Electricity Load Forecasting with Collective Echo State Networks

Short-term load forecasting (STLF) is one of the most important tasks in power grid systems. Despite thorough research of STLF on community scales, predicting individual power consumption remains challenging. With the deployment of advanced metering infrastructure (AMI) technology, it is more convenient to access and operate load usage data on an individual scale. This work first investigates a novel deep learning method, Collective Long Short-term Memory (LSTM), which can fully utilize correlated meter information from AMI. Then, Collective Echo State Networks (ESN) are proposed as a simplified form of Collective LSTM using the reservoir method. Experimental results show that Collective ESN can achieve state-of-the-art prediction accuracy with extremely fast speed and fewer computational resources.