Business Results

  • Higher reliability of service from proactive maintenance

  • Lower operational costs and efficiency gains from on-site visits

  • Up to 15% reduction in training time

  • Up to 30% faster inference time

  • Up to a 25% increase in prediction accuracy compared to current siloed manual processes

author-image

By

View All Reference Kits

Background

Utility service reliability depends on the health of the assets in the field. As energy consumption continues to grow worldwide, distribution assets in the field are also expected to grow. In the US, for example, the failure of any of the 170 million utility poles could result in considerable service disruptions and potentially risk creating wildfires in vulnerable areas.

Without accurate visibility of the health of the field assets, utilities make unnecessary service calls or potentially delay needed maintenance of assets. These challenges drive utility companies to adopt better predictive maintenance solutions.

 

Solution

Predict the Health of Utility Assets With Higher Accuracy

In collaboration with Accenture*, Intel has developed an AI reference kit to help utilities predict the health of utility poles and probability of failure. Each toolkit includes:
 

  • Training data
  • An open source, trained, predictive asset analytics model
  • Libraries
  • User guides
  • oneAPI components

These AI models are optimized for 95 percent predictive accuracy, faster training performance and testing, and lower inferencing costs that result in higher data science success.

This predictive analytics model was trained on utility poles with 34 attributes and over 10 million data points. Data includes asset age, mechanical properties, geospatial data, inspections, the manufacturer, prior repair and maintenance history, and outage records. The predictive asset maintenance model continuously learns as new data is provided, such as a new pole manufacturer and outages.

Data scientists can use this machine learning model with their own data, and apply the results to prioritize maintenance schedules, find cost efficiencies, and improve system reliability.

Technology

Optimized on Intel® oneAPI Data Analytics Library (oneDAL) for Better Performance

The predictive asset analytics model was trained using XGBoost optimized by the oneDAL for better performance. oneDAL allows you to reuse your XGBoost model development code without any changes for training, and minimal code changes for inferencing without loss of quality.

The Intel® AI Analytics Toolkit (AI Kit) was chosen to build training and inference models across a heterogeneous XPU architecture. Components of the AI Kit, including Intel® Distribution of Modin* and XGBoost optimized by Intel, were tested on Amazon EC2* M6i using 3rd generation Intel® Xeon® processors to optimize the solution.

Benefits

For the data scientists, faster model training means more asset health models can be built and trained. With Intel® oneAPI toolkits, little to no code changes are required to attain the performance boost. Additional machine learning models could include overhead utility assets like poles, switches, and transformers, and underground assets like cables and transformers. Faster inference time results in less compute time and fewer costs to produce predictive asset health forecasts for hundreds of thousands of assets.

For utilities, an AI predictive asset analytics model can prioritize which assets a field crew visits based on the highest probability of failure. With a map of those assets and their health index, crews can proactively plan maintenance schedules with efficiencies in mind, resulting in a higher level of service reliability and longer life expectancy for assets.

Download Kit

Product and Performance Information

1

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.