Oct. 31, 2023

Unleashing the Power of Data: Eneco's Energy Predictions Get a 10% Boost

We worked with Eneco to improve prediction of long term energy use of its clients. We were able to show a 10% improvement compared to current methods and integrate the machine learning predictions into their operational framework. After this successful first step we will continue working with Eneco to improve the scope and model performance of the current setup.

Energy's Evolution: Adapting to a Rapidly Changing Landscape

In recent times, the energy landscape has undergone a remarkable transformation. For decades people's electricity and gas consumption were relatively stable. But then came solar panels, electric cars and innovative electric appliances like electric boilers and heat exchangers. In a matter of years, what was once a relatively predictable use case; “estimating the electricity demand for a given area at a specific hour throughout the year” has turned into a logistical nightmare using traditional tools. Even worse, volatility in electricity and gas prices caused by global events has further complicated the situation.

Role of Grid Operators and Energy Suppliers

In the Netherlands (and most neighboring countries), grid operators play a crucial role in maintaining grid stability throughout the day. They ensure that the electricity generated matches the consumption. On the other hand, energy suppliers, such as Eneco and Oxxio, are responsible for predicting the energy usage of their customers. They order energy from the grid operators who make sure the supply meets these future demands. Errors can lead to costly consequences and pose potential risks to the grid's stability. The accuracy of the usage predictions is therefore of paramount importance.

5 Objectives for improving Long-Term Energy Predictions

Eneco faced the challenge of adapting their long-term usage planning to cope with the dynamic changes in the energy market. To address this, a large-scale project was initiated, enlisting the support of BigData Republic to develop automated systems and machine learning-based predictions. The project's primary focus was to improve the accuracy of long-term usage predictions for both gas and electricity consumption.

We established a few crucial objectives to achieve this goal:

  1. Creating a benchmark: The project aimed to implement a benchmark that accurately replicates the existing tool-specific forecast methods for electricity and gas usage. This benchmark would serve as a starting point for evaluating improvements.
  2. Proof of Concept Prediction Model: Developing a proof of concept prediction model was crucial to demonstrate improved forecasting capabilities compared to the benchmark. This model would form the basis for further refinement and scaling.
  3. Scaling to a Proof of Value Set-Up: In this stage we aim to provide forecasts for all industrial-scale users which could be ingested in the current systems.
  4. Integration into an Automation Platform: After the model has proved its value in a real life use case we want to further integrate this into an automated system.
  5. Full Automation and Long-Term Monitoring: The ultimate goal was to achieve full automation, providing accurate long-term forecasts for the entire user base. Additionally, a robust monitoring system would be put in place to ensure ongoing performance evaluation and timely adjustments.

Together with the Eneco team, we carefully crafted a roadmap that outlined the key milestones and tasks necessary to achieve the project's objectives. Each step requires a good understanding of what it means to succeed. By setting clear goals and expectations we can evaluate our progress at each point in time.

Setting Up for Success: Building Trust and Transparency

We want to develop a model and pipeline that is fit for its purposes, and we definitely want to build a system Eneco can trust. A lot of projects ultimately fail because its end-users are hesitant to put the results into practice. Trust needs to be built alongside the platform and model. Trust is built in understanding; to cultivate it at Eneco we involve our end-users closely in the development process. We discussed progress, intermediate results and design choices with dedicated contacts within the user group.This allows us to take into account any specific needs and ensure that everyone was kept informed throughout the development process. We provided interactive graphs and examples that accurately represent the current model predictions. This approach helps the end-users to visualize the progress being made and gain a better understanding of the potential outcomes. Overall it is important to be transparent about the improvements being made, but also about the limitations of the current setup. By providing open and honest communication, we were able to establish credibility and confidence in our end-users.

Reaping the rewards: from Proof of Concept to Production

The proof of concept at Eneco began with a subset of 200 industrial-scale connections, each having a minimum of 3 years of data. The primary goal was to develop a model capable of predicting energy consumption a year into the future with fine temporal accuracy. Remarkably, the tests showed a 15-20% improvement over existing methods, setting a promising foundation for the project. Encouraged by the initial success, we moved on to the proof of value stage, where larger-scale predictions were tackled in a dedicated pipeline.

This involved optimizing data flows and compute environments to handle significant data requests and parallel processing. It also meant handling a more diverse set of connections with varying amounts of historical data. A reduction in data quality and an increase in complexity often means performance will go down. Therefore, it serves as a true test run; how does the model setup hold up in a real-life situation. The proof of value produced positive results, showcasing over 10% improvement compared to the current setup. Though slightly lower than the proof of concept's improvement, this result was expected due to the more complex conditions. Nevertheless, a 10% increase in accuracy is a remarkable achievement, particularly in an industry with high risks and large scales. With confidence in the model's performance, we are now poised to fully integrate its predictions into Eneco's day-to-day work. The enhanced accuracy will empower the sourcing and pricing team to make informed decisions, optimize energy sourcing, and refine pricing strategies.

In overview

Eneco's ambitious project aimed to enhance long-term energy use prediction, and we gladly took on the challenge. Throughout the journey, we achieved significant milestones, successfully replicating the current methodology, establishing an efficient prediction process, and seamlessly integrating it with the existing data flows. During the proof of value phase, our model showcased an impressive 10% improvement, paving the way for its implementation on a large scale. This achievement stands as a testament to the power of data science in transforming the energy sector and propelling Eneco towards a future of enhanced accuracy and efficiency.