Topic 12

Prediction of water consumption on the level of water distribution area by implementing machine learning techniques

Introduction:

Increasing the efficiency of water distribution is crucial for coping with the increasingly emphasized water distribution problems such as droughts or scarcity of drinking water. Water supply system management is often based on decision support systems, which in turn rely heavily on the water consumption forecasting system. The availability of accurate water consumption forecasting contributes to the reduction of the overall costs of the water supply system, the reduction of the load on the system components, and the increase in the quality of water delivered to users. It is necessary to develop machine learning based regression models for prediction of water consumption at an hourly and daily resolution up to 7 days in advance, and then using daily predictions as supervisors when learning the models to improve the accuracy of their consumption predictions on an hourly resolution.

 

 

Tasks:

  1. Study the methodology of machine learning regression techniques with special emphasis on the feature engineering of time-series data and implement a simple regression model using the Python environment.
  2. Develop regression models to predict hourly and daily water consumption at the water supply area level in the Python environment, using data from water supply areas in Tavira, Portugal, and Almeria, Spain.
  3. Study and implement a version of the feature selection algorithm (e.g. some filter algorithm) to sort and find the most relevant features for modeling of water consumption data. Compare the prediction accuracies of models with all features and the models with n most relevant features.

  4. Improve the tuned models for generating predictions on hourly resolution by tuning them with the supervision of pre-calculated daily predictions and compare the obtained results with the previous models that don't utilize the predicted daily consumption data.

Additional information:

  • Number of students: 1
  • Keywords: water demand prediction, machine learning regression, filter features selection algorithms, ensemble methods
  • historical water consumption measurements of Tavira, Portugal and Almeria, Spain