Generalized Regression Neural Network Model Research Articles

Vaccine rate forecast for COVID-19 in Africa using hybrid forecasting models.

The public health sectors can use the forecasting applications to determine vaccine stock requirements to avoid excess or shortage stock. This prediction will ensure that immunization protection for COVID- 19 is well-distributed among African citizens. The aim of this study is to forecast vaccination rate for COVID-19 in Africa. The method used to estimate predictions is the hybrid forecasting models which predicts the COVID-19 vaccination rate (CVR). HARIMA is a hybrid of ARIMA and the Linear Regression model and HGRNN is a hybrid of Generalized Regression Neural Network (GRNN) and the Gaussian Process Regression (GPR) model which are used to improve predictive accuracy. In this study, standard and hybrid forecasting models are used to evaluate new COVID-19 vaccine cases daily in May and June 2021. To evaluate the effectiveness of the models, the COVID-19 vaccine dataset for Africa was used, which included new vaccine cases daily from 13 January 2021 to 16 May 2021. Root Mean Squared Error (RMSE) and Error Percentage (EP) are used as evaluation measures in this process. The results obtained showed that the hybrid GRNN model performed better than the hybrid ARIMA model. HGRNN model provides accurate daily vaccinated case forecast, which helps to maintain optimal vaccine stock to avoid vaccine wastage and save many lives.

Assessment of Using Artificial Neural Network and Support Vector Machine Techniques for Predicting Wave-Overtopping Discharges at Coastal Structures

Coastal defence structures play a crucial role in protecting coastal communities against extreme weather and flooding. This study investigates artificial neural network-based approaches, such as multilayer perceptron neural network (MPNN), cascade correlation neural network (CCNN), general regression neural network (GRNN), and support vector machine (SVM) with radial-bias function for estimating the wave-overtopping discharge at coastal structures featuring a straight slope ‘without a berm’. The newly developed EurOtop database was used for this study. Discriminant analysis was performed using the principal component analysis method, and Taylor diagram visualisation and other statistical analyses were performed to evaluate the models. For predicting wave-overtopping discharge, the GRNN yielded highly accurate results. As compared to the other models, the scatter index of the GRNN (0.353) was lower than that of the SVM (0.585), CCNN (0.791), and MPNN (1.068) models. In terms of the R-index, the GRNN (0.991) was superior to the SVM (0.981), CCNN (0.958), and MPNN (0.922). The GRNN results were compared with those of the previous models. An in-depth sensitivity analysis was conducted to determine the significance of each predictive variable. Furthermore, sensitivity analysis was conducted to select the optimal validation method for the GRNN model. The results revealed that both the validation methods were highly accurate, with the leave-one-out validation method outperforming the cross-validation method by only a small margin.

Machine learning techniques for estimating wave-overtopping discharges at coastal structures

Inspired by the development of machine-learning methods, this study employed artificial neural-network-based techniques using different algorithms—such as the multilayer perceptron, cascade correlation, and general regression neural networks (GRNN), and a support vector machine with radial-bias functionality—to estimate the wave-overtopping discharge at coastal structures featuring a straight slope “without a berm.” The newly developed EurOtop database was used in this study. The predictive performance of each model was assessed using six statistical features. In terms of predicting the wave-overtopping discharge, the GRNN yielded highly accurate results. The obtained results were compared with those of the previous models. An in-depth sensitivity analysis was conducted to determine the significance of each predictive variable. Moreover, sensitivity analysis was also conducted to select the optimal validation method for the GRNN model. The results showed both validation methods to be highly accurate, with the leave-one-out validation method outperforming the cross-validation method by a small margin.

