Study and Analysis on Addressing Present Drawbacks of Traditional Surge Protection Devices (SPDs) using Machine Learning

Surge Protection Devices (SPDs) are being extensively used at present to safeguard electronic equipment from lightning generated transient over-voltages. Despite SPDs being employed to protect electronic equipment, every year millions worth damages are being reported. Hence, isolation from the power grid would be considered as the best solution to prevent the infiltration of harmful energy contained in the transient over-voltages. But isolation cannot be performed by humans as they are not sensitive to imminent lightning discharges nor fast enough to respond post lightning events. Thus, there should be an extra-fast mechanism to detect imminent lighting discharge and perform a change-over from the utility supply to a local power supply. This study aims to device a machine learning solution which could be used to overcome such limitations in traditional SPDs. For the convenience of analysis, reported impulses were categorized into three signature types. Namely, pulse-burst, unipolar and bipolar. A data sample was taken which represents all above said signature types, was processed and fed into the Azure Machine Learning Studio in order to train a linear regression model. Such model yielded an R² value of 0.7547. The strong positive correlation between the strength of the electric field and the magnitude of the induced voltage was thereby confirmed. The deployed solution had a mean accuracy of 87.82% of its predictions, confirming its ability to accurately predict the magnitude of the induced voltages to take proactive action and thereby safeguard electrical and electronic equipment if an incoming induced voltage is beyond the threshold.