Combining knowledge graph with deep adversarial network for water quality prediction.

Area of Science:

• Environmental science
• Artificial intelligence
• Data science

Background:

• Accurate water quality prediction is crucial for smart water management and pollution control.
• Current models primarily rely on sensor data, limiting their predictive power.
• Integrating diverse knowledge sources can improve prediction accuracy.

Purpose of the Study:

• To develop an advanced water quality prediction model by combining data-driven and knowledge-driven approaches.
• To enhance the accuracy and reliability of predicting water quality parameters, such as total nitrogen.
• To address the limitations of existing data-only models in smart water applications.

Main Methods:

• A framework integrating a knowledge graph with deep adversarial networks was developed.
• Knowledge extraction utilized a deep adversarial joint model to build a water quality knowledge graph.
• A weighted CNN-LSTM model with adversarial learning was employed for total nitrogen prediction, incorporating fused parameter importance.

Main Results:

• The proposed model demonstrated a significant improvement in water quality prediction accuracy.
• Experimental results on data from the Juhe River validated the model's effectiveness.
• The integration of knowledge and data effectively improved the prediction of total nitrogen levels.

Conclusions:

• The novel approach of combining knowledge graphs and deep adversarial networks offers a powerful tool for water quality prediction.
• This integrated method enhances predictive accuracy beyond traditional data-driven models.
• The findings support the application of this framework in smart water management for pollution control.