QSPR-based model extrapolation prediction of enthalpy of solvation

Enthalpy of solvation (ΔHsolv) is an important thermodynamic parameter in chemical, biological and environmental science. In the current submission a quantitative structure–property relationship (QSPR) model was developed for a large data set of 6106 enthalpies of solvation in 68 solvents, by applying generalized regression neural network (GRNN) using Dragon descriptors for both the solutes and solvents. The optimal GRNN model with the smoothing factor σ = 0.16 was based on 3082 enthalpies of solvation in 34 solvents in the training set, and validated with 3024 enthalpies of solvation in other 34 solvents in the test set. The optimal GRNN model has correlation coefficients of 0.978 and 0.943, and root mean square errors of 4.153 and 6.088 kJ/mol for the training and test sets. In predicting the enthalpies of solvation, the optimal GRNN model is comparable to other published QSPR models reported in the literature, however, the number of ΔHsolv in the test set is 4 times more than that from the latest reported literature study predicting enthalpies of solvation (3024 to 632).

Evaluation of skid resistance of asphalt pavement covered with aeolian sand in desert areas

Particulate contaminants adhering to a desert road can significantly affect the contact characteristics of tires and pavements. The effects of sand quantity, temperature and wearing on the skid resistance were analyzed through the laboratory tests. Image analysis and generalized regression neural networks (GRNN) were used to obtain the macrostructure, followed by reconstructing a three-dimensional model and predicting the texture depth of asphalt pavements. Finally, decay prediction models of skid resistance were established. The results show that the 3D model reconstructed by digital images can clearly show the sand distribution on the road. Furthermore, the GRNN model was found to be reliable, with an average relative error of only 3.4%. Aeolian sand amount, temperature and wearing cycles all affect the skid resistance, while the sand has the greatest influence. The skid resistance model can predict the friction capacity well and it provides a reference for determining the maintenance time for asphalt pavements in desert areas.

Study of color reproduction in pigment digital printing

In pigment ink digital printing, the color difference between the original and the printed product can be visible due to the varied color gamut of devices and the impact of fabric characteristics on the color-presenting effect. To improve the color reproduction accuracy of digital printing, two methods, the full-process optimized (FPO) method and the neural network aided original adjustment (NNAOA) method, were proposed to reduce printing color difference during the color-transferring process. The FPO method, based on accounting for the fabric type and the characteristics of the printing machine, was proposed by optimizing the printing parameter settings in the color management process. Here, the optimal single-channel ink limit was determined and dot gain was compensated using the inverse function correction method, and then the International Color Consortium profile was generated. After using the FPO method, the color difference between the printed sample and the original was reduced from 20.8 to 5.254 compared with performing the original color management. To reduce the reliance on personal experience in color adjustment, the NNAOA method was proposed and implemented for 20 color patches with the largest color differences in the FPO method, allowing the color of the originals to be adjusted with the establishment of the general regression neural network model. This method predicted the input CMYK values of the original using measured L*a*b* values for the printed sample. The results showed that the color difference of the 20 color patches was reduced from 8.54 to 6.60. Consequently, both methods offered distinguishable improvements in the color reproduction effect.

Data-Driven Exhaust Gas Temperature Baseline Predictions for Aeroengine Based on Machine Learning Algorithms

The exhaust gas temperature (EGT) baseline of an aeroengine is key to accurately analyzing engine health, formulating maintenance decisions and ensuring flight safety. However, due to the complex performance characteristics of aeroengine and the constraints of many external factors, it is difficult to obtain accurate non-linear features between various operating factors and EGT. In order to diagnose and forecast aeroengine performance quickly and accurately, four data-driven baseline prediction frameworks for EGT are proposed. These baseline frameworks took engine operating conditions and operating state control parameters as input variables and EGT as predicted output variables. The original data were collected from CFM56-5B engine ACARS flight data. Four typical machine learning methods, including Generalized Regression Neural Network (GRNN), Radial Basis Neural Network (RBF), Support Vector Regression (SVR) and Random Forest (RF) are trained to develop the models. Four aeroengine EGT baseline models were validated by comparing the after-flight data of another engine. The results show that the developed GRNN models have the best accuracy and computational efficiency compared with other models, and their RE and CPU calculation time on the verification set are 1.132 × 10−3 and 3.512 × 10−3 s, respectively. The developed baseline prediction frameworks can meet the needs of practical engineering applications for airlines. The methodologies developed can be employed by airlines to predict the EGT baseline for the purpose of engine performance monitoring and health management.

