A power battery SOC prediction method and device based on improved i-elm

Abstract

The invention discloses a power battery SOC prediction method and device based on the improved I-ELM, wherein the method includes: collecting training samples, normalizing the charging and discharging data and SOC data of the power battery; The samples are input into the improved I-ELM network for training, and the network model parameters are determined; the improved I-ELM network is a network that adds bias to the output matrix of the hidden layer on the basis of the I-ELM network; the improved I-ELM network is established based on the network model parameters. The I‑ELM network model of the power battery is based on the improved I‑ELM network model, and the power battery charge and discharge data collected on site are normalized and input into the improved I‑ELM network model to determine the state of charge of the power battery. This method has the advantages of fast learning speed and better prediction accuracy. It can accurately predict the SOC of the battery, reduce the loss of the battery due to repeated charging and discharging, and prolong the service life of the battery.

Description

technical field [0001] The present invention relates to the technical field of power battery state prediction, in particular to a power battery SOC prediction method and device based on an improved I-ELM. Background technique [0002] In recent years, lithium iron phosphate batteries, which have the advantages of long life, high specific energy, no environmental pollution, and good safety, have been widely used as on-board energy sources for new energy vehicles. Accurate SOC prediction is a key technical prerequisite for efficient management of lithium iron phosphate batteries. However, due to the complex internal reaction of lithium iron phosphate battery and its complex nonlinear and time-varying characteristics during operation, it is difficult to accurately describe the characteristics of the battery state only by the equivalent circuit, which seriously affects the battery life. The prediction accuracy of the state of charge (SOC) affects the utilization efficiency and ...

AI Technical Summary

Problems solved by technology

However, since the input weights and the bias of hidden layer neurons are randomly selected, there may be some hidden layer neurons whose output weights are too small, making their contribution to the network output very small, and there is a problem of invalid neurons. It not only makes the network more complex, but also reduces the stability of the network

[0006] Nowadays, due to the complex internal reaction of lithium iron phosphate battery and its complex nonlinear and time-varying characteristics during operation, the ampere-hour measurement method, open circuit voltage method, internal resistance method, and Kalman filter method etc. Modeling the SOC prediction of lithium iron phosphate power battery, the SOC prediction accuracy of lithium iron phosphate battery is not high, it is difficult to meet the actual requirements

Traditional neural networks such as error backpropagation (Back Propagation, BP) neural networks have complex mechanisms, different structure choices, large amount of calculation, very slow learning speed, overfitting, and poor generalization ability when dealing with small sample data , When the amount of data is too large, it is easy to fall into local minimum and other shortcomings, and the support vector machine is difficult to implement for large-scale training samples, so it is difficult to be popularized in the actual SOC prediction modeling of lithium iron phosphate power batteries

When I-ELM predicts and models the SOC of lithium iron phosphate power battery, because its input weights and the bias of hidden layer neurons are randomly selected, there may be some output weights of hidden layer neurons that are too small. It makes little contribution to the network output, and there is a problem of invalid neurons, which not only makes the network more complex, but also reduces the stability of the network

Images