Battery cell state determination apparatus and method of operating same

Abstract

A battery cell state determination apparatus according to an embodiment of the present disclosure may include a first module configured to generate first state information about a battery cell based on a time series data set about the battery cell, and a second module configured to generate second state information about the battery cell based on the first state information. And a second module configured to generate second state information about the battery cells based on the time series data set, wherein the first module is further configured to perform learning using teacher enforcement based on the second state information.

Description

Technical Field

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to and the benefit of Korean Patent Application No. 10-2022-0126528 filed in the Korean Intellectual Property Office on October 4, 2022, the entire contents of which are incorporated herein by reference. Technical Field

[0004] Embodiments disclosed herein relate to a battery cell state determination apparatus and an operating method thereof. Background Art

[0005] Recently, the research and development of secondary batteries has been actively carried out. As a rechargeable / dischargeable secondary battery, it can include all conventional nickel (Ni) / cadmium (Cd) batteries, Ni / metal hydride (MH) batteries, etc., and recent lithium-ion batteries. Lithium-ion batteries have a much higher energy density than conventional Ni / Cd batteries, Ni / MH batteries, etc. In addition, lithium-ion batteries can be made small and lightweight, so that lithium-ion batteries have been used as power sources for mobile devices, and recently, their scope of use has been expanded to power sources for electric vehicles, attracting attention as a next-generation energy storage medium.

[0006] Batteries tend to deteriorate as they are repeatedly charged and discharged. For example, as batteries are repeatedly charged and discharged, their capacity and resistance may deteriorate, and their remaining life may decrease. In addition, the degree of degradation and remaining life of the battery may vary depending on usage conditions.

[0007] When the remaining life of a battery decreases rapidly, safety problems may occur in the use of the battery. Therefore, a method is needed to prevent risks caused by a sudden decrease in battery performance by determining the state of the battery. Summary of the invention

[0008] Technical issues

[0009] The embodiments disclosed herein are directed to providing a device including a first module for determining a state of a battery cell based on a time series data set of the battery cell and an operating method of the device.

[0010] The embodiments disclosed herein are also intended to provide a device and an operating method of the device for performing learning based on an experimental data set collected under preset experimental conditions and determining a battery state based on a time series data set.

[0011] Technical problems of the embodiments disclosed herein are not limited to the above-mentioned technical problems, and other unmentioned technical problems will be clearly understood by those of ordinary skill in the art from the following description.

[0012] Technical Solution

[0013] According to an embodiment of the present disclosure, a battery cell state determination device may include a first module and a second module, wherein the first module is configured to generate first state information about the battery cell based on a time series data set about the battery cell, and the second module is configured to generate second state information about the battery cell based on the time series data set, wherein the first module is also configured to perform learning using teacher forcing based on the second state information.

[0014] According to an embodiment, a time series data set may include a plurality of time series tokens, which are continuous data having time characteristics.

[0015] According to an embodiment, the first module may include: an encoder block configured to extract context information of the time series dataset from a first time series token included in the time series dataset; a decoder block configured to generate prediction data based on the context information of the time series dataset and a second time series token included in the time series dataset; and a determination block configured to generate first state information based on the prediction data and a third time series token included in the time series dataset.

[0016] According to an embodiment, the first module may further be configured to correct the prediction data based on the second state information.

[0017] According to an embodiment, the third time series token may be data collected from the battery cell at a point in time when the state of the battery cell is determined.

[0018] According to an embodiment, the third time series token may be data predicted to be collected from the battery cell when the battery cell is in a normal state.

[0019] According to an embodiment, the second module may include a long short-term memory block configured to extract a first feature based on a time series data set, a convolution block configured to extract a second feature based on the time series data set, and a combination block configured to generate second state information by combining the first feature with the second feature.

[0020] According to an embodiment, the long short-term memory block may include a recurrent layer configured to recurrently process a time series data set and a memory layer configured to store a processing result of the recurrent layer.

[0021] According to an embodiment, the convolution block may include a convolution layer configured to perform a convolution operation on the time series data set.

[0022] According to an embodiment, the convolution block may include a compression layer configured to compress an operation result of the convolution layer and an active layer configured to correct the operation result of the convolution layer based on the operation result of the compression layer.

[0023] According to an embodiment, the second module may be further configured to perform learning based on an experimental data set, and the experimental data set may be a data set in which data of an experimental battery cell collected under preset experimental conditions and state information of the experimental battery cell match each other.

[0024] According to another embodiment of the present disclosure, an operating method of a battery cell state determination device includes: collecting a time series data set of battery cells; generating first state information about the battery cells based on the time series data set; generating second state information about the battery cells based on the time series data set; and performing teacher forcing based on the second state information.

[0025] According to another embodiment, generation of the first state information may include: extracting context information of the time series dataset from a first time series token included in the time series dataset; generating prediction data based on the context information of the time series dataset and a second time series token included in the time series dataset; and generating the first state information based on the prediction data and a third time series token included in the time series dataset.

[0026] According to another embodiment, the generation of the second state information may include: extracting a first feature based on a time series data set; extracting a second feature based on the time series data set; and generating the second state information by combining the first feature with the second feature.

[0027] According to another embodiment, the extracting of the first feature may include: cyclically processing the time series data set; saving the cyclic processing result on the time series data set; and extracting the first feature based on the saved cyclic processing result.

[0028] According to another embodiment, the extracting of the second feature may include: performing a convolution operation on the time series data set; compressing a result of the convolution operation; and extracting the second feature by correcting the result of the convolution operation based on the compressed result.

[0029] According to another embodiment, the operating method may further include learning a generation method of the second state information based on an experimental data set, wherein the experimental data set is a data set in which data of an experimental battery cell collected under preset experimental conditions and the state information of the experimental battery cell match each other.