Parameter estimation in unbalanced three-phase distribution lines using general regression neural networks with inconsistent data handling capacity

The feeder parameters of a power system play an important role in its operation and analysis. However, in many situations, the values used in the calculations differ from the actual parameters, resulting in large errors. Therefore, in this study, a highly accurate feeder parameter estimation method for a three-phase distribution line was proposed. The proposed method uses the measured node voltages and power quantities from the two ends of the distribution line. A precise electrical model of the three-phase distribution line was used. An improved type of neural network, namely general regression neural networks (GRNN), is used to solve the complex non-linear equation and then estimate the actual parameters. However, the proposed estimation process does not require synchronised phasors or phasor-measurement units installed at the point of measurement. Nonetheless, the estimation process is also capable of handling incorrect data information, noisy datasets, and measurement errors while maintaining accuracy at the desired level. The estimator process is examined and analysed using four different IEEE unbalanced three-phase systems. In addition, the proposed methodology is compared with standard algorithms of radial basis function networks, quasi-Newton, and multi-run optimisation methods on the basis of maximum absolute percentage error and other standard error functions. The hyperparameter-tuned GRNN model was also tested using several statistical significance tests. In all scenarios, the performance and accuracy of the proposed methodology and model exceeded those of the other methods.

Prediction of the Sound Absorption Coefficient of Three-Layer Aluminum Foam by Hybrid Neural Network Optimization Algorithm

The combination of multilayer aluminum foam can have high sound absorption coefficients (SAC) at low and medium frequencies, and predicting its absorption coefficient can help the optimal structural design. In this study, a hybrid EO-GRNN model was proposed for predicting the sound absorption coefficient of the three-layer composite structure of the aluminum foam. The generalized regression neural network (GRNN) model was used to predict the sound absorption coefficient of three-layer composite structural aluminum foam due to its outstanding nonlinear problem-handling capability. An equilibrium optimization (EO) algorithm was used to determine the parameters in the neuronal network. The prediction results show that this method has good accuracy and high precision. The calculation result shows that this proposed hybrid model outperforms the single GRNN model, the GRNN model optimized by PSO (PSO-GRNN), and the GRNN model optimized by FOA(FOA-GRNN). The prediction results are expressed in terms of root mean square error (RMSE), absolute error, and relative error, and this method performs well with an average RMSE of only 0.011.

Machine Learning-Assisted In Vitro Rooting Optimization in Passiflora caerulea

In vitro rooting as one of the most critical steps of micropropagation is affected by various extrinsic (e.g., medium composition, auxins) and intrinsic factors (e.g., species, explant). In Passiflora species, in vitro adventitious rooting is a difficult, complex, and non-linear process. Since in vitro rooting is a multivariable complex biological process, efficient and reliable computational approaches such as machine learning (ML) are required to model, predict, and optimize this non-linear biological process. Therefore, in the current study, a hybrid of generalized regression neural network (GRNN) and genetic algorithm (GA) was employed to predict in vitro rooting responses (rooting percentage, number of roots, and root length) of Passiflora caerulea based on the optimization of the level of auxins (indole-3-acetic acid (IAA), indolebutyric acid (IBA), and 1-naphthaleneacetic acid (NAA)) and the type of explant (microshoots derived from leaf, node, and internode). Based on the results, the GRNN model was accurate in predicting all in vitro rooting responses of P. caerulea (R2 > 0.92) in either training or testing sets. The result of the validation experiment also showed that there was a negligible difference between the predicted-optimized values and the validated results demonstrating the reliability of the developed GRNN-GA model. Generally, the results of the current study showed that GRNN-GA is a reliable and accurate model to predict and optimize in vitro rooting of P. caerulea.

