Digital Transformation Quality Control of Power Grid Production Project Management Based on Deep Learning

Aiming at the dynamic response lag and adaptability of complex operating conditions faced by quality control in digital transformation of power grid production project management, an intelligent evaluation framework based on LSTM-Transformer hybrid model (LSTM is the abbreviation of Long Short-Term Memory Networks) is proposed in this paper. By integrating the production data of 15 provincial companies of the State Grid from 2019 to 2025, covering the three business types of UHV (Ultra High Voltage) construction, distribution network transformation and new energy grid connection, a quality control system including 8-dimensional indicators such as equipment abnormality occurrence rate and voltage qualification rate is established. The model adopts a dual-channel feature extraction mechanism and integrates nonlinear risk indicators with CRITIC (Criteria Importance Through Intercriteria Correlation) weighting method. The prediction accuracy, F1 score and mAP value of 92.7%, 90.3% and 87.9% are achieved on the spatio-temporal isolation validation set, which improves the accuracy of 20.7% compared with the single LSTM model. In the application of C city power supply project, the model warns abnormal partial discharge 21 days in advance, the average response time is 4.7 hours, and the unplanned outage rate of the line is reduced to 0.23 times/100 km•year, which is significantly better than the assessment standard of the State Grid. The empirical results show that this framework provides technical support for the transformation of power grid digitalization from passive disposal to active prevention and control.