[0030] Beneficial Effects

[0031] A battery cell state prediction device and an operating method thereof according to embodiments disclosed herein may predict the state of a battery cell based on time series data.

[0032] The battery cell state prediction device and the operating method thereof according to the embodiments disclosed herein can verify the prediction result based on time series data on the basis of experimental data collected under preset conditions.

[0033] The battery cell state prediction device and the operating method thereof according to the embodiments disclosed herein may generate state information corresponding to the state of the battery cell after time series data is collected.

[0034] The battery cell state determination device and the operation method thereof according to the embodiments disclosed herein can improve the accuracy of prediction of state information based on experimental data.

[0035] Furthermore, various effects directly or indirectly recognized from the disclosure can be provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a block diagram of a battery cell state determination apparatus according to an embodiment disclosed herein.

[0037] Figure 2 is a block diagram illustrating a first module according to an embodiment disclosed herein.

[0038] Figure 3 is a block diagram illustrating a second module according to an embodiment disclosed herein.

[0039] Figure 4 is a flow chart of a method for determining a battery cell state of a first module according to an embodiment disclosed herein.

[0040] Figure 5 is a flow chart of a method for determining a battery cell state of a second module according to an embodiment disclosed herein.

[0041] Figure 6 is a flowchart of an operating method of a battery cell state determination device according to an embodiment disclosed herein.

[0042] Figure 7 is a flowchart of a first state information generating method according to an embodiment disclosed herein.

[0043] Figure 8 is a flowchart of a method for generating second state information according to an embodiment disclosed herein.

[0044] Fig. 9 is a flowchart of a method for generating second state information according to another embodiment disclosed herein.

[0045] Fig.10 is a block diagram illustrating a hardware configuration of a computing system for executing an operating method of a battery cell state determining apparatus according to an embodiment disclosed herein. DETAILED DESCRIPTION

[0046] Hereinafter, the embodiments disclosed in this document will be described in detail with reference to the exemplary drawings. When adding the drawing tokens for the components of each drawing, it should be noted that the same components are given the same drawing tokens even if they are indicated in different drawings. In addition, when describing the embodiments disclosed in this document, when it is determined that the detailed description of the related known configuration or function interferes with the understanding of the embodiments disclosed in this document, its detailed description will be omitted.

[0047] To describe the components of the embodiments disclosed herein, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only used to distinguish one component from another component and do not limit the components to the nature, order, sequence, etc. of the components. The terms (including technical and scientific terms) used herein have the same meanings as those commonly understood by those skilled in the art, as long as these terms are not defined differently. Generally, terms defined in commonly used dictionaries should be interpreted as having the same meanings as the contextual meanings of the relevant technology, and should not be interpreted as having ideal or exaggerated meanings unless they are clearly defined in this application.

[0048] Figure 1 is a block diagram of a battery cell state determination apparatus according to an embodiment disclosed herein.

[0049] refer to Figure 1 , a battery cell state determination device 1 according to an embodiment disclosed herein may include a first module 10 and a second module 20 .

[0050] The first module 10 and the second module 20 may be connected to a battery management system 30 that transmits a time series data set about battery cells, and the second module 20 may be connected to an experiment database 40 that transmits an experiment data set.

[0051] The first module 10 may generate first status information about the battery cell based on a time series data set about the battery cell.

[0052] The first state information about the battery cell may be information obtained by predicting the state of the battery cell by the first module 10. The first state information may include information about the determination of whether the battery cell is in a normal state or an abnormal state.

[0053] The first module 10 may generate prediction data about the state of the battery cell at a preset time point based on the collected time series data set, and determine the state of the battery cell based on the prediction data.

[0054] Will refer to Figure 2 The battery cell state determination method of the first module 10 is described in detail.

[0055] The second module 20 may generate second status information about the battery cell based on the time series data set about the battery cell.

[0056] The second state information about the battery cell may be state information of the battery cell corresponding to the collected time series data set and includes information about the determination of whether the battery cell is in a normal state or in an abnormal state, similar to the first state information.

[0057] Will refer to Figure 3 The battery cell state determination method of the second module 20 is described in detail.

[0058] The battery management system 30 may collect information about the battery cells in real time from the battery cells as the state determination target. The information about the battery cells collected in real time may be a time series data set about the battery cells and may be provided to the first module 10 and the second module 20.

[0059] The time series data set about the battery cell may include a plurality of time series tokens. According to an embodiment, the time series token may be data specific to a time window, wherein a preset time window is applied to data collected from the battery cell in real time.

[0060] According to another embodiment, the time series token may be data specific to a time window, wherein feature values ​​of data collected from battery cells in real time may be calculated, and a preset time window is applied to the calculated feature values.

[0061] According to an embodiment, the battery management system 30 may be connected to a battery module including battery cells and collect data about the battery cells from the battery module. The data collected by the battery management system 30 may be raw data of the battery cells, such as voltage, current, temperature, internal resistance, impedance, etc.

[0062] The battery management system 30 can calculate the capacity of the battery cell, the health state (SOH) of the battery cell, the state of charge (SOC) of the battery cell, the remaining useful life (RUL) of the battery cell, the charge change value of the voltage change, etc. from the raw data. The calculation result can be a statistical calculation value of the battery cell.

[0063] The battery management system 30 may be included in the vehicle. The battery management system 30 may apply a time window to data collected in actual driving of the vehicle and send the data as a time series data set to the first module 10 and the second module 20. According to another embodiment, the battery management system 30 may calculate feature values ​​of the data collected in actual driving of the vehicle, apply a preset time window to the calculated feature values, and send the calculated feature values ​​as a time series data set to the first module 10 and the second module 20. The data to which the time window is applied may be a time series token as data specific to the time window.

[0064] The experimental database 40 may store data of experimental battery cells collected under preset experimental conditions, and transmit the data of the experimental battery cells to the second module 20 as an experimental data set.

