Abstract: In order to predict seam destroyed floor depth accurately, firstly, main factors affecting destroyed floor depth were identified to futher determine the evaluation indicators on the basis of a comprehensive analysis. ELM optimizing the input weights and hidden layer bias by PSO was chosen as a base predictor, Boosting as a ensemble learning framework, and a strong predictor model was built for destroyed floor depth based on PSO-ELM-Boosting theory. Then, effect of combination of ELM hidden layer nodes and base predictor number on the prediction result was studied, and the most superior combination of both was obtained. Besides, the "weight expansion" issues were avoided by controlling the weight of sample. Finally, according to the 64 available data from destroyed floor depth measurement, a series of experiments were carried out to compare and analyze the prediction results obtained by using the PSO-ELM-Boosting model and others. The experimental results show that the PSO-ELM-Boosting model has lower mean absolute percentage error(4.54%), mean squared error(0.429 2 m²) and higher R² (0.956 5),which illustrates the effectiveness of the PSO-ELM-Boosting model.

Key words: destroyed floor depth, Boosting algorithm, ensemble learning, extreme learning machine (ELM), particle swarm optimization(PSO)