is composed of magnetic nanoparticles with the size of 5-15nm immersed in a base fluid (such as water, oil, etc.). Although the amount of thermal conductivity of the magnetic nanoparticles is lower than that of metallic and metallic oxide nanoparticles, their constructability by magnetic field makes them ideal to be used in heat transfer applications. In this study, the heat transfer coefficient (HTC) of the Fe3O4 nanoparticles dispersed in water under constant and alternating magnetic field is investigated by artificial intelligence methods and CFD simulation. Multilayer feed-forward neural network, group method of data handling type neural network, support vector regression model and adaptive neuro-fuzzy inference system are developed to predict the HTC of the Fe3O4-water ferrofluid under magnetic field. Volume fraction of nanoparticle, intensity of the magnetic field, frequency of the magnetic field, Reynolds number and dimensionless distance of the tube are selected as input variables of the networks and the HTC is selected as output variable of the network. The results show that artificial intelligence methods can successfully predict the target with very good accuracy.

Application of Group Method of Data Handling Type Neural Network for Significant Wave Height Prediction

The estimation of wave parameters is of great importance in coastal activities such as design studies for harbor, inshore and offshore structures, coastal erosion, sediment transport, and wave energy estimation. For this purpose, several models and approaches have been proposed to predict wave parameters, such as empirical, numerical-based approaches, and soft computing. In this study, the group method of data handling type neural network (GMDH-NN) was presented for significant wave height prediction in an attempt to suggest a new model with superior explanatory power and stability. The GMDH-NN results were compared with the field data and with a multilayer perceptron neural networks (MLPNN) model. The results indicate that the prediction accuracy and avoidance of over-fitting of the GMDH-NN method were superior to those of the MLPNN method. The percentage improvement in the root mean square error and the mean absolute percentage error of the GMDH-NN model over the MLPNN model were 72.92% and 81.02%, respectively. Also, according to the indices, the GMDH-NN model performs the best for predicting the Hs of all of the wave height ranges. That is, the GMDH-NN model is capable of predicting wave heights for different ranges. The results of the analysis suggest that the GMDH-NN-based modeling is effective in predicting significant wave height.

Modeling of a single cell micro proton exchange membrane fuel cell by a new hybrid neural network method

In this research, a novel hybrid model is developed based on neural network with high capability to predict the behavior of a micro proton exchange membrane fuel cell (micro-PEMFC) at various operational conditions. The proposed model is a combination of Group Method of Data Handling type neural networks and Genetic Algorithm (GMDH-GA). Genetic algorithm was used to optimize the correlation parameters to improve the accuracy of model. Input variables including humidity, temperature, electrical current and the oxygen and hydrogen flow rates were considered to predict cell performance. First, the GMDH-GA model was trained using the experimental input and output data set, then the trained model was tested using an independent data set. The model was implemented to determine the performance curves of a micro-PEMFC at different operating settings. The model could obtain the optimized values for the input variables corresponding to the value of objective function. Results showed a consistency between experimental data and the data made by the model. Therefore, it is indicated that the GMDH-GA method is an effective method, which can predict the performance of micro-cell with high accuracy.

Elasticity modulus of clayey deposits estimation using group method of data handling type neural network

Elasticity modulus (Es) is used to predict the elastic settlement of foundations. Plate load test (PLT) is one of the most commonly applied methods to determine this parameter. Given that the method is relatively costly and time-consuming, especially in depths, various empirical equations have been proposed to relate the Es to the results of standard penetration test (SPT) which is one of the most frequently used tests during the geotechnical investigation. Considering that the existing relationships are not able to estimate Es properly, in the present research, Group Method of Data Handling (GMDH) type neural network (NN) is used to estimate the Es of clayey deposits based on a database including 131 plate load test (PLT) and standard penetration test (SPT) from geotechnical investigation sites in Qazvin, Iran. The comparison between proposed models and existing empirical equations indicate that the GMDH models exhibit higher performance (32–42% improvement) with respect to the other correlations. The sensitivity analysis of the developed models has been conducted to determine the contributions of the parameters affecting Es and shows that N60 is the most influential parameter on the GMDH-based models to predict Es. Also, it has been demonstrated that the Es predicted by the best model is significantly affected by error in the measurement of the LL value. Therefore, measuring this parameter in the laboratory requires high accuracy.

