Wenchang Wei

This study proposes a machine learning‐based model for predicting the molecular weight distribution (MWD) of polypropylene, aiming to optimize the performance of dielectric materials for film capacitors. A database was established by collecting 233 sets of experimental data from published literature, with 10 influencing factors as input variables and molecular weight distribution as the output variable. Six machine learning algorithms—Gradient Boosting Regression (GBR), AdaBoost Regression (ABR), Bagging Regression (BAGR), Support Vector Regression (SVR), Random Forest Regression (RFR), and K Neighbors Regressor (KNR)—were employed to predict and analyze polypropylene molecular weight distribution. Results demonstrate that the constructed machine learning models effectively learn from complex data and accurately predict molecular weight distribution. Among these, the Gradient Boosting Regression model exhibited the highest predictive performance, achieving a coefficient of determination ( R ² ) of 0.9380 and a mean absolute error (MAE) of only 0.1545. This research provides an efficient tool for optimizing the molecular weight distribution of polypropylene and lays a theoretical foundation for enhancing the performance of film capacitors.

Nano-modified insulating oils have great potential to enhance the dielectric and thermal performance of offshore wind transformers, but nanoparticles are affected by temperature fluctuations to change the oil properties and microstructure, which restricts their large-scale application in offshore wind power equipment. In this study, γ-aminopropyltriethoxysilane (APTES) modified graphene oxide/natural ester composite insulating oil is developed through a surface functionalization strategy. The thermal conductivity and dielectric loss characteristics are systematically evaluated within a wide temperature range, revealing temperature-nanoparticle synergistic interaction mechanisms. Additionally, a model of natural ester insulating oil doped with nanoparticles is constructed based on molecular dynamics (MD) to elucidate the performance enhancement mechanisms, and the performance enhancement mechanism is clarified by calculating the mean square displacement (MSD), free volume (FFV), and radial distribution function (RDF) of the model. Results show that when the temperature rises, the modified nanoparticles can efficiently suppress the diffusion behavior of molecules in the oil, significantly reduce the free volume fraction by filling the gaps between oil molecules, and form hydrogen bonds with water to maintain the number of hydrogen bonds. This work provides theoretical foundations and design guidelines for developing high-temperature-resistant nanocomposite insulating oils.

The frequency domain spectrum (FDS) method can effectively evaluate the aging state of cross-linked poly-ethylene (XLPE) cable insulation. To more accurately achieve a quantitative analysis of information in the FDS, this paper combines the concepts of fractional element (Cap-Resistor) and fractional calculus, inspired by the Debye model and the multi-branch improvement method of the Cole-Cole (C-C) model, and establishes an improved fractional Zener model based on the existing Cap-Resistor model, and combines the improved fractional Zener model with the Elite Genetic Algo-rithm (EGA) to propose a new method for analyzing the re-laxation polarization information of XLPE dielectric. Six sets of XLPE cable samples with different aging times (including a validation set) are prepared at an aging temperature of 150 °C, and the FDS of the thermal aging samples are meas-ured. Then the FDS are fitted using a modified fractional Ze-ner model: firstly, the least squares method is used to estab-lish the optimized objective function, and the characteristic parameters are extracted and preferentially selected using EGA for analysis, and the fitting equations between aging time and characteristic parameters are established, and the validation results manifest that the error of the characteristic parameters on the aging time characterization of XLPE cable insulation samples is 0.18465 %, which is excellent. The eval-uation method’s accuracy has been validated. This paper in-troduces a feasible approach for analyzing relaxation polari-zation information in XLPE dielectric using an improved frac-tional Zener model and EGA.

The moisture intrusion of high-voltage oil-immersed paper (OIP) bushing may cause insulation deterioration. Polarization and depolarization current (PDC) technology is an effective and nondestructive method to evaluate the moisture state of OIP bushings’ insulation. However, the current PDC assessment methods are difficult to evaluate the nonuniform moisture status of OIP bushings. To address this issue, this article studies the PDC characteristics of OIP bushings under radial nonuniform moisture distribution. Initially, by combining with the laboratory PDC measurements and finite element method (FEM) simulation, the influence laws of the dampest insulation layer’s moisture content and position on the overall PDC curve of OIP bushings are explored. The results show that the moisture content and position of the dampest insulation layer have a significant influence on the overall polarization current curve. The peak value, steady-state value, and integral value of the polarization current are negatively correlated to the position and positively correlated to the moisture content of the dampest insulation layer. The moistest layer’s position of the radial nonuniformly damped bushing can be determined by polarization current integral difference under positive and negative PDC link-wire ways. This research provides the theoretical basis and technical support for the nonuniform moisture state evaluation of OIP bushing insulation.

Currently, the method of detecting methanol in transformer oil by surface-enhanced Raman spectroscopy (SERS) possesses the disadvantages of high cost, long time, and difficulty in field testing. To tackle this problem, a novel method was proposed for the combination of extraction and silver nanosubstrate to detect dissolved methanol in transformer oil. Specifically, it was determined first by Gauss view and Gaussian software that pure water and silver nanoparticles (Ag NPs) were applicable for the detection of methanol content by the Raman spectroscopy. Then, multilayer Ag NPs were synthesized on the surface of the copper sheet by a displacement reaction. The electric field distribution on the surface of Ag NPs was analyzed by COMSOL software, which further revealed that the Ag NPs could improve the Raman signal of methanol molecules. Finally, the pure water and Ag NPs were used for Raman measurement of methanol transformer oil solution with different concentrations based on the experimental platform of Raman spectroscopy. The results pinpointed that the minimum detection content of methanol in transformer oil was 1.8 mg/L. This method is simple, fast, low-cost, and applicable to the detection of trace methanol.

Under the combined effect of the electric and temperature fields, bubbles will be formed on the transformer oil-paper insulation system, which will seriously endanger the insulation strength. However, the current study usually divided the formation and motion of bubbles into independent parts, lacking research on the whole process of bubble evolution. This paper experimentally investigated and theoretically analyzed the whole process of the formation, deformation, and motion of bubbles in the oil-paper insulation system, and explored the influence of the extremely inhomogeneous electric field (EIEF) on bubbles evolution. The results show that with the enhancement of the electric field, the bubble inception temperature (BIT) and bubble size increase significantly, while the number of bubbles decreases. Under the stretching and pushing effect of the electric field force, the bubbles will deform and move toward the low electric field region. In addition, this paper studies the influence of bubbles on the transformer electric field distribution through multi-physical field simulation.

Debye and its modified models are widely used to explain the relaxation polarization process in dielectrics at present. However, such models require prior knowledge of the number of polarization branches in them, so it is difficult to build an equivalent circuit model that matches the actual characteristics of relaxation polarization. Given this issue, a new method is proposed to analyze the relaxation polarization process in dielectrics by the distribution of relaxation time (DRT). In this paper, cross-linked polyethylene (XLPE) cables are selected as the research object and firstly aged at 160 °C for 24, 72, 144, 240, and 360 h, and the corresponding frequency domain spectrum (FDS) curves are obtained by experiment. Then, the DRT curves are solved by mathematical transformation, and the characteristic parameters are extracted. Finally, the quantitative relationship between aging duration and characteristic parameters was established by fitting a Logistic model. The results show that DRT can adequately reflect different types of polarization information, and its characteristic parameters can be applied to study the aging state of XLPE cables. The innovation of this paper is to propose a possible method based on DRT to study relaxation polarization.