Improved Particle Swarm Optimization and Stacking Learning Target Power Grid Load Outliers

Researchers have developed a machine learning method designed to identify fluctuation outliers in power grid load data, combining an improved particle swarm optimization algorithm with an integrated stacking learning framework. The work, published open access in Energy Informatics on 21 September 2026, addresses what the authors describe as shortcomings in traditional power load anomaly recognition, which they characterize as inefficient, strongly subjective, and heavily dependent on human factors.

The method uses the feedback mechanism of the Elman function to modify the deep neural network training architecture. Improved particle swarm optimization is then applied to tune the weights and thresholds of the deep neural network model, replacing manual parameter selection with an automated search process.

On top of this optimized network, the researchers introduce an integrated learning framework. For load curves associated with abnormal electricity usage patterns, they construct what they describe as an abnormal load data domain, which serves as the basis for detecting fluctuating outliers in grid load data.

According to the authors’ experimental results, the design method achieved a load identification effect they judge to be good, with coverage performance remaining stable above 90 percent. They also report that the accuracy of the calculation score is higher than with the approaches they compared against, concluding that the method can meet the accuracy requirements of modern power systems for recognizing outliers.

The study is presented as early-access, peer-reviewed accepted research carrying a permanent DOI, with the publisher noting that this version remains subject to further edits before the final Version of Record replaces it.

Subject of Research: A machine learning method using improved particle swarm optimization and stacking integrated learning to identify fluctuation outliers in power grid load data.

Article Title: A method for identifying outliers of power grid load data fluctuations based on IPSO-Stacking integrated learning

Article References: Zhang, J. (2026). A method for identifying outliers of power grid load data fluctuations based on IPSO-Stacking integrated learning. Energy Informatics. https://doi.org/10.1186/s42162-026-00684-z

Image Credits: AI Generated

DOI: 10.1186/s42162-026-00684-z

Keywords: power grid, load data, outlier detection, particle swarm optimization, stacking ensemble, deep neural network, Elman network, anomaly recognition, Energy Informatics, machine learning, method, identifying