关键词: 地下厂房, 施工通风, 多目标优化, 代理模型, 最小二乘支持向量回归

Abstract: Formulating a reasonable construction ventilation scheme is the key to ensuring safety and efficiency in underground powerhouse construction. Most of the previous studies on ventilation scheme optimization started from a single optimization objective such as ventilation smoke dissipation time and average pollutant concentration; traditional numerical simulation methods have the shortcomings of high modeling cost and low computational efficiency. This paper presents a new multi-objective optimization method for the construction ventilation schemes of an underground powerhouse based on an improved Least Squares Support Vector Regression (LSSVR) surrogate model of Improved Dung Beetle Optimizer (IDBO). First, a mathematical model for multi-objective optimization of the schemes is constructed, taking ventilation effect and ventilation cost as optimization objectives, and selecting ventilation parameters as design variables, such as fan airflow and the distance from the duct opening to the palm surface. Then, an IDBO-LSSVR surrogate model is constructed for prediction of the ventilation effect by combining the advantage of LSSVR in predicting small-sample data; the IDBO-improved LSSVR regularization parameter is used to optimize the LSSVR regularization parameter γ and kernel parameter σ, thereby overcoming the difficulty in model hyperparameter specification and achieving a fast prediction of the ventilation effect target. And combined with the NSGA-II algorithm, the surrogate model gives a multi-objective optimization solution. Finally, this method is applied to an underground plant project at the Luoning pumped storage power station, achieving the optimized construction ventilation scheme and a fast and accurate prediction of ventilation effect. The results show that the optimized scheme increases the ventilation and dust removal rate by 20.01%, and reduces the ventilation cost by 9.52%.

Key words: underground powerhouse, construction ventilation, multi-objective optimization, surrogate model, least squares support vector regression