KR20220077186A - Lithium ion battery pack control method and apparatus using cell balancing neural network model - Google Patents

Abstract

An embodiment of the present invention provides a lithium-ion battery (LIB) pack control method using a cell balancing neural network model. In a lithium-ion battery (LIB) pack control method using a cell balancing neural network model, a cell balancing neural network model in which a plurality of neural network models is combined is trained using input data about the state of each cell in the LIB pack transmitted from the battery controller. . Thereafter, the state of each cell in the LIB pack is predicted using the learned cell balancing neural network model. According to the prediction, a control signal for correcting the state of charge of each cell is transmitted to the battery controller to correct the state of charge of each cell.

Description

Lithium ion battery pack control method using cell balancing neural network model and control device thereof

The present invention relates to a method for controlling a lithium ion battery pack using a cell balancing neural network model and an apparatus therefor, and more particularly, to a degraded cell or a cell to be degraded using a learned cell balancing neural network model and predicts and detects and predicts a state of charge The present invention relates to a method for controlling a lithium ion battery pack using a cell balancing neural network model for controlling a lithium ion battery pack through calibration, and an apparatus therefor.

Among the core technologies of the 4th industry, the use of Lithium Ion Battery (LIB) is continuously increasing in various fields such as electric vehicles (EVs), drones, electronic devices, and ESS (Energy Storage System). Accordingly, interest in improving the stability and durability of LIB and the scale of the industry are increasing.

In general, many LIB cells are connected in series or in parallel to use high power. At this time, voltage imbalance occurs between cells during charging and discharging depending on the arrangement and structure of each cell. This causes problems such as overcharging, overdischarging, and reduced efficiency of the entire LIB pack (Lithium Ion Battery Pack), and further shortens the lifespan of the entire system, and threatens safety by causing explosion and fire of the battery.

To solve this problem, cell balancing technologies that allow a plurality of cells inside the LIB pack to be uniformly charged and discharged in each field are being actively studied in various ways.

1 is a diagram illustrating a conventional passive cell balancing method.

Passive cell balancing, which is one of the commercialized technologies, connects and discharges a resistor to a cell that is relatively more charged when the state of charge (SoC) of each cell inside the LIB pack is unbalanced, as shown in the upper part of FIG. 1 . This is a technique to uniformly match the SoC of each cell (see the lower figure of FIG. 1).

This method has the advantage of simple configuration and low cost, but has a fatal disadvantage that a normal rechargeable battery must be consumed through a resistor, and the heat generated thereby threatens the safety of the entire system.

The technical task of the present invention is to predict and detect deteriorated cells in the LIB pack using a complex neural network-based cell balancing neural network model, and perform charge state correction to achieve cell balancing Lithium using a cell balancing neural network model To provide a method for controlling an ion battery pack and an apparatus therefor.

The technical problems to be achieved by the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned can be clearly understood by those of ordinary skill in the art to which the present invention belongs from the description below. There will be.

In order to achieve the above technical problem, an embodiment of the present invention provides a lithium-ion battery (LIB) pack control method using a cell balancing neural network model. In a lithium-ion battery (LIB) pack control method using a cell balancing neural network model, a cell balancing neural network model in which a plurality of neural network models is combined is trained using input data about the state of each cell in the LIB pack transmitted from the battery controller. . Thereafter, the state of each cell in the LIB pack is predicted using the learned cell balancing neural network model. According to the prediction, a control signal for correcting the state of charge of each cell is transmitted to the battery controller to correct the state of charge of each cell.

In an embodiment of the present invention, in the step of learning the cell balancing neural network model, the input data regarding the charge/discharge state of each cell of the LIB pack transmitted from the battery controller is transmitted through the Convolutional Neural Network (CNN) neural network. learning and extracting spatial characteristics and cycle shapes between input data; learning and extracting charging/discharging characteristics according to time of each cell using a Long Short Term Memory (LSTM) neural network for the input data; and learning and extracting a correlation coefficient between the voltage statistical characteristics of each cell at the same time using the input data through a recurrent neural network (RNN) neural network.

