How does Baseform's event detection benefit from AI?

Monitor's AI ‘algorithm’ is permanently learning the collective water use behavior of large groups of people, in order to predict where it will be in the next minutes, hours or days.

Baseform Monitor is our real-time operational environment where all sensorization converges – dedicated to event detection, fast response, tracking of supply, consumption and losses, hydraulic simulation, and direct everyday O&M. Event detection is a game-changer enabled by computational methods. Multiple factors are at play in the competent detection of network events. We are often asked where exactly is AI at play inside this environment. Event detection is based on the ability to detect values that deviate from normality in a data stream; and then decide whether this deviation is likely to be worth the time of the utility crews that will look into it. In water distribution, normal behavior can be many things – from simple residential consumption (which in itself varies considerably from one neighborhood to the next, to the myriad different behaviors that may be generated by industrial, commercial, leisure and so many other human activities.

Establishing this normality is a challenge, particularly when trying to detect events automatically. This is where machine-learning and other AI capabilities come in handy. An easy-to-understand, if distant, parallel: the “algorithm” used by social networking apps such as Instagram or Youtube is none other than a set of observational capabilities that track each user's behavior, trying to determine what is to be expected – i.e., the user's normal behavior and interests – in order to serve them content that targets that behavior. Monitor's AI ‘algorithm’ is permanently learning the collective water use behavior of large groups of people – through an aggregate measurement of their water consumption – in order to predict where it will be in the next minutes, hours or days. From there, it is better equipped to recognize what it sees when it gets there: does it looks like normal consumption, or not?

This normal will include a large number of past behaviors that have either been spontaneously learnt by the machine, or taught to it through the utility’s continuous processing of events: e.g., what filling up a swimming pool looks like; or sprinklers coming on at night when summer kicks in.

We’ve been deploying detector configurations in over 400 water systems comprising some 6000 water balance sectors (supply zones, pressure zones, DMAs). Having our engines detect all of those types of events in such numbers has taught us, and our machine-learning algorithms, a lot.

Baseform is the vDMA expert software. Can you afford to trust your vDMAs to second-best?

Get in touch to find out how your utility can also benefit from Baseform.