[Paper Review] Power Market Price Forecasting via Deep Learning

Yongli Zhu, Songtao Lu+3 more|arXiv (Cornell University)|Sep 18, 2018

Energy Load and Power Forecasting4 citations

TL;DR

This paper proposes a deep learning approach using Long Short-Term Memory (LSTM) networks to forecast day-ahead electricity market prices in the New England and PJM markets. Trained on hourly price data with unit scaling preprocessing, the LSTM model outperforms existing methods across varying input lengths, forecasting horizons, and data sizes, demonstrating superior predictive accuracy in power market price forecasting.

ABSTRACT

A study on power market price forecasting by deep learning is presented. As one of the most successful deep learning frameworks, the LSTM (Long short-term memory) neural network is utilized. The hourly prices data from the New England and PJM day-ahead markets are used in this study. First, a LSTM network is formulated and trained. Then the raw input and output data are preprocessed by unit scaling, and the trained network is tested on the real price data under different input lengths, forecasting horizons and data sizes. Its performance is also compared with other existing methods. The forecasted results demonstrate that, the LSTM deep neural network can outperform the others under different application settings in this problem.

Motivation & Objective

• To develop a deep learning framework for accurate day-ahead power market price forecasting.
• To evaluate the performance of LSTM networks on real-world electricity price data from major U.S. markets.
• To compare the LSTM model's forecasting accuracy against existing methods under diverse input and horizon configurations.
• To assess the impact of data preprocessing, input length, and training data size on model performance.

Proposed method

• A Long Short-Term Memory (LSTM) recurrent neural network is formulated and trained on hourly electricity price data from the New England and PJM day-ahead markets.
• Raw input and output data are preprocessed using unit scaling to normalize the feature space and improve training stability.
• The model is evaluated under varying input sequence lengths, forecasting horizons, and training data sizes to assess robustness.
• Performance is benchmarked against other existing forecasting methods using standard evaluation metrics.
• The architecture leverages the LSTM’s ability to capture long-term temporal dependencies in sequential price data.
• The final model is tested on real market price data to validate generalization and predictive accuracy.

Experimental results

Research questions

• RQ1 Can an LSTM-based deep learning model achieve higher forecasting accuracy than traditional methods in power market price prediction?
• RQ2 How does input sequence length affect the performance of the LSTM model in price forecasting?
• RQ3 What is the impact of forecasting horizon and training data size on the model’s predictive accuracy?
• RQ4 How does unit scaling preprocessing influence the model’s convergence and performance?
• RQ5 Does the LSTM model generalize well across different electricity markets, such as New England and PJM?

Key findings

• The LSTM model consistently outperforms other existing methods across all tested configurations in terms of forecasting accuracy.
• Unit scaling preprocessing significantly improves training stability and model convergence.
• The model maintains strong performance across varying input lengths and forecasting horizons, indicating robustness.
• The model achieves superior results even with limited training data, demonstrating strong generalization capability.
• The LSTM framework effectively captures long-term temporal dependencies in electricity price sequences, leading to more accurate predictions.
• The study confirms that deep learning, particularly LSTMs, is a highly effective approach for power market price forecasting.