In an embodiment of the present invention, the step of training the cell balancing neural network model includes the charging/discharging voltage data of each cell of the LIB pack transmitted from the battery controller as the input data, the CNN neural network, the LSTM neural network and parallel learning through the RNN neural network.

In an embodiment of the present invention, in order to learn the cell balancing neural network and predict the state of each cell, the input data is two-dimensional charge/discharge cycle data with an x-axis as time and a y-axis as a cell or 3 It may be provided after being converted into dimensional data.

In an embodiment of the present invention, in the step of learning and extracting the correlation coefficient between the voltage statistical characteristics of each cell at the same time through the RNN neural network,

[Correlation coefficient formula]

The RNN neural network is trained with the input data, and the voltage values of each cell in the LIB pack at the same time t are substituted into the correlation coefficient formula, and the output p _x,y value is used as the LSTM input in the RNN. , it is possible to learn and extract a correlation coefficient between the voltage statistical characteristics of each cell at the same time t.

In an embodiment of the present invention, the learned cell balancing neural network model predicts the state of charge (SoC) of each cell and the correlation coefficient in the step of predicting the state of each cell in the LIB pack in an arbitrary cycle It is possible to detect a cell to be degraded in

In an embodiment of the present invention, in the step of training the cell balancing neural network model, each weight (W _C,R,V, B _C,R,V ) learned by the CNN neural network, the LSTM neural network, and the RNN neural network )Is,

[Weight Combination Formula]

C _t = σ(W _C * X + B _C ) : Weights learned by CNN

R _t = σ(W _R * X + B _R ) : Weights learned by RNN

V _t = σ(W _V * X + B _V ) : Weights learned with VDCNN

P _t = ReLU(Ct * h _(t-1) + Vt * h _(t-1) )

Final output h(t) = (1-K) R _t * h _(t-1) + KP _t * h _(t-1) (0 < K < 1)

(* symbol is convolution operation, K is weighting factor between each neural network, σ is sigmoid function (activation function), W is weight, X is input, B is bias value)

The cell balancing neural network model can be trained by combining by the weight combination formula.

In an embodiment of the present invention, through learning of the CNN neural network, a cycle shape and a cycle pattern can be extracted based on the inflection point, the maximum value, and the minimum value of the charge/discharge curve of the charge/discharge voltage data.

In order to solve the technical problem of the present invention, another embodiment of the present invention provides an apparatus for controlling a lithium ion battery (LIB) pack using a cell balancing neural network model. The control device of the lithium ion battery (LIB) pack using the cell balancing neural network model learns the input data about the charge/discharge state of each cell in the LIB pack transmitted from the battery controller through CNN operation, and the spatial characteristics and cycle of the input data a pattern extraction neural network unit for extracting a cycle shape and a cycle pattern; a state of charge learning neural network unit for learning state-of-charge characteristics according to time of each cell through LSTM operation on the input data; a correlation coefficient learning neural network unit for learning a correlation coefficient between the voltage statistical characteristics of each cell at the same time through RNN operation on the input data; and integrating the results learned from the pattern extraction neural network unit, the state of charge learning neural network unit, and the correlation coefficient learning neural network unit to generate a cell balancing neural network model, and predicting cells to be degraded using the generated cell balancing neural network model, , a predictor configured to transmit a state of charge correction parameter to the battery controller to achieve cell balancing.

In an embodiment of the present invention, further comprising an input unit that receives the input data from the battery controller and provides it to the pattern extraction neural network unit, the state of charge learning neural network unit, and the correlation coefficient learning neural network unit, wherein the input unit provides or The input data provided by the input unit may be two-dimensional charge/discharge cycle data in which an x-axis is time and a y-axis is a cell, or three-dimensional data composed of serial and parallel cells.