Modeling and prediction of metallic powder behavior in explosive compaction process by using genetic programming method based on dimensionless numbers

Explosive compaction process of metallic powders has been studied using a semi-empirical method. This method utilizes dimensional analysis along with genetic programming approach to obtain an expression relating the final density of compacts to the effective parameters during compaction process such as shock compaction energy, properties of metallic powder, and geometry of the problem and explosive charge. Dimensionless numbers have been constructed based on the effective parameters using a complete set of input–output experimental data. The obtained dimensionless numbers then have been applied as input–output data pairs for genetic programming optimization process considering modeling error as the objective. The obtained results show that the proposed model using dimensional analysis method along with genetic programming can predict the final density of compacts with 99.8% accuracy. Also, the outputs of the proposed model have been compared with those obtained by group method of data handling type neural network in the literature. Consequently, genetic programming method has much less root mean square error than group method of data handling model and can be successfully used for modeling and prediction of the complex process behavior.

Evolutionary design of generalized group method of data handling-type neural network for estimating the hydraulic jump roller length

Hydraulic jumps generally occur subsequent to structures such as ogee spillways, control gates, and weirs. The jump roller length is considered one of the main hydraulic jump parameters. In this study, the roller length of a hydraulic jump on a rough channel bed is predicted using a novel, evolutionary, generalized structure design of a group method of data handling (GS-GMDH)-type neural network. The topology of GMDH is designed with a genetic algorithm . Initially, the three most important non-dimensional parameters affecting hydraulic jump roller length, including the Froude number upstream of a hydraulic jump \(\left( {Fr} \right) \), the ratio of sequent depths \(\left( {{h_2 }/{h_1 }} \right) \), and the relative roughness \(\left( {{ks}/{h_1 }} \right) \) were used to generate four different GS-GMDH models, and the most accurate model is identified. The best new GS-GMDH model prediction statistics, including RMSE, MARE, and correlation coefficient are 1.816, 0.081, and 0.966, respectively, while the scatter index and BIAS values are 0.084 and 1.45, respectively. A partial derivative sensitivity analysis of the input parameters for the new model is also performed. The new model predictions are then compared with predictions of a number of other models. The superior performance of the new GS-GMDH over these existing models is illustrated.

Pressuremeter Modulus and Limit Pressure of Clayey Soils Using GMDH-Type Neural Network and Genetic Algorithms

Pressuremeter modulus ( $$E_{M}$$ ) and limit pressure ( $$P_{L}$$ ) are used for the calculation of the settlement and bearing capacity of foundation respectively. As the determination of these parameters from pressuremeter test (PMT) is relatively time-consuming and expensive, various empirical correlations have been proposed to correlate the $$E_{M}$$ and $$P_{L}$$ to other soil parameters. For the existing equations are incapable of estimating these PMT parameters well, in present research group method of data handling type neural network is used to estimate the $$E_{M}$$ and $$P_{L}$$ of clayey soils. The $$E_{M}$$ and $$P_{L}$$ were modeled as a function of three variables including the moisture content ( $$\omega$$ ), plasticity index and corrected SPT blow counts ( $$N_{60}$$ ). A database containing 51 data sets have been used for training and testing of the models. The performances of proposed models are compared with those of existing empirical equations. The results demonstrate that appreciable improvement with respect to the other correlations has been achieved. At the end, sensitivity analysis of the obtained models has been performed to study the influence of input parameters on model outputs and shows that the $$N_{60}$$ is the most influential parameter on the PMT parameters.

New neural network-based response surface method for reliability analysis of structures

In this paper, a new algorithm is introduced for reliability analysis of structures using response surface method based on a group method of data handling-type neural networks with general structure (GS-GMDH-type NN). A multilayer network of quadratic neurons, GMDH, offers an effective solution to modeling nonlinear systems without an explicit limit state function. In the proposed method, the response surface function is determined using GMDH-type neural networks. This is then connected to a reliability method, such as first-order or second-order reliability methods (FORM or SORM) or Monte Carlo simulation method to predict the failure probability (Pf). In the proposed method, the use of the GMDH-type neural network with general structure, where all neurons from previous layers are used to produce neurons in the new layer, can improve the limit state function. In addition, the structure of the neural network and its weight are simultaneously optimized by genetic algorithm and singular value decomposition. As a result, the obtained model has no significant error, despite its simplicity. Moreover, the obtained limit state function is explicit and allows direct use of FORM and SORM methods. To determine the accuracy and efficiency of the proposed method, four numerical examples are solved and their results are compared to other conventional methods. The results show that the proposed method is simply applicable to analyzing the reliability of large complex and sophisticated structures without an explicit limit state function. The proposed approach is a high accurate method that can significantly reduce computing time compared with direct Monte Carlo method.