[0065] The experimental data set may be a data set in which data of the experimental data cell and state information of the experimental battery cell match each other.

[0066] For example, the experimental data set may be a data set in which experimental data tokens obtained by applying a time window to data collected from an experimental battery and state information of an experimental battery cell corresponding to each experimental data token match each other. The state information of the experimental battery cell may include information on the determination of whether the experimental battery cell is in a normal state or an abnormal state.

[0067] The experimental database 40 may receive data of the experimental battery cells from a plurality of experimental battery cells operating under preset experimental conditions. State information of the experimental battery cells may be calculated by an experimental battery management system including the experimental battery cells or by the experimental database 40.

[0068] The experiment database 40 may send the experiment data set to which the status information matches to the second module 20 .

[0069] The second module 20 may learn a state information generation method based on a time series data set of battery cells on the basis of the received experimental data set.

[0070] That is, the second module 20 may learn a state information generation method based on a data set of experimental battery cells whose state information matches therewith, and generate second state information from the time series data set through the learned state information generation method.

[0071] The second status information generated by the second module 20 can be the basis for teacher enforcement of the first module 10.

[0072] Teacher forcing can be a way to improve the accuracy of learning by coming up with ground truth values ​​for input values ​​during module learning together.

[0073] According to an embodiment, the second state information of the time series data set can be generated by a second module 20 that has completed learning based on the experimental data set, and the second state information can be input into the first module 10 as a benchmark true value of the time series data set, thereby improving the accuracy of the first state information generated by the first module 10.

[0074] The first module 10 may receive a time series data set and second state information generated from the time series data set, thereby improving the accuracy of generating prediction data used for generating the first state information.

[0075] The first module 10 may receive the second state information from the second module 20 and may perform learning based on the received second state information.

[0076] According to an embodiment, the first module 10 can improve the accuracy of the predicted data by applying the second state information to the predicted data generation. The first module 10 can learn the predicted data generation method to generate the first state information that matches the input second state information, thereby improving the accuracy of the predicted data generated based on the time series data set. That is, the first module 10 can perform teacher forcing by applying the second state information to the first state information.

[0077] The learning performed by the second module 20 based on the experimental data set may be performed before the second state information is generated based on the real-time data set. In addition, the learning performed by the second module 20 based on the experimental data set may be performed before the learning performed by the first module 10.

[0078] Figure 2 is a block diagram illustrating a first module according to an embodiment disclosed herein.

[0079] The first module 10 may include an encoder block 100 that extracts context information of a time series dataset, a decoder block 200 that generates prediction data based on the context information of the time series dataset, and a determination block 300 that generates first state information based on the prediction data.

[0080] The first module 10 may be a Transformer model that learns and predicts continuous data based on attention. Attention may be a data processing scheme that applies weight values ​​based on relationships among input tokens to improve the accuracy of data processing.

[0081] The first module 10 may receive a time series data set including a plurality of time series tokens.

[0082] According to an embodiment, the time series token may be time window-specific data generated by applying a preset time window to time series data collected from the battery cells by the battery management system 30 .

[0083] The first module 10 may generate prediction data by using a plurality of time-series data tokens, and generate first state information about the battery cell based on the generated prediction data.

[0084] According to an embodiment, the first module 10 may generate the first state information based on the first time series token, the second time series token, and the third time series token included in the time series data set.

[0085] The encoder block 100 included in the first module 10 may extract context information of the time series dataset from a first time series token included in the time series dataset.

[0086] The first time series token may include multiple tokens for a preset window to extract contextual information of the entire time series data set.

[0087] According to an embodiment, the first time series token may be differently set according to a learning method of the first module 10. The first time series token may share a token for a certain time window with a second time series token input to the decoder block 200.

[0088] According to another embodiment, the first time series token may include a token for a time window preceding the second time series token.

[0089] The first time series token may include a continuous token having a time characteristic or sequence to extract contextual information of the time series data set. For example, the first time series token may be obtained by applying a preset time window to continuous data collected from a battery cell undergoing state determination. Thus, the first time series token may be continuous sequence data of the battery cell undergoing state determination.

[0090] The encoder block 100 may include: an encoder input layer 110 that converts input tokens into dimensions that can be learned or processed; a position encoder layer 120 that includes the relative positions of the converted tokens of the encoder input layer 110; and an encoder layer 130 that extracts contextual information from the tokens.

[0091] The encoder input layer 110 may quantize the plurality of tokens included in the first time series tokens through feature extraction. More specifically, the encoder input layer 110 may vectorize the plurality of tokens and map each token onto a geometric space.

[0092] According to an embodiment, the encoder input layer 110 may be an embedding layer, and the tokens processed in the encoder input layer 110 may be referred to as embedding vectors.

[0093] The position encoding layer 120 may add position information to the embedding vector to apply position information of a plurality of tokens included in the first time series tokens to learning or data processing.

[0094] The first module 10 may receive time series data as a basis for learning or processing at one time, and reflect position information between tokens through the position encoding layer 120. That is, the position encoding layer 120 may apply the relationship among tokens to learning and prediction.

[0095] The encoder block 100 may include a plurality of encoder layers 130, the number of which may be a tuning value for optimization as a hyperparameter.

[0096] The encoder layer 130 may perform self-attention on the vector corresponding to the first time series token of the input, and pass the context information obtained by the self-attention to the decoder layer 220 .

[0097] The encoder layer 130 may include a self-attention layer 131 , a normalization layer 132 , a feed-forward layer 133 , and a normalization layer 134 .

[0098] The self-attention layer 131 may linearly transform a vector of tokens included in the input first time series tokens to generate a query vector, a key vector, and a value vector.

[0099] The self-attention layer 131 may perform attention based on tokens included in the first time series tokens.

[0100] The self-attention layer 131 may perform attention on all tokens included in the first time series tokens.

