Intelligent prediction of main-works construction duration for pumped storage power stations using small-sample machine learning
Bo Wang, Qikai Li, Tianyu Fan, Hao Dong, Anlan Li, Xiangtian NiePurpose
Pumped storage has become widely recognized as one of the major technical options for power system regulation, and its development is considered important, necessary, and urgent across the industry. Construction duration prediction is a key component of schedule management for pumped-storage power station projects. However, owing to the limited number of completed stations and confidentiality restrictions, existing studies perform poorly under small-sample conditions.
Design/methodology/approach
This study proposes an intelligent prediction model for the construction duration of the main works of pumped-storage power stations by integrating similarity-based project selection with small-sample machine learning. First, highly relevant factors are selected from a large number of candidate factors affecting construction duration. Second, based on the kernelized angular measure method, both scale differences and structural proportional differences are jointly considered to identify similar projects. Finally, an intelligent prediction model for the construction duration of the main works is developed based on Improved Harris Hawks Optimization and Least Squares Support Vector Machine.
Findings
This study identifies seven major factors affecting the construction duration of pumped-storage power stations, including installed capacity. Case studies further demonstrate that the proposed intelligent construction duration prediction method for small samples achieves a prediction accuracy of 95.46%, demonstrates good predictive performance on the test dataset, meets the accuracy requirements for construction duration planning, and significantly outperforms traditional construction duration prediction algorithms.
Originality/value
The main contribution of this study lies in integrating similarity-based computation with multiple machine learning methods to propose a construction duration prediction framework for pumped-storage power stations under small-sample conditions, enabling accurate capture of the complex nonlinear relationships between influencing factors and construction duration, significantly improving the accuracy and efficiency of construction duration prediction, and providing a useful reference for the rational planning of construction organization in pumped-storage power station projects.