Multi-gene Genetic Programming Model Research Articles

Unconfined compressive strength of geopolymers based soil: Model development using gene expression programming and a comparison with other computer-based approaches

The present work aims to propose a model for calculating unconfined compressive strength (UCS) of soil stabilized by geopolymer based on gene expression programming (GEP). The new model is compared with earlier models developed from artificial neural networks (ANN), multivariable regression analysis (MVR), multi-gene genetic programming (MGGP), and support vector machines (SVM) models. The results indicate that the GEP model is superior to MVR and MGGP models and quite comparable to ANN and SVM models. The study demonstrates the robustness of the GEP model through parametric analysis. Additionally, this model is simpler and easier to use in practice.

Data-Driven Identification of Crane Dynamics Using Regularized Genetic Programming

The meaningful problem of improving crane safety, reliability, and efficiency is extensively studied in the literature and targeted via various model-based control approaches. In recent years, crane data-driven modeling has attracted much attention compared to physics-based models, particularly due to its potential in real-time crane control applications, specifically in model predictive control. This paper proposes grammar-guided genetic programming with sparse regression (G3P-SR) to identify the nonlinear dynamics of an underactuated crane system. G3P-SR uses grammars to bias the search space and produces a fixed number of candidate model terms, while a local search method based on an l0-regularized regression results in a sparse solution, thereby also reducing model complexity as well as reducing the probability of overfitting. Identification is performed on experimental data obtained from a laboratory-scale overhead crane. The proposed method is compared with multi-gene genetic programming (MGGP), NARX neural network, and Takagi-Sugeno fuzzy (TSF) ARX models in terms of model complexity, prediction accuracy, and sensitivity. The G3P-SR algorithm evolved a model with a maximum mean square error (MSE) of crane velocity and sway prediction of 1.1860 × 10−4 and 4.8531 × 10−4, respectively, in simulations for different testing data sets, showing better accuracy than the TSF ARX and MGGP models. Only the NARX neural network model with velocity and sway maximum MSEs of 1.4595 × 10−4 and 4.8571 × 10−4 achieves a similar accuracy or an even better one in some testing scenarios, but at the cost of increasing the total number of parameters to be estimated by over 300% and the number of output lags compared to the G3P-SR model. Moreover, the G3P-SR model is proven to be notably less sensitive, exhibiting the least deviation from the nominal trajectory for deviations in the payload mass by approximately a factor of 10.

Prediction of effective equivalent linear temperature gradients in bonded concrete overlays of asphalt pavements

PurposeThe time-varying equivalent linear temperature gradient (ELTG) significantly affects the development of faulting and must therefore be accounted for in pavement design. The same is true for faulting of bonded concrete overlays of asphalt (BCOA) with slabs larger than 3 x 3 m. However, the evaluation of ELTG in Mechanistic-Empirical (ME) BCOA design is highly time-consuming. The use of an effective ELTG (EELTG) is an efficient alternative to calculating ELTG. In this study, a model to quickly evaluate EELTG was developed for faulting in BCOA for panels 3 m or longer in size, whose faulting is sensitive to ELTG.Design/methodology/approachA database of EELTG responses was generated for 144 BCOAs at 169 locations throughout the continental United States, which was used to develop a series of prediction models. Three methods were evaluated: multiple linear regression (MLR), artificial neural networks (ANNs), and multi-gene genetic programming (MGGP). The performance of each method was compared, considering both accuracy and model complexity.FindingsIt was shown that ANNs display the highest accuracy, with an R2 of 0.90 on the validation dataset. MLR and MGGP models achieved R2 of 0.73 and 0.71, respectively. However, these models consisted of far fewer free parameters as compared to the ANNs. The model comparison performed in this study highlights the need for researchers to consider the complexity of models so that their direct implementation is feasible.Originality/valueThis research produced a rapid EELTG prediction model for BCOAs that can be incorporated into the existing faulting model framework.

