Search

• Analyze the correlation between power load data and relevant search engine indices. • Take search engine indices as input variables to forecast electricity load. • Use Hodrick-Prescott filter to decompose the time series of search engine indices. • Propose a LSTM model based on quantum-behaved particle swarm optimization algorithm. Accurate electricity load forecast plays an important role in the operation of a power system. Many factors influence the electricity load data such as air temperature, humidity and holidays, and they are taken as the explanatory variables in load forecasting cases traditionally. Search engine indices, a variable may be related to load data which has been never considered before in load forecasting, is discussed and utilized to increase the accuracy of power load prediction in this paper. Spearman's correlation coefficients and Granger test results verify the correlation between Google Trends (GT) and electricity load data. A methodology for processing GT time series with Hodrick-Prescott filter is proposed. To forecast electricity load with an adaptive network model in such a novel situation, we propose a long short-term memory neural network model based on quantum particle swarm algorithm. The performance of load forecast for Long Island region taking GT and weather data as input variables is compared with that taking only weather data as input variables, which shows that the introduction of GT improves short-term forecasting effectiveness significantly. [ABSTRACT FROM AUTHOR]