Generalized models for predicting the critical properties of pure chemical compounds

Critical properties of chemical compounds are of great importance as they are necessary for estimating large variety of thermodynamic and physical properties by equations of state or theorem of corresponding states. These properties are important in optimum design, operation and control of every piece of equipment in petroleum and chemical industries. In this communication, four intelligent models including radial basis function (RBF) and multilayer perceptron (MLP) neural networks, least square support vector machine (LSSVM) and hybrid group method of data handling type neural network (GMDH-PNN) are used to develop predictive models for estimation of the critical properties of pure compounds including critical temperature, critical pressure, and critical volume. A comprehensive data set of 563 pure compounds from 13 different chemical groups has been used to develop rigorous and reliable predictive models. Three easily measurable properties (normal boiling point, molecular weight, and specific gravity) were chosen as input parameters of the proposed models. The results illustrate that all of the developed models have high accuracy in predicting the critical properties, while the developed RBF models are the most accurate ones and GMDH-PNN models are the simplest models in the form of simple mathematical correlations. Comparison of the proposed models in this study with the previously developed models shows the high superiority of the proposed models. The influence degree of each input parameter on the models' outputs was investigated quantitatively using the relevancy factor. Finally, the Leverage approach was used to find out applicability domain and probable outliers of the developed models. It was found that only small percent of the data are located out of the applicability domain of the developed models.

Modeling and optimization of dew-point evaporative coolers based on a developed GMDH-type neural network

A precise model of a counter-flow indirect dew-point evaporative cooler was developed using the group method of data handling-type neural network while the network was trained by extracted data from a validated numerical model. After validating the model, it was employed in a multi-objective optimization problem that implements the non-dominated sorting genetic algorithm-II method. The system was optimized in diverse climatic conditions. In this regard, Yazd, Masjed-Soleiman and Ahvaz were chosen as representatives of cities with hot, hot semi-humid and hot-humid climates in Iran. In each city, optimum values of channel length, channel gap, inlet air velocity and return to intake air ratio were found so that these decision variables maximize the average coefficient of performance and minimize the specific area of the cooler, simultaneously. The results indicated that the developed model can predict the supply air temperature accurately with less than 1°C error and due to its quick calculation process it is possible to optimize the design based on hourly climate data without any needs to very fast processors. Moreover, using the optimization, the coefficient of performance and specific area for the system designed to be used in Yazd were improved 36.3% and 30.9%, respectively. This figure at Masjed-Soleiman and Ahvaz was 16% and 7.92% improvement in the specific area at the cost of 2.63% and 2.19% reduction in the coefficient of performance, respectively. These improvements allowed the system to reach its full potential and making dew-point evaporative coolers as a suitable cooling system in diverse climatic conditions.

Compressibility indices of saturated clays by group method of data handling and genetic algorithms

Compression index (C c) and recompression index (C r) are used to estimate the consolidation settlement of fine-grained soils. As the determination of these indices from oedometer test is relatively time-consuming, in present research group method of data handling-type neural network optimized using genetic algorithms is used to estimate the compressibility indices (C c and C r) of saturated clays. C c and C r were modeled as a function of three variables including the initial void ratio (e 0), liquid limit (LL) and specific gravity (G s). Three hundred data sets collected from multiple sites in the province of Mazandaran, Iran, were used for the training and testing of the models. The predicted compressibility indices were compared with those of experimentally measured values to evaluate the performances of the proposed models. The results showed that appreciable improvement toward other correlations has been achieved. At the end, sensitivity analyses of the obtained models were carried out to evaluate the influence of input parameters on model outputs and showed that e 0 and LL are the most influential parameters on C c and C r, respectively. Also, it has been demonstrated that the compressibility indices predicted by models are considerably influenced by changing measured G s (uncertainty). In other words, the mean absolute percent error values increase greatly by G s variation. Therefore, it needs more accuracy to measure this parameter in the laboratory.

Modeling and multi-objective optimization of an M-cycle cross-flow indirect evaporative cooler using the GMDH type neural network

A model was presented to determine product air properties of dew-point indirect evaporative coolers with cross flow heat exchanger, M-cycle CrFIEC. In this regard, the most powerful statistical method known as the group method of data handling-type neural network (GMDH) was employed. Then the developed GMDH model was implemented for multi-objective optimization of a prototype CrFIEC and the average annual values of coefficient of performance (COP) and cooling capacity (CC) were maximized, simultaneously, while working to air ratio (WAR) and inlet air velocity were decision variables of optimization. Accordingly, features of the proposed system were optimized at twelve diverse climates of the world based on Koppen–Geiger's classification. Results implied that the optimized inlet air velocity for all climates varied between 1.796 and 1.957 m.s−1, while the optimum WAR was 0.318 for “A” class cities. Moreover, the mean values of the COP and CC were improved 8.1% and 6.9%, respectively.