Successive Model-Agnostic Meta-Learning for Few-Shot Fault Time Series Prognosis

Hai Su,Jiajun Hu,Songsen Yu
2023-11-04
Abstract:Meta learning is a promising technique for solving few-shot fault prediction problems, which have attracted the attention of many researchers in recent years. Existing meta-learning methods for time series prediction, which predominantly rely on random and similarity matching-based task partitioning, face three major limitations: (1) feature exploitation inefficiency; (2) suboptimal task data allocation; and (3) limited robustness with small samples. To overcome these limitations, we introduce a novel 'pseudo meta-task' partitioning scheme that treats a continuous time period of a time series as a meta-task, composed of multiple successive short time periods. Employing continuous time series as pseudo meta-tasks allows our method to extract more comprehensive features and relationships from the data, resulting in more accurate predictions. Moreover, we introduce a differential algorithm to enhance the robustness of our method across different datasets. Through extensive experiments on several fault and time series prediction datasets, we demonstrate that our approach substantially enhances prediction performance and generalization capability under both few-shot and general conditions.
Machine Learning,Artificial Intelligence
What problem does this paper attempt to address?
The paper primarily addresses the few-shot problem in time series fault prediction by proposing a new method called Successive Model-Agnostic Meta-Learning (SMAML). The paper aims to solve the following key issues: 1. **Low feature utilization efficiency**: Existing meta-learning methods for time series prediction tasks often rely on random or similarity-based task partitioning, which leads to low feature utilization. 2. **Suboptimal task data allocation**: Current methods are not optimized in task data allocation, which limits the model's generalization ability on new tasks. 3. **Limited robustness in small sample scenarios**: The performance of existing meta-learning methods is often unstable when the data volume is small. To address the above issues, the authors propose the SMAML method, whose core features include: - **Pseudo-meta task construction**: A novel "pseudo-meta task" partitioning scheme is adopted, treating continuous time periods as a meta-task composed of multiple consecutive short time periods. This approach helps extract more comprehensive features and relationships from the data, thereby improving prediction accuracy. - **Differential algorithm to enhance robustness**: A differential algorithm is introduced to enhance the robustness of the method across different datasets. Through extensive experiments on various fault and time series prediction datasets, the paper demonstrates that the proposed SMAML method significantly improves prediction performance and generalization ability, performing well under both few-shot and general conditions. In summary, this research aims to improve task partitioning strategies within the meta-learning framework to tackle the few-shot challenge in time series fault prediction, and empirically proves the effectiveness and superiority of the proposed method.