Deep learning can optimise waste stream logistics

By Dr. Beatriz Royo, ZLC Associate Professor.

Leer versión en español


The need to re-use, repurpose, recycle or responsibly dispose of wastes of all kinds has in the past few decades moved from an aspiration to an imperative. But achieving this remains a huge challenge.

That has been the focus of OPTIWASTE, a project sponsored by Spain’s Ministry of Industry, Commerce and Tourism, which ZLC and our partners have recently completed. The project looks specifically at three objectives: environmental sustainability, in terms of minimising the output of CO2 per kg of waste managed and of other pollutants; economic sustainability – reducing the costs of transport and final disposal, and therefore minimising the impact on the cost of new products; and increased ‘circularity’ – improving the recyclability of wastes and their reintegration into the supply chain (which might be at any level from the reconditioning of complete items to the recovery and reuse of raw materials), by optimising the logistics and custody cycle so that there is the least possible degradation of items and materials.

The project focussed on that most difficult of waste streams – Waste Electrical and Electronic Materials (WEEE). Optimising this particular waste stream is important in a number of ways, beyond the need to handle large and rapidly growing volumes. On the one hand, WEEE often contains elements such as gold, copper, and rare earths which are valuable in themselves and whose primary extraction is often both environmentally damaging and exploitative.

Meanwhile, the uncontrolled storage and disposal of WEEE creates many significant hazards: combustion risks from damaged batteries that contain (waste batteries and accumulators have a separate management flow), groundwater contamination from the leaching of heavy metals and other compounds, the creation of microplastics, to name but three.

The issue has long been recognised – the EU Directive on WEEE came into law some twenty years ago. The Directive and associated legislation set mandatory requirements for recovery (in kg per head of population) and for the recyclate content of new products, and for cost-free routes for consumers to return WEEE. However, the reverse logistics required to achieve these targets is far from simple. WEEE can arise anywhere – sometimes with major manufacturers, users, or retailers that generate wastes or take used goods back. Most WEEE  (from individual households or other sites), ends up with distributors that take goods back on behalf of manufacturers, at collection/concentration points run by local authorities in an attempt to divert WEEE from the general waste stream, or at a variety of small shops and other businesses.

Typically, when enough waste has accumulated, the waste management company collects the waste, transports it to a waste management plant, and separates it into products that can have their useful life extended, streams that offer material recovery opportunities, and a residue of waste (quite possibly hazardous) that must be stored pending safe and responsible disposal. The nature, volume, and frequency of consignment from any particular point appears to be quite unpredictable, thus challenging attempts at transport and logistics optimisation.

However, ZLC and our OPTIWASTE partners Predictland and Pronet-ISE have access to a database of transactional and collection data across the Iberian Peninsula, the Balearics and the Canaries extending back 17 years and more. Can we use modern data techniques to enable waste managers to optimize service quality by reducing the lag between a client ordering a collection and that occurring, and the tie taken to treat the waste at the recovery plant, optimize the costs and emissions of the transport element, and increase the proportion of material that can be reintroduced into the production cycle?

Predictland was able to predict anticipate when the orders for collecting WEEE will take place by using big data and artificial intelligence techniques. In the case of ZLC, we analysed this data to try to determine the best location for a waste treatment plant, and the number of containers that would be required in the transport network to serve this. We looked at the logistics network data for the previous year, and found that while some regions are self-contained, in that the waste generated in the region is also treated there, for 10 out of the 19 regions there was an imbalance between generation and treatment capacity. We focussed on these areas as potential locations for the new plant.

The next step  was to devise and apply multicriteria decision analysis. That isn’t entirely straightforward – what do we mean by ‘optimal location’. We put a set of experts, with varying profiles but of great experience, through an iterative process, not only defining the criteria but deciding the relative weights to be placed on them. A range of social, economic, environmental, legislative and other considerations were included in the mix.

Actually finding values for each criterion required a mix of quantities from the database, third party data, and some more subjective input from the experts.

This learning was run through an algorithm and a sensitivity analysis (and as it happens, Aragon was deemed the best location, followed by Castilla Leon).

It was then fairly straightforward to estimate the number of different types of container required for the proposed new situation.

We have effectively shown how data forms the basis of well-informed decision-making for strategic choices such as the location of a new plant. Additionally, by utilizing cutting-edge data analytics and artificial intelligence capabilities, the consortium was able to mitigate the unpredictable nature of WEEE orders. The findings will highlight the potential cost savings, enhanced service quality, and increased “circularity” of products throughout their life cycles, all while lowering unfavourable logistical externalities, particularly greenhouse gas emissions.

For more information contact Dr. Beatriz Royo, ZLC Associate Professor at [email protected]