Frequency domain dielectric spectroscopy (FDS) is widely used for condition diagnosis of oil-paper insulated power equipment due to its high measurement accuracy and ease of operation. However, in winter in Northeast and Northwest China, the temperatures remain below -40℃ for extended periods, and rapid internal cooling of equipment during maintenance can lead to water crystallization and partial solidification of transformer oil, severely affecting accuracy of FDS results. In order to improve the accuracy of oil-paper insulation condition assessment, it is necessary to perform temperature normalization of the test results. However, traditional "master curve" methods are unsuitable for low-temperature environments as they produce significant errors in high and low frequency ranges. Existing research rarely focuses on the dielectric response characteristics and assessment methods of oil-paper insulation in extremely low-temperature environments. Therefore, this paper studies the FDS results of oil paper insulation at different temperature, and establishes a new temperature normalization model by Havriliak-Negami (H-N) model. This model improves the temperature normalization accuracy, filling the gap in the assessment of oil-paper insulation condition in low-temperature environments.

Firstly, starting from the physics of dielectrics, derive the effects of temperature on the relaxation processes. to obtain the formula for temperature normalization parameters. Samples of oil paper insulation with different moisture contents (0.41%~3.91%) are prepared, and an experimental platform for high and low temperature dielectric response testing is set up. By measuring the frequency dielectric spectra (1 mHz~5 kHz) of samples with different moisture contents at various temperatures (-40~30℃), it is found that as the moisture content gradually increases, the dielectric loss values also increase. Additionally, the decrease in temperature tends to make the high-frequency FDS results more consistent the relaxation peaks less distinct, which increases the difficulty of assessing moisture content. In order to understand the changes in the internal water morphology of oil-paper insulation at low temperatures, the distribution of moisture within insulation paper is studied using isothermal adsorption experiments, revealing a substantial amount of free water attached to cellulose fibers. At low temperature, this part of water crystallizes and precipitates, which affects internal polarization processes of oil paper insulation. Using thermally stimulated depolarization current (TSDC), it is discovered that concentration polarization, interfacial polarization, and dipole polarization are the main three polarization processes in oil-paper insulation. Based on the extended derivative method, it is found that as the temperature decreases, the intensity of concentration polarization gradually weakens, and the relaxation time of interfacial polarization decreases.

In order to study different polarization processes separately, the improved Havriliak-Negami (H-N) model is used to decompose FDS results, extracting characteristic parameters of each relaxation process. It is discovered that as temperature decreases, concentration polarization diminishes and disappears below 0℃, temperature only changes the relaxation time of interfacial polarization without altering its strength, and dipole polarization intensifies due to reduced molecular thermal motion. Moreover, the conductivity process, influenced by both ionic and electrophoretic conductivity, gradually decreases and stabilizes. At the same time, temperature normalization parameters for each relaxation process are extracted.

Finally, a new temperature normalization method is proposed based on the characteristics of each polarization process. Compared to the traditional “master curve” method, this method has higher accuracy in low temperature environments and in conditions with high moisture content. In low temperature, this method maintains high accuracy with a goodness of fit of 0.975 7, compared to 0.952 6 with the traditional method, In samples with high moisture content, the goodness of fit is 0.982 2. At the same time, 5 to 7 more frequency data points are added and full-frequency range correction is achieved, solving the issues of large low-temperature correction errors and insufficient frequency data in the “master curve” method.