group method of data handling (gmdh)

Low impedance differential relays are widely used in the protection systems of power transformers. While being highly reliable, differential relays can misidentify the inrush currents generated during the switching of power transformers as faults and issue a tripping command when one is not needed. Therefore, these protection systems need a mechanism to differentiate between inrush currents and faults in order to prevent unnecessary activation. Accordingly, this paper presents a new method based on a group method of data handling (GMDH) neural network for differentiating faults from inrush currents. The proposed method can quickly detect a wide variety of faults that may occur simultaneously with inrush currents and is perfectly noise-resistant. The proposed method is compared with the conventional methods used in the industry, namely second harmonic and zero-crossing methods. The results demonstrate the ability of the proposed method to outperform conventional methods under a wide variety of operating conditions.

Utilizing rubber shreds in civil engineering industry such as geotechnical structures can accelerate generated waste tire recycling process in an economical and environmentally friendly manner. However, understanding the rubber grains strength parameters is required for engineering designs and can be acquired through experimental tests. In this study, small and large direct shear test was implemented to specify shear strength parameters of five rubber grains group which are different in gradation and size. Moreover, artificial neural networks (ANN) are developed based on the test results and optimized networks which best captured the shear stress (τ), and vertical strain (εv) behavior of rubbers, are introduced. Additionally, a prediction model using the combinatorial algorithm in group method of data handling (GMDH) is proposed for the shear strength and vertical strain in the arrangement of closed-form equations. The performance and accuracies of the proposed models were checked using correlation coefficient (R) between the experimental and predicted data and the existing mean square error (MSE) was evaluated. R-values of the modeled τ and εv are equal to 0.9977 and 0.9994 for ANN, and 0.9862 and 0.9942 for GMDH models, respectively. The GMDH proposed models are presented as comparatively simple explicit mathematical equations for further applications.

Sediment transport is a commonly occurring vital process in fluvial and coastal environments, and “incipient motion” is an issue inseparably bound to this topic. In this study, the existing equations were first examined by use of three different sources of experimental data through statistical indexes. The powerful method of Group Method of Data Handling (GMDH) was used for estimating the densimetric Froude number (Fr) and the Singular Value Decomposition (SVD) method was used to compute the leaner coefficient vectors. A novel equation is proposed, devolved through utilizing a combination of coding methods including GMDH and genetic algorithms. The studies conducted indicate that the presented equation is fairly accurate (RMSE= 0.19 & MAPE= 7%) in predicting incipient motion.

Sedimentation is a problematic issue concerningsewer system design. In order to reducesediment deposition in sewersystems, two new equationsarepresented with asmoothing function and Group Method of Data Handling (GMDH) to estimateminimum flowvelocity. For this purpose, dimensional analysis is used to determine the factors affecting sediment transport at limit of deposition. These factors are categorized in five different groups: transport, transport mode, flow resistance, sediment and motion. Six different models are presented for predicting the densimetric Froude number (Fr) using the smoothing function and GMDH. Themodels presented withthese two methods are compared with existing equations. The results indicate that the equations proposedusing thesmoothingfunction (MAPE= 5.05, RMSE= 0.24 & AIC= -43.04) and GMDH (MAPE= 5.39, RMSE= 0.3 & AIC= 72.78) are more accurate than existing models. Furthermore, sensitivity analysis is performed to examine the efficacy of each of the dimensionless parameters presented by the best model in estimating Fr.

Choice modelling is valuable for understanding and predicting customer behaviour. This study introduces the Group Method of Data Handling (GMDH) into choice modelling and applies this new technique to model consumer choice in the longdistance communication market. When we compare the GMDH with the Arti cial Neural Network (ANN) and logit models, the results show that the new model provides better predictions of customer choice than the ANN and logit models. In addition, the new model can identify the important explanatory variables that a ect customer choice, and reveal how the variables a ect this choice, which cannot be directly accomplished using the ANN model. This advantage will help rms to better analyse the behaviour of their customers and, thereby, develop suitable marketing strategies.