Just-in-time learning with trend similarity for offshore wind power forecasting

Meteorological factors, such as wind speed, wind direction, and atmospheric pressure, affect wind power. Forecasting these factors is more difficult under offshore climate conditions because offshore climate is more volatile than onshore climate. Traditional causal model is not suitable for offshore wind power forecasting. In addition, most existing forecasting methods employ offline models that remain fixed after training, making it difficult to update them efficiently as new data accumulate. These issues pose a challenge for offshore wind power forecasting. To overcome this problem, an autoregressive exogenous (ARX) model using a just-in-time learning (JITL) algorithm based on trend nearest neighbors is proposed in this study. The JITL algorithm selects similar modelling samples for different predictions. Based on this, the proposed method considered the wind power trend that employed a new similarity index for selecting similar samples in the format of a time series. The case study performed using the proposed algorithm demonstrated that the stable selection of similar samples is obtained using the similar trends. Furthermore, the proposed JITL method based on trend nearest neighbors was found to outperform three other time series methods, in terms of average mean absolute percentage error (MAPE), R-square and computational time.