In an embodiment of the present invention, the correlation coefficient learning neural network unit,

[Correlation coefficient formula]

The RNN neural network is trained with the input data, and the voltage values of each cell in the LIB pack at the same time t are substituted into the correlation coefficient formula and the output p _x,y value is used as the LSTM input in the RNN. At time t, a correlation coefficient between the voltage statistical characteristics of each cell may be learned.

In an embodiment of the present invention, the prediction unit,

[Weight Combination Formula]

C _t = σ(W _C * X + B _C ) : Weights learned by CNN

R _t = σ(W _R * X + B _R ) : Weights learned by RNN

V _t = σ(W _V * X + B _V ) : Weights learned with VDCNN

P _t = ReLU(Ct * h _(t-1) + Vt * h _(t-1) )

Final output h(t) = (1-K) R _t * h _(t-1) + KP _t * h _(t-1) (0 < K < 1)

(* symbol is convolution operation, K is weighting factor between each neural network, σ is sigmoid function (activation function), W is weight, X is input, B is bias value)

Each of the weights (W _C,R,V, B _C,R,V ) learned by the pattern extraction neural network unit, the state of charge learning neural network unit, and the correlation coefficient learning neural network unit is combined by the weight combination formula to the cell A balancing neural network model can be created.

According to an embodiment of the present invention, the SoC of each cell is predicted through the cell balancing neural network model learned by receiving charge/discharge data of each cell in the LIB pack from the battery control module for cell balancing, A method and apparatus for controlling a lithium ion battery pack using a cell balancing model can be provided, in which early detection and corrected SoC parameters are transmitted back to the battery control module to control all cells to have the same SoC.

According to an embodiment of the present invention, a lithium ion battery pack using a cell balancing model that replaces the conventionally used passive cell balancing technique and improves the stability and lifespan of the LIB pack by eliminating the process of consuming power as a resistor. A control method and an apparatus therefor can be provided.

It should be understood that the effects of the present invention are not limited to the above-described effects, and include all effects that can be inferred from the configuration of the invention described in the detailed description or claims of the present invention.

1 is a diagram illustrating a conventional passive cell balancing method.
2 is a flowchart illustrating a method for controlling a lithium ion battery pack using a cell balancing neural network model according to an embodiment of the present invention.
3 is a diagram for explaining a process of achieving cell balancing using a complex neural network.
4 is a diagram illustrating a control device for a lithium ion battery pack using a cell balancing neural network model based on a complex neural network according to an embodiment of the present invention.
5 is a diagram illustrating two-dimensional charge/discharge cycle data as an example of input data of a cell balancing neural network model based on a complex neural network according to an embodiment of the present invention.

Since the present invention can have various changes and can have various forms, specific embodiments are illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to the specific disclosed form, it should be understood to include all modifications, equivalents and substitutes included in the spirit and scope of the present invention. In describing each figure, like reference numerals have been used for like elements.

Unless defined otherwise, all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. does not

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the accompanying drawings.

2 is a flowchart illustrating a method for controlling a lithium ion battery pack using a cell balancing neural network model according to an embodiment of the present invention. 3 is a diagram for explaining a process of achieving cell balancing using a complex neural network.

The lithium ion battery pack control method using the cell balancing neural network model of this embodiment learns a cell balancing neural network model using information detected from a lithium ion battery pack (hereinafter, LIB pack), and the learned cell balancing It can be applied to a control method by predicting the state of charge (hereinafter, SoC) of each cell of the LIB pack using a neural network model, detecting and correcting the deteriorated cell.

In the lithium-ion battery pack control method using a cell balancing neural network model, first, a cell balancing neural network model in which a plurality of neural network models is combined is trained using input data from each cell in the LIB pack collected from the battery controller (S10).

Thereafter, the state of each cell in the LIB pack is predicted using the learned cell balancing neural network model (S20).

Thereafter, a control signal for correcting the state of charge of each cell according to the prediction is transmitted to the battery controller to correct the state of charge of each cell in the LIB pack (S30).

Hereinafter, each process will be described in detail.