Multi-objective optimization of particle gluing operating parameters in particleboard production based on improved machine learning algorithms

Particle gluing operating parameters in particleboard (PB) production have an important influence on the mechanical properties of PBs. This study developed a multi-objective optimization model based on support vector regression (SVR) optimized by the non-dominated sorted genetic algorithm-II (NSGA2) to realize the multi-objective accurate prediction of PB mechanical properties (modulus of elasticity (MOE), modulus of rupture (MOR), and internal bonding (IB) strength) by adjusting particle gluing operating parameters. The NSGA2-SVR multi-objective prediction model was trained by 496 groups of experimental data of particle gluing operating parameters and PB mechanical properties. The prediction results of the NSGA2-SVR multi-objective prediction model were evaluated by 124 groups of experimental data and compared with the prediction results of the back propagation neural network (BPNN) model, general regression neural network (GRNN) model, and SVR model. The mean absolute percentage errors (MAPEs) of the NSGA2-SVR model were 49.11%, 33.64%, and 24.20% lower than that of the BPNN model, GRNN model, and SVR model, respectively. The Theil’s inequality coefficients (TICs) of the NSGA2-SVR model were 40.93%, 27.39%, and 18.58% lower than that of the BPNN model, GRNN model, and SVR model, respectively. The results showed that the multi-objective prediction model based on NSGA2-SVR has a superior fitting and higher prediction accuracy for the prediction performance of particle gluing operating parameters, and the NSGA2-SVR model can be applied to the multi-objective synchronous prediction of particle gluing operating parameters in the PB production line.

Surrogate-based multiobjective optimization to rapidly size low impact development practices for outflow capture

The optimal implementation of low impact development (LID) practices remains an open research issue. Multiobjective optimization (MOO) based on physically-based models facilitates LID sizing but requires large computational budgets. We developed and validated a surrogate MOO framework based on machine learning (ML) models to obtain satisfactory optimal solutions with an affordable computational cost. First, a calibrated and validated storm water management model (SWMM) was driven by a 2-year, 120 min Chicago pattern storm with random LID areas for generating 10,000 samples. Second, ML models, including multiple linear regression (MLR), generalized regression neural network (GRNN), and backpropagation neural network (BPNN), were trained and validated to predict catchment total outflow (CTO) and catchment peak outflow (CPO). Third, using the outperformed ML models as surrogates, the LID areas were optimized by non-dominated sorting genetic algorithm-II to minimize construction costs, CTO, and CPO. The results obtained for an urban catchment in Fengxi, China, demonstrated that: (1) BPNN outperformed MLR and GRNN in terms of accuracy, while the training time (∼40 min) was ∼6857 times and ∼706 times that of MLR (∼0.35 s) and GRNN (3.4 s) models, respectively. (2) The optimization process using BPNN models as surrogates converged well, indicating the efficiency of the proposed framework. (3) Under the marginal effect, the permeable pavement was identified as the leading practice for most Pareto solutions, followed by bioretention cell and green roof. Moreover, a subcatchment with a higher runoff coefficient was deemed preferable for LID implementation. (4) Outflow reduction by the LIDs was significant (CTO and CPO were reduced by 28.6–56.9 % and 27.5–48.3 %, respectively) but negligible for retarding the time of CPO. (5) The optimization time was reduced by ∼95.82 % using the surrogate-based MOO instead of the SWMM-based model when the training time of the BPNN models was ignored. This study provides insights into the efficient optimization of LIDs.

Comparison of GRNN and RF algorithms for predicting heat transfer coefficient in heat exchange channels with bulges

In the research of heat transfer, heat exchanger plays an important role in highly integrated and high-precision thermal management to ensure the thermal balance. In previous studies, traditional experiments and CFD simulations consume lots of time and computational resources, while the heat transfer correlations have large errors. Hence, this research aims to establish a reliable method to predict the Heat Transfer Coefficient (HTC) of heat exchange channels more quickly and accurately. In this paper, General Regression Neural Network (GRNN) and Random Forests (RF) models, which are trained by hundreds of CFD simulation results, are adopted to predict the heat-exchange performances of channels with different height bulges. The prediction results show that the HTC of channels with different bulge arrangements are accurately predicted, supported by R2 > 0.97 in both training and validation sets. Also, it shows that GRNN is more applicable to heat exchange channels than RF. Besides, it can be inferred from the prediction results that the front-end bulge has a significant impact on the overall HTC, as does the uniformity of the bulge height. In conclusion, machine learning algorithms have great potential in predicting the HTC of channels, and the GRNN algorithm may perforce better when calculate other complex heat transfer problems.

