Optimisation of energy utilisation in Umgeni Water Works

Energy in the 21st century has been seen to escalate in cost and will continue to be one of the key cost drivers for municipalities and water utilities that operate and maintain water treatment plants. It is part of every stage of the life-cycle of water supply, i.e. abstraction, water treatment, distribution, use and disposal. Umgeni Water (UW) conducted a baseline energy assessment of eight large water works in 2013. One of the recommendations of that study was to develop an energy optimisation model that can be used for regular energy assessments by Process Engineers to improve the operational protocol as well as for input into new plant designs. A case study was conducted at a large water works which receives its raw water from two sources. The baseline energy assessment of the water works indicated that 39.9% of the total energy consumed is by the raw water abstraction works, 46.2% by the distribution pump-stations and 13.9% by the treatment works. Engineering fundamentals for an energy balance were used for the energy optimisation, where the required head was determined for the backwash pumps. An energy optimisation model for evaluating the backwash performance of a set of rapid gravity filters was developed using Artificial Neural Network (ANN) software. MATLAB ANN Simulink programme was employed to predict the optimal backwash time and filter cycle time based on using the phenomenon called feed-forward back-propagation training through its simulation. The data used to train the network consists of raw water quality data, filter run time, turbidity profiles and the final water quality data. It was seen that the MATLAB Simulink can predict the backwash time at 7% lower than the actual backwash time which is higher than the desirable 5% allowable accuracy. However, the filter cycle time was predicted at 3% higher than the actual filter run cycle time. It is concluded from this study that more backwash profiles and a larger data collection equivalent to at least six months work is required to validate this research. It was also discovered that in the presence of data that cannot be readily linked, Artificial Neural Networks can establish and clarify relationships with the different data variables. The Table 4.2 shows that indeed energy savings are possible for the operations at UW Water Works G for Filter Plant 2. The minimal amount of energy required to get the turbidity to the required specification using ANN was calculated from simulated backwash time. This lead to the average energy savings with a monetary worth only R1161.55 using power ratings on the pump while it was as high as R416 301.10 where power input calculations were used considering the head and fluid flow happening inside the pump. The filter backwash and cycle run times can be predicted and optimised to target a turbidity of less than 10 NTU, specifically 9.9 NTU as the backwash turbidity target used for the purposes of this study. Further work is recommended on gathering six months’ worth of data to further verify the model simulation results.