Online monitoring data for marine microplastics often suffers from biases or missing values, significantly impacting the analysis of regional microplastic distribution. To enhance the accuracy of data repair, we propose a T2V-Transformer deep learning repair method based on the Mutli-source Monte Carlo. Drawing from previously collected data on microplastics in the Beibu Gulf, we have established a Monte Carlo model for predicting the spatiotemporal transmission of microplastics, extracting the root mean square error (RMSE) between estimated and observed values as the error sequence. The error values serve as the output of the transformer, with inputs including microplastic presence characteristics, latitude and longitude range, meteorological data types, number of stations, and seawater hydrological data parameters. Time2Vec, a method to encode temporal information into the model and introduce a unique time information tagging signal. The T2V-Transformer is able to handle high-dimensional time series data and capture inter-factor dependencies through the self-attention mechanism Analysis results indicate that the Monte Carlo-based T2V-Transformer model, which calculates the decision coefficient (R²) through the error sequence, achieves higher predictive accuracy than traditional machine learning models. The established model predicted the microplastic abundance and characteristics for several hours ahead with an explained variance score （EVS）of 0.903, which is higher than the XGBOOST, MLP, and the Decision Tree model. The microplastic spatiotemporal transmission prediction model we have developed demonstrates good performance in microplastic prediction and data repair, offering a novel approach for the repair of missing microplastic data across various maritime regions.

Microplastics , Data repair Mutli-source Monte Carlo, Deep Learning