Referring to FIG. 3 , a lithium-ion battery pack control method using a cell balancing neural network model uses a learned cell balancing neural network model based on a complex neural network. That is, a cell balancing neural network model in which a plurality of neural networks are combined is trained using input data from each cell in the LIB pack collected from the battery controller (S10).

In order to realize such a process, for example, the lithium ion battery pack control device using the cell balancing neural network model of this embodiment (hereinafter, the LIB pack control device) collects charge/discharge data of each cell in the LIB pack from the battery controller. can do. The LIB pack controller can control the state of charge of the LIB pack using a cell balancing neural network model based on a complex neural network. The LIB pack controller can continuously learn, update, or upgrade the cell balancing neural network model using the collected charging and discharging data. The cell balancing neural network model used by the LIB pack controller will be described in more detail below.

Thereafter, the state of each cell in the LIB pack is predicted using the learned cell balancing neural network model (S20). For example, the LIB pack controller having the cell balancing neural network model prepared as described above may receive current charge/discharge data of each cell from the battery controller. By using the received current charging/discharging data as an input of the cell balancing neural network model, the current SoC state of each cell in the LIB pack may be detected, and then, the change may be predicted or estimated.

Next, a control signal for correcting the state of charge of each cell according to the prediction is transmitted to the battery controller to correct the state of charge of each cell in the LIB pack (S30).

For example, the LIB pack controller predicts the SoC of each cell in the LIB pack as described above, detects a degraded cell or a cell to be degraded early, and provides a control signal (eg, SoC correction parameter) according to the prediction. It can be transmitted to the battery controller. The battery controller may control all cells to have a uniform SoC by applying the SoC correction parameter to each cell according to the control signal.

As described above, the LIB pack control method and control apparatus according to the present embodiment uses a prediction model (cell balancing neural network model) based on a complex neural network, and detects deteriorated cells early through prediction through this, and the detected deteriorated Cell balancing can be achieved by correcting the state of charge by reflecting the cells.

4 is a diagram illustrating a control device for a lithium ion battery pack using a cell balancing neural network model based on a complex neural network according to an embodiment of the present invention. 5 is a diagram illustrating two-dimensional charge/discharge cycle data as an example of input data of a cell balancing neural network model based on a complex neural network according to an embodiment of the present invention.

The cell balancing neural network model (see FIG. 4 ) used by the LIB pack control method of this embodiment may be a complex neural network model that learns using a plurality of neural networks and integrates them. The plurality of neural networks may include a Convolutional Neural Network (CNN) neural network, a Long Short Term Memory (LSTM) neural network, and a Recurrent Neural Network (RNN) neural network.

Specifically, the cell balancing neural network model learns spatial characteristics and cycle types between input data through a convolutional neural network (CNN) with input data from a plurality of cells of the LIB pack delivered from the battery controller in learning. and can be extracted (S11). For example, in the CNN learning, a cycle shape and a pattern may be extracted based on an inflection point and a maximum/minimum value of a charge/discharge curve (see FIG. 5 ).

Also, it is possible to learn and extract characteristics according to time from the charge/discharge voltage data of each cell using the LSTM neural network for the input data (S12).

In addition, it is possible to learn and extract the correlation coefficient between the voltage statistical characteristics of the cells at the same time through the RNN neural network using the input data (S13).

For this complex neural network learning, the LIB pack control device is, for example, as shown in FIG. 4 , an input unit 100, a pattern extraction neural network unit 200, a charged state learning neural network unit 300, a correlation coefficient learning neural network unit ( 400) and a prediction unit 500 .

In order to learn the cell balancing neural network (S10) and predict the state of each cell (S20), the input data is provided as two-dimensional charge/discharge cycle data (x-axis: time, y-axis: cell) or three-dimensional data ( It can be provided by converting to serial-parallel cells).

The input unit 100 may receive charge/discharge data of each cell in the LIB pack from the battery controller as, for example, two-dimensional voltage cycle data (refer to FIG. 5 ).

