Neural Tucker Convolutional Network for Water Quality Analysis

TL;DR

Water quality data are often missing due to sensor failures, creating high-dimensional sparse datasets. The authors propose NTCN, which learns embeddings for stations, indicators, and time slices, builds a high-order interaction tensor via outer product, then applies two layers of 3D convolution to extract spatiotemporal features and an MLP to predict missing values. The model uses L2 regularization and standardization to handle sparsity and prevent overfitting. Experiments on three real-world Hong Kong datasets show that NTCN achieves lower RMSE and MAE than several state-of-the-art imputation methods, demonstrating effective capture of cross-mode interactions. The approach offers a scalable, accurate solution for water quality monitoring data, with potential extensions to other ecological sensing tasks.

Abstract

Water quality monitoring is a core component of ecological environmental protection. However, due to sensor failure or other inevitable factors, data missing often exists in long-term monitoring, posing great challenges in water quality analysis. This paper proposes a Neural Tucker Convolutional Network (NTCN) model for water quality data imputation, which features the following key components: a) Encode different mode entities into respective embedding vectors, and construct a Tucker interaction tensor by outer product operations to capture the complex mode-wise feature interactions; b) Use 3D convolution to extract fine-grained spatiotemporal features from the interaction tensor. Experiments on three real-world water quality datasets show that the proposed NTCN model outperforms several state-of-the-art imputation models in terms of accuracy.

Figures (1)

• Figure 1: Architecture of the NTCN model for water quality data imputation. The numerical labels in the figure correspond to the parameter configurations employed in the experiments, including kernel size, stride, and number of channels.