Advertisement

Household Packaging Waste Management

  • João A. Ferreira
  • Manuel C. FigueiredoEmail author
  • José A. Oliveira
Conference paper
  • 1.4k Downloads
Part of the Lecture Notes in Computer Science book series (LNCS, volume 10405)

Abstract

Household packaging waste (HPW) has an important environmental impact and economic relevance. Thus there are networks of collection points (named “ecopontos” in Portugal) where HPW may be deposited for collection by waste management companies.

In order to optimize HPW logistics, accurate estimates of the waste generation rates are needed to calculate the number of collections required for each ecoponto in a given period of time.

The most important factors to estimate HPW generation rates are linked to the characteristics of the population and the social and economic activities around each ecoponto location.

We developed multiple linear regression models and artificial neural networks models to forecast the number of collections per year required for each location. For operational short term planning purposes, these forecasts need to be adjusted for seasonality in order to determine the required number of collections for the relevant planning period. In this paper we describe the methodology used to obtain these forecasts.

Keywords

Forecasting Household packaging waste Waste collection Recycling Seasonality 

1 Introduction

Recycling of waste materials became a very important issue for society, as the environment benefits greatly from any advances made in direction of a cleaner future. Process collecting for recycling involves teams of workers and vehicles. One of the main problems lies in finding optimal collection routes, where a set of collection points is targeted, and each point is given a priority level. This problem can be described as the Vehicle Routing Problem (VRP). However more flexibility is needed when it comes to choose only a part of the collection points to be visited, instead of the whole set. Thus, a more fitting description of the selective waste collection process may be the Team Orienteering Problem (TOP).

In this context, the TOP can be described as the problem of designing the routes to be assigned to a fleet of vehicles that perform the collection of different types of waste stored along a network of collection points. Each one of these collection points contains a certain amount of waste that is directly linked to the respective priority level. The collection routes have maximum durations or distances, and consequently, the selection of collection points to be visited by the vehicles is made by balancing their priorities and their contributions for the route duration or distance. The objective is to maximize the total amount of waste collected by all routes while respecting the time or distance constraints.

Aside from the routing problem, there are other issues related to the process of waste collection for recycling, especially when dealing with real scenarios and the activity of real waste collection companies. One of these issues is the estimate of waste material quantities generated over time at each collection point in a given collection network, which enables the estimation of a waste generation rate (WGR) for each collection point. Estimating WGRs is crucial for designing collection networks since collection points are located according to these estimates. Moreover, forecasts for waste generated along a network are used to design collection routes as each collection point is given a priority according to their WGR, which translates into the frequency of service by collection vehicles.

Considering the goals Portugal has to fulfil for the recycling and recovery of HPW, there is a permanent need for increased efficiency in waste collection. Thus, one important task is the development of models to forecast the quantities of waste generated at collection points.

The paper is structured in 6 sections. Section 2 presents the problem. A literature review is presented in Sect. 3. The forecasting models developed are described in Sect. 4. Seasonal adjustments to the base forecasts are described in Sect. 5. Finally, in Sect. 6, the results and main conclusions of this study are presented.

2 The Problem

The main objective is to solve a real-world problem faced by an intermunicipal waste management company, in six municipalities: Braga, Vieira do Minho, Vila Verde, Povoa do Lanhoso, Amares and Terras de Bouro, in Northern of Portugal (Fig. 1).
Fig. 1.

Collection area of HPW

The problem focus is on HPW collection, and a major step to achieve a good performance is to obtain accurate forecasts for waste generation rates at the waste collection points in order to determine the frequency of collection.

In Portugal, household packaging waste (HPW) is disposed of in collection points called ecopontos. Each ecoponto can have three types of containers, identified with different colors: glass (green), paper/cardboard (blue), and plastic/metal (yellow), as can be seen in Fig. 2.
Fig. 2.

Ecopontos illustration (Color figure online)

The company currently operates a network of more than 1,200 ecopontos located across. These six municipalities are characterized by a mix of urban and rural areas, which require different strategies concerning waste management.

The company vehicles do not visit all ecopontos every workday. It is necessary to select a subset of ecopontos to visit, according to their estimated fill rate, each time the route planning is done. Thus, given a planning horizon, for example a week, the company must decide which ecopontos must be visited (because they are near full) and which can be skipped (because they are near empty) by the collection fleet in order to design efficient routes for the selective collection of HPW. Thus the priority level of an ecoponto is highly related to the estimated amount of waste it holds during the route planning phase, Fig. 3.
Fig. 3.

