Algorithm 2 The proposed hybrid missing data imputation method

Input: The original dataset $X_{ORG}$

Output: The imputed complete dataset $X_{IMP}$

1. Begin

2. Unfolding data along the batch dimension, get the 2D dataset X;

3.Classifying the missing data into five categories: transient isolated missing values, short-term missing variables, long-term missing variables, short-term missing samples and long-term missing samples;

4. Splitting dataset X, get $X=[X_1, X_1^{*}, \dots ,X_k, X_{k^{*}}^{*}, \dots , X_{K-1},X_{K-1}^{*}, X_K]$;

5.Imputing transient isolated missing values in each data segment $X_k$ using single-dimensional interpolation models;

6. ${X_k}^{\left(1\right)}(k=1, ..., K)$ ← The imputed data segments;

7. Standardize each data segment;

8.Imputing long-term missing variables in each data segment $X_k$ using the iterative imputation based on multivariate regression model, and imputing short-term missing variables in each data segment $X_k$ using the combination model based on single-dimensional interpolation and multivariate regression;

9. ${X_k}^{\left(2\right)}(k=1, ..., K)$ ← The imputed data segments;

10.Imputing short-term missing samples and long-term missing samples (i.e., the missing data segments $X_{k^{*}}^{*}$) using LSTM model;

11. ${X_{k^{*}}^{*}}^{\left(1\right)}(k^{*}=1, \dots , K-1)$ ← The imputed data segments;

12. Complete dataset $X_{IMP}$ ← De-standardize, and transform 2D data to 3D data;

13. End