KR20230166196A - Method and system for preprocessing battery data for diagnosing condition of battery, and battery condition prediction system - Google Patents

Abstract

The present invention relates to a battery data preprocessing method and system for diagnosing battery status, and a battery status prediction system, wherein outliers are removed from battery data including at least one of voltage, current, and temperature, or the outliers are preset. At least one of outlier preprocessing to replace with a 1 value, missing value preprocessing to remove missing values or replace the missing values with a preset second value, correlation coefficient reporting processing to calculate the correlation between data, and noise reduction processing to recover data mixed with noise. A data analysis unit that performs one preprocessing; and a data size conversion unit that normalizes and normalizes the battery data preprocessed by the data analysis unit and inputs it into a pre-trained artificial intelligence model that predicts the state of the battery.

Description

Battery data preprocessing method and system for battery condition diagnosis, and battery condition prediction system {METHOD AND SYSTEM FOR PREPROCESSING BATTERY DATA FOR DIAGNOSING CONDITION OF BATTERY, AND BATTERY CONDITION PREDICTION SYSTEM}

The present invention relates to a preprocessing system, and more specifically, to a battery data preprocessing method and system for battery status diagnosis, and a battery status prediction system.

In recent years, the demand for portable electronic products such as laptops, video cameras, and portable phones has rapidly increased, and as the development of energy storage batteries, robots, and satellites has begun, the need for high-performance secondary batteries capable of repeated charging and discharging has increased. Research is actively underway.

Currently commercialized batteries include nickel cadmium batteries, nickel hydrogen batteries, nickel zinc batteries, and lithium-ion batteries. Among these, lithium-ion batteries have the advantage of being lighter in weight, longer lifespan, and higher energy density than other batteries. However, lithium-ion batteries have not been used often due to the risk of ignition and explosion, but recently, stability has been achieved by maintaining cell voltage through a battery management system (BMS), making them rechargeable batteries for various products and devices. The use rate of lithium-ion batteries is increasing.

However, as mentioned above, lithium-ion batteries have a high energy storage density, so fires and explosions frequently occur due to carelessness in use, such as external shock and overcharging, and manufacturing defects, such as defective batteries or protection circuits, which poses major safety problems. , Research using artificial intelligence is continuing to predict the condition and defective state of the battery.

Due to various factors such as the sensor and environment that measure each product, the measured data such as parameters (voltage, current, and temperature, etc.) to determine the status, life, and defect of the battery, etc., include the type and number of parameters and the method of measurement. , sampling frequencies, etc. are different.

Therefore, directly using measured data without any processing in an artificial intelligence model to diagnose battery status reduces the accuracy of battery status diagnosis, so a data preprocessing system tailored to the characteristics of each product must be introduced.

Domestic Published Patent No. 10-2022-0043059 Domestic Published Patent No. 10-2022-0012534

This specification was designed to solve the problems described above, and provides a battery data preprocessing method and system for battery status diagnosis that can be applied to battery data collected in different environments, and a battery status prediction system. There is a purpose.

The purpose is to provide a battery data preprocessing method and system that can increase the accuracy of battery condition diagnosis, and a battery condition prediction system.

To achieve this purpose, according to an embodiment of the present specification, the battery data preprocessing system for battery status diagnosis according to the present specification removes outliers from battery data including at least one of voltage, current, and temperature. or outlier preprocessing that replaces the outlier with a preset first value, missing value preprocessing that removes the missing value or replaces the missing value with a preset second value, correlation coefficient reporting processing that calculates the correlation between data, and data mixed with noise. a data analysis unit that performs at least one preprocessing during the noise reduction process for recovery; and a data size conversion unit that normalizes and normalizes the battery data preprocessed by the data analysis unit and inputs it into a pre-trained artificial intelligence model that predicts the state of the battery.

Preferably, the data analysis unit sorts each feature of the battery data in ascending order, divides each feature into 4 equal parts, and then calculates the difference between the 75% point and the 25% point among the divided values as the interquartile range ( Interquartile range), the maximum value is determined as the interquartile range multiplied by 1.5 plus the 75% point value, and the minimum value is determined as the interquartile range multiplied by 1.5 plus the 25% point value. , and the value exceeding the maximum value and the minimum value is determined as the outlier.

Preferably, the data analysis unit detects the outlier based on machine learning, which finds values that fall outside the distribution according to the characteristics of the data.

Preferably, the data analysis unit is characterized in that among the features included in the battery data, a value that does not arrive at the actual signal time or is not a desired value is determined as the missing value.

Preferably, the data analysis unit analyzes each feature of the battery data, remaining capacity (State of Charge), capacity life (State of Health), output life (State of Power), and state of balance. It is characterized by calculating the correlation between any one of them.