The prediction model (cell balancing neural network model) presented in FIG. 4 is a newly devised complex neural network by combining three neural networks, and it is possible to detect and predict deteriorated cells and SOC through the following process.

The pattern extraction neural network unit 200 may extract a pattern of a cycle shape by learning two-dimensional voltage cycle data (input data) through CNN operation ((S11).

For example, the pattern extraction neural network unit 200 converts two-dimensional input data into a neural network consisting of n layers of q filters of p x p size, spatial characteristics and cycle shape between input data through convolution operation. ) can be learned and extracted. Typically, the inflection point of the charge/discharge curve and the maximum and minimum values determine the cycle shape, and by learning these characteristics, the performance of the final prediction can be further improved. At this time, each parameter can be flexibly changed in consideration of the learning environment (filter size (p), number of filters (q), number of layers (n), learning time, hardware, prediction performance, etc.).

The state of charge learning neural network unit 300 may learn the SoC of each cell by learning the two-dimensional voltage cycle data (input data) through RNN operation, specifically, through an LSTM neural network (S12). The state-of-charge learning neural network unit 300 may learn and extract characteristics according to time from the charge/discharge voltage data of each cell by using an LSTM that exhibits performance conforming to time series prediction with the same initial input data.

Correlation coefficient learning The neural network unit 400 may learn a correlation coefficient between voltage statistical characteristics of cells through RNN operation on two-dimensional voltage cycle data (input data) (S13). For example, train the RNN with the same initial input data. In more detail, the p _x,y value output by substituting the voltage values of each cell in the LIB pack at the same time t into the following correlation coefficient equation is used as the LSTM input in the RNN. Through this, it is possible to learn and extract the voltage statistical characteristics between cells at the same time t.

[Correlation coefficient formula]

Referring to the correlation coefficient formula, the correlation coefficient learning neural network unit 400 learns through RNN operation with input data, but substituting the voltage values of each cell in the LIB pack at the same time t into the correlation coefficient formula The output p _x,y values can be used as LSTM inputs in RNN. Accordingly, the correlation coefficient learning neural network unit 400 may learn and extract the correlation coefficient between the voltage statistical characteristics of the cells at the same time t.

The above-described process of learning the cell balancing neural network model may be learned by parallel learning through the CNN, LSTM and RNN neural networks on charge/discharge voltage data (input data) of the LIB pack transmitted from the battery controller.

The prediction unit 500 generates a prediction model by integrating the results learned from the pattern extraction neural network unit 200, the state of charge learning neural network unit 300, and the correlation coefficient learning neural network unit 400, and is to be deteriorated based on this. cells can be predicted. For example, a deteriorated cell may be identified according to the predicted voltage of the second RNN and a correlation coefficient derived from the third neural network.

In generating a cell balancing neural network model by parallel learning the plurality of neural networks, the weights (W _{C, R, V,} B _{C, R, V} ) can be optimized.

[Weight Combination Equation] Equation for combining the learning weights of three neural networks

C _t = σ(W _C * X + B _C ) : Weights learned by CNN

R _t = σ(W _R * X + B _R ) : Weights learned by RNN

V _t = σ(W _V * X + B _V ) : Weights learned with VDCNN

P _t = ReLU(Ct * h _(t-1) + Vt * h _(t-1) )

Final output h(t) = (1-K) R _t * h _(t-1) + KP _t * h _(t-1) (0 < K < 1)

* Symbol is a convolution operation, K is a weighting factor between each neural network, σ is a sigmoid function (activation function), W is a weight, X is an input, and B is a bias value.

Each weight (W _C,R,V , B _C,R,V ) learned by the CNN, LSTM and RNN neural networks is combined by the weight combination formula to form the complex neural network (cell balancing neural network model). . The prediction model (cell balancing neural network model) learned through the above process can extract correlation coefficients between spatial characteristics of input data, temporal characteristics, and voltage statistical characteristics of each cell.

