Using Machine Learning for carbon emission optimization in Transport and Logistics

This article analyzes the diverse capabilities of Artificial Intelligence and Machine Learning methods for the reduction of greenhouse gas emissions in the Transport & Logistics sector. It furthermore provides a selection of best practices and a weighted outlook in regards to the efficiency of such methods.

Published in

shipzero

9 min readJan 21, 2020

The negative impact of greenhouse gas (GHG) emissions on the global climate has been an ubiquitous topic for the recent years, and finally also concrete activities are being implemented in order to reduce these emission — may it be on a political level as for example the passed Climate Agreement in Germany and the announcement of the European Green Deal from the EU, or on a corporate level with global initiatives towards less corporate carbon emissions like the Carbon Disclosure Project (CDP) and the science-based targets initiative.

Apart from organizations and political regulations, new and innovative technologies are driving the future of GHG emission levels for private persons as well as for corporate organizations.

Especially Artificial Intelligence (AI) and its cognitive domain Machine Learning (ML) are expected to be leading technologies in the mitigation of climate change.

A report by the International Telecommunication Union on the UN activities and projects in 2019 highlights the diverse capabilities of AI use cases: more than 35 UN entities describe how they are increasingly using AI to meet many of the world’s most urgent challenges, from responding to humanitarian crises to tackling climate change.

On a corporate level in specific, emissions are summarized in Scope 1 to Scope 3 in a wide range of activities. Equally, there are a lot of possibilities along the entire value chain to reduce GHG emissions.

Categories of Scope 1, 2 and 3 emissions

Particularly the logistic sector, which has its most substential emission factors within Scope 1 and Scope 3 via upstream and downstream activities and that accounted to 27% of the total EU-28 GHG emissions in 2017, is under pressure: If the 2050 target for this sector (Decrease of emissions from 1990 standard by two thirds) shall be fulfilled, more efficient (technological) solutions for carbon neutrality need to be introduced. Especially, because Transport & Logistics is the industry with the most negative development of carbon emissions since 1990.

Total CO2 Emissions per Industry (1990–2017)

Thinking of ML tools for the Transport & Logistics sector in specific, many of us would primarily connect the technology with the most well-known application: autonomous vehicles. Object and pattern detection already today improve the security of passengers in Tesla’s or other autonomous vehicles. This feature is indeed very relevant for features such as security and driving behaviour, though is it also relevant for climate change? Lets take a brief look at autonomous driving and other opportunities arising from a higher level of automation and data-driven decision-making in the context of carbon emissions:

Reducing and improving transport activity

Bundling shipments together is an efficient method to reduce transport activity and emissions accordingly. As of today, roughly 30% of transports are (partly) empty runs due to demand imbalances, time constraints or insufficient planning. ML methods are applicable for geographical clustering of shippers and the destination of goods as well as for the detection of transport way disruptions. ML can also help in determining the complex combination of shipment size, transport mode, routing coverage and service features. A successful use of ML algorithms in that case could lead to less empty runs of containers / shipments and trips. In terms of the general feasibility, we need to look at the healthy balance of competition coming along with a certain degree of information arbitrage and fully centralized information and control. Many players in the freight forwarding sector are tackling this issue with big budgets to pave the way towards reliable and digitized information.

Furthermore, ML algorithms can improve transportation infrastructure by predicting mobility patterns and recommending more fluent transportation paths. This development is based on the rise of new types of sensors that have become available in the past years, with IoT technology as the infrastructure to connect devices and sensors. Additionally, in Computer Vision ML algorithms are trained in order to detect traffic patterns via object detection and recognition — as an example, vehicles can be detected in high-resolution satellite images. Large transportation companies are focusing on an advancement in this sector — e.g. the logistics and packaging giant UPS fosters partnerships with self-driving vehicle startups like TuSimple — as well as young startups with new technologies like Embark Trucks gain momentum in the market.

Recently, researchers have also been working on more precise and short-termed weather forecasts, which could not only be significantly important for crisis management, but also for the determination of transporting routes and a prevention of shipments being stuck in heavy weather conditions. Specifically Google is making progress in this field and is reportedly able to forecast rainfalls precisely to up to six hours ahead of time in order to avoid transport emissions through short-term detours. The method is based on predictions about simple radar data (and the detection of clusters or classifications) instead of trying to analyze the motion of clouds or the physics based simulation of weather systems, which have been the two existing methods on the market so far.

Transport efficiency

ML algorithms are also used for predictive analytics of machinery and accurate assembly of components, which can result in an improved design or functionality of vehicle features and a longer product life-time.

This process, called Predictive Maintenance, is a common method to prevent failures of machinery components and is accordingly effective in using less material (and energy) than in regular production and maintenance processes. Further improvements in energy consumption through ML tools can be generated for the field of autonomous vehicles, as it was already indicated in the introduction: Economic driving, the avoidance of traffic congestion or the efficiency of platooning effects (close distance of driving through autonomous vehicles while vehicles are allowed to brake and accelerate simultaneously) are only some advantages in carbon emissions compared to regular vehicles.The more data a vehicle including all its components provides, the less emissions will be emitted: Fully connected vehicles will pollute less through a most-efficient way of driving, and the possibility of broken machineries (and coherently more energy emissions through additional productions) can be minimized through a better understanding of all data, referring to the concept of digital twins.