Preferably, the data analysis unit determines that the closer the correlation coefficient is to 0, the less correlated the feature is, and the closer the correlation coefficient is to 1 or -1, the more correlated the feature.

According to another embodiment of the present specification, the battery data preprocessing method for battery status diagnosis according to the present specification includes a battery data preprocessing system removing outliers from battery data including at least one of voltage, current, and temperature. Outlier preprocessing that replaces the outlier with a preset first value, missing value preprocessing that removes the missing value or replaces the missing value with a preset second value, correlation coefficient reporting processing that calculates the correlation between data, and recovery of data mixed with noise. performing at least one preprocessing of noise reduction processing; And the battery data pre-processing system normalizes and normalizes the pre-processed battery data, and inputs it into a pre-trained artificial intelligence model that predicts the state of the battery.

According to another embodiment of the present specification, the battery state prediction system according to the present specification collects battery data including at least one of voltage, current, and temperature from a battery management system, and converts the sampling period of the collected battery data. a data collection unit that does; For battery data whose sampling cycle has been converted by the data collection unit, outlier preprocessing to remove outliers or replace the outliers with a preset first value, missing values to remove missing values or replace the missing values with a preset second value A data preprocessor that performs at least one of preprocessing, correlation coefficient reporting processing to calculate correlation between data, and noise reduction processing to recover data mixed with noise, and normalizes and normalizes the preprocessed battery data; and a battery state prediction unit that predicts the state of the battery by inputting the battery data normalized and standardized by the data preprocessor into a pre-trained artificial intelligence model.

As described above, according to the present specification, for battery data including at least one of voltage, current, and temperature, outlier preprocessing to remove outliers or replace the outliers with a preset first value, remove missing values, or replace the outliers with a preset first value. Perform at least one preprocessing of missing value preprocessing to replace with a preset second value, correlation coefficient reporting processing to calculate the correlation between data, and noise reduction processing to recover data mixed with noise, and normalize and normalize the preprocessed battery data. By standardizing and providing a battery data preprocessing method and system for diagnosing battery status that is input to a pre-trained artificial intelligence model that predicts the status of the battery, and a battery status prediction system, a method or parameter value of data is measured for each battery product. Even if the data is different, data of similar quality can be obtained.

In addition, a battery data preprocessing method and system that can increase the accuracy of battery condition diagnosis, and a battery condition prediction system can be provided.

1 is a block diagram showing a schematic internal configuration of a battery state prediction system according to an embodiment of the present invention;
Figure 2 is a block diagram showing an example of the functions of the data analysis unit according to an embodiment of the present invention;
3 is a block diagram showing an example of hardware capable of realizing the function of the battery state prediction system according to the embodiment of the present invention, and
Figure 4 is a flowchart showing a battery data preprocessing method for battery condition diagnosis according to an embodiment of the present invention.
Figure 5 is a diagram illustrating an example of noise reduction results according to an embodiment of the present invention.

It should be noted that the technical terms used in this specification are only used to describe specific embodiments and are not intended to limit the present invention. In addition, the technical terms used in this specification, unless specifically defined in a different way in this specification, should be interpreted as meanings commonly understood by those skilled in the art in the technical field to which the present invention pertains, and are not overly comprehensive. It should not be interpreted in a literal or excessively reduced sense. Additionally, if the technical terms used in this specification are incorrect technical terms that do not accurately express the spirit of the present invention, they should be replaced with technical terms that can be correctly understood by those skilled in the art. In addition, general terms used in the present invention should be interpreted according to the definition in the dictionary or according to the context, and should not be interpreted in an excessively reduced sense.

Additionally, as used herein, singular expressions include plural expressions, unless the context clearly dictates otherwise. In the present application, terms such as “consists of” or “comprises” should not be construed as necessarily including all of the various components or steps described in the specification, and some of the components or steps may include It may not be included, or it should be interpreted as including additional components or steps.

In addition, the suffixes “module” and “part” for components used in this specification are given or used interchangeably only considering the ease of writing the specification, and do not have distinct meanings or roles in themselves.

Additionally, terms including ordinal numbers, such as first, second, etc., used in this specification may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, a first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component without departing from the scope of the present invention.

In an embodiment of the present invention, the status of the battery refers to a diagnosis of SOx, which is an indicator expressing the condition of the battery. Here, SOx may include remaining capacity (State of Charge, SOC), capacity life (State of Health, SOH), output life (State of Power, SOP), and balance life (State of Balance).

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings. However, identical or similar components will be assigned the same reference numbers regardless of the reference numerals, and duplicate descriptions thereof will be omitted.

Additionally, when describing the present invention, if it is determined that a detailed description of related known technologies may obscure the gist of the present invention, the detailed description will be omitted. In addition, it should be noted that the attached drawings are only intended to facilitate easy understanding of the spirit of the present invention, and should not be construed as limiting the spirit of the present invention by the attached drawings.

