Search

A dependable evaluation of the stability of slopes is a prerequisite in many construction projects. Although machine learning models have been satisfactorily used for this purpose, combining them with metaheuristic optimizers has resulted in a larger accuracy. This study, therefore, suggests the use of equilibrium optimization (EO) and vortex search algorithm (VSA) for optimizing a multi-layer perceptron neural network (MLPNN) employed to anticipate the factor of safety of a single-layer soil slope. Two hybrid models, as well as the regular MLPNN, are fed by a total of 630 data acquired from finite element simulations. The results, first, showed the applicability of artificial intelligence in this field. Next, reducing the training root mean square error (RMSE) of the MLPNN (from 0.4715 to 0.3891 and 0.4383 by the EO and VSA, respectively) revealed the efficiency of the used algorithms in remedying the computational weaknesses of this model. Moreover, the testing RMSE declined from 0.5397 to 0.4129 and 0.5155, which indicates a higher generalization ability of the hybrid models. Furthermore, due to the larger accuracy of the EO-based ensemble, this algorithm outperformed the VSA in optimizing the MLPNN.

To achieve an efficient methodology for approximating pan evaporation (EP), this study offers two metaheuristic-integrated predictors. Shuffled complex evolution (SCE) and electromagnetic field optimization (EFO) are two of the fastest metaheuristic algorithms that are synthesized with artificial neural network (ANN). By doing this, the ANN is optimized in a noticeably shorter time compared to its integration with other metaheuristic techniques. Five-year climatic data of the Bakersfield station (California, USA) with an 80:20 ratio are used for developing and testing the methods. The proposed hybrids are implemented with appropriate population sizes (20 and 35 for the SCE and EFO, respectively) and their results are compared to a single ANN. Accuracy evaluation (correlation coefficients > 0.99) professed that the neural network with both conventional and sophisticated trainers is a competent approach for the EPsimulation. Besides, it was observed that the error of prediction by the ANN-SCE and ANN-EFO is 6.02 and 9.27% lower than the single ANN, respectively. Therefore, the used strategies can enhance the applicability of the ANN. The time elapsed in the optimization using SCE and EFO was 479.0 and 281.9 s, respectively. A comparison between these algorithms revealed that the EFO is both a faster and more accurate optimizer. The ANN-EFO is accordingly recommended as a new efficient model for predicting the EP.

The crucial significance of proper management of heating, ventilating, and air conditioning systems in energy-efficient buildings were the main reason for dedicating this study to test a novel approach for this task. Shuffled complex evolution (SCE) is an efficient metaheuristic technique that is used to optimize the performance of a multi-layer perceptron neural network (MLP) for accurate prediction of cooling load (CL). The CL information of 768 residential buildings, obtained from a vast computer simulation in the published literature, is used to train and validate the performance of the proposed model. The results showed that the SCE could properly surmount the computational drawbacks of the MLP, as its learning and prediction accuracies are enhanced by 19.52 and 22.84%, respectively. Also, the SCE outperformed two benchmark optimizers of moth–flame optimization and optics inspired optimization in both training and testing phases. Another advantage of the tested SCE-MLP was the considerably simpler structure, and consequently, shorter computation time (722 vs. 1050 and 46,192 s). Therefore, the proposed model can be promisingly used in practice for the early prediction of CL in energy-efficient buildings.

Recent improvements achieved using nature-inspired optimizers encouraged the authors to employ a novel type of metaheuristic algorithms, namely crow search algorithm (CSA) in this study. The CSA is employed for optimizing a feed-forward artificial neural network (ANN) in predicting the soil compression coefficient (SCC). The SCC is one of the most crucial geotechnical parameters that the early prediction of it can increase the safety and cost-effectiveness of a project. For more reliability, the used data are collected from a real-world project. After developing the CSA–ANN hybrid, the most proper values for the algorithm parameters, including flock size, flight length, and awareness probability are found by sensitivity analysis (to be 400, 2, and 0.1, respectively). A comparison between the results of the typical ANN and the CSA-trained version revealed that the proposed algorithm can effectively reduce the mean absolute error (MAE) in both learning and predicting the SCC pattern (by 8.25 and 7.29%, respectively). Moreover, the increase of the coefficient of determination (R2) from 70.64 to 74.83% in the training phase, and from 73.74 to 76.19% in the testing phase proves the efficiency of the CSA in enhancing the ANN. The suggested CSA–ANN, therefore, can be an efficient model for the early prediction of the SCC in civil/geotechnical engineering projects.

