Two IITKGP students propose a smart water management system and win kudos at an International Conference on ML and Soft Computing in Vietnam
Smart cities need smart water management and distribution system. In advanced countries, telecommunications, computing, computer-based modelling, AI, Machine Learning, Data analysis and processing have changed the way water resources are managed. In fact, they have given rise to what is called hydroinformatics systems. A similar change may happen in India as well.
Two students of IIT Kharagpur, Stuti Modi and Aditi Kambli recently presented their paper on an intelligent water management and distribution system based on data-driven models at the Third International Conference on Machine Learning and Soft Computing (ICMLSC) in Da Lat, Vietnam. They won the Best Presentation Award at the conference held in January 2019.
Stuti and Aditi proposed the use of two data-driven models – recurrent neural networks (RNN) and fuzzy-logic based models. With the use of these models, they demonstrated how daily average water demand can be predicted, how drought/flood could be predicted and an optimum level can be maintained in the dam reservoir, how the water level in reservoirs in houses and localities can be controlled, how the drinkability of water can be judged and water treatment can be planned.
Previous works in this field have concentrated on any one aspect of the water management process, for example, on either water demand prediction or treatment. “What we offer is a complete one-package solution. Our model is segmented in such a way that in case of any breakdown, we can easily track the root cause for it and get it fixed. All the individual models here come together and form a platform for water management,” said Aditi.
Aditi is from the Department of Ocean and Naval Architecture and Stuti belongs to the Department of Electrical Engineering. They had worked under the guidance of Prof. Sudhir Kumar Barai of the Department of Civil Engineering on a course of Soft Computing that he teaches to all departments.
For the prediction of daily average water demand, the duo work on the presumption that smart water meters are installed in all houses or localities. Data on water use of individual households, collected by the sensors of the smart water meters, are sent to a central medium. This data is used to train an LSTM based RNN, which then predicts the water usage for the upcoming day.
To validate their assumption, they used the time series data of daily water usage for the last four years. Water usage of the past 30 days was provided as an input to the network. The training of the RNN was performed using Keras.
In the case of drought/flood prediction and the optimum dam reservoir level, the fuzzy inference system is implemented using percentile storage, percentage full storage capacity of the reservoir and rainfall to predict the condition of the dam reservoir. Too much water in the dam threatens dam safety and could cause flood and too little could cause drought. The fuzzy inference system predicts what should be the optimum water level in the reservoir and the Fuzzy PI Controller controls the reservoir water level by releasing or not releasing water.
Such fuzzy logic controller could also man how much water needs to be present in individual water tanks of residential houses. It is often seen that the inlet of water into the tanks is not proportional to the outlet rate, causing the water tank to empty out faster than it fills up. To obtain the water level, however, an ultrasonic water level sensor has to be installed on the top of the reservoir to monitor input and output flows. The water reservoir also has to be modelled as a tank in Simulink whose input and output rates can be controlled.
The water quality model is divided into two parts in this research. One, the assessment of water quality, and two the treatment of moderate/non-drinkable water in a water treatment plant.
For water to be drinkable, various parameters – such as pH, dissolved oxygen, alkalinity, coliform etc – have to be within safe limits. Stuti explains, “Suppose the turbidity of the water is poor and the DO (dissolved oxygen) is moderate, but if the pH and coliform component is rated good, the water is drinkable. This means, all the parameters do not necessarily have to be good for water to be drinkable.” Their fuzzy expert system creates a water quality index that is easy to understand, logical and useful for common people.
Once the fuzzy expert system grades a water to be moderate or non-drinkable through the data collected by sensors and assessed by MATLAB program installed on the computers of the water treatment plant, the water ought to be sent for treatment. For treatment again, the fuzzy logic based system sets ideal values, and based on the error, the correction methodology is powered.
Aditi and Stuti know that the ideas proposed can be improved further with actual sensor data and constant updating of standard points. They would also like to create separate models for each region, since the same parameters don’t work for every region.
Prof. Barai says, “Aditi and Stuti need to implement their model on a large scale hardware model to observe the results and fulfil their goal to implement an intelligent system that conserves and uses water efficiently.”
Graphics : Suman Sutradhar