Abstract: A missing inventory data interpolation model based on improved matrix decomposition is designed for inventory missing data. According to the characteristics of inventory data, the unit root test and Nemenyi post-hoc multiple comparison was adopted to analyze the data stationarity and significance. For missing data with different missing patterns, a time regularizer long short-term memory neural network was introduced to obtain the time dependence in time series data, and a space regularizer graph Laplacian was used to consider the spatiotemporal characteristics by taking advantage of the spatial correlation among network sensors. Meanwhile, an Adam optimizer was added to achieve high-performance interpolation of inventory missing data. According to the data characteristics, the RMSE evaluation metric was adopted for model evaluation. Through comparative studies with advanced methods, it is proved that the model has superior interpolation performance.