CN-121978531-A - SOC detection method and system for battery pack

CN121978531ACN 121978531 ACN121978531 ACN 121978531ACN-121978531-A

Abstract

The invention discloses a method and a system for detecting the SOC of a battery pack, which relate to the technical field of battery pack health management, and the method comprises the steps of determining a main battery pack and at least one auxiliary battery pack; the method comprises the steps of executing parallel connection on a secondary battery pack by a main battery pack, sending a synchronous detection instruction to acquire an electric signal monitoring result of each battery pack in real time, extracting coupling duration and spatial gradient by establishing an association relation between the main battery pack and the secondary battery pack, inputting the coupling duration and the spatial gradient into an SOC state identification model, outputting an estimated SOC label of each battery pack, introducing a particle swarm algorithm into the SOC state identification model based on a regression tree combination framework, adjusting characteristic weights and standard abnormal thresholds according to a control mode and an effective coupling period, executing sliding window weighted average on the estimated SOC label, generating an SOC state score, introducing a machine learning model, and predicting the next fault event.

Inventors

Shui Shengjing
LIU HUANYANG
LIU CHANHUI
GUO SHUANGFENG
ZHANG JIE
KANG CHEN

Assignees

西安宇驰特能防务装备研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. The method is characterized by comprising the steps of determining a main battery pack and at least one auxiliary battery pack, wherein the main battery pack performs parallel connection on the auxiliary battery packs and sends a synchronous detection instruction to acquire the electric signal monitoring result of each battery pack in real time; Based on the electric signal monitoring result, establishing an association relation between the main battery pack and the auxiliary battery pack, extracting coupling duration and spatial gradient based on the association relation, inputting the extracted coupling duration and spatial gradient into an SOC state identification model, and outputting estimated SOC labels of the battery packs, wherein the SOC state identification model is based on a regression tree combination architecture, introducing a particle swarm algorithm, and adjusting characteristic weights and standard abnormal thresholds according to a control mode and an effective coupling period; and performing sliding window weighted average on the estimated SOC label, generating SOC state scores, introducing a machine learning model, and predicting the next fault event.
2. The SOC detecting method of claim 1, wherein the electric signal monitoring result includes a shunt current and a terminal voltage of each battery pack.
3. The SOC detection method of claim 2, wherein establishing an association of the primary battery pack and the secondary battery pack comprises: the method comprises the steps of mapping a main battery pack into a root topological node, mapping each auxiliary battery pack into a slave topological node, identifying cell branches in the slave topological node, establishing association relations between the root topological node and each cell branch, and configuring equivalent resistance and polarization inductance on the topological side of each cell branch; Meanwhile, electric signal monitoring results of all nodes under the same time stamp are obtained through synchronous sampling, a first product of equivalent resistance and shunt current and a second product of polarization inductance and current change rate are calculated, and a two-dimensional correlation space with the ratio of the first product to terminal voltage as a horizontal axis and the ratio of the second product to terminal voltage as a vertical axis is established.
4. The SOC detection method of claim 3, wherein extracting the coupling duration and the spatial gradient comprises: The method comprises the steps of monitoring the communication state of a topological edge, identifying the contact detection result of a parallel relay corresponding to a secondary battery pack and a high-voltage bus, and determining the access start time of the secondary battery pack; And drawing a state evolution track according to the two-dimensional correlation space, and identifying the ascending gradient and the descending gradient of the state evolution track in the coupling time so as to determine the spatial gradient.
5. The SOC detection method of the battery pack of claim 4, wherein determining the coupling duration further comprises: obtaining all shunt currents of a main battery pack and a secondary battery pack in the coupling time length to form corresponding current sequences, wherein the corresponding current sequences comprise a first current sequence and a second current sequence, the first current sequence corresponds to the main battery pack, and the second current sequence corresponds to the secondary battery pack; And identifying the cross correlation coefficient of the first current sequence and the second current sequence, screening and extracting the coupling duration of which the cross correlation coefficient is larger than the standard coupling threshold value, and obtaining the effective coupling period.
6. The SOC detection method of claim 5, wherein the SOC identification model comprises: Extracting the association of each topological node and a cell branch, wherein each branch at least carries coupling duration and spatial gradient; The method comprises the steps of setting a state function set of a topological node, obtaining a corresponding feature solution matrix through analyzing the state function set, generating an abnormal score through weighting mapping of the feature solution matrix and feature weights of node splitting under a current access mode, meanwhile, comparing the abnormal score with a standard abnormal threshold, selecting a spatial gradient as a node splitting feature if the abnormal score is larger than the standard abnormal threshold, and selecting a coupling time length as the node splitting feature if the abnormal score is smaller than or equal to the standard abnormal threshold, wherein the standard abnormal threshold is a dynamic value.
7. The SOC detecting method of claim 6, wherein the access mode includes a direct connection access and a plug access.
8. The SOC detection method of claim 7, wherein introducing a particle swarm algorithm adjusts feature weights and standard anomaly thresholds based on an access mode and an effective coupling period, comprising: constructing a high-dimensional target search space; The method comprises the steps of taking a characteristic weight and a standard abnormal threshold value as parameters to be optimized, dynamically searching an optimal parameter vector in a target search space, providing evolution constraint conditions for population evolution according to a current access mode and an effective coupling period, adjusting a position vector by taking the access mode as a physical constraint term of a fitness function, taking the effective coupling period duty ratio as a search constraint term, adjusting a speed vector until the maximum iteration number is reached, and respectively mapping the optimal parameter vector into the gain of the characteristic weight of regression tree node splitting and the offset of the standard abnormal threshold value by taking a minimized residual error as a target.
9. The SOC detection method of the battery pack of claim 8, wherein the machine learning model embeds the first branch and the second branch, comprising: after outputting the estimated SOC label, taking the reciprocal of the residual error as a weight coefficient to determine the SOC state score; Extracting a mean value, a standard deviation and a time stamp from the SOC state score through a first branch to construct a first feature vector, simultaneously, calling an electric signal monitoring result under the same time stamp, and counting time sequence features with extremely poor voltage to generate a second feature vector; The fusion feature is input to the second branch, predicting the next failure event.
10. The system is characterized by comprising a target sensing module, a target detection module and a control module, wherein the target sensing module is used for determining a main battery pack and at least one auxiliary battery pack; The detection analysis module establishes an association relation between the main battery pack and the auxiliary battery pack based on the electric signal monitoring result, extracts coupling duration and spatial gradient based on the association relation, inputs the extracted coupling duration and spatial gradient into the SOC state identification model, and outputs estimated SOC labels of the battery packs, wherein the SOC state identification model is based on a regression tree combination architecture, a particle swarm algorithm is introduced, and characteristic weights and standard abnormal thresholds are adjusted according to a control mode and an effective coupling period; And the fault output module is used for executing sliding window weighted average on the estimated SOC label, generating SOC state scores, introducing a machine learning model and predicting the next fault event.