Electric vehicles

The impact of electric vehicles on carbon emission efficiency is supposed to be of major importance, with a positive trend the longer the car and the respective battery is in use. ML algorithms can support the recharging process, as failures in charging can be prevented by an early detection of anomalies, or charging patterns can support the adequate installation of new charging stations. Equally, the production process of batteries as well as the maintenance can be implemented more efficiently and more rapidly, as failures can be detected easier.

The electric car manufacturer Tesla is dominating the market of electric and autonomous vehicles via two technological advantages: its full self-driving car computer technology that uses two own-produced AI processors that are supposed to be more powerful than the market standards, and its own battery production centers that let them produce electric cars more cost-efficient. ML methods support this cheaper production process via a continuously learning progress from performance data such as cell temperature, charge currents and swelling levels.

Load security, time of arrivals and identification of trucks in facilities

Accumulated data from IoT sensors and / or telematic systems can be used in shipyards, logistic centers and other facilities in order to granularly track vehicles. This may improve the required time a human eye needs for identifying a truck on a facility and accordingly prevents possible mistakes in the organization of freight collection and unloading. Furthermore, better estimations of transport duration and time of arrivals through classification and cluster patterns can not only save money, but can also contribute to a more efficient work process at shipyards and logistic centers. This may not seem relevant on the first glance, but especially because emissions on logistic sites have a low relevance in media yet and may be underrated, the reduction of such emissions will become more relevant soon too.

Another factor ML methods can help is in the prevention of accidents: On the way to its destination, image recognition tools may detect transport hazards (such as overloaded trucks or vehicle instability) and prevent accidents and traffic congestion that cause additional GHG emissions more easily. Once again, IoT sensor and / or Computer Vision technology are the required infrastructure in order to realize such prevention. This factor is compliant with the possibility to predict the most efficient way of transport in a transparent market via ML methods: The climate footprint of corporates and their service providers will play a more dominant role, thus pricing will likely not be the most dominating factor in tendering processes anymore. Consequently, even an intermodal shift for the transport way may be relevant in case the ML application predicts a faster and / or a more sustainable arrival via other modes such as rail, ship or urban e-mobility modes.

Downsides of ML applications for carbon emissions

All these suggestions sound very helpful, but what are the downsides of ML applications?

Despite the extensive and diverse possibilities to use ML methods in order to reduce carbon emissions in the Transport & Logistics sector, ML methods are also being criticized for the extensive computing power that is necessary in order to train the respective algorithms. The carbon footprint of training AI models is often overseen, but once datasets get bigger, the cost in energy (and environmental impact) are going to become substential as well. This is especially true for handling large amounts of real-time data coming from sensors and telematics systems of large vehicle fleets.

In order to accomplish a lower carbon intensity in the Supply Chain, not only a sophisticated technological infrastructure via IoT sensor and Computer Vision technology has to be guaranteed. Also, the application of ML methods itself (lower amount of needed data to train algorithms) and the power generation (higher use of renewable energy sources for servers) need to develop further. It is certainly important that the industry progresses in the development of more sustainable training solutions, and tools that are being developed in order to raise awareness show that this movement is taken seriously.

Additionally, intermodal friction may be a challenge in the collection of data: Different transport modes and respective logistic and telematic service providers likely provide different technological data standards and tools, which may create more effort to gather all relevant information for the emission calculation by product. This aspect is connected with the level of digitization that may be insufficient for certain small and medium companies and that account for a large part of the transportation market. The handling of intermodal data provision, possibly different software systems and API management require intenal resources that are more likely to be found in large companies.

ML in the German Transport & Logistics sector

Looking at the current market situation in Germany, there is still a lot potential for increasing the innovation and penetration level. Young and innovative companies like Evertracker (offering an AI based platform based control tower for multiple activities within the Supply Chain) have made a successful entry to the market, but the market share of digital native companies or services are yet extremely low - the logistic industry has to catch up with the current state of ML applications though in order to accomplish the ambitious carbon emission targets.

Key Take-aways

The Transport & Logistics sector needs a strong turnaround towards more sustainable processes in order to meet future climate targets
Opportunities to reduce GHG emissions via Machine Learning methods are diverse, but are approached on a low scale in the sector of Transport & Logistics yet
The more data we have, the better we can predict and coherently prevent failures and carbon emissions through additional energy use
Energy usage for training of ML algorithms and for handling real-time sensor data is a relevant factor and can be tackled via sustainable energy sources for server and a more efficient use of
For corporate companies, the selection of their logistic partners will be based more on the accuracy of carbon emission data and the final footprint of the service providers
Small and medium sized companies (logistic and non-logistic) often do not have the capabilities to generate efficient data; telematic systems and aggregators may help though by platform solutions.

Sources:

Using Machine Learning for carbon emission optimization in Transport and Logistics

Written by Martin Jacobs