Chaotic Time Series Prediction Based on IBH-LSSVM and Its Application to Short-term Prediction of Dynamic Fluid Level in Oil Wells

In order to improve prediction accuracy, the problem of selecting model parameters is typically addressed using combinatorial optimization. However, the prediction results are often influenced by uncertain parameters in the optimization algorithm. To solve this problem, we propose a prediction model based on an improved black-hole algorithm and a least-squares support vector machine. In this model, we consider the parameters of the embedding dimension, time delay, regularization, and kernel function as the combinatorial optimization targets. The optimization algorithm is not affected by any subjective parameters. We use the online verification method based on the fast leave-one-out technique to prevent over-fitting. By improving the iterative searching mechanism, better prediction effects are achieved. The case study results confirm the significance of the proposed model in the practical application of making short-term predictions of the dynamic fluid level in oil wells.