[0101] Attention can be the following operation: for each token included in the first time series token, a query vector, a key vector and a value vector are generated using a weight value matrix, and an output vector having attention values ​​of all tokens included in the input first time series token is generated by using the query vectors, key vectors and value vectors of all tokens.

[0102] The self-attention layer 131 may calculate a weight value matrix and vector of input tokens to generate a query vector, a key vector, and a value vector.

[0103] The weight value matrices used to generate the query vector, key vector, and value vector may be parameters that are updated during the learning process.

[0104] The self-attention layer 131 may calculate similarities for a query vector corresponding to a corresponding token with all key vectors corresponding to the corresponding token, and apply the similarities as weight values ​​to each value vector mapped to each corresponding key vector. The sum of the vectors to which the weight values ​​are applied may be an output vector having attention values ​​for all tokens.

[0105] According to an embodiment, the self-attention layer 131 can generate multiple output vectors by performing several attentions on the first time series token in parallel. Attention can be performed by different weight value matrices. The operation of performing several attentions in parallel can be called multi-head attention. The number of times of attention performed in parallel can be a tuning value for optimization.

[0106] The self-attention layer 131 may concatenate output vectors generated as a result of the parallel attention and multiply them by an additional weight value matrix.

[0107] The normalization layer 132 may add the vector input to the self-attention layer 131 to the vector output from the self-attention layer 131 and normalize the sum vector to prevent information loss. That is, the normalization layer 132 may perform residual connection and layer normalization on the output of the self-attention layer 131.

[0108] The feedforward layer 133 may be a fully connected layer including a plurality of hidden layers. A vector input to the feedforward layer 133 may be output under the influence of a weight value applied between hidden layers included in the feedforward layer 133. The size of the hidden layer included in the feedforward layer 133 may be a tuning value for optimization.

[0109] The normalization layer 134 may add the vector input to the feedforward layer 133 and the vector output from the feedforward layer 133 and normalize the sum vector to prevent information loss. That is, the normalization layer 134 may perform residual connection and layer normalization on the output of the feedforward layer 133.

[0110] These operations may be repeated as many times as the number of encoder layers 130. A vector output from the encoder layer 130 may have the same size as a vector input to the encoder layer 130 from the position encoding layer 120.

[0111] The encoder block 100 may pass the output value of the encoder layer 130 to the decoder block 200 as context information.

[0112] The encoder block 100 may perform the aforementioned operations and update the context information each time when the time series data set is input from the battery management system 30 to the first module 10 .

[0113] According to an embodiment, the context information may be a set of key vectors and value vectors obtained from an input time series dataset.

[0114] The decoder block 200 may generate an output sequence corresponding to the prediction data based on the second time series tokens included in the time series dataset and the context information output from the encoder block 100 .

[0115] The decoder block 200 may include a decoder input layer 210 that converts input tokens into a dimension that can be learned or processed; a decoding layer 220 that performs decoding based on the converted tokens of the decoder input layer 210; and a linear mapping layer 230 for generating an output sequence.

[0116] The decoder input layer 210 may quantize a plurality of tokens included in the second time series tokens input to the decoder block 200 through feature extraction like the encoder input layer 110. More specifically, the decoder input layer 210 may vectorize the plurality of tokens and map each token onto a geometric space.

[0117] The second time series token may include a plurality of tokens for the entire preset window.

[0118] According to an embodiment, the second time series token may be differently set according to the learning method of the first module 10. The second time series token may share a token for a certain time window with the first time series token input to the encoder block 100.

[0119] According to another embodiment, the second time series token may include a token for a time window after the first time series token and include a token for a time window before the third time series token.

[0120] According to an embodiment, the decoder input layer 210 may be an embedding layer, and the tokens processed in the decoder input layer 210 may be referred to as embedding vectors.

[0121] The decoder layer 220 may decode the context information output from the encoder block 100 and learn features of the time series dataset input to the first module 10 based on the decoded context information and the embedded second time series tokens.

[0122] The decoder block 200 may include a plurality of decoder layers 220, the number of which may be a tuned value for optimization as a hyperparameter.

[0123] The decoder layer 220 may infer prediction data, which is data of a battery cell for a time window after the second time series token, based on the learned features of the time series data set.

[0124] The decoder layer 220 may include a masked self-attention layer 221 , a normalization layer 222 , an encoder-decoder attention layer 223 , a normalization layer 224 , a feed-forward layer 225 , and a normalization layer 226 .

[0125] The masked self-attention layer 221 may linearly transform the vectors of tokens included in the second time series tokens to generate a query vector, a key vector, and a value vector.

[0126] The masked self-attention layer 221 may perform attention based on the tokens included in the second time series tokens. However, the masked self-attention layer 221 may mask the tokens corresponding to the time window after each token is obtained when performing attention, and then perform attention.

[0127] The mask may be an operation of preventing the decoder block 200 from referring to a token for a time window after a time window in which learning or processing is performed. The attention except the mask may be substantially the same as the attention of the self-attention layer 131 included in the encoder block 100.

[0128] For example, the masked self-attention layer 221 may perform attention based on tokens included in the second time series tokens input to the decoder block 200, and the attention performed by the masked self-attention layer 221 may be multi-head attention.

[0129] The normalization layer 222 may add the vector input to the masked self-attention layer 221 to the vector output from the masked self-attention layer 221 and normalize the sum vector to prevent information loss. That is, the normalization layer 222 may perform residual connection and layer normalization on the output of the masked self-attention layer 221.

[0130] The encoder-decoder attention layer 223 can perform multi-head attention like the masked self-attention layer 221 or the self-attention layer 131, but can perform attention using both the output value of the encoder block 100 and the output value of the masked self-attention layer 221.

[0131] More specifically, the encoder-decoder attention layer 223 may perform attention using context information output from the encoder block 100 and a vector based on a token included in the second time-series token.

