A new dual weights optimization incremental learning algorithm for time series forecasting

摘要

In this paper, a novel dual weights optimization Incremental Learning (w2IL) algorithm is developed to solve Time Series Forecasting (TSF) problem. The w2IL algorithm utilizes IELM as the base learner, while its incremental learning scheme is implemented by employing a newly designed Adaptively Weighted Predictors Aggregation (AdaWPA) subalgorithm to aggregate the existing base predictors with the ones generated upon the new data. There exist two major innovations within w2IL, namely, the well-designed Adaptive Samples Weights Initialization (AdaSWI) and AdaWPA subalgorithms. The AdaSWI subalgorithm initializes the samples’ weights adaptively based on the generated base models’ prediction errors, and fine-tunes the samples’ weights based on the distances from the samples to the clustering centers of base models’ training datasets, achieving more appropriate samples weights initialization. While the AdaWPA algorithm adaptively adjusts base predictors’ weights based on prediction instances and integrates the base predictors employing these adjusted weights. Besides, the AdaWPA subalgorithm makes use of Fuzzy C-Means (FCM) clustering algorithm for distance measurement, further reducing computational complexity and storage space of the algorithm. The w2IL algorithm constructed in this way possesses significantly superior prediction performance compared with other existing good algorithms, which has been verified through experimental results on six benchmark real-world TSF datasets.