Estimating Unconfined Compression Strength of Fly Ash and Cement Stabilised Clayey Soil using Artificial intelligence Techniques

The study aims to develop proposed predictive formulas for determining the unconfined compression strength (UCS) of fly ash and cement stabilised clayey soil based on Multi-Gene Genetic Programming (MGGP) and Artificial Neural Network (ANN) techniques. Thirteen parameters, including the soil characteristics, the binder types, the binder contents, the curing period, the mixing method, and the fly ash characteristics, such as calcium oxide (CaO) content, CaO/SiO2 ratio, loss of ignition, were considered as the independent variables in the model. The results show that the selected optimal ANN and MGGP models can predict the target values with high correlation coefficients (R-value approximately 0.994 and 0.973, respectively), and low errors. The performances of the MGGP and ANN models were compared based on statistical parameters and several external criteria. The study finds that both models show their generalisation capabilities with robust, powerful, and accurate prediction ability; however, the ANN model slightly outperforms the MGGP model. The proposed predictive equations formulated from the selected optimal MGGP and ANN models could help engineers and consultants to choose the suitable binder and a reasonable amount of fly ash in the pre-planning and pre-design period.

Comparative study of artificial intelligence based multi‐modelling approach and optimization of photoreactor

AbstractThis study presents a generic methodology for modelling and optimizing a reactor with complex and poorly understood kinetics. Here, a photoreactor is considered which performs photodegradation of sodium oxalate salt in spent Bayer liquor. Multiple data‐driven modelling methods, including artificial neural networks (ANN), genetic programming (GP), hybrid genetic programming‐grey wolf optimization (GP‐GWO), and multi‐gene genetic programming (MGGP), were used to model the reactor's performance based on experimental data. The input parameters considered for modelling were initial solution pH, power of the lamp, total organic carbon, and catalyst loading. The models were evaluated based on their predictability, explainability, complexity, and adherence to the reactor's phenomenology. The MGGP model was found to be the most effective and was used to generate surface plots showing the parity between experimental results and model predictions. Additionally, the MGGP model was optimized using GWO to determine the process conditions that maximize the reaction rate.

A new model developed by multigene genetic programming for the temporal evolution of bridge pier scour

Forecasting the time development of scour depth at bridge pier foundations is of great significance to mitigate or avoid the potential failure of bridges. Presently, several models have been developed to predict the scour depth at the base of bridge piers in the case of flood events. This study summarizes existing models for the temporal evolution of bridge pier scour and divides these studies into semiempirical models and empirical models, as well as artificial intelligence models. Several experimental data sets collected from previous studies, 665 points in total, are used to develop a new multigene genetic programming (MGGP) model for temporal scour depth at a circular bridge pier. In addition, independent data, 899 points in total, from previous studies and new physical modeling tests are applied to evaluate the behaviours of existing models, as well as the newly developed MGGP model. It is shown that the MGGP model has good prediction capability when compared with existing empirical and mathematical models.

Modeling of pertinent parameters influence on the time dependent mass transfer coefficient of particulate matter under the sink effect

One of the most crucial evaluation metrics for the performance of particle sink purification technology is the time-dependent mass transfer coefficient (TDMTC). Therefore, it is very helpful for designers and developers to accurately describe the functional relationship between different influence parameters and the TDMTC. In this paper, four influence parameters (the applied voltage (V), interelectrode distance (dc), porosity (P), and shape (n) of the collecting electrode) were considered, and then non-linear multiple regression (NLMR) and multi-gene genetic programming (MGGP) methods were used to establish prediction models of the TDMTC. Results showed that V and n were the most significant factors, followed by dc and P. Both multi-factor models could accurately predict the TDMTC under the sink effect with a maximum prediction error of 20% and 15%, respectively. Moreover, for particulate matter (PM) with different size fractions, MGGP models could improve the prediction accuracy by 5–10% compared to NLMR models.

Analytical Model for Progressive Collapse of RC Frame Beam-Column Substructures Using Multi-Gene Genetic Programming

Establishing a concise and accurate analytical model is the key to developing a feasible progressive collapse design for engineering practice. However, existing models either focused on an individual force mechanism or required complicated computer programming. Among existing machine learning (ML) techniques, multi-gene genetic programming (MGGP) can be trained to obtain explicit formulas for engineering problems. In this study, a comprehensive database was established by data collection, Latin hypercube sampling and structural design, and was used to train the mathematical model for quantifying progressive collapse resistance of reinforced concrete (RC) beam-column substructures under middle column removal scenarios. Further, an energy-based error index was proposed to validate the accuracy of the MGGP model among others. The research outcomes can provide references for the development of simplified analytical models for calculating the progressive collapse progress of RC frame structures, and promote the development of the practical design method.