Estimated amount of waste held

Taking into account its priority level, an ecoponto may or may not be selected to be visited during the established planning horizon, Fig. 4.
Fig. 4.

House packaging waste collection for recycling

Thus, to improve route planning, reliable forecasts of the amount of waste generated at each ecoponto are necessary. In a previous phase of this study [5], significant factors for HPW generation rates were identified in order to develop forecasting models to predict the number of times each ecoponto should be visited during a certain period of time. However, due to a variety of seasonal factors, such as holidays, periodic events, weather conditions, etc., waste generation rates need to be seasonally adjusted. For example, the number of collection visits required in this study increases significantly during summer months. In this paper we present a method for making these seasonal adjustments that allow a significant reduction of the number of unnecessary visits to “nearly empty” ecopontos and therefore saving resources and reducing the logistics costs of the operation, Fig. 5.
Fig. 5.

Optimizing collection routes

3 Literature Review

Forecasting the generation rates of municipal solid waste (MSW) is important for both, the development of waste management infrastructures, and the implementation of the logistics processes for waste collection.

In a review, Beigl et al. [2] concluded that MSW generation is best predicted by time series analysis and regression methods. According to Denafas et al. [4] seasonal factors may have a strong impact on MSW generation and should be taken into account in the decision-making processes concerning waste management. Concerning waste generation contributing factors, Wang and Nie [11] identified the growth of urban population and gross domestic product (GDP) as the most important ones. Grossman et al. [6] considered factors such as: increase of population, income level and housing type. Later studies pointed out that waste generation can be related to production and consumption levels [3, 9]. More detailed analyses showed that the growth of the urban population had a greater impact than GDP on the total amount of MSW produced. Also, with factors like increasing income, MSW seems to change more in composition rather than in total amount generated. Other factors that may influence the generation and composition of waste are climate, living habits, level of education, religious and cultural beliefs, and social and public attitudes [1, 8, 10].

Frequently, time series forecasting models may be a good choice to estimate MSW generation when there is access to a substantial amount of historical data. Based on the comparison of several forecasting methods, Beigl et al. [2] imply that a forecasting tool based on socio-economic variables was more suitable than single time series analyses. In most cases, the application of modelling methods such as regression analyses, and group comparisons, seems to be the best option when the goal is to test the relationship between the level of affluence and the generation of total MSW or a material-related fraction, and to identify significant effects of waste management activities on recycling quotas. The application of time series analyses and input-output analyses is advantageous when there is a need for special information (i.e., assessment of seasonal effects for short-term forecasts).

After this review, it became clear that most of the previous research in this area focused on a different level of analysis that did not match our purpose of predicting waste generation at each collection point in order to determine when it needs to be emptied.

4 Forecasting Models

The available data consisted on all the waste collections performed by the company in all the six municipalities it operates from 2013. It contained monthly records showing how many times each ecoponto in the network was emptied during each month of the year.

Our aim was to forecast the number of times each ecoponto needs to be emptied each year. Therefore, this number of collections per year (and per ecoponto), hereafter referred as CPY, was set as the dependent variable considered in the forecasting models developed. In the next subsections, a brief description of the developed models is given, followed by a detailed description of the seasonal adjustments made to estimate the required number of collections per month.

The forecasting models for CPY use data from waste collection records. The factors used to estimate waste generation rates, were determined using several sources for demographic information and socioeconomic indicators. In Table 1, a list with all explaining factors for HPW generation is presented.
Table 1.

Factors used to estimate waste generation rates.

Factor

Description

Acronym

1

Number of Ecopontos in the area

NE

2

Population Density in the area

PD

3

Number of Inhabitants per Ecoponto

NRE

4

Ecoponto Density

ED

5

Ecoponto Type (street level or underground) – qualitative

ET

6

Ecoponto Position (closed or open area) – qualitative

EP

7

Ecoponto Capacity

EC

8

Number of Ecopontos within a 300 ms radius

NE300

9

Demographic Factor (household density around each ecoponto in a 300 m radius) - qualitative

DF

10

SocioEconomic Factor (based on the number of schools, businesses, local attractions, leisure and sports infrastructures, restaurants, etc.)

SEF

The data used consisted of information on these factors for all the ecopontos (185) from two municipalities: Amares and Vila Verde.

