Research On The Prediction And Control Strategy Optimization Inversion Of Trailing Suction Hopper Dredger Loading Production Based On Deep Learning

As the main vessel for land reclamation,dredging waterways,and port construction,the trailing suction hopper dredger(TSHD)has been widely used in the dredging industry.In recent years,with the rapid development of deep learning technology,intelligent dredging has become a new direction for the development of the dredging industry.In traditional dredging construction,control parameters are often adjusted based on manual experience,which leads to difficulties in maintaining high levels of dredging efficiency for dredgers.Therefore,predictive analysis and control parameter optimization of the TSHD’s loading process have significant guiding significance for improving dredging efficiency and reducing energy consumption.This article aims to address the real-time demand for production forecasting in the TSHD loading process,the insufficient generalization ability of traditional deep learning models for multiple operating conditions,and the difficulty in selecting control strategies in the loading process.Based on the mechanism of the suction dredging operation and a large amount of historical monitoring data,this article conducts research on deep learning production prediction for the loading process,parameter transfer for production and energy consumption during multiple operating conditions,and optimization of the loading process control strategy.The main contents of this research are as follows:1.Research on production prediction for the TSHD loading process based on deep learning.This article conducts production prediction for the TSHD loading process using data-driven and deep learning methods based on different model depths,different control parameter feature selection methods,and different attention mechanisms,and employs GA optimization algorithms,Dropout mechanisms,and EarlyStopping mechanisms to optimize the deep learning model structure.The effectiveness of this method is validated for single-step and long-term production prediction.2.Research on production and energy consumption prediction for the TSHD loading process based on parameter transfer.This article solves the problem of weak generalization performance and low prediction accuracy of traditional deep learning models for multiple operating conditions by using parameter transfer methods for production and energy consumption data during different dredging conditions.Based on the principles of transfer learning and statistical methods,this article first determines the pre-training model structure for production prediction under a benchmark dredging condition in the Yangtze River Estuary.It then conducts production and energy consumption transfer prediction research under different freezing strategies,different learning methods,different learning rates,and different stepping lengths for multiple dredging conditions(Yangtze River Estuary,Lianyungang,and Xiamen),and demonstrates that transfer learning methods can achieve better generalization performance and sample fit accuracy than traditional deep learning models.3.Research on the optimization design of the suction dredger loading process control strategy.This article employs MDS feature dimension reduction,response surface methodology,and RBF neural networks to establish the interaction relationships between the suction head depth,suction head angle to the ground,sailing speed,and production.The optimization design of loading strategies under small sample construction data is achieved.In addition,this article designs a control parameter optimization inversion model based on a closed-loop network structure,and through mathematical statistics and deep learning methods,identifies the optimal operating point control parameters and the relationship between loading production,leading to an improved loading process control parameter combination design scheme that greatly improves the efficiency of suction dredger loading construction.