A hybrid statistical regression technical for prediction wastewater inflow

Abstract One of the appropriate methods for parameter-less prediction is estimation of wastewater flow using simulation parameters by Monte Carlo approach. Appropriate numerical and analytical methods have been developed to simulate wastewater discharge, but none of them can be used reliably before calibration. Therefore, it seems necessary to build a model that can predict the discharge of wastewater more quickly and with general information. Among the new methods used to model the prediction of phenomenon behavior are intelligent data-driven methods. In the present paper, a hybrid multi-objective algorithm called Group Method of Data Handling (GMDH) was used. This method was used to estimate the flow rate discharge for two stations, Zabol and Zahedan. Upon analyzing and learning the hidden relations between the inputs, the GMDH-based neural network logically presented the optimal model and predicted the output variables. Using Fourier and Monte Carlo analysis, flow-type data were constructed for these networks, in turn by which, the flow over the desired interval times was calculated. According to the results of the two stations, among the two models of GMDH and GA + GMDH, the hybrid model of genetic algorithm with GMDH has accounted for the best accuracy. As for the best accuracy, at the Zabol station in October, the statistical parameters obtained for RMSE, d, EF, CI and rp were 5.19, 0.997, 0.991, 0.985 and 0.99, respectively, and as for the Zahedan station in April in terms of statistical parameters it was estimated about 1.11, 0.998, 0.992, 0.999 and 0.99, respectively. At the same time, Taylor’s diagram confirms this accuracy.