Prediction of the influence of castor oil–ethanol–diesel​ blends on single-cylinder diesel engine characteristics using generalized regression neural networks (GRNNs)

With some inferior fuel properties such as cetane number, flash point and calorific value, in this study the ethanol–diesel blend was improved through the use of neat castor oil. According to a preliminary study on blending stability and lubricity, a fuel blend was formulated with 10% ethanol, 10% castor oil and 80% diesel fuel (D80E10C10) which was selected to test in the diesel engine. The generalized regression neural network (GRNNs), which is one of the artificial intelligence algorithms that is suitable for a small data set, was applied to create a model for predicting the influence of castor oil on the characteristics of the diesel engine fueled with an ethanol–diesel blend. The experimental results of fuel properties were firstly analyzed and revealed that the presence of 10% castor oil can improve the defective fuel properties of ethanol–diesel to keep them under the limitations of the diesel fuel specification. A lower thermal efficiency was obtained with the combustion of the castor oil blend. A higher peak of in-cylinder pressure and the earlier start of combustion was achieved with increasing engine operating load and compression ratio. A lower peak of heat release rate was attained with the combustion of the castor oil blend. Hydrocarbons and carbon monoxide were increased with the castor oil blend, while oxides of nitrogen were decreased and no significant change in smoke emissions was found with respect to diesel fuel. Subsequently, the results of the prediction model concluded that the response surface methodology was a good method to optimize the smoothing parameter of the GRNNs model in the case of multi-output factors with different levels of importance. The GRNNs model was achieved with relatively high performance to predict BSFC, thermal efficiency, NOx, HC, CO and smoke emissions, with mean absolute percentage error (MAPE) values in the range of 1.562%–8.181% and coefficient of determination (R2) in the range of 0.708-0.993.

A real-time driver fatigue identification method based on GA-GRNN.

It is of great practical and theoretical significance to identify driver fatigue state in real time and accurately and provide active safety warning in time. In this paper, a non-invasive and low-cost method of fatigue driving state identification based on genetic algorithm optimization of generalized regression neural network model is proposed. The specific work is as follows: (1) design simulated driving experiment and real driving experiment, determine the fatigue state of drivers according to the binary Karolinska Sleepiness Scale (KSS), and establish the fatigue driving sample database. (2) Improved Multi-Task Cascaded Convolutional Networks (MTCNN) and applied to face detection. Dlib library was used to extract the coordinate values of face feature points, collect the characteristic parameters of driver's eyes and mouth, and calculate the Euler Angle parameters of head posture. A fatigue identification model was constructed by using multiple characteristic parameters. (3) Genetic Algorithm (GA) was used to find the optimal smooth factor of Generalized Regression Neural Network (GRNN) and construct GA-GRNN fatigue driving identification model. Compared with K-Nearest Neighbor (KNN), Random Forest (RF), and GRNN fatigue driving identification algorithms. GA-GRNN has the best generalization ability and high stability, with an accuracy of 93.3%. This study provides theoretical and technical support for the application of driver fatigue identification.

The Application of the Generalized Regression Neural Network Model Based on Information Granulation for Short-Term Temperature Prediction

The Application of the Generalized Regression Neural Network Model Based on Information Granulation for Short-Term Temperature Prediction

Predicting the thermal performance of double pipe heat exchanger using the generalized regression neural network model

A heat transfer device in which heat exchange takes place between two fluids at different temperatures. The performance of any heat exchanger is evaluated by its heat transfer rate. Such heat transfer rate directly depends on the difference in temperature between two fluids and the mass flow rate. In this study, an attempt has been made to predict the rate of heat transfer between two fluids and the exit temperature using the generalized regression neural network (GRNN) model. For this, two GRNN models are designed using 15 real-time experimental datasets among which the GRNN-1 model is effectively designed to predict the exit temperature of both the hot and the cold fluids flowing through the heat exchanger with an accuracy of 97.48%. Similarly, the GRNN-2 model is designed to predict the rate of heat transfer between the fluids with an accuracy of 94.04%.

