Water quality assessment of deep learning-improved comprehensive pollution index: a case study of Dagu River, Jiaozhou Bay, China.

In the past few decades, with the country's rapid development, water pollution has become a significant problem many countries face. Most of the existing water quality evaluation uses a single time-invariant model to simulate the evolution process, which cannot directly describe the complex behavior of long-term water quality evolution. In addition, the traditional comprehensive index method, fuzzy comprehensive evaluation, and gray pattern recognition have more subjective factors. It can lead to an inevitable subjectivity of the results and weak applicability. Given these shortcomings, this paper proposes a deep learning-improved comprehensive pollution index method to predict future water quality development. As a first processing step, the historical data is normalized. Three deep learning models, multilayer perceptron (MLP), recurrent neural network (RNN), and long short-term memory (LSTM), are used to train historical data. The optimal data prediction model is selected through simulation and comparative analysis of relevant measured data, and the improved entropy weight comprehensive pollution index method is applied to evaluate future water quality changes. Compared with the traditional time-invariant evaluation model, the feature of this model is that it can effectively reflect the development of water quality in the future. Moreover, the entropy weight method is introduced to balance the errors caused by subjective weight. The result shows that LSTM performs well in accurately identifying and predicting water quality. And the deep learning-improved comprehensive pollution index method can provide helpful information and enlightenment for water quality change, which can help improve the water quality prediction and scientific management of coastal water resources.