1
$\begingroup$

I am starting a project to predict the generation of urban waste.

I have found very little information on this topic on the internet. I would be very useful advice on how to approach this topic, and what techniques you would use.

I have found academic articles that make predictions with feedforward neural networks. But they seem basic or old. I would expect the use of recurrent networks, but all the examples I find on the internet are about climate prediction, or economics. And they are very different topics.

It would be very useful to make the prediction of waste generation based on an existing model, and not something invented by me. On what topic would you recommend that I take as a base?

I appreciate any advice. regards

$\endgroup$
  • $\begingroup$ Do revise your question,check your grammar....etc... $\endgroup$ – quintumnia Sep 30 '18 at 8:45
2
$\begingroup$

It is correct that climate and economic models are distinct from waste models.

A memory based model is the correct approach because the time domain is key in prediction based on existing trend data. However, the RNN is not a practical production model with others that have proven more productive. Although this article is trendy and not accurate in every technical respect, it introduces the more effective models and their relationship to RNNs. If you study the article and the terms it introduces on the web, the right solution and possibly an open source project that can get you started can be found.

Basis for Prediction of Urban Waste

Whether modelling solid or liquid waste, the following model is recommended by two consultants to several large European municipal clients. All the variables and distributions are time dependent.

$v(t) = n(t) \; d\big(I(t), A(t), t\big) \; s(t) \; r(\vec{x(t)}, t)$

  • $v$ is the waste rate in metric tons per week
  • $n$ is the population of the collection region
  • $d$ is the average demographically based disposal rate in Kg per person – Income and age affect consumption and the waste products of consumption, affecting disposal.
  • $I$ is the distribution of income
  • $A$ is the distribution of age
  • $s$ is the dimensionless seasonal factor
  • $f$ is the dimensionless regional factor – The granularity of regions depends on variance considerations.
  • $\vec{x}$ is a vector of non-demographic factors that are likely to correlate with regional waste rates, such as number of university students, size of recycling receptacles, or number of fast food restaurants

Although some of the above factors can be combined in a single training process, there are advantages of leaving them as separate factors.

  • Demographics is measured at the municipal level and has its own model that can be trained based on immigration, emigration, births, deaths, and aging. Proven models exist for this already.
  • Separation of training models allows for component wise system development and the understanding of relationships in data that would remain opaque if training was aggregated.

Each of the distributions can be obtained from municipal data. Labeled data for $n, d, s,$ and $r$ can be measured by implementing the appropriate municipal data collection processes. If it is already available, that would be unusual, and cleaning the data (data sanitation) would be an important consideration.

Once labeled data is available for training and the artificial network design is selected, training can occur, followed by testing, followed by validation. As the system is deployed, care would need to be taken to continuously refine the artificial network hyper-parameters for monthly, quarterly, or annual training and continue to improve data collection accuracy and reliability.

High accuracy and reliability in such prediction could be expected for short term. Longer term prediction may have considerably lower accuracy and reliability because of the higher probably of the occurrence of new types of social and economic shift around which the model could not have been advanced trained.

| improve this answer | |
$\endgroup$
  • $\begingroup$ I think "$f$ is the dimensionless regional factor" needs to be replaced by "$r$ is the dimensionless regional factor"? Other than that, it is very interesting to see a real world model including factors such as explainability and injected domain knowledge. $\endgroup$ – Neil Slater Oct 1 '18 at 17:16
  • 1
    $\begingroup$ @FelicityC I love you. Thank you very much. This is very useful to me. Could you give me some reference to the model you named? I didn't find it, and it would be very useful to me. $\endgroup$ – user60108 Oct 21 '18 at 15:25
  • 1
    $\begingroup$ @user60108, the only reference is my brain and the bragging of two contractors at a consulting agency's holiday party. But everything they said jived with my own experience in a similar domain of solar cell recycling, where I mined and profiled some global data for a proposal. I doubt urban planning academia is up to speed. This answer being Creative Commons may be the first of further documents with improved treatments over mine. If you discover more, you can add your own answer to your question and comment back so I see it. $\endgroup$ – FelicityC Oct 26 '18 at 16:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.