# A Long-Time Series Forecast Method for Wind Turbine Blade Strain with Incremental Bi-LSTM Learning

## TL;DR

This paper introduces a new method using incremental learning to improve long-term strain prediction in wind turbine blades, enhancing accuracy over existing models.

## Contribution

The novel incremental Bi-LSTM method with error-supervised feedback and experience replay for long-time series forecasting.

## Key findings

- The method achieves a 24% accuracy improvement over Bi-LSTM.
- It outperforms Transformer by 4.6% in long-time strain prediction.
- The approach offers a new technical framework for long-time series forecasting.

## Abstract

This article presents a novel incremental forecast method to address the challenges in long-time strain status prediction for a wind turbine blade (WTB) under wind loading. Taking strain as the key indicator of structural health, a mathematical model is established to characterize the long-time series forecast forecasting process. Based on the Bi-directional Long Short-Term Memory (Bi-LSTM) framework, the proposed method incorporates incremental learning via an error-supervised feedback mechanism, enabling the dynamic self-updating of the model parameters. The experience replay and elastic weight consolidation are integrated to further enhance the prediction accuracy. Ultimately, the experimental results demonstrate that the proposed incremental forecast method achieves a 24% and 4.6% improvement in accuracy over the Bi-LSTM and Transformer, respectively. This research not only provides an effective solution for long-time prediction of WTB health but also offers a novel technical framework and theoretical foundation for long-time series forecasting.