Data complexity of daily natural gas consumption: Measurement and impact on forecasting performance

Abstract Data complexity has a great impact on daily natural gas consumption forecasting. However, due to the existence of irregular data, complex periodic change, and volatility data, the conventional methods, such as Lyapunov exponent and sample entropy, are failed to assess the complexity of the consumption data. Thus, this paper proposes a hybrid method of complexity measure, named CMLS. The novel method combined correlation coefficient analysis, missing data detect, Lyapunov exponent, and skewness analysis. Compared with Lyapunov exponent and sample entropy, CMLS is more stable and insensitive to the length of data in complexity measures. Additionally, for revealing the relationship between data complexity and forecasting performance, we design three case studies including 56 sets of daily natural gas consumption, and forecast with three advanced models. The results show that the forecasting performance various a lot in different complexity level. Particularly in very hard level, the daily natural gas consumption data is very hard to be forecasted and the R2 of forecasts are all negative. This paper serves as an initial study seeks to reveal the impact of data complexity on forecasting performance. The findings can help forecasters to evaluate the performance and difficulty of natural gas consumption forecasting.