Uncertainty quantification and sensitivity analysis of the efficiency of abiofilm reactor with a moving bed using a combination of Monte Carlosimulation and artificial neural network.

In this study, the uncertainty and sensitivity analysis of the efficiency of a biofilm reactorwith a moving bed were conducted using a combination of Monte Carlo simulation and artificialneural network. The qualitative prediction of the quality of petroleum source was estimated usingartificial neural networks, and the removal efficiency of COD and TPH pollutants was predictedusing a set of laboratory data.This dataset included parameters such as TPHin, CODin, F/M, Filing Ratio, MLVSS/MLSS,DO, CODout, and TPHout, which are considered influential in the quality of petroleum sources. A۴-layer neural network with ۷ neurons and an error rate of ۴.۰۶% was designed for COD removalefficiency prediction, and a ۳-layer neural network with ۸ neurons and an error rate of ۳.۹۵%was designed for TPH removal efficiency prediction. By incorporating the mathematical modelas an activation function in the Monte Carlo method, the uncertainty and reliability of theestimated COD and TPH removal efficiencies were determined.In the Monte Carlo method,using the Latin Hypercubes algorithm, the removal efficiencies were obtained as intervals with a۹۵% confidence level, yielding satisfactory results. This is important as it reduces the risk ofpollutant removal during managerial decision-making and covers sudden changes in the qualityof petroleum sources that affect the removal efficiency.In previous random sampling methods, arandom number was extracted from the probability distribution space of a data. However, thismethod divides the input's normal distribution into rows and columns, and then selects a randomsample from among the created sections so that the realization of values with low probability isalso practical. One of the advantages of using this method is obtaining confidence levels for riskswith high effects but low probabilities [۱].

کلیدواژه ها:

MBBR ، Artificial Neural Network ، Monte Carlo