Gearbox Fault Diagnosis Based on a Sparse Principal Component-Generalized Regression Neural Network

The fault diagnosis of urban rail transit gearboxes has the characteristics of complex vibration signals and large amounts of data. The daily scheduled maintenance cannot meet the needs of gearbox maintenance, so it is necessary to predict the fault types in advance. In this paper, a method of gearbox fault prediction based on sparse principal component analysis and a generalized regression neural network is presented, and the result of fault prediction can provide a reference for making maintenance plans. Based on principal component analysis (PCA), a sparse principal component is obtained by adding the LASSO penalty term, which reduces the risk of overfitting of PCA while obtaining a sparse solution. Then, the sparse reduced dimension principal component is input into the generalized regression neural network model for fault diagnosis. The results show that the fault diagnosis method based on the sparse principal component-generalized regression neural network model has high accuracy and is time-consuming.

Daily Groundwater Level Prediction and Uncertainty Using LSTM Coupled with PMI and Bootstrap Incorporating Teleconnection Patterns Information

Daily groundwater level is an indicator of groundwater resources. Accurate and reliable groundwater level (GWL) prediction is crucial for groundwater resources management and land subsidence risk assessment. In this study, a representative deep learning model, long short-term memory (LSTM), is adopted to predict groundwater level with the selected predictors by partial mutual information (PMI), and bootstrap is employed to generate different samples combination for training many LSTM models, and the predicted values by many LSTM models are used for the uncertainty assessment of groundwater level prediction. Two wells of different climate zones in the USA were used as a case study. Different significant predictors of GWL for two wells were identified by PMI from candidate predictors incorporating teleconnection patterns information. The results show that GWL is significantly affected by antecedent GWL, AO, Niño 3.4, Niño 1 + 2, and precipitation in humid areas, and by antecedent GWL, AO, Niño 3.4, Niño 3, Niño 1 + 2, and PNA in arid areas. Predictor selection can assist in improving the prediction performance of the LSTM model. The relationship between GWL and significant predictors were modeled by the LSTM model, and it achieved higher accuracy in humid areas, while the performance in arid areas was poorer due to limited precipitation information. The performance of LSTM was improved by increasing correlation coefficient (R2) values by 10% and 25% for 2 wells compared to generalized regression neural network (GRNN). Three uncertainty evaluation metrics indicate that LSTM reduced the uncertainty compared to GRNN model. LSTM coupling with PMI and bootstrap can be a promising approach for accurate and reliable groundwater level prediction for different climate zones.

Analysis on the susceptibility of environmental geological disasters considering regional sustainable development.

Environmental geological disasters seriously threaten human lives and property. A reasonable analysis of the susceptibility to environmental geological disasters is the basis for disaster prevention and mitigation and can promote the sustainable development of regional economy. This study analyzes the susceptibility of environmental geological disasters, such as collapses, landslides, and debris flows in Helong City, China. Through investigation and comprehensive analysis, ten environmental, geological disaster causing factors, including stratum lithology, distance from the fault, elevation, slope, aspect, rainfall, distance from the water system, NDVI, distance from the road, and profile curvature, were extracted. Combined with GIS, a vulnerability analysis database of environmental geological disasters was established, and vulnerability zoning prediction was performed by using two models of information amount and a generalized regression neural network (GRNN). Then, disaster-vulnerability factors such as population density, road density, GDP, and land use type were added. The results show that the predicted results of the two models are similar to the actual survey results.. The environmental geological disasters in the study area are mainly low and not prone to occur, and the northeast and central areas are highly prone to environmental geological disasters, which are the key prevention and control areas in the study area. The coverage rate of high-vulnerability areas with a high degree of economic development is 8.63%, and the prediction results of the GRNN model are mostly distributed in spots and strips, which is more conducive to accurate disaster prevention and mitigation and cost reduction, promotes regional sustainable development, and has guiding significance for disaster prevention and control and urban planning and construction.