Hybrid Models for Short-Term Load Forecasting Using Clustering and Time Series Wael

Abstract

Surface electromyogram (sEMG) is a bioelectric signal that can be cap- tured non-invasively by placing electrodes on the human skin. The sEMG is capa- ble of representing the action intent of nearby muscles. The research of myoelectric control using sEMG has been primarily driven by the potential to create human- machine interfaces which respond to users intentions intuitively. However, it is one of the major gaps between research and commercial applications that there are rarely robust simultaneous control schemes. This paper proposes one classification method and a potential real-time control scheme. Four machine learning classifiers have been tested and compared to find the best configuration for different potential applications, and non-negative matrix factorisation has been used as a pre-processing tool for per- formance improvement. This control scheme achieves its highest accuracy when it is adapted to a single user at a time. It can identify intact subjects hand movements with above 98% precision and 91% upwards for amputees but takes double the amount of time for decision-making