[0132] The encoder block 100 may send context information to the encoder-decoder attention layer 223 , and the context information may include a key vector and a value vector for an output value of the encoder layer 130 .

[0133] The encoder-decoder attention layer 223 can generate an output value having attention values ​​of all tokens included in the second time series tokens based on a query vector generated from the vector output by the masked self-attention layer 221 and the normalization layer 222, and a key vector and a value vector for the output value of the encoder layer 130.

[0134] According to an embodiment, the encoder-decoder attention layer 223 may perform multi-head attention on all tokens included in the second time series tokens, perform parallel attention, and connect the generated output vectors to multiply by an additional weight value matrix.

[0135] The normalization layer 224 may add the vector input to the encoder-decoder attention layer 223 and the vector output from the encoder-decoder attention layer 223 and normalize the sum vector, thereby preventing information loss. That is, the normalization layer 224 may perform residual connection and layer normalization on the output of the encoder-decoder attention layer 223.

[0136] The feedforward layer 225 may be a fully connected layer including a plurality of hidden layers. A vector input to the feedforward layer 225 may be output under the influence of a weight value applied between hidden layers included in the feedforward layer 225. The size of the hidden layer included in the feedforward layer 225 may be a tuning value for optimization.

[0137] The normalization layer 226 may add the vector input to the feedforward layer 225 and the vector output from the feedforward layer 225 and normalize the sum vector to prevent information loss. That is, the normalization layer 226 may perform residual connection and layer normalization on the output of the feedforward layer 225.

[0138] These operations may be repeated as many times as the number of decoder layers 220. A vector output from the decoder layer 220 may have the same size as a vector input to the decoder layer 220 from the decoder input layer 210.

[0139] The linear mapping layer 230 may be a fully connected layer in which the battery management system 30 determines prediction data to be collected from the battery cell when the battery cell is in a normal state.

[0140] That is, the prediction data may be data predicted to be collected by the battery management system 30 when the battery cell is in a normal state when the first module 10 generates the state information of the battery cell.

[0141] The prediction data may be data having the same dimensionality as the tokens included in the time series data set input to the first module 10 .

[0142] The determination block 300 may generate first state information about the battery cell based on the prediction data and the third time series token.

[0143] The third time series token may be time series data actually collected from the battery cell by the battery management system 30 when the state of the battery cell is determined.

[0144] The determination block 300 may compare the predicted data with the third time series token to generate the first state information. For example, the determination block 300 may calculate the difference between the predicted data and the third time series token, and when the calculated difference exceeds a preset value, determine that the battery cell is in an abnormal state.

[0145] According to an embodiment, the determination block 300 may train the encoder block 100 and the decoder block 200 based on the second state information received from the second module 20, thereby improving the accuracy of the generated prediction data.

[0146] The determination block 300 may compare the first state information with the second state information and correct the learning parameters of the encoder block 100 and the decoder block 200 based on the comparison value for generating prediction data. By correcting the learning parameters, the accuracy of prediction data generation may be improved.

[0147] Figure 3 is a block diagram illustrating a second module according to an embodiment disclosed herein.

[0148] The second module 20 may perform learning based on an experimental data set that is data of experimental battery cells collected under preset experimental conditions, and may generate second state information based on a time series data set about the battery cells.

[0149] The second module 20 may include a long short-term memory block 400 , a convolution block 500 , and a merging block 600 .

[0150] The long short-term memory block 400 may include a shuffle layer 410 , a long short-term memory layer 420 , and a dropout layer 430 .

[0151] The convolution block 500 may include a convolution layer 510 and a pooling layer 520 .

[0152] The experimental data set may be a data set in which an experimental data token for a preset experimental condition and state information of an experimental battery cell match each other.

[0153] The long short-term memory block 400 may output a first feature based on the received time series data set. The convolution block 500 may output a second feature based on the time series data set.

[0154] The merging block 600 may generate second state information by merging the first feature with the second feature.

[0155] The long short-term memory block 400 may include a shuffle layer 410 , a long short-term memory layer 420 , and a dropout layer 430 .

[0156] The shuffle layer 410 can facilitate the processing of multivariate time series data, and improve the processing speed and prevent overfitting. The short-term / long-term memory layer 420 can be a network for processing sequentially input time series data, and can include: a recurrent layer 421, which is used to cyclically process the time series data set by taking the output of the input token as input; and a memory layer 422, which is used to save the processing results of the recurrent layer on the previous input token. That is, the long short-term memory layer 420 can be a long short-term memory (LSTM). According to an embodiment, the long short-term memory layer 420 can be an attention LSTM layer that performs attention.

[0157] The dropout layer 430 may be a layer for preventing overfitting and randomly removing some connections from the fully connected layer.

[0158] The convolution block 500 may include a convolution layer 510 and a pooling layer 520. The convolution block 500 may include a plurality of convolution layers 510.

[0159] The convolution layer 510 may include a convolution layer 511 that performs a convolution operation on a time series data set, a compression layer 512 that reduces the dimension of an output of the convolution layer 511, and an active layer 513 that normalizes the output of the compression layer 512 to apply a weight value.

[0160] The pooling layer 520 may be a layer that downsamples the operation result of the convolution layer 510 to reduce the size, and the second feature may be output through the pooling layer 520 .

[0161] The merging block 600 may determine the state of the battery cell from the input time series data set by merging the first feature with the second feature. The state of the battery cell determined by the merging block 600 may be the second state information.

[0162] For example, the second state information may include information about the determination of whether the battery cell is in a normal state or an abnormal state. The second state information may be state information of the battery cell at a point in time when the time series data set is input, and more specifically, may be state information of the battery cell at a point in time when a third time series token included in the time series data set is collected.

[0163] The second module 20 may perform learning based on the experimental data set.

