Attention-ResNet-AR-LSTM: An intelligent method for PCCP deformation prediction via structure monitoring based on distributed fiber optics

Abstract

The prestressed concrete cylinder pipeline (PCCP) is extensively employed in long water diversion projects due to its commendable performance and cost-effectiveness. Nonetheless, the continuous application of load on the prestressed steel wire during the production phase can amplify the creep deformation in both the concrete and steel cylinder of the PCCP. This phenomenon may potentially lead to the development of cracks at the socket end of the PCCP, posing a significant risk to the safe operational integrity of the pipeline. In this study, the strain development prediction model of PCCP concrete and steel cylinder is constructed through the integration of ResNet, AR, and LSTM modules, and the attention mechanism is also applied. Based on the prototype test of PCCP structure monitoring, the strain values of steel cylinder and inner concrete were recorded by pre-set distributed fiber optic sensors on the 1st, 3rd, 5th, 7th, 9th, 11th, 13th, 15th, 17th, and 20th day after the production of PCCP. Then an Attention-ResNet-AR-LSTM model is constructed to predict the strain development of PCCP steel cylinder and concrete based on monitoring data on the 20th day using the data of 1st to 11th days. For steel cylinder and concrete structures, the R2 of the proposed models is above 98%, indicating that the proposed models can capture the strain changes of different structures well. Compared with other deep models and machine learning models, the proposed model also shows a significant predictive ability. This model introduces a novel approach to forecast strain evolution during the manufacturing of PCCP.