Day-ahead electricity market price forecasting and analysis: a BiLSTM-SHAP approach

In the context of high growth rates of renewable energy bringing more complexity and instability to the electricity market, many machine learning models have been proposed to predict market prices accurately. However, though the ML model has high accuracy on prediction and the capability to handle large amounts of data, the 'black box' nature results in a lack of interpretability, credibility, and the inability to analyse price-driving factors, especially for market participants. Explainable Artificial Intelligence (XAI) offers new approaches to address these issues. In this paper, an explainable model based on Shapley Additive exPlanations (SHAP) is proposed with a prediction model based on a Bidirectional Long Short-Term Memory (BiLSTM) neural network to locally analyse and predict the influence weights of driving factors crossing the time domain. By demonstrating the proposed method on the UK's day-ahead market database, the results show that the model can analyse and predict the impact of drivers like demand and renewable energy on price variations by Shapley Value.