[0164] The experimental data set may be a data set in which the experimental data tokens for the preset experimental conditions and the state information of the experimental battery cells match each other. The second module 20 may perform learning to input the experimental data tokens included in the experimental data set to the long short-term memory block 400 and the convolution block 500 to extract features corresponding to the experimental data tokens from each block, and may perform learning to cause the merging block 600 to merge the features to predict the state of the experimental battery cells.

[0165] More specifically, the second module 20 may perform learning by comparing the state information of the experimental battery cell matched with the experimental data token with the information predicted by the long short-term memory block 400 , the convolution block 500 , and the merging block 600 , thereby improving the accuracy of learning.

[0166] After performing learning based on the experimental data set, the second module 20 may receive the time series data set and predict the state of the battery cell based on the received time series data set.

[0167] Figure 4 is a flow chart of a method for determining a battery cell state of a first module according to an embodiment disclosed herein.

[0168] For convenience, description will be made using an example in which a time series data set includes first to sixth tokens T1 , T2 , T3 , T4 , T5 , and T6 , and the state of a battery cell is determined for a time window in which the sixth token T6 is collected.

[0169] The first module 10 can select a first time series token for extracting context information of the time series data set from the time series data set. For example, to determine the state of the battery cell in the time window of collecting the sixth token T6, the first time series token can include the first to fourth tokens T1, T2, T3 and T4.

[0170] The first time series tokens may include tokens (eg, T1, T2, T3, etc.) collected in a time series prior to token T6 collected in a time window that is a standard for determining the state of the battery cell.

[0171] The encoder block 100 may receive the first time series token and extract context information about the time series dataset from the received first time series token.

[0172] Each time when the time series data set is input from the battery management system 30 to the first module 10 , the encoder block 100 may update the extracted context information.

[0173] The method of extracting context information has been referred to Figure 2 has been described, and thus will not be redundantly described.

[0174] The decoder block 200 may receive an output value of the encoder block 100 as context information and perform attention by applying the context information to the encoder-decoder attention layer 223 included in the decoder layer 220 .

[0175] The decoder block 200 may select a second time series token for generating prediction data from the time series dataset.

[0176] The second time series token may include tokens T5 collected just before the collection of tokens T6 collected in the time window that is the criterion for determining the state of the battery cell.

[0177] According to an embodiment, the second time series token may include multiple tokens T4 and T5. The tokens T4 and T5 included in the second time series token may be continuous tokens with time characteristics. That is, the second time series token may include tokens collected for continuous time windows, such as the fourth token and the fifth token.

[0178] The decoder block 200 may generate prediction data T5' and T6' based on the context information and the second time series token. According to an embodiment, the determination block 300 may compare tokens T5 and T6 corresponding to the same time window as the prediction data T5' and T6' to improve the accuracy of the decoding algorithm of the decoder block 200.

[0179] The prediction data generated by the decoder block 200 may be continuous data corresponding to a token input to the decoder block 200 .

[0180] According to another embodiment, the second time series tokens input to the decoder block 200 may include only the fifth token T5 collected just before the time window that is the standard for determining the state of the battery cell, and the decoder block 200 may generate prediction data T6' for one time window based on the fifth token T5.

[0181] The prediction data T5 ′ and T6 ′ output from the decoder block 200 may be data that a battery cell in a normal state is predicted to have in a corresponding time window.

[0182] The determination block 300 may compare the tokens T6 actually collected from the battery cells in the time window as a criterion for state determination with the prediction data T6 ′ to generate the first state information S1 .

[0183] The determination block 300 may receive the second state information S2 from the second module 20. The determination block 300 may use the received second state information S2 as a reference true value of the input time series data set (T1, T2, T3, T4, T5, and T6), and adjust the tuning values ​​of the encoder block 100 and the decoder block 20 to improve the accuracy of the first state information S2 generation algorithm.

[0184] The encoder block 100 and the decoder block 200 may tune the learning algorithm based on the second state information S2 , thereby improving the accuracy of the generated prediction data T5 ′ and T6 ′.

[0185] The first module 10 can learn a prediction data generation method to generate first state information S1 that matches the input second state information S2, thereby improving the accuracy of prediction data T5' and T6' generated based on the time series data set.

[0186] Figure 5 is a flow chart of a method for determining a battery cell state of a second module according to an embodiment disclosed herein.

[0187] For convenience, description will be made using an example in which a time series data set includes first to sixth tokens T1 , T2 , T3 , T4 , T5 , and T6 , and the state of a battery cell is determined for a time window in which the sixth token T6 is collected.

[0188] The time series dataset may be input to the long short-term memory block 400 and the convolution block 500 in parallel.

[0189] The long short-term memory block 400 may extract a first feature F1 based on consecutive tokens T1 to T6 included in the time series dataset.

[0190] The convolution block 500 may extract a second feature F2 based on consecutive tokens T1 to T6 included in the time series dataset.

[0191] The operation methods of the long short-term memory block 400 and the convolution block 500 have been referred to Figure 3 has been described, and thus will not be redundantly described.

[0192] The merging block 600 may generate second state information S2 by merging the first feature F1 with the second feature F2.

[0193] The second module 20 may receive the time series data set after completing learning based on the experimental data set, so that the accuracy of state information determination may be higher than that of the first module 10 .

[0194] The experimental data set may be a data set in which the experimental data token for the preset experimental condition and the state information of the experimental battery cell match each other, so that the second module 20 can improve the accuracy of the state information determination based on the state information of the experimental battery cell.

[0195] Therefore, the first module 10 can perform teacher forcing by taking the second state information S2 generated by the second module 20 as a reference true value.

[0196] Figure 6 is a flowchart of an operating method of a battery cell state determination device according to an embodiment disclosed herein.

[0197] In operation S100 , the second module 20 may learn a method of generating second state information based on an experimental data set.

[0198] The experimental data set may be data in which the experimental data token and the state information on the experimental battery cell match each other.

