Lithium battery SOC estimation method based on competitive generative adversarial neural network

Abstract

The invention discloses a lithium battery SOC estimation method based on a competitive generative adversarial neural network. The network is composed of a plurality of generators and a discriminator. Collecting various feature data of the lithium battery, generating a plurality of new data sets from the original data set according to different feature combinations, and respectively inputting the new data sets into each generator for training; and alternately iterating internal parameters of the generators and the discriminators by utilizing a mutual game between the generators and the discriminators, and meanwhile, eliminating the generator with the lowest SOC estimation precision at a certain training batch interval. And estimating the SOC of the lithium battery by using the generator which is reserved to the last. According to the method, a feature selection mechanism is realized by utilizing the competition between the features, a generator mutual promotion mechanism and a generator elimination mechanism are realized by utilizing the competition between the generators, and a weight updating mechanism of each generator is realized by utilizing the competition between the generators and the discriminator, so that the SOC estimation precision of the lithium battery is improved.

Description

technical field [0001] The invention relates to the technical field of lithium batteries, in particular to a method for estimating the SOC of lithium batteries based on a competitive generative confrontation neural network. Background technique [0002] Lithium batteries have the advantages of high energy density and long service life, so they are widely used in various fields, and high-precision estimation of battery state of charge (SOC) is an important basis for scientific use of lithium batteries. [0003] At present, the methods of lithium battery SOC estimation mainly include ampere-hour integration method, model method and data-driven method. The ampere-hour integration method is easy to operate and the principle is simple, but it requires high measurement accuracy of the current sensor; the model method can simulate the internal chemical changes of the battery and has a high SOC estimation accuracy, but the complex modeling process requires a large amount of calculat...

AI Technical Summary

Problems solved by technology

The ampere-hour integration method is easy to operate and simple in principle, but it requires high measurement accuracy of the current sensor; the model method can simulate the internal chemical changes of the battery and has high SOC estimation accuracy, but the complex modeling process requires a large amount of calculation and is not It is conducive to the online estimation of lithium battery SOC; the data-driven method does not need to consider the internal electrochemical reaction of lithium-ion batteries, and the estimation of lithium battery SOC can be realized directly through the data measured during battery operation and machine learning methods

[0004] The current data-driven method is extremely sensitive to the selected features, and it is difficult to select suitable features

At present, the mainstream method is to select features through algorithms such as Pearson method and gray correlation analysis method, but the feature correlation will fluctuate greatly with the change of lithium battery type.

In addition, the existing SOC estimation method based on a single model is prone to unstable estimation results, and the joint use of multiple models can make up for the problem of low robustness caused by the small number of models to a certain extent.

However, the mutual cooperation of existing multi-models is often realized in a single form by using the output of the previous model as the input of the latter model, and it has not gone deep into the mutual cooperation of each model training stage, so there is still a low SOC estimation accuracy. and the problem of unstable estimation results

Images