Artificial Neural Network Approach for Predicting the Water Turbidity Level Using Optical Tomography

Abstract

Water pollution can occur with a variety of reasons such as the change in water colour, the presence of harmful bacteria and toxic waste spills. This paper presents an application of an optical tomography system based on artificial neural network (ANN) to predict the turbidity level of water sample. The system made use of the independent component analysis algorithm to calculate the K value, which indicates the attenuation value of the water turbidity level. The K value then is utilized by ANN to estimate the turbidity level. The optical tomography system can be used to evaluate the water turbidity level in the pipeline without disturbing the flow process. Evaluation of the mean square error (MSE), sum square error (SSE) and regression analysis (R) also enabled us to determine the network performance which demonstrated that the neural network is effective in inspecting the water turbidity level. The best neurone structure is revealed when two hidden layers with 20 and 10 neurones in the first and the second layer, respectively, are used. The training result shows $${9.7147 \times 10^{-7}}$$ for MSE, 0.1432 for SSE and 0.99911 for regression. For the testing part, the result for the neurone structure is $${8.1473 \times 10^{-5}}$$ for MSE, 0.7509 for SSE and 0.98525 for regression. The results revealed that the performance of ANN demonstrated a good prediction capability when the turbidity level changed. Thus, an optical tomography system with ANN proved to be an efficient tool to classify the water quality level and is beneficial to the water industry.