A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Estimating the compressive strength of roller compacted concrete using a novel swarm-optimised light gradient boosting machine
This study proposes a data-driven method for predicting the compressive strength (CS) of roller-compacted concrete (RCC). The new method is an integration of light gradient boosting machine (LightGBM) and particle swarm optimization (PSO). The former is used to generalize the mapping relationship between the CS and its influencing factors. The latter is employed to optimize the learning phase of LightGBM. A dataset, consisting of 408 laboratory tests and eight predictor variables, has been collected from the literature to train and verify the proposed data-driven method. Experimental results, supported by statistical hypothesis tests, point out that the proposed approach achieves outstanding performance with a root mean squared error of 4.53 and a mean absolute percentage error of 13.2%. The new model, named PSO-LightGBM, is capable of explaining roughly 90% of the variation in the CS of RCC mixes. In addition, an asymmetric loss function is applied in the training phase of LightGBM to reduce the proportion of overestimated CS values from 53% to 32%. This reduction in overestimations can help enhance the reliability of the data-driven method. Therefore, the newly developed model can serve as a decision-supporting tool in designing RCC mixtures.
Estimating the compressive strength of roller compacted concrete using a novel swarm-optimised light gradient boosting machine
This study proposes a data-driven method for predicting the compressive strength (CS) of roller-compacted concrete (RCC). The new method is an integration of light gradient boosting machine (LightGBM) and particle swarm optimization (PSO). The former is used to generalize the mapping relationship between the CS and its influencing factors. The latter is employed to optimize the learning phase of LightGBM. A dataset, consisting of 408 laboratory tests and eight predictor variables, has been collected from the literature to train and verify the proposed data-driven method. Experimental results, supported by statistical hypothesis tests, point out that the proposed approach achieves outstanding performance with a root mean squared error of 4.53 and a mean absolute percentage error of 13.2%. The new model, named PSO-LightGBM, is capable of explaining roughly 90% of the variation in the CS of RCC mixes. In addition, an asymmetric loss function is applied in the training phase of LightGBM to reduce the proportion of overestimated CS values from 53% to 32%. This reduction in overestimations can help enhance the reliability of the data-driven method. Therefore, the newly developed model can serve as a decision-supporting tool in designing RCC mixtures.
Estimating the compressive strength of roller compacted concrete using a novel swarm-optimised light gradient boosting machine
Hoang, Nhat-Duc (author)
2023-01-28
Article (Journal)
Electronic Resource
English
A compressive strength model for roller-compacted concrete with fly ash
| Online Contents | 2004
Strength of Roller Compacted Concrete
| British Library Online Contents | 2006
Strength of Roller Compacted Concrete
| British Library Conference Proceedings | 2006
| British Library Conference Proceedings | 2005