Photo: Fred Hochstaedter used under creative commons licence
MathWorks – a global company that develops mathematical computing software for engineers and scientists, has got into the IoT business with ThingSpeak, a cloud-based service that aggregates, visualises and analyses live data streams in the cloud.
In particular ThingSpeak is integrated with MathWorks’ MATLAB computing environment widely used by scientists and engineers.
MathWorks argues that these two technologies combined with rich data sources obtainable from IoT devices enable machine learning techniques to be applied to produce insights and forecasts that traditionally require specialised systems.
An example is elaborated in a white paper by MathWorks’ Robert S Mawrey is a system for forecasting wind induced tidal surges.
As he explains “Water depth varies with the tides, but it is also significantly influenced by the strength, duration, and direction of the wind. Forecasting wind-driven water levels typically requires sophisticated hydrodynamic models as well as detailed knowledge of the shape of the local bay and ocean floor.
“[In the US] NOAA [National Oceanic and Atmospheric Administration] and other organisations use these resources to forecast water depth in major harbours, but minor ports and bays cannot justify the expense involved.”
Tidal forecasting on the cheap
His white paper explains how MathWorks has been able to produce accurate forecasts for minor ports and bays at costs that make such a service economically feasible by using neural networks and low-cost hardware devices rather than computationally intensive hydrodynamic models and complex web infrastructure.
The example detailed in his white paper uses data captured in a bay in Cape Cod, Massachusetts. The system uses historical tide and wind data combined with wind speed and direction forecasts and astronomical tide forecasts and current tide levels measured by sensors.
MathWorks MatLab software takes all these data sources and “runs tidal prediction and neural network algorithms that forecast the tide surge and generate an on-demand tide surge forecast plot.”
After the event, the forecast wind-induces tidal surges are compared with the observed surges to refine the algorithms.
Mawrey says: “We have demonstrated prototyping and deploying sophisticated IoT analytics without web development or the deployment of web infrastructure. We found that the ability to use MATLAB functions and tools such as Neural Network Toolbox apps reduces the overall effort and makes it possible for a single engineer to implement a working IoT analytics system without specialised web development skills or highly specialised statistics and machine learning knowledge.
He concludes: “This IoT analytics workflow can easily be applied to other data analytics and machine learning applications such as predictive maintenance or power load forecasting.”
As more IoT devices are installed, initially for basic monitoring applications, they will create rich sources of data that can be mined, combined and analysed. It seems highly likely that many new use cases for machine learning and analytics will be developed that, like the tidal forecasts, today are either not possible or prohibitively expensive.
