Distribution network risk prediction based on data mining and improved PSO fused with SVM

Area of Science:

• Electrical Engineering
• Data Science
• Artificial Intelligence

Background:

• Ubiquitous Power Internet of Things (Upiot) environments generate multi-source heterogeneous data.
• Challenges include feature redundancy and poor risk prediction accuracy with small sample sizes.

Purpose of the Study:

• To propose a robust distribution network risk prediction method.
• To address challenges of data complexity and small sample sizes in Upiot.

Main Methods:

• Utilized Kernel Principal Component Analysis (KPCA) for feature reduction.
• Employed an improved Particle Swarm Optimization (PSO) with logistic mapping for parameter optimization.
• Developed a hybrid Support Vector Machine (SVM) model (DM-IS).

Main Results:

• Achieved high-precision fitting with a median error of 0.022 under small-sample constraints.
• Outperformed baseline models by narrowing error distribution bandwidth.
• Reduced average fault detection time by 77.90% and O&M costs by 42.65%.

Conclusions:

• The DM-IS model effectively overcomes premature convergence in parameter optimization.
• Demonstrated the efficacy of multi-source heterogeneous data fusion for forecast robustness.
• Provides quantitative support for transitioning power systems to predictive maintenance.