Multigene Genetic Programming Based Prediction of Concrete Fracture Parameters of Unnotched Specimens

This study explores the fracture energy of notched and unnotched concrete specimens subjected to the classical three-point bend test, instantiating a gradational step in the continued development of concrete fracture mechanics. An experimental campaign involving 18 notched test specimens and nine unnotched specimens of three different grades of concrete, an examination of the existing literature models for unnotched specimens, and a novel Multigene Genetic programming (MGGP) based concrete fracture energy model for unnotched specimens are integral to this study. As a salient result, the multiple approaches to quasi-brittle materials adopted in the study, highlighted the criticality of the determination of fracture energy, tensile strength and characteristic length for the crack width study. The failure modes of notched and unnotched specimens were found to be similar. The reported literature has mainly focused on a limited number of fracture energy influencing parameters. Therefore, six impact parameters have been chosen and incorporated into the present study to provide a more acceptable explanation of concrete fracture behaviour. A sensitivity analysis of the parameters and an error analysis of the model undertaken have established the accuracy and robustness of the developed MGGP model. Doi: 10.28991/CEJ-2023-09-02-011 Full Text: PDF

Convective drying of unripe plantain: a comparative response surface methodology and genetic algorithm optimization study, certainty and sensitivity analysis

The requirement for reduction of post-harvest losses increased production, and cost-effectiveness of foods are driving continuous food process investigations. In this study, Response Surface Methodology (RSM) was utilized to understand, model, and optimize the effect of selected process factors on the moisture content (MC) of convectively dried unripe plantain fruit. For technical accuracies, Multi Gene Genetic Programming (MGGP) was also used to model the process and both MGGP and RSM models were statistically compared. Furthermore, Monte Carlo Simulation (MCS) was used to conduct sensitivity analysis of unripe plantain’s MC to each selected process factor. Results showed that increased sample thickness increased the MC while increased drying temperature and drying time decreased the MC of unripe plantain. RSM model had Chi-square, MBE, t-value, RMSE and R2 values of 15.2131, 0.7531, 7.6170, 0.9193 and 0.9674, respectively; while MGGP model had 3.0415, 0.2563, 2.6871, 0.4111 and 0.9956, respectively. Sensitivity analysis showed that sample thickness, drying temperature and drying time had +89.5 %, -10.2 % and -0.3 % contribution to the variances of MC, respectively. These results are useful in unripe plantain drying process prediction, optimization, and product standardization.

Development of correlations for steam condensation over a vertical tube in the presence of noncondensable gas using machine learning approach

Steam condensation is an important phenomenon encountered in nuclear reactor under severe accidents. Even though many correlations for predicting steam condensation heat transfer coefficient (HTC) in the presence of noncondensable gas (NCG) have been proposed over the past decade, a more reliable and accurate model is still required. Thus, in this study, multigene genetic programming (MGGP), a biologically inspired machine learning method, is applied to develop new correlations for condensation HTC of steam-NCG mixture over a vertical tube in turbulent free convection regime. To this end, a consolidated database with 1440 data points from 18 sources is compiled. Then, using the database, both a new empirical correlation and a MGGP model are developed for better comparison. The performance of the MGGP-based correlation selected using Pareto tournaments strategy is compared with the new developed empirical correlation and another 20 relevant correlations. The results reveal the superiority of the MGGP-based correlation. In addition, it is found that the tube length is excluded in the best-trained correlation, even though it is used as the input of MGGP, which agrees well with the results of previous theoretical and experimental studies. The present study demonstrates that MGGP is promising in developing explicit, accurate, and compact models for the complex heat transfer and multiphase flow phenomena such as steam condensation in the presence of NCG.

Modeling monthly reference evapotranspiration process in Turkey: application of machine learning methods.

In this study, the predictive power of three different machine learning (ML)-based approaches, namely, multi-gene genetic programming (MGGP), M5 model trees (M5Tree), and K-nearest neighbor algorithm (KNN), for long-term monthly reference evapotranspiration (ET0) prediction were investigated. The input data consist of monthly solar radiation (Rs), maximum air temperature (Tmax), and wind speed (Ws) derived from 163 meteorological stations in Turkey. Different input combinations were created and analyzed. The model's performance was evaluated using criteria such as Nash-Sutcliffe efficiency, Kling-Gupta efficiency, relative root mean squared error, mean absolute percentage error, and determination coefficient. Moreover, Taylor, radar, and boxplot diagrams were created. It was determined that the MGGP model outperformed both the M5Tree and the KNN models. The equation obtained from the MGGP model, for the best-performed combination of Rs-Tmax-Ws, was presented. The best weather conditions were obtained as 0.029 to 31.814MJ/m2, - 5.8 to 45.7°C, and 0.140 to 5.086m/s for Rs, Tmax, and Ws, respectively. It was also found that the Rs was the most potent input variable for ET0 estimation while Ws was the weakest.

Comparison of neural network, Gaussian regression, support vector machine, long short-term memory, multi-gene genetic programming, and M5 Trees methods for solving civil engineering problems

In this study, it was investigated that how machine learning (ML) methods show performance in different problems having different characteristics. Six ML approaches including Artificial neural networks (ANN), gaussian process regression (GPR), support vector machine regression (SVMR), long short-term memory (LSTM), multi-gene genetic programming (MGGP) and M5 model tree (M5Tree) were utilized to analyze three independent civil engineering problems belonging to construction management, geotechnical engineering, and hydrological engineering sub-disciplines. Mean absolute percentage error (MAPE), root mean square error (RMSE), coefficient of determination (R2), relative root means square error (RRMSE), Nash–Sutcliffe efficiency (NSE), Kling–Gupta efficiency (KGE), and overall index of model performance (OI) criteria were used to evaluate the performances of the models. Besides performance criteria, the relative performances of the six ML models were assessed using Taylor diagram, Violin diagram and One-Tailed Wilcoxon Signed-Rank Test. For each of the problem considered in this study, the effectiveness of the input parameters on the output parameter has been defined using the Relief Method and Correlation Coefficient. The results show that ANN and MGGP models yielded the most successful estimations for three different problems considered. The best prediction was achieved by MGGP model for hydrological engineering problem. For the construction management, geotechnical engineering problems, the best results were obtained using the ANN model. All models were reliable to solve the geotechnical engineering and hydrological engineering problems while LSTM and SVMR models are not reliable to solve the construction management problem. The most and least effective input parameters on output parameter were contract cost (CC) and work definition number (WDN) for the managerial data set. On the other hand, the most and least effective input parameters on the output parameters for the experimental and natural data sets have been obtained as width of the pile (B), rotation degree (R) and minimum temperature (Tmin), streamflow (Q) data, respectively. The number of data and data selection have a significant effect on the homogeneity of the data set and its representativeness of the problem. The error values obtained in test stage are affected from this condition. The equations to calculate the outputs of each of the problem considered were obtained using MGGP and M5Tree models.

Designing a Multi-Stage Expert System for daily ocean wave energy forecasting: A multivariate data decomposition-based approach

