Crude oil time series prediction model based on LSTM network with chaotic Henry gas solubility optimization

Estimating the price of crude oil, which is seen as an important resource for economic development and stability in the world, is a topic of great interest by policy makers and market participants. However, the chaotic and nonlinear characteristics of crude oil time series (COTS) make it difficult to estimate crude oil prices with high accuracy. To overcome these challenges, a new crude oil price prediction model is proposed in this study, which includes the long short-term memory (LSTM), technical indicators such as trend, volatility and momentum, and the chaotic Henry gas solubility optimization (CHGSO) technique. In the proposed model, features based on trend, momentum and volatility technical indicators are utilized. The features are obtained by using the trend indicators such as exponential moving average (EMA), simple moving average (SMA) and Kaufman's adaptive moving average (KAMA), the momentum indicators such as commodity channel index (CCI), rate of change (ROC) and relative strength index (RSI), and the volatility indicators such as average true range (ATR), volatility ratio (VR) and highest high-lowest low (HHLL). These indicators are obtained separately for the West Texas Intermediate (WTI) and Brent COTS. Especially, including the volatility indicator in the model is important in terms of the robustness of the proposed model. The features based on EMA, SMA, KAMA indicators are composed by changing the period values between 3 and 10, the features based on ROC indicator is created by changing the period values between 5 and 12, and the features based on CCI, RSI, ATR, VR and HHLL indicators are formed by changing the period values between 5 and 20. The features are selected by CHGSO algorithm based on the logistic chaotic map, which is successful in avoiding local optima and balancing exploitation and exploration in the search space. Both Theil's U and the mean absolute percentage error (MAPE) values are utilized in the optimization algorithm as the objective function. The results show that the proposed prediction model copes with the chaoticity and nonlinear dynamics of both WTI and Brent COTS.