The continuous increase in the rate of industrialization in developing countries, in recent times, calls for continuous industrial water quality assessment and prediction. This is to create more awareness and ensure cleaner and sustainable industrial production. Water quality for industrial uses is often described in terms of corrosion and scaling potentials (CSPs). In this paper, optimized artificial intelligence models (e.g. multiple regressions (MR), hierarchical clusters (HCs), and artificial neural networks (ANNs)) for assessing and predicting the CSPs of water resources were developed, for Ojoto suburb (SE Nigeria). Indices used in evaluating the CSPs are chloride–sulphate mass ratio (CSMR), Larson-Skold index (LSI), Langelier index (LI), aggressive index (AI), Ryznar stability index (RSI), Puckorius scaling index (PSI), and Revelle index (RI). This work is the first of its kind to utilize predictive models in simultaneously predicting the industrial water quality indices. Prior to the predictive modeling, R-mode HCs, correlation, principal component, and factor analyses were used to analyze the relationships between the physicochemical variables (pH–T–EC–TDS–TH–Ca–HCO3–Cl–SO4–Fe–Zn–Pb) and the CSP indices. Q-mode HCs effectively identified the spatiotemporal water corrosion/scaling risk classification and distribution in the area. Both MR and ANN models suitably predicted the CSP indices. However, both models better predicted AI, LI, RSI and PSI than LSI, RI and CSMR. The MR models’ performances were analyzed using R, R2, adjusted R2 and F-ratio values whereas the ANN models were verified using parity plots, R2, RMSE, and residual error plots. For the ANN modeling, scaled-conjugate-gradient optimizer outperformed gradient-descent optimizer. Also, ANN models outperformed the MR models. The practical implications of the present research findings were also discussed.
Read full abstract