Accurate forecasting of the wave energy is crucial and has significant potential because every wave meter possesses an energy amount ranging from 30 to 40 kW along the shore. By harnessing, it does not produce toxic gases, which is a better alternative to the energies that use fossil fuels. In this research, a multi-stage Multivariate Variational Mode Decomposition (MVMD) integrated with Boruta-Extreme Gradient Boosting (BXGB) feature selection and Cascaded Forward Neural Network (CFNN) (i.e., MVMD-BXGB-CFNN) is proposed to forecast daily ocean wave energy in the regions of Queensland State, Australia. The modelling outcomes were benchmarked via three other robust intelligence-based alternatives comprised of Multigene Genetic Programming (MGGP), Least Square Support Machine (LSSVM), and Gradient Boosted Decision Tree (GBDT) models hybridized with MVMD and BXGB (i.e., MVMD-BXGB-MGGP, MVMD-BXGB-LSSVM, and MVMD-BXGB-GBDT), and their counterpart standalone CFNN, GBDT, LSSVM, and MGGP models. To develop the multi-step hybrid intelligent systems, first, the primary input signals were simultaneously decomposed into intrinsic mode functions (IMFs) and residual components using the MVMD pre-processing technique. Next, the significant lags at the t-1 and t-2 timescales computed using the cross-correlation function were imposed on the decomposed components and further filtered by the BXGB feature selection to identify the best IMFs and reduce the computational cost and enhance the accuracy. Finally, the filtered IMFs were incorporated into the machine learning (ML) models to forecast the wave energy. Forecasting performance of all the provided models (hybrid and counterpart standalone ones) was evaluated during the testing phase by several well-known metrics, infographic tools, and diagnostic analysis. The results showed that the MVMD-BXGB-CFNN technique, as a capable expert system, outperformed the other hybrid and counterpart standalone methods and has an adequate degree of reliability to forecast the daily wave energy in coastal regions.

Stability improvement of the PSS-connected power system network with ensemble machine learning tool

Stability is a primary requirement of the electrical power system for its flawless, secure, and economical operation. Low-frequency oscillations (LFOs), commonly seen in interconnected power systems, initiate the possibility of instability and, therefore, require sophisticated care to deal with. This paper proposes an original approach to tuning the parameters of the power system stabilizer (PSS), which plays a crucial role in the power system networks to dampen unwanted oscillations. The ensemble method combines multiple machine learning techniques and has been used for tuning the PSS parameters in real-time for two PSS-connected power system networks. The first system is a single-machine infinite bus power system, while the second is a unified power flow controller (UPFC) device. The backtracking search algorithm (BSA) based proposed ensemble model is formed by combining three machine learning (ML) techniques, namely the extreme learning machine (ELM), neurogenetic (NG) system, and multi-gene genetic programming (MGGP). To validate the stability of the network, Eigenvalues, well-recognized statistical parameters, and minimum damping ratios were analyzed, besides the time-domain simulation results. Furthermore, results for various loading conditions were prepared to check the robustness of the proposed model. A comparative study of the proposed approach with NG, ELM, MGGP models, and two reference cases along with the conventional method will validate the superiority of the employed ML approach.

Prediction of Solar Radiation Using Data Driven Models

The local climatic conditions play a significant role in the amount of Solar Radiation that reaches the Earth’s surface. In India, most of the locations, receive ample Solar Radiation. The exact evaluation of SR is most essential for engineering studies on energy and climate. For the present investigations, the meteorological data for Hyderabad city in Telangana state for the period 2009 - 2020 is adopted. Five climatological variables viz., Max. Temp. (Tmax), Min. Temp. (Tmin) and Mean Air Temp. (Tmean), Sunshine hours (n), and Relative Humidity (RH) were collected from the weather forecast. In the present investigations, two conventional empirical equations viz., Abdalla, and Angstrom are adopted for analytical estimation of Solar Radiation (SR). Furthermore, two data-driven models via., Artificial Neural Networks (ANN) and Multi Gene-genetic Programming (MGGP) are used to develop an equation for estimating Solar Radiation (Rs). The model equations were developed based on the parameters adopted for each conventional equation. These results are compared with conventional equation results. The present investigation exhibited that, the expected results of the Solar radiation of the arrangement formed by the data are favourable and also simplify the testing data. The corresponding R2 values for random data are 0.976 for the training dataset, while the value for the testing dataset is 0.964. Based on the performance analysis for each equation, the efficiency and reliability of the proposed MGGP and ANN model are validated and predicted for the year 2021. The predicted results were found to be in good agreement with the analytical results evaluated from three properly used equations.

