FM-GRU: A Time Series Prediction Method for Water Quality Based on seq2seq Framework

Jianlong Xu,Kun Wang,Che Lin,Lianghong Xiao,Xingshan Huang,Yufeng Zhang

DOI: https://doi.org/10.3390/w13081031

IF: 3.53

2021-04-09

Water

Abstract:Water quality prediction plays a crucial role in both enterprise management and government environmental management. However, due to the variety in water quality data, inconsistent frequency of data acquisition, inconsistency in data organization, and volatility and sparsity of data, predicting water quality accurately and efficiently has become a key problem. This paper presents a recurrent neural network water quality prediction method based on a sequence-to-sequence (seq2seq) framework. The gate recurrent unit (GRU) model is used as an encoder and decoder, and a factorization machine (FM) is integrated into the model to solve the problem of high sparsity and high dimensional feature interaction in the data, which was not addressed by the water quality prediction models in prior research. Moreover, due to the long period and timespan of water quality data, we add a dual attention mechanism to the seq2seq framework to address memory failures in deep learning. We conducted a series of experiments, and the results show that our proposed method is more accurate than several typical water quality prediction methods.

environmental sciences,water resources

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to solve several key challenges in water quality prediction. Specifically, water quality data has the following characteristics:

1. **Indicator diversity**: Water quality data contains multiple indicators, such as 24 basic pollutants and other chemical elements and organic compounds, with a total of 109 different indicators. This increases the complexity of parallel computing.

2. **Unfixed sampling frequency**: Different monitoring indicators have different sampling frequencies (hourly, daily, monthly), which brings additional difficulties to data processing and model construction.

3. **Complex correlations between indicators**: There may be complex causal relationships between different indicators of water quality data. If these relationships cannot be accurately extracted, the accuracy of the prediction model will be greatly reduced.

4. **Long time span**: The time span of water quality monitoring data may be as long as several years, which challenges the long - term information processing ability of the model.

Due to the above challenges, traditional water quality prediction models (such as linear regression) often cannot obtain high - precision prediction results, and are especially prone to large errors in the case of unbalanced data. To solve these problems, this paper proposes a new time - series prediction method - FM - GRU, which is based on the seq2seq framework and combines the Factorization Machine (FM) and the Gated Recurrent Unit (GRU), as well as the dual - attention mechanism.

### Main contributions of the FM - GRU model

1. **Introduction of Factorization Machine (FM)**: It is used to extract high - dimensional feature information in time series, solving the problem that traditional manual feature engineering is complex and time - consuming.

2. **Addition of dual - attention mechanism**: It solves the problem of information loss in the traditional seq2seq framework when dealing with long - time series, and improves the model's prediction ability for long - span time series.

3. **Improved encoding - decoding structure**: By inputting the high - dimensional features output by the FM module into the improved encoder - decoder model, effective prediction of water quality data is achieved.

In short, the goal of this paper is to improve the accuracy and efficiency of water quality prediction through the FM - GRU model, so as to better support environmental management decisions of enterprises and governments.