IoT Based Online Load Forecasting

Abstract

Load forecasting is a data intensive statistical method. Internet of things (IoT) based online load forecasting (LF) collects those data from internet on demand and then performs fast statistical and optimization methods for forecasting efficiently. IoT based online LF not only depends on power systems properties, but also internet, machine-to-machine (M2M) connections, communications and computation facilities. Limitations for better utilization, reliability, stability and control of smart grid technologies make it different from traditional load forecasting. Power systems are typically large, complex and distributed. In this study, load data is collected from smart meters and stored as historical load data. However, weather data at a given geographical location including temperature, humidity, wind speed, wind direction, heat, sunlight, solar radiation, rainfall and so on with good accuracy are collected from internet on demand. Computations are done in two steps: first neural network (NN) training to map the dynamics of load and then an optimization on the NN weights to improve overall forecasting error. NN is an effective mathematical tool for mapping complex relationships. On the other hand, particle swarm optimization (PSO) is used because it is the most promising swarm based optimization tool. Results show the effectiveness of the proposed online short term load forecasting in IoT.