Multi-gene genetic programming extension of AASHTO M-E for design of low-volume concrete pavements

The American Association of State Highway and Transportation Officials Mechanistic-Empirical Pavement Design Guide (AASHTO M-E) offers an opportunity to design more economical and sustainable high-volume rigid pavements compared to conventional design guidelines. It is achieved through optimizing pavement structural and thickness design under specified climate and traffic conditions using advanced M-E principles, thereby minimizing economic costs and environmental impact. However, the implementation of AASHTO M-E design for low-volume concrete pavements using AASHTOWare Pavement ME Design (Pavement ME) software is often overly conservative. This is because Pavement ME specifies the minimum design thickness of concrete slab as 152.4 ​mm (6 in.). This paper introduces a novel extension of the AASHTO M-E framework for the design of low-volume joint plain concrete pavements (JPCPs) without modification of Pavement ME. It utilizes multi-gene genetic programming (MGGP)-based computational models to obtain rapid solutions for JPCP damage accumulation and long-term performance analyses. The developed MGGP models simulate the fatigue damage and differential energy accumulations. This permits the prediction of transverse cracking and joint faulting for a wide range of design input parameters and axle spectrum. The developed MGGP-based models match Pavement ME-predicted cracking and faulting for rigid pavements with conventional concrete slab thicknesses and enable rational extrapolation of performance prediction for thinner JPCPs. This paper demonstrates how the developed computational model enables sustainable low-volume pavement design using optimized ME solutions for Pittsburgh, PA, conditions.

Stress Analysis of 2D-FG Rectangular Plates with Multi-Gene Genetic Programming

Functionally Graded Materials (FGMs) are designed for use in high-temperature applications. Since the mass production of FGM has not yet been made, the determination of its thermo-mechanical limits depends on the compositional gradient exponent value. In this study, an efficient working model is created for the thermal stress problem of the 2D-FG plate using Multi-gene Genetic Programming (MGGP). In our MGGP model in this study, data sets obtained from the numerical analysis results of the thermal stress problem are used, and formulas that give equivalent stress levels as output data, with the input data being the compositional gradient exponent, are obtained. For the current problem, efficient models that reduce CPU processing time are obtained by using the MGGP method.

Multi-perspective analysis on rainfall-induced spatial response of soil suction in a vegetated soil

In this study, an intelligent monitoring platform is established for continuous quantification of soil, vegetation, and atmosphere parameters (e.g. soil suction, rainfall, tree canopy, air temperature, and wind speed) to provide an efficient dataset for modeling suction response through machine learning. Two characteristic parameters representing suction response during wetting processes, i.e. response time and mean reduction rate of suction, are formulated through multi-gene genetic programming (MGGP) using eight selected influential parameters including depth, initial soil suction, vegetation- and atmosphere-related parameters. An error standard–based performance evaluation indicated that MGGP has appreciable potential for model development when working with even fewer than 100 data. Global sensitivity analysis revealed the importance of tree canopy and mean wind speed to estimation of response time and indicated that initial soil suction and rainfall amount have an important effect on the estimated suction reduction rate during a wetting process. Uncertainty assessment indicated that the two MGGP models describing suction response after rainfall are reliable and robust under uncertain conditions. In-depth analysis of spatial variations in suction response validated the robustness of two obtained MGGP models in prediction of suction variation characteristics under natural conditions.

Intelligent Prediction Models For UCS Of Cement/Lime Stabilized QLD Soil

The study aims to develop proposed predictive formulas for determining the unconfined compression strength (UCS) of cement/lime stabilized Queensland soil based on Multi-Gene Genetic Programming (MGGP) and Artificial Neural Network (ANN). The models evaluate the effect of three independent variables, including the binder type (cement and lime), the binder content, and the curing time, on the UCS of the stabilized soil. The results show that the selected optimal MGGP and ANN models can predict the target values with high correlation coefficients (R-value approximately of 0.992 and 0.998, respectively), and low errors (e.g., RMSE and MAE). The sensitivity analysis of the MGGP and ANN models provide the same results, in which the curing time has the greatest influence on the UCS value, followed by the binder content and binder type. The performances of the MGGP and ANN models are compared based on statistical parameters, several external criteria, and distribution properties. The study finds that both models show their generalization capabilities with robust, powerful, and accurate prediction ability; however, the ANN model slightly outperforms the MGGP model. The proposed predictive equations formulated from the selected optimal MGGP and ANN models could help engineers and consultants to choose the suitable binder and the reasonable amount of binder in the pre-planning and pre-design period.