A SOC estimation method for a power battery

Abstract

The invention discloses a method for estimating the SOC of a power battery. Sort by time; build a four-layer network architecture of input layer, LSTM layer, DaNN layer, and fully connected layer; use the source field to train the model and save the model parameters; obtain the discharge data under actual use as the target field and use it for the target field Perform data preprocessing on the discharge data; read the model parameters, input the input features of the source domain, the input features of the target domain and the SOC of the source domain to retrain the model to obtain the SOC of the target domain. The present invention uses labeled data to train unlabeled data, and can complete the adjustment of model parameters even when there is no accurate training label under actual working conditions.

Description

technical field [0001] The invention relates to a method for estimating the state of charge of a battery, in particular to a method for estimating the SOC of a power lithium-ion battery, and more specifically, it can realize the self-adaptive adjustment of the estimation model. Background technique [0002] In recent years, with the improvement of GDP and people's living standards, my country's automobile industry has developed rapidly. There is no doubt that the automobile industry has become the mainstay of national economic development. However, while the popularity of automobiles facilitates people's lives, their exhaust emissions seriously threaten human health, causing air pollution, global warming and other issues. Compared with traditional fuel vehicles, electric vehicles use power batteries as the power source, and basically do not emit harmful gases in the process of driving the vehicle, which solves the two main problems of energy consumption and exhaust emissions...

AI Technical Summary

Problems solved by technology

[0009] The technical problem to be solved by the present invention is to overcome the defect that the Kalman filter method has a large amount of computation and its accuracy depends on the accuracy of the equivalent circuit model in the prior art, and its accuracy decreases when the state of the battery changes; the neural network method needs to have accurate The SOC value of the model is used as a model training label, and the accuracy depends on the training data sample and the defect that the model parameters need to be adjusted in time when the use conditions change. A power battery SOC prediction method based on LSTM (long short-term memory)-DaNN is provided. The domain adaptive network (DaNN) in transfer learning introduces the SOC estimation method to form the LSTM-DaNN algorithm, and uses the labeled data to train the unlabeled data, so as to solve the difficulty of the traditional machine learning algorithm in the SOC estimation method due to the lack of training labels in engineering applications. landing problem

Among them, LSTM is a cyclic neural network, which considers the relationship between battery data time and time, and solves the timing problem of lithium battery SOC estimation; DaNN is a domain adaptive network, when the actual driving conditions of the vehicle are quite different from the training data of the model At this time, it is necessary to combine the actual driving condition data of the vehicle to retrain the model and adjust its network parameters. However, at this time, the SOC in the data is directly parsed from the BMS (Battery Management System) message, and it is impossible to determine whether it is an accurate SOC. , that is, it cannot be used as a label when training the model. In order to solve the problem of unlabeled training, the present invention proposes for the first time to add the DaNN network in transfer learning to the deep learning framework for estimating SOC, which can realize the use of existing correct labels (SOC) The data, together with the unlabeled data, trains the model to achieve the purpose of retraining the model and adaptively adjusting the model parameters in combination with the actual driving condition data of the vehicle

Images