[0199] The second module 20 may generate the second state information based on the time series data set of the battery cells by learning the state information generation method in advance.

[0200] In operation S200, the battery management system 30 may collect a time series data set of the battery cells. The time series data set may be time series data collected by the battery management system 30 until a time point when a state determination of the battery cells is required.

[0201] The time series data set may include a plurality of tokens obtained by applying a preset time window to the time series data obtained from the battery cells.

[0202] Tokens included in a time series dataset may be continuous data with temporal characteristics.

[0203] In operation S300 , the first module 10 may generate first status information based on the time series data set.

[0204] The first module 10 may include an encoder block 100, a decoder block 200, and a determination block 300, and the first state information may be state information of the battery cell at a time point when a state determination of the battery cell is required. For example, the state information of the battery cell may include information on the determination of whether the battery cell is in a normal state or an abnormal state.

[0205] The second module 20 may include a long short-term memory block 400 , a convolution block 500 , and a merging block 600 , and like the first state information, the second state information may be state information of the battery cell at a point in time when state determination of the battery cell is required.

[0206] The second module 20 may generate the status information of the battery cells in a different manner than the first module 10 .

[0207] In operation S500, the first module 10 may perform teacher forcing on the first state information generating algorithm based on the second state information.

[0208] The first module 10 can improve the accuracy of the prediction data generation algorithm by using the second state information as a reference true value.

[0209] Figure 7 is a flowchart of a first state information generating method according to an embodiment disclosed herein.

[0210] In operation S310 , the encoder block 100 included in the first module 10 may extract context information of the time series dataset from a first time series token included in the time series dataset.

[0211] The first time series token may include multiple tokens for a preset window to extract contextual information of the entire time series data set.

[0212] According to an embodiment, the context information may include a key vector and a value vector for an output value of the encoder block 100 .

[0213] In operation S320 , the decoder block 200 included in the first module 10 may generate prediction data based on context information of the time series dataset and a second time series token included in the time series dataset.

[0214] The second time series token may include a token collected just before a time point at which the state determination of the battery cell is required. The second time series token may include a token having a time characteristic.

[0215] The prediction data may be data that is predicted to be collected from the battery cell when the battery cell is in a normal state at a point in time when the state determination of the battery cell is required.

[0216] The decoder block 200 may generate prediction data by using a query vector obtained from the second time series token and a key vector and a value vector for an output value of the encoder block 100 .

[0217] In operation S330 , the determination block 300 may generate first state information based on the prediction data and a third time series token included in the time series data set.

[0218] The third time series token may be actual data of the battery cell obtained from the battery cell at a point in time when the state determination of the battery cell is required.

[0219] The determination block 300 may compare the prediction data with the third time series token to determine whether the battery cell is in a normal state and generate first state information.

[0220] That is, when the battery cell is in a normal state, the determination block 300 may compare the predicted data, which is the data predicted that the battery cell will have, with the third time series token, which is the actual data of the battery, to determine the state of the battery cell.

[0221] Figure 8 is a flowchart of a method for generating second state information according to an embodiment disclosed herein.

[0222] In operation S410, the long short-term memory block 400 included in the second module 20 may extract a first feature based on a time series data set.

[0223] The long short-term memory block 400 may be an LSTM block including a recurrent layer 421 and a memory layer 422 .

[0224] In operation S420, the convolution block 500 included in the second module 20 may extract a second feature based on the time series dataset.

[0225] The convolution block 500 may include a convolution layer 511 that performs a convolution operation.

[0226] In operation S430, the merging block 600 included in the second module 20 may generate second state information by merging the first feature with the second feature.

[0227] The first module 10 may perform teacher enforcement based on the second status information.

[0228] Fig. 9 is a flowchart of a method for generating second state information according to another embodiment disclosed herein.

[0229] In operation S411, the long short-term memory block 400 included in the second module 20 may cyclically process the time series data set.

[0230] Recurrent processing of the time series dataset may be performed by the recurrent layer 421 included in the long short-term memory block 400 .

[0231] In operation S412 , the long short-term memory block 400 may save the loop processing result on the time series data set.

[0232] The loop processing results of the time series data set can be saved by the memory layer 422 included in the long short-term memory block 400.

[0233] In operation S413 , the long short-term memory block 400 may extract a first feature based on the saved loop processing result.

[0234] In operation S421, the convolution block 500 included in the second module 20 may perform a convolution operation on the time series data set.

[0235] The convolution operation may be performed in a convolution layer 511 including a plurality of hidden layers.

[0236] In operation S422, the convolution block 500 may compress the result of the convolution operation.

[0237] Compression may be performed in a compression layer 512 included in the convolution block 500 .

[0238] In operation S423, the convolution block 500 may extract the second feature by correcting the result of the convolution operation based on the result of the compression.

[0239] The merging block 600 may generate second state information based on the extracted first features and second features.

[0240] Fig.10 is a block diagram illustrating a hardware configuration of a computing system for executing an operating method of a battery cell state determining apparatus according to an embodiment disclosed herein.

[0241] refer to Fig.10 , the computing system 1000 according to the embodiments disclosed herein may include an MCU 1010 , a memory 1020 , an input / output I / F 1030 , and a communication I / F 1040 .

[0242] The MCU 1010 may be a processor that executes various programs stored in the memory 1020 (e.g., a battery cell voltage or current collection program, a control program for a relay included in a battery pack, a battery cell remaining life calculation program, a battery cell capacity degradation diagnosis program, a battery cell resistance degradation determination program, etc.), processes various information including battery cell remaining life information, battery cell capacity degradation information, and battery cell resistance degradation information through these programs, and executes Figure 1 The above functions of the battery cell state determination device shown in .

