| City: |
Zagreb, Croatia
|
| Organization: |
Faculty of Agriculture, University of Zagreb, Croatia
|
| Project Start Date: |
1 Unknown
|
| Project End Date: |
1 May 2019
|
| Reference: |
Ribic B, Pezo L, Sincic D, Loncar B, Voca N. Predictive model for municipal waste generation using artificial neural networks—Case study City of Zagreb, Croatia. Int J Energy Res. 2019;43:5701–5713. https://doi.org/10.1002/er.4632
|
| Problem: |
o In order to meet the European Union’s environmental legislation related to environmental protection and integration of circular economy principles, the Republic of Croatia needs to implement an integrated and sustainable waste management system.
o The problem that this study aims to address is the need to predict with high degree of accuracy, the amount of municipal waste generated by the City of Zagreb for the year of 2017 in order to optimize the waste management operations of the City.
|
| Technical Solution: |
o Approach:
Taking into account the local waste management trends and the changes in socio-economical factors, the organization aims to develop and optimize models for the estimation of generated municipal waste in Zagreb by using a combination of neural network models.
o Methodology:
Step 1 - Data collection
• Data for the years of 2013-2016 was collected to train and test the models ahead of the year 2017’s predictions
Step 2 – Statistical analysis
• Three different neural networks were developed to analyse the impact of different indicators to waste generation
• ANN1/1 aimed to predict the socio-economic indicators that would be used for 2017, based on the testing made for years 2013-2016
• ANN1/2 aimed to predict the waste management indicators that would be used for 2017, based on the socio-economic indicators available for the years 2013-2016
• ANN2 aimed to predict waste management indicators for 2017, based on the socio-economic indicators predicted by ANN1/1
Step 3 – ANN Modelling
• By using a multilayer perceptron (MLP), and after normalizing the data previously obtained, the ANN models were iteratively created.
• The experimental data was randomly divided into training, cross-validation, and testing data. This allowed to determine the optimal number of neurons and their respective weights.
• Convergence was achieved by using the sum of squares
Step 4 – Determining the Accuracy of the models
• The accuracy was determined by obtaining the coefficient of determination, chi-squre, mean bias-error, root-mean-square error, and mean percentage error
|
| Datasets Used: |
o Dataset 1: Statistical Yearbooks of the City of Zagreb for years 2013, 2014, 2015, 2016
Socio-Economic indicators (social and economic activities of the city as well as geographical, meteorological, and basic data about the city)
This data set includes over 25 chapters and more than 1000 indicators
o Dataset 2: Annual reports for the waste management amount of the period 2013 – 2016, by Cistoca
Waste Management Indicators (waste shares, place of disposal, total number of bins)
|
| Outcome: |
o Before this study, no efforts were made by the City of Zagreb to predict the amount of waste generated by the City
o The developed neural networks showed a good generalization capability to calculate the relevant socio-economic and waste management variables for the City of Zagreb for the period of 2013-2016
For ANN1/1 (socio-economic indicators based on years 2013-2016)
• The optimal number of neurons in the hidden layer was 8, for a r2 value of 0.997 for the training period
• Three indicators were obtained: total number of tourists, total number of households, salaries)
For ANN1/2 (waste management indicators based on years 2013-2016)
• The optimal number of neurons in the hidden layer was 7, for a r2 value of 0.826 for the training period
• 14 indicators were were predicted to be useful indicators
For ANN2 (waste management indicators based on socio-economic indicators from ANN1/1)
• the optimal number of neurons in the hidden layer was 9, for a r2 value of 0.710 for the training period
• 14 waste management indicators were predicted to be useful indicators
o Number of households, number of tourists, and salaries can be efficiently used to predict the amount of waste (paper and cardboard, municipal solid waste, and bulky waste) for the City of Zagreb
|
| Issues that arose: |
o Additional data would be required to increase the accuracy of the model
|
| Status: |
Terminated
|
| Entered by: |
30Oct2020: Pedro Torres-Basanta, pedro.torresbasanta@mail.utoronto.ca
|