Copula-based simulation and am-bilstm for missing value imputation of water monitoring data

CN122196361APending Publication Date: 2026-06-12NANJING HYDRAULIC RES INST +1

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: NANJING HYDRAULIC RES INST
Filing Date: 2026-02-06
Publication Date: 2026-06-12

AI Technical Summary

Technical Problem

Existing water conservancy project monitoring data suffers from insufficient accuracy, poor interpretability, and difficulty in selecting hyperparameters in scenarios with continuous data loss. Traditional methods struggle to fully capture the complex nonlinear relationships between monitoring indicators and have limited interpolation accuracy in high-dimensional data scenarios.

Method used

A method for imputing missing values in water conservancy monitoring data based on Copula simulation and AM-BiLSTM is adopted. By constructing a multi-index joint statistical model and generating twin data, and combining a bidirectional long short-term memory network with feature attention mechanism, hyperparameters are automatically searched and optimized to achieve high-precision imputation of missing values.

Benefits of technology

It improves the accuracy and reliability of interpolation of water conservancy monitoring data under multivariate, high-dimensional and continuous missing conditions, reduces the dependence on manual parameter adjustment and external simulation data, and enhances the interpretability and stability of the interpolation process.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122196361A_ABST

Patent Text Reader

Abstract

The application discloses a water conservancy monitoring data missing value imputation method based on Copula simulation and AM-BiLSTM. First, the auxiliary variables related to the target variable are screened from the multi-source monitoring indexes to construct a model input feature set; then, based on the edge distribution of each variable, a joint distribution structure is established by using a Copula function, and a twin monitoring data set is generated by inverse transformation sampling. A bidirectional long short-term memory network imputation model with a feature attention mechanism is introduced on the twin data, a Bayesian method is used for hyperparameter optimization search, the optimal structure parameter is determined, and after the model construction is completed, the predictive imputation of the missing values of the real monitoring data is carried out, and the complete data sequence and the corresponding attention weight matrix are output. The application combines statistical dependence modeling and deep learning methods, and improves the accuracy, reliability and interpretability of the reconstruction of water conservancy monitoring data under complex missing conditions.

Need to check novelty before this filing date? Find Prior Art