# Intelligent diagnosis method for river and lake ecosystem health based on improved slime mold algorithm-optimized SVR

## TL;DR

A new intelligent model using an improved slime mold algorithm optimizes water quality predictions for river and lake ecosystems.

## Contribution

Proposes an improved slime mold algorithm-optimized SVR model for enhanced accuracy and adaptability in ecosystem health diagnosis.

## Key findings

- The optimized model achieved an R2 of 0.976 and RMSE of 0.022, outperforming comparison models.
- The model showed good prediction performance for parameters like pH, dissolved oxygen, and chemical oxygen demand.
- The model's memory usage and sensitivity to outliers were better than existing methods.

## Abstract

In the diagnosis of river and lake ecosystems, there are complex nonlinear relationships among water quality parameters, and their dynamic change mechanisms are rather complicated. Traditional statistical analysis methods have limitations in providing precise assessment and timely early warning. To address the bottlenecks of traditional methods in accuracy, timeliness, and applicability, an intelligent diagnostic model based on improved slime mold algorithm-optimized support vector regression is proposed. This model improves its parameter optimization ability through dynamic weight strategy and adaptive search mechanism, and combines LightGBM feature selection to construct a combined model, effectively solving the problems of high-dimensional data modeling and dynamic adaptability. The experimental findings reveal that the optimized model improves indicators such as RMSE, MAE, and R2 compared to the comparative model. The RMSE is 0.031, the MAE is 0.021, and the R2 is 0.942. The prediction accuracy of the final proposed combination model is further optimized, with an RMSE of 0.022, an MAE of 0.016, and an R2 of 0.976. In addition, the average memory usage of the combined model is 120.5MB. The average sensitivity to outliers was 0.2, and the values were all better than those of the comparison models. At the same time, the prediction effects on pH value, dissolved oxygen, permanganate index, total phosphorus index, ammonia nitrogen index and chemical oxygen demand are relatively good. The research provides efficient and accurate methods for water quality prediction and ecosystem health diagnosis. The results show that the model proposed in the study has superior performance in the diagnosis of river and lake ecosystems and a good actual prediction effect. The intelligent diagnostic method proposed in the study enhances the ecological risk prevention and control capabilities of rivers and lakes, and promotes the digital transformation of water environment management.