Description

SOC detection method and system for battery pack Technical Field The invention relates to the technical field of battery pack health management, in particular to a battery pack SOC detection method and system. Background With the rapid development of energy storage systems, energy management technology of lithium battery packs has become the core of industry research, and particularly, a parallel system formed by a plurality of battery packs consisting of a main battery pack and a plurality of auxiliary battery packs has been widely used for high-voltage and high-capacity energy demands; In the prior art, the conventional parallel battery system has the defects that on one hand, the conventional detection method generally relies on measuring the total current of the parallel system and supposing that the current is uniformly distributed among all branches, but in the actual working condition, even if the parallel main and auxiliary battery packs show the same terminal voltage, the impedance deviation is caused by the difference of the equivalent internal resistances of the main and auxiliary packs, so that the branch current is often unevenly distributed, the calculation of the SOC of each branch produces serious asynchronous drift, serious electric quantity imbalance can be caused under long-term running, on the other hand, in the transient process of parallel connection, if obvious SOC gradient exists between the main and auxiliary packs, for example, the SOC of the main battery pack is 80%, the auxiliary battery pack is 20%, huge circulation current can be generated at the moment of parallel connection, a large amount of sampling noise is easy to generate, the prior SOC identification algorithm often lacks effective robust design when processing such high-noise saturated data, the initial deviation which is difficult to eliminate is introduced at the initial stage of parallel connection, the deep optimization of the battery performance and health management of full life cycle are unfavorable, and in addition, the prior parallel scheme adopts a hard connection mode, the SOC is difficult to dynamically identify the transient state of charge and has great accuracy after the transient state of charge is greatly reduced. Disclosure of Invention (One) solving the technical problems Aiming at the defects of the prior art, the invention provides a method and a system for detecting the SOC of a battery pack, wherein in the parallel connection process, the association relation of a main battery pack and a secondary battery pack is established by determining the electric signal monitoring results of the main battery pack and the secondary battery pack, a two-dimensional association space is established, a battery pack topology network is established, a regression tree combination-based SOC state identification model is combined, node splitting characteristics are selected according to the magnitude of abnormal scores, a particle swarm algorithm is introduced, the optimization characteristic weight and a standard abnormal threshold are searched, an estimated SOC label is output, SOC detection precision is improved, a machine learning model is introduced, and the next early warning event is predicted, so that the problems in the background technology are solved. (II) technical scheme In order to achieve the above purpose, the invention is realized by the following technical scheme: the application provides a method for detecting the SOC of a battery pack, which comprises the steps of determining a main battery pack and at least one auxiliary battery pack, wherein the main battery pack performs parallel connection on the auxiliary battery packs and sends a synchronous detection instruction so as to acquire the electric signal monitoring result of each battery pack in real time; Based on the electric signal monitoring result, establishing an association relation between the main battery pack and the auxiliary battery pack, extracting coupling duration and spatial gradient based on the association relation, inputting the extracted coupling duration and spatial gradient into an SOC state identification model, and outputting estimated SOC labels of the battery packs, wherein the SOC state identification model is based on a regression tree combination architecture, introducing a particle swarm algorithm, and adjusting characteristic weights and standard abnormal thresholds according to a control mode and an effective coupling period; and performing sliding window weighted average on the estimated SOC label, generating SOC state scores, introducing a machine learning model, and predicting the next fault event. Further, the electric signal monitoring result includes a shunt current and a terminal voltage of each battery pack. Further, establishing an association relationship between the main battery pack and the auxiliary battery pack includes: the method comprises the steps of mapping a main battery pack into a root topolog