Research on cable condition evaluation method based on multi-sensor information fusion

Abstract

High-voltage cables undertake core power transmission tasks in urban power grids, and multi-dimensional distributed sensors collect electrical, temperature, vibration, and environmental monitoring data synchronously. The raw monitoring data suffer from inconsistent sampling timestamps, magnitude difference, abnormal noise, and long-term missing values, which makes traditional single-sensor evaluation unable to capture faint early insulation defects and leads to low evaluation accuracy. Targeting this fundamental problem, this paper constructs an independent multi-sensor hierarchical information fusion framework only for static real-time cable state quantification. A complete heterogeneous data preprocessing pipeline is proposed, including time sequence alignment, hybrid outlier detection, segmented missing value filling, and Z-score normalization. A Transformer encoder is adopted to extract cross-modal deep coupling features, and a three-layer fusion mathematical model covering data layer, feature layer, and decision layer is established. Combined with deep belief network and fuzzy comprehensive evaluation, the standardized cable health index is obtained, and a 1D-CNN-LSTM (one-dimensional convolutional neural network-long short-term memory network) few-shot learning branch is developed to identify early-stage hidden cable faults. Offline simulation based on actual power grid single-point sensor samples verifies that the abnormal data recognition accuracy reaches 87.2%, the missing value filling error is controlled within 6.4%, and the single cable evaluation delay is less than 326 ms, thereby meeting real-time monitoring requirements. This work focuses on static instantaneous state perception for high-voltage cables and provides quantitative health assessment results for cable operation management.