Short-term Electricity Price Prediction Using Kernel-based Machine Learning Techniques

The consumer behavior so as the load pattern frequently changes in a dynamic power system environment. In the case of a deregulated market, it leads to a complex price profile. To make the supplier as well as the consumer a perfect balance should be maintained. Forecasting the market-clearing price (MCP) in these kinds of markets is the most common and essential task to maximize the benefit of both supplier and consumer. This can be done with the help of neural network-based prediction algorithms. It can map the complex interdependencies between electricity price, historical load, internal and external factors. In this work historical data of the Australian market is taken into consideration for short-term price prediction. The most sought after method Extreme learning machine (ELM) is compared with an advanced Kernel-based technique. To give the prediction algorithm more weight interval prediction is also focused on this work. The volume of data is carefully chosen keeping in mind to avoid premature convergence and overfitting. The performance indices like error measurement units in case of point prediction and width and probability assessment units in case of interval prediction are considered in this study. To make this analysis more extensive seven kernel functions are compared with the ELM algorithm.