1 is a block diagram showing a schematic internal configuration of a battery state prediction system according to an embodiment of the present invention.

Referring to FIG. 1, the battery state prediction system according to the present invention may include a data collection unit 110, a data preprocessing unit 120, and a battery state prediction unit 130.

The data collection unit 110 collects battery data including at least one of voltage, current, and temperature from a battery management system (BMS).

Meanwhile, as the length of the signal input to the artificial intelligence model increases, the amount of calculation of the artificial intelligence model increases and overfitting problems may occur. Additionally, it is necessary to change different sampling periods equally for each battery management system. Therefore, the data collection unit 110 according to the present invention may include a sampling period conversion unit 112 that converts the sampling period of the collected battery data. The sampling period converter 112 can convert the sampling period using up-sampling and down-sampling methods. For example, the sampling period may be in units of minutes, seconds, milliseconds, or microseconds.

Upsampling is a method of making the sampling period shorter. Upsampling is performed using interpolation methods such as linear interpolation, cubic interpolation, bilinear interpolation, and bicubic interpolation. It can be done.

Down sampling is a method of making the sampling period longer, and the sampling period can be made longer through decimation, a digital signal processing.

The data pre-processing unit 120 performs pre-processing on the battery data collected by the data collection unit 110. To this end, the data pre-processing unit 120 may include a data analysis unit 122 and a data size conversion unit 124.

The data analysis unit 122 performs outlier preprocessing to remove outliers or replace outliers with preset first values, missing value preprocessing to remove missing values or replace missing values with preset second values, and data interconnection for the collected battery data. At least one preprocessing is performed among correlation coefficient reporting processing to calculate correlation and noise reduction processing to recover data mixed with noise. The detailed configuration of the data analysis unit 122 according to the present invention will be described in FIG. 2.

Meanwhile, as a preprocessing step before inputting some of the collected battery data into the artificial intelligence model, feature scaling is important for learning the artificial intelligence model and diagnosing its status. In other words, it is necessary to adjust the value range of different variables to a certain level and size. Accordingly, the data size conversion unit 124 normalizes and standardizes the battery data preprocessed by the data analysis unit 122.

Normalization is a process of transforming the sizes of variables in different ranges to unify them. It refers to adjusting the data distribution by converting the data range between 0 and 1, and can be expressed as the following equation 1.

Here, x represents the corresponding value, x _min represents the minimum value, and x _max represents the maximum value.

Standardization is the process of converting variables of different ranges into values with a Gaussian normal distribution with a mean of 0 and a variance of 1. When variables with different value ranges exist, it has the effect of eliminating size differences between variables. . Standardization can be expressed in Equation 2 below.

Here, μ represents the mean and σ represents the standard deviation.

Therefore, the data size conversion unit 124 according to the present invention can avoid problems that may occur due to different ranges or units of variable values (for example, when assuming that the battery data collected from the battery management system contains only voltage, current, and temperature values). , mean, variance, maximum, and minimum values are all different), and it is possible to prevent artificial intelligence models from being biased toward specific data.

The battery state prediction unit 130 inputs battery data normalized and normalized by the data size conversion unit 124 into a pre-trained artificial intelligence model to predict the state of the battery.

That is, the battery state prediction unit 130 calculates the actual operating period of the battery by inputting normalized and normalized battery data into the neural network and training the neural network. In addition, the condition of the battery is predicted through diagnosis of SOX, an indicator expressing the condition of the battery. That is, the battery state prediction unit 130 measures the similarity with a preset data pattern for each normalized and normalized battery data using a learned neural network, and determines the similarity of a predetermined number of data among the normalized and standardized data. It is determined whether there is a section higher than the set threshold, and if there is a section where the similarity of a predetermined number of data is higher than the preset threshold, the section is determined to be a section in which the battery is operating, and all of the sections are added together to determine the battery Outputs the actual operating period or performs battery status prediction. For example, the battery state prediction unit 130 divides the normalized and normalized data into a first pattern (state of use) and a second pattern (state of non-use) through unsupervised learning, and a section corresponding to the first pattern. can be output as the actual period of operation of the battery, or can be used to predict battery status. In addition, the battery state prediction unit 130 divides the normalized and normalized battery data into a first pattern and a second pattern, and compares the corresponding patterns with a predetermined pattern through supervised learning, thereby corresponding to a pattern similar to the predetermined pattern. The section can be output as the actual operating period of the battery, or the state of the battery can be predicted. In other words, status prediction includes various battery states or concepts of battery condition, including the actual period of use of the battery, the degree of battery aging, the future lifespan of the battery, and the degree of power storage capacity compared to existing new batteries.

