EPR is used here to attempt to derive an expression for an annual regeneration rate and not the risk or magnitude of any potential discolouration event occurring. All the variables in Figure 6 (apart from Flush2 which is used in the calculation of the annual regeneration rate) were initially used, i.e., 13 candidate inputs in total. Input and output test (or cross validation) data were not used, due to the limited data points available for this case study and the principal goal being knowledge discovery. The seven model structures described in Giustolisi & Savic (2006) were applied, although the use of an inner function (such as logarithm, exponential, etc.) did not provide any additional accuracy and merely led to longer model run times. Results are presented here for standard polynomial structured equations produced by EPR. The MOGA process ran for 7,020 generations. Table 2 provides five models produced along with the coefficient of determination (CoD) which is based on the sum of squared errors and reflects model accuracy.

Table 2

Selected Pareto optimal regeneration rate estimation models identified by EPR

Model structure .	CoD .
	0.40
	0.49
	0.58
	0.67

Model structure .	CoD .
	0.40
	0.49
	0.58
	0.67

View Large