How to use Artificial Intelligence to obtain environmental sustainability

Why is important?

Proactive management of climate and environmental issues is a core element of Ericsson’s sustainability strategy. forms one of three main focus areas for sustainability – the other two are responsible business and digital inclusion (read more here). With the growing threat of global warming, the negative impact of carbon emissions is a pressing global concern. The pressure on companies to accelerate climate action and limit global warming has never been greater, and the corporate world is making commitments to achieve its goal of becoming Net Zero across its entire value chain. According to the US Environmental Protection Agency’s 2019 Climate Change Report, transportation is responsible for an alarming 29 percent of global greenhouse gas (GHG) emissions. Here at Ericsson, we take the need for decarbonization seriously. To address the threat head-on, Ericsson has committed to achieving net-zero emissions in our value chain by 2040. Ericsson is already working on a first major milestone to reduce emissions in the supply chain and portfolio by 50 percent by 2030 and become net zero at the same time in our own operations. One of the most important activities to reduce emissions in the supply chain is product transportation.

Instead of the above, and to achieve our ambitious goal of reducing CO2 equivalent emissions, the Group IT AI and automation experts, together with Supply, have defined a plan to optimize product transport – including monitoring, forecasting and reduction – with the goal Using AI to make the unimaginable possible.

How to measure and analyze CO2 equivalent emissions from transport

But how could this strategic plan be implemented to actually make a difference in traffic emissions and the businesses that are outside of a company’s direct control? First, we needed the information to understand the full extent of emissions in the supply chain through measurable data and transparent reporting, referred to as the ‘monitoring phase’.

This phase helped to map the current CO2 equivalent emissions scenario in the organization, including multiple complex product transport flows such as customer supply chain, product supply chain, local transport and various processes. The main challenge in developing such a solution was the lack of availability of data, linked data from different sources and the development of the exact logic to calculate the CO2 equivalent emission. Using various analysis techniques and fuel-based, distance-based and cost-based methods, we were able to calculate the emissions associated with transportation. After several trials and errors, the distance-based method proved to be the most suitable approach for Ericsson Transport Management. We derived the CO2 equivalent emission by modeling common parameters such as the volume of goods purchased, the distance traveled, the standard emission factor for the given transport mode and/or type, etc. The model was built generic enough to be suitable for most similar transport services. A simplified version of the CO2 equivalent calculations across different modes of transport is as follows:

Y=Σ (mass of goods purchased (tonnes or volume) × distance traveled on the transport route (km) × emission factor of the mode of transport or vehicle type (kg CO2e/tonne or volume/km))

With the CO2e emissions algorithm and the web-based LowCode visualization dashboard that allows global users to interact simultaneously, we were able to provide a complete monitoring solution based on the data model, a reference snapshot of the dashboard as below:

Carbon Emissions Dashboard - Ericsson Global and Unit by Unit.  Note: All numbers shown are dummy data.

Figure 1: CO2e emissions dashboard – Ericsson Global and Units wise. Note: All numbers shown are dummy data.

CO2e Emissions Dashboard Ericsson Global - Mode of Transport.

Figure 2: CO2e Emissions Dashboard Ericsson Global – Mode of Transport.

In addition to measuring and monitoring C02e emissions across different modes of transport, we were able to improve data quality. By analyzing the data collected, the company was able to identify specific areas where the quality of the data was low and drive initiatives with data stewards to improve the quality of the data collected and initiatives at an operational level to capture the right data. This iterative process of improving data quality will gradually help business drivers make sensible decisions.

Transforming data into future insights: prediction phase

With the data and fundamental analysis in hand, the next logical step was to capture key patterns and trends in order to predict future business, referred to as the “forecast phase”.

The forecast of the shipping volume and weight for different transport routes was a complex process due to the high proportion of disaggregated freight flows. The uncertainty in transportation services, diverse processes and non-standard procedures made it more difficult to achieve optimal utilization of transportation resources and distribution planning.

By applying machine learning (ML) techniques such as regression, clustering, deep learning, etc., and using historical and transactional data, we have developed a more accurate long-term and short-term shipment weight forecast than we have obtained from manual predictions.

Applying such approaches not only reduces the need for manual forecasting, but also helps logistics service providers (LSPs) achieve better delivery accuracy, resulting in improved rates and therefore lower costs. The forecast will ensure the availability of transport capacity and significantly reduce throughput time.

With a good forecast, the LSPs can identify the most important drivers for freights across regions and transport routes as well as the impact on the entire product transport chain.

The modeling part of the solution consists of several boosting algorithms with a wide range of hyperparameter tunings for features like learning rate, max_depth, n_estimators, subsample. Due to volatility and inconsistencies in the data, no single model could provide results, so an ensemble of machine learning models with different hyperparameters was developed. The framework was designed in such a way that the best models (for technically savvy people this was implemented by the lowest WMAPE, Weighted Mean Absolute Percentage Error) are captured in a dynamic mode at runtime and used to predict the associated weight/volume.

Action Prediction – Reduction Phase

After measuring and analyzing the CO2e emissions results and having a good long and short-term forecast, it is now time to make plans and implement methods to reduce CO2e emissions, also known as the “reduction phase”.

Continuous dunning processes can certainly help to optimize CO2e emissions iteratively with the help of monitoring and forecasting.

  • Reduce and improve transport activities
    • Fleet Optimization – Higher filling level in trucks, reduction of unnecessary air transport and updated fleet.
    • Improved planning of packaging and transport material, better cooperation with suppliers.
  • Improving transportation efficiency
    • Avoid short lead times by using forecasts
    • Use well-organized navigation
    • Predictive analytics to prevent vehicle failures and effectively use less energy (predictive maintenance)

In short, the logistics transportation sector is a major consumer of fossil fuels and therefore a major contributor to total greenhouse gas (GHG) emissions, which make up total GHG emissions, according to the US Environmental Protection Agency’s 2019 Climate Change Report. From our own operations, we have learned that AI can be used to reduce the use of transport vehicles by optimizing vehicle flow, providing more efficient navigation and facilitating shared transport.

Based on our insights into the potential of AI and the ongoing evolution in industrial automation, this blog post has highlighted our three-phase strategic approach – monitor, predict and mitigate.

In addition to improving the visibility and transparency of CO2e emissions, as well as improving data quality, operational efficiency and customer satisfaction, transport volume forecasting aims to reduce both CO2e emissions and operational expenses.

We believe that the environmental sustainability challenges could be addressed through this three-phase approach along with the strong domain knowledge. We have no doubt that if used wisely, AI will accelerate our sustainability efforts.

AI is already emerging as the key to empowering governments, organizations and individuals to make more conscious decisions and work towards creating a healthier planet. At Ericsson, we are working on this cause and are proud to show how applied AI can create lasting change.

The severity of climate change combined with the potential of artificial intelligence makes it too important not to try, don’t you think?

Learn more

For more AI/ML use cases, see Artificial Intelligence/Machine Learning (internal link)