Toptal increases HVAC energy model prediction accuracy by 27% with machine learning.

Toptal partnered with an HVAC industry leader to overhaul its energy usage prediction models, improving accuracy, scalability, and cost-effectiveness through advanced machine learning techniques and open-source technologies.

Client

A global HVAC and refrigeration company delivering intelligent climate and energy solutions for buildings and cold chain.

Employees

58,000+

Revenue

$23B

Industry

Industrials

Delivered Services

Get a free consultation now

Schedule a Call

Challenge

The company struggled with inaccurate and expensive models for predicting energy usage, compounded by the complexity of processing petabytes of real-time sensor data.

Solution

Scalable Model Implementation

Toptal transitioned the outdated model to a machine learning framework that utilized distributed computing with Kubernetes and Dask, enabling the parallel training of more than 500,000 lightweight, personalized models for enhanced scalability and speed.

Hybrid Modeling Strategy

Toptal combined physics-based modeling with machine learning, allowing for the creation of adaptive and interpretable models tailored to individual users, which improved the usability and actionability of the energy predictions.

Outcome

Accuracy Lifted by 27%, Reaching 85%

The new models increased prediction accuracy from 67% to 85%, making the energy usage predictions more reliable and actionable for the client.

Cost-effective Technology Transition

By moving from Amazon SageMaker to open-source technologies like Kubeflow and Dask, Toptal reduced the costs associated with training and deploying the models, while also improving performance and scalability.

Share

Get a free consultation now

Schedule a Call