A multivariate temporal communication framework with novel coarse-fine-grained imputation for missing data packets in ship-shore collaboration systems

Information transmission in ship-shore cooperative systems is a critical foundation for intelligent navigation systems. However, the high packet loss rate and limited transmission frequency of the BeiDou satellites severely degrade the data transmission performance. To address this challenge, we propose a multivariate temporal communication (MTC) framework for intelligent vessels that incorporates advanced imputation technology. We analyzed the characteristics of multiattribute and multiconditional time series and develop an irregular sampling principal component analysis to differentiate between attributes. We then developed a novel coarse-fine-grained imputation (CFGI) model that is implemented using a port monitoring communication server. Specifically, the coarse-grained stage involves the use of an exponentially weighted moving average control chart with estimation error, which enables the adaptive detection of communication anomalies in BeiDou short messages. In the fine-grained stage, an attribute-related stacked autoencoder is used to learn attribute-related features from ship gas turbines while modifying the loss function and gradient computation to adjust variable weights. The port communication server applies the CFGI model to detect and impute lost packets, thereby eliminating the need for traditional data retransmission. Experimental validation on the BeiDou-3 satellite communication platform demonstrates the effectiveness of the proposed framework in packet imputation and remote fault diagnosis.