[John Blatchford]

Hi Ratanak,

Our team was able to develop a binary classification model to predict vehicle contact hits based on historical records. We created a XGBOOST model in Python and collected free datasets readily available on the internet. We had several different outputs for the model, one of which was to classify each pole based on risk and place them into one of three categories (High/Medium/Low). The model was visualized both using the Folium Python module and Power BI.

Depending on your use case the model outputs could be used in different ways. We found that classifying risk on a per pole basis was not the most beneficial to our business. When we talked to the teams who would be doing the pole risk mitigations, they preferred having risk based on roads/or territories. This would require us to group our poles and determine which “group” had the most risk. We have explored grouping using various geospatial techniques, primarily from open source libraries.

For your reference/brainstorming on the data side, we found that pole proximity (to the road), vertices (curve of the road), proximity to points of interest (schools, shopping centers, etc), proximity to road intersections, and average road speed were the most relevant variables in our model. We are working on updating the model by bringing new datasets, for example, we would like to differentiate poles that are protected by guard rails or other protections to further improve the model, but we have not been successful in gathering this data from free data sources.

Good luck on your project! 

——————————
John Blatchford
IT Associate
San Diego Gas & Electric
8588771865
——————————
——————————————-
Original Message:
Sent: 01-08-2021 09:30
From: Ratanak Heng
Subject: Has anyone done any analytics around Pole Hits?

Hello everyone!

Our teams are starting to evaluate how analytics can be leveraged to mitigate and reduce the number of pole hits on the system. Some of our OpCos have reliability programs to relocate poles and/or install pole reflectors – and we are looking at what analytics tools/model should be developed to help us better identify where to put those investments. The capabilities our engineers are asking for:

• Descriptive Analytics: Dashboard to display location of pole hits, asset attributes, historical events, key environmental factors (i.e. proximity to roads, bars, etc.)
• Predictive Analytics: Identify and label poles with the highest risks and attributes/features contributing to the risk
• Prescriptive Analytics: Recommend poles that could benefit from pole reflectors or relocation (based on risks and factors contributing to the risks)

We’re still in the brain storming phase, and would love to hear what has been done in this space already. Thanks!

——————————
Ratanak Heng
Manager, Advanced Analytics
Exelon
——————————