Abstract: Abstract:The distribution network is directly oriented to terminal users, which can guarantee the reliability of residential and industrial production power supply. The distribution transformer is an important equipment of the distribution network. It has a wide distribution range and a large quantity，so the safe and reliable operation of the distribution network is the key to guaranteeing users' normal production and living. However, judging from the accidents that have occurred in recent years, many transformer accidents do not have any symptoms before they occur. This shows that the current routine test projects and test cycles still have certain limitations, and some accident precursor information cannot be captured in time. Taking into account the limitations of traditional detection methods, it is necessary to add effective monitoring methods to distribution transformers in a timely manner. This paper proposes a transformer oil condition monitoring method based on multi-frequency ultrasound. We obtained oil samples from 88 transformers that are running in the substation. First, in the laboratory, the oil samples are tested for the breakdown voltage and dielectric loss factor , micro water and acid value these four parameters. According to the multi-indicator comprehensive analysis, the status of the selected transformer oil samples is classified, and then a new type of multi-frequency ultrasonic equipment is used. Separately fired a beam of ultrasonic waves of different frequencies into each set of oil samples, and various parameters of ultrasonic waves obtained by the ultrasonic receiver module in real time. Then utilize the neural network to process ultrasonic parameters (wave speed, amplitude and phase angle of 240-dimensional data of the three-phase ultrasonic waves at 40 frequency points) and establish the relationship between ultrasonic spectrum parameters and the transformer oil state. The mapping relationship enables real-time, accurate and comprehensive monitoring of the state of transformer insulation oil. This article will introduce the following aspects: 1.Introduction Introduces the research background, data sources, data scale and implementation method of this paper. 2.Experimental platform introduction 2.1 Principle of experiment The principle, characteristics and application fields of ultrasonic detection technology.And the internal principles and implementation ideas of the multi-frequency ultrasound instrument applied in this experiment. 2.2 Experimental equipment introduction The experimental instrument contains the various modules , the functions and relationships of each module. 3.Experimental oil sample state assessment Select four indicators of transformer oil sample:breakdown voltage, dielectric loss, moisture content and acid value. With the new transformer oil indicators and national standard values as the boundary, the status of transformer oil is classified according to these four parameters by specified method. 4.Training neural network to establish correlation model and prediction of ultrasonic parameters and transformer oil status level Introduce neural network structure, parameter setting and implementation steps. 5.Conclusion The transformer oil multi-frequency ultrasonic on-line monitoring technology based on neural network is an effective transformer oil condition monitoring method. Under the condition of real-time monitoring of the transformer oil status level, it is not affected by the electromagnetic environment at the scene and has good real-time performance and high accuracy. It is not necessary to separately prepare the sample to completely realize on-line measurement. It is a method of transformer oil condition monitoring that is worth promoting.