The learned cell balancing neural network model predicts the correlation coefficient between the state of charge (SoC) and voltage statistical characteristics of each cell in the process of predicting the state of each cell in the LIB pack (S20), A cell or a cell to be deteriorated can be detected. For example, the prediction unit 500 of the LIB pack control device generates a cell balancing neural network model by merging the learning results of each learning unit, and predicts the correlation coefficient between the voltage statistical characteristics of each cell and the SoC in an arbitrary cycle. A degraded cell or a cell to be degraded can be detected.

Thereafter, a control signal for correcting the state of charge of each cell according to the prediction is transmitted to the battery controller to correct the state of charge of each cell in the LIB pack (S30). As a result of the correction, the state of charge of the cells becomes uniform, so that cell balancing can be achieved.

The present invention relates to a complex neural network-based deteriorated cell detection and SoC correction prediction model for cell balancing in a LIB pack, which replaces the conventional passive cell balancing technique and further eliminates the process of consuming power with a resistor to ensure safety of the LIB pack And it has the effect of life improvement.

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention.

100: input unit
200: pattern extraction neural network
300: state of charge learning neural network
400: correlation coefficient learning neural network unit
500: prediction unit

Claims (12)

In a lithium-ion battery (LIB) pack control method using a cell balancing neural network model,
training a cell balancing neural network model in which a plurality of neural network models are combined using input data on the state of each cell in the LIB pack transmitted from the battery controller;
predicting the state of each cell in the LIB pack by using the learned cell balancing neural network model; and
Lithium-ion battery using a cell balancing neural network model, comprising the step of correcting the state of charge of each cell by transmitting a control signal for correcting the state of charge of each cell according to the prediction to the battery controller How to control the pack.
The method according to claim 1,
The step of training the cell balancing neural network model comprises:
Learning and extracting the input data regarding the charge/discharge state of each cell of the LIB pack transmitted from the battery controller through a CNN (Convolutional Neural Network) neural network, spatial characteristics and cycle shape between the input data of each cell ;
learning and extracting charging/discharging characteristics according to time of each cell using a Long Short Term Memory (LSTM) neural network for the input data; and
A lithium ion battery using a cell balancing neural network model, comprising: learning and extracting a correlation coefficient between the voltage statistical characteristics of each cell at the same time using the input data through a Recurrent Neural Network (RNN) neural network How to control the pack.
3. The method according to claim 2,
The step of training the cell balancing neural network model comprises:
Using the charge/discharge voltage data of each cell of the LIB pack delivered from the battery controller as the input data, parallel learning is performed through the CNN neural network, the LSTM neural network and the RNN neural network, cell balancing neural network A method of controlling a lithium-ion battery pack using a model.
4. The method according to claim 3,
In order to learn the cell balancing neural network and predict the state of each cell, the input data is two-dimensional charge/discharge cycle data with the x-axis as time and the y-axis as the cell, or 3D data composed of serial and parallel cells. Lithium-ion battery pack control method using a cell balancing neural network model, characterized in that.
5. The method according to claim 4,
In the step of learning and extracting the correlation coefficient between the voltage statistical characteristics of each cell at the same time through the RNN neural network,
[Correlation coefficient formula]

