Monitoring flooding of wastewater systems with real-time alerts and data analysis

Abstract

Kristianstad city is one of the lowest cities in Sweden, as the Swedish Civil Contingencies Agency (Herbring and Näslund -Landenmark 2011) reports, and the city is at the highest risk of flooding in Sweden. The water flooding problem is one of the important problems related to sustainable goals defined by the UN as the 2030 agenda under SDGs Goals: Goal 6: Clean water and sanitation. The municipality of Kristianstad wishes to be notified of impending floods in the city and to monitor the city's water level. The need for an alert system with intelligent solutions was the target. Machine learning technology is an excellent choice for making prediction models that learn and forecast results for the coming days to help achieve the goal.

Three models were tested and compared to get the best result to be implemented in the system as a machine learning part. The models were linear regression, multiple linear regression, and artificial neural network long short-term memory (LSTM). The model was chosen by looking at the mean squared error (MSE) in all the models in each sensor, then analysis and comparison were made for results in terms of performance. The (LSTM) gives excellent results for forecasting for ten days, depending on the historical data for three months past, and then all results saved are to be used in the front end and the database as data for future study. The alert system front end shows the water level for the past ten days and the coming ten days, and if the water level is at the risk level, the alert will be shown to the users. The database will be updated every day at midnight to get new data. The development was completed using different technologies and Long Short-Term Memory (LSTM) as a machine learning algorithm.

This work also answers some research questions: what data could be collected from the IoT portal to be analysed to get future flood risks, and the result of that shows the important data is water level, sea level, ground level, and rain level. All the data can be collected from the IoT portal and SMHI. Then, after comparing three different machine learning algorithms- linear regression, multiple regression, and LSTM; the best result was Long Short-Term Memory (LSTM). Those questions are answered by improving the result and discussing it in the result section, and in the appendix section, there are plottings showing the results for each sensor, with the best result being Long Short Term Memory (LSTM).

Date of Award	2022-Jun
Original language	English
Supervisor	Qinghua Wang (Supervisor) & Eric Chen (Examiner)