Distribution transformer heavy overload early warning method based on load data imagination convolutional neural network

Abstract

A distribution transformer heavy overload early warning method based on the load data imagination convolutional neural network comprises the following steps: 1) selecting current data extracted by anobject needing to be predicted, converting the current data into a load rate according to a load rate calculation formula, and then performing data discretization processing on the load rate data; 2)imaging continuous data; 3) establishing a convolutional neural network, inputting the graphical data, and adjusting and optimizing model parameters; 4) establishing a correction algorithm model by taking the short-term average load rate and the weather forecast meteorological data as conditional probabilities of the naive Bayes algorithm; and 5) adding a correction rule, extracting short-term average load rate characteristics of each object, and correcting by using a naive Bayes algorithm if the short-term average load rate characteristics are compared with the long-term load rate characteristics by more than 30%. The method solves the problem that heavy overload of the distribution transformer is difficult to predict and only depends on manual experience, and has the advantages of beingnovel in idea, high in economic value and suitable for popularization and application.

Description

technical field [0001] The invention belongs to the field of power system analysis and calculation, in particular to the technical field of power system distribution network transformer load forecasting. Background technique [0002] At present, the heavy overload prediction of distribution network transformers is mainly completed by manual experience. According to the short-term historical characteristics of transformers, the empirical prediction is made. The only adjustment method is to make small adjustments based on weather forecast data. However, due to the large number of distribution network transformers, manual experience is difficult. The prediction of each transformer is usually carried out in the unit of area or line, and it is impossible to synthesize relatively long-term historical information. Very different, such vague predictions are not ideal. In recent years, many companies and institutions have tried to use big data machine learning algorithms to predict ...

AI Technical Summary

Problems solved by technology

[0002] At present, the heavy overload prediction of distribution network transformers is mainly completed by manual experience. According to the short-term historical characteristics of transformers, the empirical prediction is made. The only adjustment method is to make small adjustments based on weather forecast data. However, due to the large number of distribution network transformers, manual experience is difficult. The prediction of each transformer is usually carried out in the unit of area or line, and it is impossible to synthesize relatively long-term historical information. are very different, such vague predictions are not ideal

In recent years, many companies and institutions have tried to use big data machine learning algorithms to predict the overload of distribution transformers. This prediction is a typical regression prediction based on continuous current data, but due to the stricter standards for distribution transformer overload For example, the high load rate of an oil-immersed transformer lasts for more than 2 hours before it is considered a heavy overload, and the current data is one piece every 15 minutes, which means that 8 pieces of data need to be predicted continuously and accurately to form a heavy overload prediction. Daily forecasting requires accurate forecasting of 96 pieces of data, which is a great challenge for any regression forecasting algorithm, and it is even impossible to achieve, and the key point of heavy overload forecasting is to predict the peak load rate Even if the regression algorithm can find the change trend of some transformer load rates, it is still meaningless if the trend fits better and cannot find a high peak.

Images