CN-122020045-A - Service life prediction method and system for power transformer

CN122020045ACN 122020045 ACN122020045 ACN 122020045ACN-122020045-A

Abstract

The invention discloses a life prediction method and a life prediction system for a power transformer, which belong to the technical field of transformer health monitoring, acquire the concentration of dissolved gas in oil, load factor, operating temperature and life data of the power transformer according to the historical operating condition of the power transformer, input the data into a pre-trained prediction model, and output a life prediction result. The method can comprehensively capture various factors influencing the service life of the power transformer, the parameters reflect the operation state of the transformer from different angles, for example, the concentration of dissolved gas in oil can indicate internal faults, and the load rate and the operation temperature are related to the thermal ageing and the mechanical stress of the transformer. By combining the local feature extraction of CNN and the time sequence modeling capability of LSTM, the method can more accurately capture the dynamic change features in the operation process of the power transformer. The accurate capture of the dynamic time sequence features enables the model to be better suitable for real-time changes of the running state of the transformer, and accordingly accuracy of life prediction is improved.

Inventors

SUN QING
Li Kailang
YU YING

Assignees

西安建筑科技大学

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (7)

1.A life prediction method for a power transformer, comprising the steps of: acquiring an insulating medium state characteristic parameter, an operating load characteristic parameter and a device thermal state characteristic parameter of a power transformer to be predicted; Inputting the insulating medium state characteristic parameters, the operating load characteristic parameters and the equipment thermal state characteristic parameters into a pre-trained prediction model, and outputting a life prediction result of the power transformer; the prediction model takes a CNN convolutional neural network as a basic network, and a characteristic sequence remodelling module and an LSTM time sequence modeling module are added after a pooling layer of the original CNN convolutional neural network; The LSTM time sequence modeling module captures dynamic time sequence characteristics of transformer operation data in the time sequence characteristic sequences, generates real-time health characteristic vectors of transformers corresponding to a plurality of time sequences, and aggregates the real-time health characteristic vectors to serve as input of a full-connection layer.
2. The method for predicting the life of a power transformer according to claim 1, wherein the step of inputting the insulation medium state characteristic parameter, the operation load characteristic parameter and the equipment thermal state characteristic parameter into a pre-trained prediction model and outputting the life prediction junction of the power transformer specifically comprises the steps of: The prediction model takes a CNN convolutional neural network as a basic network, and a characteristic sequence remodelling module and an LSTM time sequence modeling module are added after a pooling layer of the original CNN convolutional neural network; The original CNN convolutional neural network comprises an input layer, a convolutional layer, a pooling layer, a full-connection layer and an output layer; the LSTM time sequence modeling module comprises a plurality of LSTM units which are arranged according to time sequence; the input layer is used for mapping the concentration of dissolved gas in oil, the load rate, the running temperature and the service life data into a feature matrix matched with the feature input format of the convolution layer, inputting the feature matrix into the convolution layer, extracting the local features of the power transformer through the sliding of a convolution kernel and the feature matrix, and carrying out nonlinear transformation on the local spatial features of the power transformer through an activation function to obtain a feature matrix after nonlinear transformation; the feature sequence remolding module is used for arranging the feature matrix subjected to dimension reduction according to time dimension through a sliding window to obtain a time sequence feature sequence; Inputting the time sequence characteristic sequences into the LSTM units respectively, and independently learning long-period time sequence association in the time sequence characteristic sequences through the synergistic effect of a gating mechanism in each LSTM unit so as to capture the dynamic time sequence characteristics of the operation data of the transformer and generate real-time health characteristic vectors of the transformers corresponding to a plurality of time sequences; The full-connection layer aggregates real-time health feature vectors of the transformers with a plurality of time sequences into one-dimensional vectors, and classifies the one-dimensional vectors by using a Softmax activation function to map the one-dimensional vectors into probability distribution of different life intervals; And the output layer screens the life interval with the largest probability value based on the confidence threshold according to probability distribution of different life intervals and outputs a final life prediction result.
3. The method for predicting the life of a power transformer according to claim 2, wherein the inputting the time sequence feature sequences into the LSTM units respectively, and autonomously learning long-short time sequence correlations in the time sequence feature sequences through the synergistic effect of a gating mechanism in each LSTM unit to capture dynamic time sequence features of the operation data of the transformer, and generating real-time health feature vectors of the transformers corresponding to the time sequences, specifically includes: at each LSTM unit, determining the input of the current moment in the time sequence characteristic sequence as ; Determining the hidden state at the t-1 time as Representing real-time health characteristics of the transformer at one moment; the output of the forgetting gate is: (7) Wherein, the A retention degree value output for the forget gate; as a matrix of weights, the weight matrix, For the corresponding offset vector to be used, In the hidden state of the last moment, Inputting each LSTM unit at the current moment; the cell status updated by the input gate is: (10) Wherein, the Is the updated cell state; the memory state at the previous moment is the long-term insulation degradation trend of the transformer; the closer to 1 the addition degree value of the current time state is, the more important the current feature is; The transformer is in an intermediate state and shows a nonlinear fault trend at the moment t of the transformer; Representing element multiplication; generating hidden state of current moment through output gate As a real-time health feature vector of the transformer at the current moment: Wherein, the An output degree value of the output gate is expressed as an output coefficient between 0 and 1; As a hyperbolic tangent function; And correspondingly acquiring real-time health feature vectors of transformers corresponding to each time sequence in the feature sequences of different time sequences.
4. The life prediction method of a power transformer according to claim 1, wherein the obtaining the insulation medium state characteristic parameter, the operation load characteristic parameter and the equipment thermal state characteristic parameter of the power transformer to be predicted specifically includes: Acquiring the state characteristic parameters of an insulating medium of the power transformer, wherein the state characteristic parameters comprise the concentration of dissolved gas in oil; Acquiring operation load characteristic parameters including load rate; Acquiring equipment thermal state characteristic parameters including an operating temperature; the method further comprises the steps of preprocessing the obtained data of the concentration of dissolved gas in the oil, the load rate and the running temperature, and taking the preprocessed data as the input of a prediction model.
5. A life prediction system for a power transformer, comprising: The data acquisition module is used for acquiring insulation medium state characteristic parameters, operation load characteristic parameters and equipment thermal state characteristic parameters of the power transformer to be predicted; The life prediction module is used for inputting the insulating medium state characteristic parameters, the operating load characteristic parameters and the equipment thermal state characteristic parameters into a pre-trained prediction model and outputting a life prediction result of the power transformer; the prediction model takes a CNN convolutional neural network as a basic network, and a characteristic sequence remodelling module and an LSTM time sequence modeling module are added after a pooling layer of the original CNN convolutional neural network; The LSTM time sequence modeling module captures dynamic time sequence characteristics of transformer operation data in the time sequence characteristic sequences, generates real-time health characteristic vectors of transformers corresponding to a plurality of time sequences, and aggregates the real-time health characteristic vectors to serve as input of a full-connection layer.
6. A computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of: acquiring an insulating medium state characteristic parameter, an operating load characteristic parameter and a device thermal state characteristic parameter of a power transformer to be predicted; Inputting the insulating medium state characteristic parameters, the operating load characteristic parameters and the equipment thermal state characteristic parameters into a pre-trained prediction model, and outputting a life prediction result of the power transformer; the prediction model takes a CNN convolutional neural network as a basic network, and a characteristic sequence remodelling module and an LSTM time sequence modeling module are added after a pooling layer of the original CNN convolutional neural network; The LSTM time sequence modeling module captures dynamic time sequence characteristics of transformer operation data in the time sequence characteristic sequences, generates real-time health characteristic vectors of transformers corresponding to a plurality of time sequences, and aggregates the real-time health characteristic vectors to serve as input of a full-connection layer.
7. A computer readable storage medium, wherein the computer readable storage medium stores a computer program which, when executed by a processor, causes the processor to perform the steps of: acquiring an insulating medium state characteristic parameter, an operating load characteristic parameter and a device thermal state characteristic parameter of a power transformer to be predicted; Inputting the insulating medium state characteristic parameters, the operating load characteristic parameters and the equipment thermal state characteristic parameters into a pre-trained prediction model, and outputting a life prediction result of the power transformer; the prediction model takes a CNN convolutional neural network as a basic network, and a characteristic sequence remodelling module and an LSTM time sequence modeling module are added after a pooling layer of the original CNN convolutional neural network; The LSTM time sequence modeling module captures dynamic time sequence characteristics of transformer operation data in the time sequence characteristic sequences, generates real-time health characteristic vectors of transformers corresponding to a plurality of time sequences, and aggregates the real-time health characteristic vectors to serve as input of a full-connection layer.

Description

Service life prediction method and system for power transformer Technical Field The invention relates to the technical field of transformer health monitoring, in particular to a life prediction method and a life prediction system of a power transformer. Background The electric energy enters into the terminal equipment of the building power system from the municipal power grid for use, the transformer is an irreplaceable middle core equipment, the transformer bears the low-voltage electric energy which is used for converting high-voltage electric energy into the internal equipment of the building directly, and meanwhile bears the key functions of voltage stability adjustment, electric energy distribution and safety isolation, and the stable operation of the transformer ensures the safety and reliability of building power supply. The life prediction is a key technical means for changing the transformer from passive maintenance to active operation and maintenance, so that the power failure risk can be avoided, the operation and maintenance cost can be reduced, the data support can be provided for the intelligent and sustainable operation of the building energy system, and the life prediction is an important component for upgrading the modern building power system. Therefore, the life prediction of the transformer is a core link for guaranteeing the reliability of building power supply, reducing the operation cost and realizing intelligent operation and maintenance, and has important significance. The traditional prediction method for the life of the transformer based on the convolutional neural network CNN can extract spatial characteristics and capture spatial correlation characteristics among multidimensional parameters, but is a static characteristic extraction model in nature, ignores time sequence characteristics of dynamic changes of the operation parameters of the transformer along with time, cannot capture long-term dependency of the operation state of equipment, and is difficult to reflect progressive and cumulative effects of the aging process of the equipment, so that the life prediction precision of the transformer is affected. Disclosure of Invention Aiming at the problems in the field, the invention provides a life prediction method and a life prediction system of a power transformer, wherein the constructed prediction model accurately overcomes the inherent defects of the existing single deep learning model in feature extraction, realizes collaborative mining of spatial features and time sequence dynamic features, and remarkably improves the accuracy and reliability of life prediction of the transformer. In order to solve the technical problems, the invention discloses a life prediction method of a power transformer, which comprises the following steps: acquiring an insulating medium state characteristic parameter, an operating load characteristic parameter and a device thermal state characteristic parameter of a power transformer to be predicted; Inputting the insulating medium state characteristic parameters, the operating load characteristic parameters and the equipment thermal state characteristic parameters into a pre-trained prediction model, and outputting a life prediction result of the power transformer; the prediction model takes a CNN convolutional neural network as a basic network, and a characteristic sequence remodelling module and an LSTM time sequence modeling module are added after a pooling layer of the original CNN convolutional neural network; The LSTM time sequence modeling module captures dynamic time sequence characteristics of transformer operation data in the time sequence characteristic sequences, generates real-time health characteristic vectors of transformers corresponding to a plurality of time sequences, and aggregates the real-time health characteristic vectors to serve as input of a full-connection layer. Preferably, the method inputs the insulation medium state characteristic parameter, the operation load characteristic parameter and the equipment thermal state characteristic parameter into a pre-trained prediction model, outputs a life prediction junction of the power transformer, and specifically comprises the following steps: The prediction model takes a CNN convolutional neural network as a basic network, and a characteristic sequence remodelling module and an LSTM time sequence modeling module are added after a pooling layer of the original CNN convolutional neural network; The original CNN convolutional neural network comprises an input layer, a convolutional layer, a pooling layer, a full-connection layer and an output layer; the LSTM time sequence modeling module comprises a plurality of LSTM units which are arranged according to time sequence; the input layer is used for mapping the concentration of dissolved gas in oil, the load rate, the running temperature and the service life data into a feature matrix matched with the feature input format of the co