Data Analytics + Inventory + IoT = Low Cost Insights
A key criticism of IoT from the IoT Utility Summit was inability to justify the business cases. The most touted suggestion was changing business models.
An alternative to deliver some of the benefits of IoT can be acheived by leveraging what they already have with a solution of Data Analytics & Inventory. In-house IoT or 3rd party devices will provide the event and performance data. The inventory will allow the data to be aggregated by hierarchy. Data analytics will combine this to provide insights to segments of the network not currently managed.
A speaker from a power distribution company suggested out of the box thinking to reduce cost. One of his suggestions was to leverage NBN CPE (Customer Premises Equipment) end points to pinpoint power outages. With their network inventory we think they could and it’s something we have done before for other networks. In this case we could consume real-time CPE event data and then overlay the relationship. By aggregating isolated events to parents you can detect actionable abnormalities for segments of the network using a Data Science sequence of clustering. Topology clustering will take availability of a NBN CPE or a Smart Meter which are then clustered based upon the their parent. This pattern can then be repeated for each major component in the hierarchy. This allows each parent in the network status to be derived by the status of its children. This pattern can then be repeated for each layer of the hierarchy. This allows us to infer visibility from the CPE all the way to the top.
To implement this pattern we typically deploy a clustering technique we call:
X in Y within Z
In PSEUDOCODE this looks like:
IF (> 50% of ADDRESSES ARE OFFLINE) in (1 DISTRIBUTION TRANSFORMER) within 5 Minutes
= Actionable Abnormality
This policy can then be chained to include the next level parent. This means we could leverage the DISTRIBUTION TRANSFORMER events which were inferred to also infer status on a SUBSTATION:
IF (> 50% of DISTRIBUTION TRANSFORMERS ARE OFFLINE) in (1 SUBSTATION) within 5 Minutes
= Actionable Abnormality
The following diagram shows how endpoints are grouped together with their one-up parent to provide relative status. These statuses can be chained to give status to the next level up in the hierarchy.
These approaches have been used to provide visibility to non-active parts of customers networks. This clustering provides operational value not just to hierarchal networks such as IP, Wireless and HFC but also applications where similar events can be grouped together to reduce noise and identify actionable abnormalities.
In this example we can monitor these components but we are inferring their status based on their child status. This process can be used not only for event management but also performance management. While not as accurate as deployed sensors this approach is likely be to substantially cheaper than physical devices on scale and can be applied to almost any network to identify actionable abnormalities in real-time. This can then be used by operations to be on the front foot and to trigger corrective action.
To implement these type of policies in the past used to be difficult. But as the AIOps market matures this capability can come out-of-the-box.
If you want to talk IoT, AIOps and reducing the cost of analytics or just have a chat about leveraging your current management assets use the form below to get in touch and we’ll set up a meeting.