Prediction of Water Quality Indices by Using Artificial Neural Network Models: Prediction of Water Quality Indices

Prediction of Water Quality Indices by Using Artificial Neural Network Models: Prediction of Water Quality Indices

Prakash Chandra Mishra (Fakir Mohan University, India) and Anil Kumar Giri (Fakir Mohan University, India)
DOI: 10.4018/978-1-5225-2440-3.ch020


Conventionally, fixed techniques are used for prediction of future time-series data. Subsequently adaptive techniques are used to forecast improved future data. The adaptive techniques are essentially based on ANN and fuzzy logic techniques. It is observed that these techniques also perform poorly when the input data set available is less and when there is abrupt change in the input data set. In this paper the proposed hybrid technique is based on data farming for intermediate data generation and the ANN model for better learning and forecasting. The performance of the proposed model has been tested with actual pertaining to water quality indices of various water samples collected from different sources.
Chapter Preview

1. Introduction

Time series forecasting finds extensive applications in many areas such as financial engineering, power system engineering, pollution control, river flow, stock market etc. Conventionally, various fixed techniques are being used for forecasting purpose. These fixed techniques suffer from accurate forecasting of time series data. To alleviate this problem adaptive techniques have been proposed in the literature. These are techniques based on artificial neural network (ANN) (Werbos, 1988, Patra et al., 1997, Widro et al, 1998) and fuzzy logic (Dash et al, 1995, Dash et al, 1995). Peng et al., 1992 has applied the ANN technique for forecasting of electric load. These adaptive techniques involve more computational load and large training time. Conventionally, these techniques perform two tasks. In the first phase, an adaptive model is developed using known input/output data sets. This phase is known as training or learning phase. The model is trained by a learning algorithm so as to reduce the output mean square error (MSE). The second phase is known as testing phase in which a new input data set is applied and the estimated output is obtained from the model as the future value of the time series.

In the present paper we have evolved a new idea of generating more data so that adequate data sets are available to train the ANN based forecasting model efficiently. This technique is termed as data farming. The combination of data farming and the ANN based model is known as data farming linked neural network (DFL-ANN), which has made the overall system attractive in the sense that the forecasting of future data becomes more accurate. The new model developed in this paper has been used to predict the pollution level of drinking water, defined by the water quality index (WQI) (Satya Mohan et al, 1986). Water available from different sources of Rourkela City was collected and different water quality parameters were analyzed for five consecutive years and are used in the present study. Rourkela is an important industrial city of Orissa occupying an area of 121.7 km2, situated on the Howrah-Mumbai railway line. The main industry here is the Steel Plant and the associated industries of steel plant are cement, fertilizer, foundries, chemicals and explosive industries. In addition, more than three hundred small scale industries are also present in and around this city. Various gases, particulate matters and liquid effluents emitted by the industries enter the biosphere daily, which contaminate the drinking water in various ways. Besides, in certain places unhygienic surroundings, improper storage and mishandling of drinking water render it unfit for consumption. These consequences made it necessary to study the WQI.

The organization of the proposed paper is as follows.

Complete Chapter List

Search this Book: