Prediction Accuracy of BPN by Levenberg–Marquardt Algorithm for the Prediction of COD from an Anaerobic Reactor

Abstract

Anaerobic wastewater treatment differs from traditional aerobic treatment where no aeration is used. In this paper, a model is built using Back Propagation Neural Network (BPN) that analyzed the data from an anaerobic reactor containing the cheese dairy wastewater. Data preprocessing is a crucial task to identify the efficient parameters that contribute best solution in reducing the training time with high accuracy. Data preprocessing includes data cleaning, data transformation and data reduction. The data set is screened using k-means clustering and normalized using statistical normalization techniques viz., z-score, minmax, biweight, tanh and double sigmoidal. Among the adopted normalized techniques z-score normalization produced satisfactory prediction results in BPN. The z-score normalization eases the training of BPN to predict the value of Chemical Oxygen Demand (COD) for a concentrated cheese-dairy wastewater from an anaerobic reactor. The normalized dataset is trained with BPN using Levenberg–Marquardt algorithm. The performance of the above model is evaluated based on Mean Square Error (MSE) and Regression Coefficient R. The prediction results were close to the observed data and the model was found satisfactory with MSE = 0.53375 and R = 0.99185.