[0243] The memory 1020 may store various programs related to log information collection and diagnosis of the battery. The memory 1020 may store various information of the battery, such as current, voltage, and charge / discharge condition information, voltage information of the battery cell in a preset number of charge / discharge cycles, dQ / dV information of the battery cell in a preset number of charge / discharge cycles, etc., as time series data of the battery cell. The memory 1020 may include Figure 1 The operation algorithms of modules 10 and 20 shown in FIG.

[0244] As required, the memory 1020 may be provided in a plurality. The memory 1020 may be a volatile memory or a non-volatile memory. For the memory 1020 as a volatile memory, a random access memory (RAM), a dynamic RAM (DRAM), a static RAM (SRAM), etc. may be used. For the memory 1020 as a non-volatile memory, a read-only memory (ROM), a programmable ROM (PROM), an electrically variable ROM (EAROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), a flash memory, etc. may be used. The examples listed above of the memory 1020 are merely examples and are not limited thereto.

[0245] The input / output I / F 1030 may provide an interface for transmitting and receiving data by connecting an input device (not shown) such as a keyboard, a mouse, a touch panel, etc. and an output device such as a display (not shown) to the MCU 1010 .

[0246] The communication I / F 1040, which is a component capable of sending and receiving various data to and from a server, may be various devices capable of supporting wired or wireless communication. For example, the battery cell state determination device may receive an experimental data set of an experimental battery cell from a separately provided external server through the communication I / F 1040. The battery cell determination device may store the received experimental data set in an experimental database. According to an embodiment, the experimental database may be provided outside the battery state determination device.

[0247] Therefore, the computer program according to the embodiment disclosed herein may be recorded in the memory 1020 and processed by the MCU 1010, thereby being implemented as an execution Figure 1 Modules with the functionality shown in .

[0248] The above description is merely illustrative of the technical idea of ​​the present disclosure, and various modifications and changes will be possible for those skilled in the art to which the embodiments disclosed herein belong without departing from the basic features of the embodiments of the present disclosure.

[0249] Therefore, the embodiments disclosed herein are intended to describe rather than limit the technical spirit of the embodiments disclosed herein, and the scope of the technical spirit of the present disclosure is not limited by these embodiments disclosed herein. The protection scope of the technical spirit disclosed herein should be interpreted by the attached claims, and all technical spirits within the same scope should be understood to be included within the scope of the present disclosure.

Claims

1. A battery cell state determination device, comprising: A first module, the first module being configured to generate first status information about the battery cell based on a time series data set about the battery cell; and a second module configured to generate second status information about the battery cell based on the time series data set, Wherein, the first module is further configured to perform learning using teacher forcing based on the second state information.

2. The battery cell state determination device according to claim 1, wherein: The time series data set includes a plurality of time series tokens, and the plurality of time series tokens are continuous data with time characteristics.

3. The battery cell state determination device according to claim 1, wherein: The first module includes: an encoder block configured to extract context information of the time series dataset from a first time series token included in the time series dataset; a decoder block configured to generate prediction data based on context information of the time series dataset and a second time series token included in the time series dataset; and A determination block is configured to generate the first state information based on the prediction data and a third time series token included in the time series data set.

4. The battery cell state determination device according to claim 3, wherein: The first module is further configured to correct the prediction data based on the second state information.

5. The battery cell state determination device according to claim 3, wherein: The third time series token is data collected from the battery cell at a point in time when the state of the battery cell is determined.

6. The battery cell state determination device according to claim 3, wherein: The third time series token is data predicted to be collected from the battery cell when the battery cell is in a normal state.

7. The battery cell state determination device according to claim 1, wherein: The second module includes: a long short-term memory block configured to extract a first feature based on the time series dataset; a convolution block configured to extract a second feature based on the time series dataset; and A combination block is configured to generate the second state information by combining the first feature with the second feature.

8. The battery cell state determination device according to claim 7, wherein: The long short-term memory block includes: a recurrent layer configured to recurrently process the time series dataset; and A memory layer is configured to store a processing result of the loop layer.

9. The battery cell state determination device according to claim 7, wherein: The convolution block includes a convolution layer configured to perform a convolution operation on the time series data set.

10. The battery cell state determination device according to claim 9, wherein: The convolution block includes: a compression layer configured to compress an operation result of the convolution layer; and An active layer is configured to correct an operation result of the convolution layer based on an operation result of the compression layer.

11. The battery cell state determination device according to claim 1, wherein: The second module is further configured to perform learning based on an experimental data set, and The experimental data set is a data set in which data of an experimental battery cell collected under preset experimental conditions and state information of the experimental battery cell match each other.

12. An operating method of a battery cell state determination device, the operating method comprising: Collect time series datasets of battery cells; generating first status information about the battery cell based on the time series data set; generating second status information about the battery cell based on the time series data set; and Teacher enforcement is performed based on the second status information.

13. The operating method according to claim 12, wherein: The generating of the first state information includes: extracting context information of the time series dataset from a first time series token included in the time series dataset; generating prediction data based on context information of the time series dataset and a second time series token included in the time series dataset; and The first state information is generated based on the prediction data and a third time series token included in the time series data set.

14. The operating method according to claim 12, wherein: The generating of the second state information includes: extracting a first feature based on the time series data set; extracting a second feature based on the time series dataset; and The second state information is generated by combining the first feature with the second feature.

15. The operating method according to claim 14, wherein: The extraction of the first feature includes: cyclically processing the time series time set; Saving the loop processing results on the time series data set; and The first feature is extracted based on the saved loop processing result.

16. The operating method according to claim 14, wherein: The extraction of the second feature includes: performing a convolution operation on the time series data set; compressing the result of the convolution operation; and The second feature is extracted by correcting a result of the convolution operation based on a compression result of the convolution operation.

17. The operating method according to claim 12, further comprising learning a method for generating the second state information based on an experimental data set, in, The experimental data set is a data set in which data of experimental battery cells collected under preset experimental conditions and state information of the experimental battery cells match each other.

Images