Scholarlink

<p>An accurate early forecast of water quality indicators helps to take preventive measures for the sustainable development of the ecosystem. However, it is not easy to handle the dynamically changing and highly influential nature of water bodies. To the best of our knowledge, though the existing research has made commendable progress in water quality forecasting using Artificial Intelligence, the results of some of the existing works fall short due to insufficient treatment of missing data, negligence in spatiotemporal factors, inability to model nonlinear dependencies, or compromised sequential long-term prediction. To resolve the issues, this research proposed an advanced framework for water quality forecasting to integrate data pre-processing, feature enhancement, and hybrid modelling techniques. In the pre-processing phase, missing values are addressed using an unsupervised Generative Adversarial Network, which generates realistic synthetic data to improve dataset efficiency. Further, Data is reshaped by spatial-temporal-parameter embedding for feature enhancement, thereby significantly refining the effectiveness of a self-attention mechanism. The hybrid modelling amalgamates Multi-headed Self-attention with a Sugeno-Fuzzy Inference System. The forecasting is improved due to the effective handling of non-linearity along with multi-perspective parallel processing of long-term dependencies. The considered dataset is collected from the Bihar Pollution Control Board, India. Moreover, the performance evaluation across various time lags reveals that an 8-hour forecasting window achieves optimal performance. It effectively shows long-term dependencies over a 24-hour horizon. The proposed model outperforms existing methods in terms of RMSE, MAE, R², category accuracy of parameters, and overall forecasting efficiency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DeepFuzzy-ANDO: ammonium nitrate and dissolved oxygen based water quality forecasting using generative Attention-based fuzzy neural network

  • 期刊论文

  • 发表日期：

    2026年1月13日

查看全文
  • Arpita Narayan,
  • Ayan Kumar Das

摘要

An accurate early forecast of water quality indicators helps to take preventive measures for the sustainable development of the ecosystem. However, it is not easy to handle the dynamically changing and highly influential nature of water bodies. To the best of our knowledge, though the existing research has made commendable progress in water quality forecasting using Artificial Intelligence, the results of some of the existing works fall short due to insufficient treatment of missing data, negligence in spatiotemporal factors, inability to model nonlinear dependencies, or compromised sequential long-term prediction. To resolve the issues, this research proposed an advanced framework for water quality forecasting to integrate data pre-processing, feature enhancement, and hybrid modelling techniques. In the pre-processing phase, missing values are addressed using an unsupervised Generative Adversarial Network, which generates realistic synthetic data to improve dataset efficiency. Further, Data is reshaped by spatial-temporal-parameter embedding for feature enhancement, thereby significantly refining the effectiveness of a self-attention mechanism. The hybrid modelling amalgamates Multi-headed Self-attention with a Sugeno-Fuzzy Inference System. The forecasting is improved due to the effective handling of non-linearity along with multi-perspective parallel processing of long-term dependencies. The considered dataset is collected from the Bihar Pollution Control Board, India. Moreover, the performance evaluation across various time lags reveals that an 8-hour forecasting window achieves optimal performance. It effectively shows long-term dependencies over a 24-hour horizon. The proposed model outperforms existing methods in terms of RMSE, MAE, R², category accuracy of parameters, and overall forecasting efficiency.