Sludge volume index (SVI) modelling: Data mining approach

Abstract

In this paper, statistical models to forecast based on the sludge volume index (SVI) with the continuous measurements carried out in the period from 2013 to 2016 for waste water treatment Sitkowka-Nowiny was developed at the same, for two variants of analyses. In the first one, a model of SVI predicting based on the quality indicators of wastewater flowing into the treatment plant, i.e. Biochemical (BOD) and chemical oxygen demand (COD), the content of total nitrogen (TN) and ammonia nitrogen (NH4), total suspended solids, total phosphorus (TP) and the operating parameters of the bioreactor (pH, temperature, oxygen concentration in the nitrification chamber). In the second case, the possibility of replacing individual measurements of the quality of wastewater values calculated on the basis of daily sewage flows to the treatment plant was examined. The above mentioned models statistical analysis was performed using the method of k-nearest neighbor (k-NN), cascading neural network (CNN) and boosted tree (BT). To evaluate the predictive ability of these models the average relative error (MAE) and absolute error (MAPE) were used. The conducted analysis showed that based on the above mentioned indicators of effluent quality and technological parameters of the biological reactor it is possible to modeling of sediment volume index with satisfactory accuracy. In the case under consideration methods of lower values of the prediction error of SVI obtained using a cascade neural networks (MAE = 17.49 ml/g and MAPE = 9.80%) than for the method k-nearest neighbor (MAE = 27.85 ml/g and MAPE = 14.50%). Furthermore, based on the performed simulation, it was found that it is possible to model the analyzed work of the quality of waste water on the basis of the daily flow with reasonable accuracy, it is confirmed by the calculated value of the average and absolute and relative error, and the better ability predictive characterized by the models obtained on the basis CNN than k-NN. In examined cases, the MAP in a set of validation did not exceed 10.13%. The simulation results of quality indicators obtained by CNN were substituted in place of the explanatory variables of sludge volume index in the model for prediction index of sediment and conducted simulations SVI, set out the error MAE = 25.15 ml/g and MAPE = 15.26%. On this basis, it is possible to replace the measured values of the quality of the results of their simulation, thereby reducing the cost of testing, but also gives you continuous control of SVI and adjustments discussed in this work of technological parameters of the biological reactor.