Using Fuzzy Models and Time Series Analysis to Predict Water Quality

Water quality prediction is very important for both water resource scheduling and management. Simple linear regression analysis and artificial neural network models cannot accurately forecast water quality because of complicated linear and nonlinear relationships in the water quality dataset. An adaptive neuro-fuzzy inference system (ANFIS) that can integrate linear and nonlinear relationships has been proposed to address the problem. However, the ANFIS model can only work in scenarios where input and target parameters have strong correlations. In this paper, a fuzzy model integrated with a time series data analysis method is proposed to address the water quality prediction problem when the correlation between the input and target parameters is weak. The water quality datasets collected from the Las Vegas Wash between the years 2005 and 2010, and the Boulder Basin, Nevada-Arizona from the years 2011 to 2016 are used to test the proposed model. The prediction accuracy of the proposed model is measured by three different statistical indices: mean average percentage error, root mean square error, and coefficient of determination. The experimental results have proven that the ANFIS model combined with a time series analysis method achieves the best prediction accuracy for predicting electrical conductivity and total dissolved solids in the Las Vegas Wash, with the testing value of coefficient of determination reaching 0.999 and 0.997, respectively. The fuzzy time series analysis has the best performance for dissolved oxygen and electrical conductivity prediction in the Boulder Basin, and dissolved oxygen prediction in the Las Vegas Wash, with testing value of coefficients of determination equal to 0.990, 90975, and 0.960, respectively.