Proper approximation of friction capacity (FC) of driven piles is a noticeable issue in geotechnical engineering. Hence, the pivotal focus of the current research is on proposing reliable predictors for evaluating this parameter. Three artificial neural networks (ANNs) improved by firefly algorithm (FA), multi-tracker optimization algorithm (MTOA), and black hole algorithm (BHA) are used to estimate the FC when it is affected by four key factors, namely pile length, pile diameter, vertical effective stress, and undrained shear strength. Checking the optimization process of different population sizes revealed that the MTOA and BHA need the population twice as many as the FA does (400 vs. 200). The results showed that all three models can properly comprehend the association of FC to the mentioned parameters. According to the values of root mean square error (RMSE) as well as the coefficient of determination (R2) obtained in the prediction phase (6.9606 and 0.8827, 8.5411 and 0.8370, 6.0454 and 0.9194, respectively, for the FA-ANN, MTOA-ANN, and BHA-ANN), the BHA-ANN is more accurate than two other algorithms, however, the MTOA-ANN gained the largest accuracy of training. The suggested models proved the high efficiency of hybrid predictors and can be potential alternatives to experimental and traditional approaches.

Analyzing the thermal load is a significant task for energy-efficient buildings. Intelligent models have shown high reliability for predicting the heating load of different buildings. But computational shortcomings like local minima have yet remained a drawback that can be remedied by optimization techniques. This study investigates the efficiency of two novel metaheuristic algorithms, namely vortex search algorithms (VSA) and backtracking search algorithm (BSA), for optimizing the performance of a multilayer perceptron neural network (MLP). The methods are used to predict the heating load of a residential building. Evaluation of the results revealed that the proposed metaheuristic schemes could properly optimize the MLP. In this regard, the training error experienced around 19.99 and 5.99% reduction by synthesizing the VSA and BSA, respectively. These values were obtained 20.39 and 6.18% for the testing phase. Also, the correlation of the MLP products rose from 93.52 to 95.62 and 94.00%. Although the best-fitted BSA was around six times as fast as VSA, the VSA-base ensemble enjoys more accuracy of prediction. Overall, the findings showed that utilizing the VSA-MLP and BSA-MLP models is a promising way for the early prediction of the heating load.

This paper presents the planning of a hybrid renewable system with wind turbines and bio-waste energy units along with stationary (i.e., batteries) and mobile (i.e., electric vehicles) energy storage. This model minimizes the cost of construction, maintenance and storage degradation. In this model, the energy supply priority is given to renewable sources. Then the mentioned storage devices are used to cover the gap between the load profile and the renewable generation power. The proposed scheme addresses the uncertainties of loads, renewable power and energy consumption of mobile storage devices. Therefore, robust optimization based on information-gap decision theory (IGDT) is utilized to obtain a robust solution against the prediction error of the aforementioned uncertainties. Also, the hybrid optimization algorithm of honey bee mating and artificial bee colony is derived, showing the optimal solution with low dispersion in the final response. The proposed design is applied to the data of the city of Espoo, Finland. The obtained numerical results show the capability of the proposed design in deriving robust economic planning for the proposed hybrid system. So, the used hybrid solution algorithm is able to obtain the best solution in a lower computation time compared to non-hybrid solvers, and it has a low standard deviation around 0.94 % in the final response. The presence of the aforementioned renewable sources leads to attaining an environment-friendly hybrid system. The optimal performance of the storage devices leads to the robustness of the optimal solution against the maximum error of 18.19 % in the prediction of uncertainties. This occurs considering the smart charging management of mobile storage, but the maximum uncertainty level of 11.12 % holds to non-smart management of these storage devices in the proposed scheme. The energy management of mobile storage devices based on smart (non-smart) charging strategy also reduces (increases) the planning cost of the off-grid system by 7.62 % (39.68 %) compared to their absence.