[Paper Review] Privacy Preserving Ultra-Short-term Wind Power Prediction Based on Secure Multi Party Computation

Hang Fan, Xiaoyu Fan+7 more|arXiv (Cornell University)|Jan 31, 2023

Remote Sensing and LiDAR Applications4 citations

TL;DR

This paper proposes pwXGBoost, a privacy-preserving ultra-short-term wind power prediction framework based on secure multi-party computation (SMPC) that enables multiple wind farms to collaboratively train a XGBoost model without exposing raw data. By leveraging secret sharing and a maximum mean discrepancy (MMD)-based selection criterion, the method achieves high prediction accuracy while preserving data privacy, validated on real Inner Mongolian wind farm data with significant performance gains.

ABSTRACT

Mining the spatial and temporal correlation of wind farm output data is beneficial for enhancing the precision of ultra-short-term wind power prediction. However, if the wind farms are owned by separate entities, they may be reluctant to share their data directly due to privacy concerns as well as business management regulation policies. Although cryptographic approaches have been designed to protect privacy in the process of data sharing, it is still a challenging problem to encrypt the original data while extracting the nonlinear relationship among multiple wind farms in the machine learning process. This paper presents pwXGBoost, a technique based on the machine learning tree model and secure multi-party computation (SMPC) that can successfully extract complicated relationships while preserving data privacy. A maximum mean discrepancy (MMD) based scheme is proposed to effectively choose adjacent candidate wind farms to participate in the collaborative model training, therefore improving the accuracy and reducing the burden of data acquisition. The proposed method was evaluated on real world data collected from a cluster of wind farms in Inner Mongolia, China, demonstrating that it is capable of achieving considerable efficiency and performance improvements while preserving privacy

Motivation & Objective

• To address the challenge of sharing sensitive wind power and NWP data among privately owned wind farms due to privacy and regulatory concerns.
• To enable collaborative training of ultra-short-term wind power prediction models without exposing raw data from individual wind farms.
• To develop a secure, efficient, and accurate method that preserves data privacy while extracting complex spatial-temporal correlations across multiple wind farms.
• To reduce the data acquisition burden by intelligently selecting only the most relevant neighboring wind farms for collaboration using an MMD-based criterion.

Proposed method

• The method employs a vertical federated learning approach using secure multi-party computation (SMPC) with a 2-out-of-3 secret sharing protocol to protect data during model training.
• Each wind farm encrypts its data into secret shares and distributes them among multiple computation servers, ensuring no single party can reconstruct the original data.
• A maximum mean discrepancy (MMD)-based criterion is designed to select the most relevant neighboring wind farms for collaboration, minimizing redundant data collection.
• The framework integrates XGBoost with SMPC, enabling secure evaluation of tree-based model operations such as addition, subtraction, multiplication, comparison, and division using homomorphic secret sharing.
• Secure division is implemented via the Newton-Raphson method applied to secret-shared values, enabling non-linear operations essential for gradient boosting.
• The final model output is reconstructed only after computation, ensuring that no intermediate or final results leak private information.

Experimental results

Research questions

• RQ1 How can wind farms collaboratively improve ultra-short-term wind power prediction accuracy without sharing raw historical power or NWP data?
• RQ2 What criteria can be used to select the most relevant neighboring wind farms for secure collaborative training, minimizing data acquisition overhead?
• RQ3 Can secure multi-party computation be effectively applied to train complex tree-based models like XGBoost while preserving data privacy in wind power forecasting?
• RQ4 To what extent does the MMD-based selection criterion reduce the number of required participants without degrading prediction accuracy?
• RQ5 How does the proposed pwXGBoost model compare in performance and privacy guarantees to traditional centralized or differential privacy-based approaches?

Key findings

• The proposed pwXGBoost model achieves significant improvements in ultra-short-term wind power prediction accuracy compared to non-collaborative models, with performance gains validated on real-world Inner Mongolian wind farm data.
• The MMD-based participant selection criterion effectively reduces the number of required collaborating wind farms by identifying the most relevant spatial neighbors, lowering data acquisition costs.
• The secure computation framework successfully preserves data privacy throughout the training process, ensuring that no single party can reconstruct the original data from secret shares.
• The integration of secure division via the Newton-Raphson method enables accurate gradient computation in XGBoost under SMPC, overcoming a major challenge in applying tree models to privacy-preserving learning.
• The method demonstrates robustness and scalability in a real-world deployment scenario, showing that privacy-preserving collaborative learning is feasible and effective for wind power forecasting.