Author: Riccardo Biella, software engineer at Paradox Engineering
Annual waste production will increase by 70% to 3.4 billion tons by 2050, says the World Bank. At this growth rate, we will need about 3 worlds to absorb and dispose of our trash without compromising our health and the environment.
Cities are increasingly looking for innovative solutions to improve municipal solid waste collection and treatment to preserve both public health and urban livability. At Paradox Engineering, we have extended our flagship platform PE Smart Urban Network to include a ready-to-use Smart Waste application for City and utility managers.
The news is, we are integrating Artificial Intelligence and Machine Learning algorithms to enhance this solution and make it even more performing. While traditional algorithms convert input to output using some given rules, we chose to have Machine Learning to build an accurate predictive model based on sample data.
First step is to equip bins with Paradox Engineering-MinebeaMitsumi 6LoWPAN sensors and allow municipal waste operators to collect and monitor data about the level of filling, date and time of latest waste collection, and bin location information.
Thanks to Machine Learning techniques, we are evolving the system from a raw data collection platform to an actionable prediction solution: the bin level prediction system can receive data from smart bins’ filling sensors and process it, providing an estimate of the date when the bin will reach its capacity limit.
Did you know that bins are generally emptied when they are only 40% full? With our bin filling level prediction system, operators can dispatch trucks only when and where the bins are close to full, thus optimising solid waste collection, reducing the number of truck rolls and the mileage associated with them, which in turn decreases pollution and congestion.
The bin filling level prediction system leverages a deep neural network that is trained on a dataset containing historical data of smart bins. Features considered during software training are based on the current level of filling correlated with time data, as well as geographic location data.
More complex logic will soon be integrated to teach the bin identify relevant points of interest nearby, such as supermarkets, stadiums, stations, or hospitals – and predict waste input upon variables such as people density, calendar, festivities, holiday seasons, and more.
Continual Learning techniques allow the model to continuously evolve combining new data and accumulated knowledge. This means that, by analysing filling patterns, operators can also take data-driven decisions about the quantity, capacity, and location of containers. Bins can be positioned to fit different urban scenarios and adapt within a few days to changing conditions, ie. different residents’ habits or behaviours, like the increase of packaging-related trash due to the rise of e-commerce and home delivery in pandemic time.
Thanks to PE Smart Urban Network and Machine Learning algorithms, cities can truly improve waste collection and management, enhancing livability while protecting public health and the environment.
Learn more about our Smart Waste application: download our white paper (free registration required), and contact our Machine Learning experts