We used two different methods to develop the forecasting models: Multiple Regression (MR) and Artificial Neural Networks (ANN). MR models were developed using Forecast Pro software and ANN models were developed using Encog [7]. Regression models are widely known and used for forecasting purposes, and ANN are attaining more recently some relevance in this field, with promising results being reported in the last years.

The best performing forecasting models for the collection of cardboard at the 185 ecopontos sample are described in Eq. 1 (regression model) and Table 2 (ANN model).
Table 2.

Artificial neural network model.

Network type

Multilayer perceptron

Network structure

3 layers (4-4-1)

Learning algorithm

Levenberg-Marquardt

Activation function

Sigmoid

Training, test, validation

125, 10, 50

Training epochs

1000

Inputs

DF, ET, NE, SEF

Output

CPY

$$ {CPYi} = {18.209DF} - {0.006EC} + {8.314ED} + {21.141ET} + {0.035NRE} - {0.027PD} + \in $$
(1)
The performance of these models at forecasting the actual number of collections per year (Observed collections) in a test sample of 50 ecopontos can be seen in Fig. 6.
Fig. 6.

Forecast results for the best ANN and regression models.

The mean absolute deviation (MAD) of forecasting errors for both models is shown in Table 3.
Table 3.

Mean absolute deviation of forecasting errors.

Multiple regression

Artificial neural network

10.44

9.14

5 Seasonal Adjustments

The models described in Sect. 4 are used to estimate the number of collections per year required for each ecoponto. However, as previously stated, there are seasonal variations on the waste generation rates (WGRs). Therefore there is a need to adjust the number of collections to these variations in order to forecast the number of collections required for each month of the year. For example, in summer months, there are usually higher WGRs in this region due to emigrants returning to their home country and tourism activities resulting in significant population increases. There are also seasonal changes in social habits. For example, there is a higher consumption of beverages in plastic, metal and glass containers. Many other factors may also have contributed for seasonal variations in WGRs throughout the year.

Table 4 shows general performance indicators of the company operation from 2010 to 2014. In order to describe the seasonal adjustment procedures we will use data concerning cardboard collection in a sample of 185 ecopontos from two municipalities, Amares and Vila Verde. Table 5 shows the average monthly number of effective cardboard collections done for the 185 ecopontos in the sample.
Table 4.

Company indicators

Year

2010

2011

2012

2013

2014

Collections

Glass

11 570

11 555

11 238

11 829

11 887

Cardboard

62 294

72 431

70 234

69 704

65 571

Plastic

56 044

59 180

60 880

61 556

58 899

 

129 908

143 166

142 352

143 089

136 357

Quantity coll.

Glass (ton)

6 320

6 282

5 968

6 224

6 170

Cardboard (ton)

7 137

7 151

6 244

6 245

6 354

Plastic (ton)

1 434

1 730

1 909

1 992

2 095

Number of Ecopontos

1 131

1 170

1 208

1 114

1 159

Table 5.

Average number of cardboard collections per month.

The analysis of the data in Table 4 shows that there is not any significant trend in the number of collections from 2010 to 2014. Thus to determine the seasonal factors we may simply divide the average number of collections for each particular month by the monthly collections average. Table 6 shows the seasonal factors for each month of the year.
Table 6.

Seasonal factors

Once determined the seasonal adjustment factors, the forecasts of total collections for each month can be corrected using the corresponding seasonal adjustment factor. For example, if the forecast of CPYi for a certain ecoponto i is 120, then the average number of collections per month will be 10. To correct forecasts for example for the months of March and July, the factors 0.798 and 1.176 are used. Thus the revised forecast for required collections for ecoponto i would be 8 collections for March and 12 collections for July. Since the company operates approximately 25 days per month (there is no collection on Sundays), in operational terms this means that ecoponto i should be emptied every 3 days in March and every 2 days in July. The use of these revised forecasts at the route planning phase makes the collection routes much more efficient by avoiding visits to “nearly empty” ecopontos.

6 Discussion and Conclusions

The use of the seasonal adjustments resulted in an important reduction of the errors at forecasting the number of effective collections required per month. The impact of the revised forecast values resulted in an estimated 10% decrease in the number of visits to ecopontos not resulting in any collection. For the company involved in this study this means avoiding hundreds of unnecessary visits to ecopontos per year, Fig. 7.
Fig. 7.

House packaging waste collection for recycling

In this study we presented a real-world problem faced by a company that collects House Packaging Waste (HPW) deposited in a network of waste collection points (ecopontos).