The RNN neural network is trained with the input data, and the px,y values output by substituting the voltage values of each cell at the same time t into the correlation coefficient formula are used as the LSTM input in the RNN. A method of controlling a lithium ion battery pack using a cell balancing neural network model, characterized in that learning and extracting a correlation coefficient between the voltage statistical characteristics of each cell in the .
6. The method of claim 5,
The learned cell balancing neural network model is,
In the step of predicting the state of each cell in the LIB pack, the cell balancing neural network model, characterized in that the cell to be deteriorated in an arbitrary cycle is detected by predicting the state of charge (SoC) of each cell and the correlation coefficient. A method of controlling a lithium-ion battery pack using
4. The method according to claim 3,
In the step of training the cell balancing neural network model,
Each weight (W _C,R,V, B _C,R,V ) learned by the CNN neural network, the LSTM neural network, and the RNN neural network is,
[Weight Combination Formula]
C _t = σ(W _C * X + B _C ) : Weights learned by CNN
R _t = σ(W _R * X + B _R ) : Weights learned by RNN
V _t = σ(W _V * X + B _V ) : Weights learned with VDCNN
P _t = ReLU(Ct * h _(t-1) + Vt * h _(t-1) )
final output h(t) = (1-K) R _t * h _(t-1) + KP _t * h _(t-1) (0 < K < 1)
(* symbol is convolution operation, K is weighting factor between each neural network, σ is sigmoid function (activation function), W is weight, X is input, B is bias value)
A method of controlling a lithium ion battery pack using a cell balancing neural network model, characterized in that the cell balancing neural network model is learned by combining by the weighted coupling formula.
5. The method according to claim 4,
Lithium using a cell balancing neural network model, characterized in that through the learning of the CNN neural network, a cycle shape and a cycle pattern are extracted based on the inflection point, maximum and minimum values of the charge/discharge curve of the charge/discharge voltage data How to control an ion battery pack.
In the control device of a lithium ion battery (LIB) pack using a cell balancing neural network model,
a pattern extraction neural network unit that learns input data related to the charge/discharge state of each cell in the LIB pack transmitted from the battery controller through CNN operation and extracts spatial characteristics, cycle shape, and cycle pattern of the input data;
a state of charge learning neural network unit for learning state-of-charge characteristics according to time of each cell through LSTM operation on the input data;
a correlation coefficient learning neural network unit for learning a correlation coefficient between the voltage statistical characteristics of each cell at the same time through RNN operation on the input data; and
By integrating the results learned from the pattern extraction neural network unit, the state of charge learning neural network unit, and the correlation coefficient learning neural network unit, a cell balancing neural network model is generated, and cells to be deteriorated are predicted using the generated cell balancing neural network model, A control device for a lithium ion battery pack using a cell balancing neural network model, comprising a prediction unit configured to transmit a state of charge correction parameter to the battery controller to achieve cell balancing.
10. The method of claim 9,
Further comprising an input unit that receives the input data from the battery controller and provides the pattern extraction neural network unit, the state of charge learning neural network unit, and the correlation coefficient learning neural network unit,
The input data provided by the input unit or received by the input unit is two-dimensional charge/discharge cycle data with the x-axis as time and the y-axis as the cell, or 3D data composed of serial and parallel cells, a cell balancing neural network model. A control device for a lithium-ion battery pack used.
10. The method of claim 9,
The correlation coefficient learning neural network unit,
[Correlation coefficient formula]

The RNN neural network is trained with the input data, and the px,y values output by substituting the voltage values of each cell at the same time t into the correlation coefficient formula are used as the LSTM input in the RNN. A control device for a lithium ion battery pack using a cell balancing neural network model, characterized in that it learns a correlation coefficient between voltage statistical characteristics of each cell.
10. The method of claim 9,
The prediction unit,
[Weight Combination Formula]
C _t = σ(W _C * X + B _C ) : Weights learned by CNN
R _t = σ(W _R * X + B _R ) : Weights learned by RNN
V _t = σ(W _V * X + B _V ) : Weights learned with VDCNN
P _t = ReLU(Ct * h _(t-1) + Vt * h _(t-1) )
Final output h(t) = (1-K) R _t * h _(t-1) + KP _t * h _(t-1) (0 < K < 1)
(* symbol is convolution operation, K is weighting factor between each neural network, σ is sigmoid function (activation function), W is weight, X is input, B is bias value)
The cell balancing neural network by combining the weights (WC, R, V, BC, R, V) learned by the pattern extraction neural network unit, the state of charge learning neural network unit, and the correlation coefficient learning neural network unit by the weight combination formula A control device for a lithium-ion battery pack using a cell balancing neural network model, characterized in that it generates a model.

Images