Fig. 4: Interpretable analysis of models. | npj Clean Water

Fig. 4: Interpretable analysis of models.

From: Knowledge embedding and interpretable machine learning optimize comprehensive benefits for water treatment

Fig. 4

a Correlation grid plots of cast predicted values and input features, with the thickness and color of each line proportional to the correlation coefficient. b Global interpretation of significance results of model prediction results based on the SHAP approach. c SHAP percentage plot fan charts for the three features with the largest share and other features. d Supervised clustering of the test set distinguishing eigenvalues, with the horizontal axis being the test set sample percentage. e Modes of influence of the main characterizing variable. f Number and average depth of subtrees for the RF model. T temperature, Q flow, TUR turbidity, AN ammonia nitrogen, NP network pressure, WVC variation coefficient of water withdrawal, PE Photovoltaic electricity, WE electricity to obtain water, SE electricity for sludge. TE total electricity, EC electricity cost, CC chemical cost, TC total cost. The subscripts indicate different stages of wastewater treatment. \({X}_{R}:\) raw water. \({X}_{{IN}}:\) influent water. \({X}_{P}:\) pre-chlorination water. \({X}_{S}:\) sedimentation tank. \({X}_{D}:\) disinfection contact tank. \({X}_{{EFF}}:\) effluent water.

Back to article page