The main problem faced by the company was the high number of visits made by the collection teams to ecopontos that are “empty”. This represented an important waste of resources that translated in thousands of kilometers traveled and hours spent by the collection teams that do not resulted in any collection.

Our main goal was to improve the forecasts for the number of monthly waste collections required. We achieved this by correcting the base forecasts for the number of waste collections per year for each ecoponto (CPYi), developed with Regression and Artificial Neural Networks models, using seasonal factors for each month of the year.

These seasonal factors where determined using actual data from effective collections. The results achieved show that these seasonal adjustments reduce the number of unnecessary visits to ecopontos, thus leading to important savings in environmental and logistics costs.

Notes

Acknowledgments

This research has been partially supported by COMPETE: POCI-01-0145-FEDER-007043 and FCT – Fundação para a Ciência e Tecnologia within the Project Scope: UID/CEC/00319/2013.

References

  1. 1.
    Bandara, J., Hettiaratchi, J., Wirasinghe, S., Pilapiiya, S.: Relation of waste generation and composition to socio-economic factors: a case study. Environ. Monit. Assess. 135(1–3), 31–39 (2007)CrossRefGoogle Scholar
  2. 2.
    Beigl, P., Lebersorger, S., Salhofer, S.: Modelling municipal solid waste generation: a review. Waste Manag. 28(1), 200–214 (2008)CrossRefGoogle Scholar
  3. 3.
    Bruvoll, A., Spurkland, G.: Waste in Norway up to 2010, reports 95/8. Statistics Norway (1995)Google Scholar
  4. 4.
    Denafas, G., Vitkauskaitơ, L., Jankauskaitơ, K., Staniulis, D., Martuzeviþius, D., Kavaliauskas, A., Tumynas, D.: Seasonal variation of municipal solid waste generation and composition in four East European cities. Resour. Conserv. Recycl. 89, 22–30 (2014)CrossRefGoogle Scholar
  5. 5.
    Ferreira, João A., Figueiredo, Manuel C., Oliveira, José A.: Forecasting household packaging waste generation: a case study. In: Murgante, B., Misra, S., Rocha, Ana Maria A.C., Torre, C., Rocha, J.G., Falcão, M.I., Taniar, D., Apduhan, Bernady O., Gervasi, O. (eds.) ICCSA 2014. LNCS, vol. 8581, pp. 523–538. Springer, Cham (2014). doi: 10.1007/978-3-319-09150-1_38 Google Scholar
  6. 6.
    Grossman, D., Hudson, J.F., Mark, D.H.: Waste generation methods for solid waste collection. J. Environ. Eng. 6, 1219–1230 (1974). ASCEGoogle Scholar
  7. 7.
    Heaton, J.: Encog: library of interchangeable machine learning models for Java and C#. J. Mach. Learn. Res. 16, 1243–1247 (2015)MathSciNetzbMATHGoogle Scholar
  8. 8.
    Marquez, M.Y., Ojeda, S., Hidalgo, H.: Identification of behavior patterns in household solid waste generation in Mexicali’s city: study case. Resour. Conserv. Recycl. 52(11), 1299–1306 (2008)CrossRefGoogle Scholar
  9. 9.
    Nagelhout, D., Joosten, M., Wierenga, K.: Future waste disposal in the Netherlands. Resour. Conserv. Recycl. 4(4), 283–295 (1990)CrossRefGoogle Scholar
  10. 10.
    Oribe-Garcia, I., Kamara-Esteban, O., Martin, C., Macarulla-Arenaza, A., Alonso-Vicario, A.: Identification of influencing municipal characteristics regarding household waste generation and their forecasting ability in Biscay. Waste Manag. 39, 26–34 (2015)CrossRefGoogle Scholar
  11. 11.
    Wang, H.T., Nie, Y.F.: Municipal solid waste characteristics and management in China. J. Air Waste Manag. Assoc. 51(2), 250–263 (2001)CrossRefGoogle Scholar

Copyright information

© Springer International Publishing AG 2017

Authors and Affiliations

  • João A. Ferreira
    • 1
    • 2
  • Manuel C. Figueiredo
    • 1
    • 2
    Email author
  • José A. Oliveira
    • 1
    • 2
  1. 1.ALGORITMI Research CentreUniversity of MinhoGuimarãesPortugal
  2. 2.Department of Production and SystemsUniversity of MinhoBragaPortugal

Personalised recommendations