Article: Intelligent progress prediction for power grid engineering projects based on unstructured text data and deep learning Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.11 pp.1 - 14 Abstract: Power grid engineering progress prediction traditionally depends on structured numerical data, overlooking qualitative signals in unstructured documents. This study presents an intelligent prediction model (BERT-BiLSTM-TA) integrating BERT-based semantic extraction with bidirectional LSTM networks and temporal attention mechanisms to process daily reports, meeting minutes, and inspection records. Validation on 186 real-world projects achieves MAE of 3.67% for multi-horizon forecasts, a 28.3% reduction compared to the BiLSTM-attention baseline (MAE 5.12%). Pilot deployment across five projects detects emerging delays 3-4 weeks earlier than conventional monitoring, reducing schedule overruns from 8.3% to 3.7%. Attention visualisation enhances interpretability for practical adoption. This work establishes a methodological foundation for transforming textual information into actionable predictive insights for construction management. Inderscience Publishers - linking academia, business and industry through research

Title: Intelligent progress prediction for power grid engineering projects based on unstructured text data and deep learning

Authors: Jian Shen; Jinxia Li; Huaxing Bian; Tianmu Hu

Addresses: Material Branch, State Grid Jiangsu Electric Power Co., Ltd., Nanjing 210000, Jiangsu, China ' Material Branch, State Grid Jiangsu Electric Power Co., Ltd., Nanjing 210000, Jiangsu, China ' Material Branch, State Grid Jiangsu Electric Power Co., Ltd., Nanjing 210000, Jiangsu, China ' Jiangsu Electric Power Information Technology Co., Ltd., No. 22, Beijing West Road, Gulou District, Nanjing City, 210000, Jiangsu, China

Abstract: Power grid engineering progress prediction traditionally depends on structured numerical data, overlooking qualitative signals in unstructured documents. This study presents an intelligent prediction model (BERT-BiLSTM-TA) integrating BERT-based semantic extraction with bidirectional LSTM networks and temporal attention mechanisms to process daily reports, meeting minutes, and inspection records. Validation on 186 real-world projects achieves MAE of 3.67% for multi-horizon forecasts, a 28.3% reduction compared to the BiLSTM-attention baseline (MAE 5.12%). Pilot deployment across five projects detects emerging delays 3-4 weeks earlier than conventional monitoring, reducing schedule overruns from 8.3% to 3.7%. Attention visualisation enhances interpretability for practical adoption. This work establishes a methodological foundation for transforming textual information into actionable predictive insights for construction management.

Keywords: power grid engineering; progress prediction; unstructured text data; deep learning; BERT; temporal attention.

DOI: 10.1504/IJRIS.2026.152728

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.11, pp.1 - 14

Received: 12 Nov 2025
Accepted: 30 Jan 2026
Published online: 07 Apr 2026 *

Title: Intelligent progress prediction for power grid engineering projects based on unstructured text data and deep learning

Keep up-to-date