Here, the battery state prediction unit 130 uses a dilated convolution neural network (hereinafter referred to as 'dilated CNN') as a basic structure to analyze the correlation between long time series data, and uses dilated CNN and batch normalization ( An entire artificial neural network can be constructed by repeatedly stacking Batch Normalization (Batch Normalization) technology. Additionally, the battery state prediction unit 130 may use 1D-CNN (Convolution Neural Network) to reduce the dimensionality of long time series data.

Dilated convolution (a` trous convolution) is a type of convolution originally developed for wavelet decomposition. For example, [Holschneider, M.; Kronland-Martinet, R.; Morlet, J.; and Tchamitchian, Ph., A Real-Time Algorithm for Signal Analysis with the Help of the Wavelet Transform in Wavelets : Time-Frequency Methods and Phase Space, JM Combes et al., eds., pp. 286-297 (1987)]. However, it has been especially applied to semantic segmentation to obtain complex features. See, for example, [Yu, Fisher and Koltun, Vladlen, Multi-scale context aggregation by dilated convolutions , 2016 Int'l Conference on Learning Representations (ICLR) (hereinafter, "Yu et al. 2016")].

In a pure CNN consisting of convolutional layers without pooling, feature maps can be generated by convolving pixels of adjacent data in the input, so the receptive field of a unit only grows linearly layer by layer. You can. Feasible ways to increase the receive field are to convolve the input data over a larger area. This may be similar to using a 'dilation kernel' in dilation convolution, instead of using an existing dense kernel for conventional convolution.

Assuming that F is a discrete function, K is the convolution kernel, and the expanded convolution * _d is a generalized version of a typical convolution, as defined by equation 3 below, the conventional convolution is a simple 1-expansion It can be a convolution (i.e. when d=1). Here, d is the dilation factor.

One advantage of applying expanded convolution in CNN is that the expanded version has a larger receiving field. A dilated convolutional filter can be obtained by upsampling the original filter. In other words, it can be obtained by inserting 0 between the elements of the original filter. Therefore, a filter extended by design can have a structured pattern of zero components. Compared with weight pruning, where the zero elements have random patterns and positions, the extended filter can have a structured pattern for the zero weights, reducing computational complexity in hardware and software. could be much more useful. In particular, when many values are sequentially input to a neural network as in the present invention, this extended convolution can be used to reduce computational complexity but not reduce the accuracy of the result.

Batch normalization refers to the process of normalizing (creating a normal distribution) the activation value or output value of an activation function. Specifically, batch normalization normalizes the mini-batch data by calculating the mean and standard deviation for each feature. Since this batch normalization is learned by normalizing in mini-batch units, it can reduce internal covariate shift that can occur during artificial neural network training.

Meanwhile, the present invention is not limited to constructing an entire artificial neural network by repeatedly stacking dilated CNN and batch normalization technology, and can construct an artificial neural network through other methods.

A multi-layer perceptron (Multi-Layer-Perceptron) or a 1D or 2D CNN that stacks several perceptron neurons in several layers, or a recurrent neural network (Recurrent neural network) in which the connections between units have a circular structure, or an encoder using only attention. You can create a decoder structure so that the learning speed is Transformer architecture, or you can construct an artificial neural network by learning the first half with CNN and the second half with RNN.

Additionally, the battery state prediction unit 130 inputs part of the normalized and normalized battery data into a neural network to predict the actual usage of the battery or the state of the battery.

Additionally, the battery state prediction unit 130 may divide normalized and normalized battery data into specific intervals (eg, time and cycle, etc.) and input them into a neural network.

Figure 2 is a block diagram showing an example of the functions of the data analysis unit according to an embodiment of the present invention.

Referring to FIG. 2, the data analysis unit 122 according to the present invention may include an outlier pre-processing unit 210, a missing value pre-processing unit 220, a correlation coefficient reporting unit 230, and a noise reduction unit 240. .

The outlier preprocessor 210 detects outliers in each feature of the collected battery data, removes the outliers, or replaces the outliers with a preset first value. To this end, the outlier pre-processing unit 210 may include an outlier detection unit 212 and an outlier deletion and replacement unit 214.

The outlier detection unit 212 determines a value greater than or less than a certain threshold for each feature as an outlier. Specifically, the outlier detection unit 212 can detect outliers using a method using an interquartile range (IQR) and a method using machine learning. That is, the outlier detection unit 212 sorts each feature in ascending order, divides each feature into four equal parts, and defines the difference between the 75% point and the 25% point among the divided values as the interquartile range. The value obtained by multiplying the interquartile range by 1.5 plus the 75% point value is determined as the maximum value, the value obtained by multiplying the interquartile range by 1.5 plus the 25% point value is determined as the minimum value, and the value exceeding the maximum and minimum values is determined as the maximum value. It can be judged as an outlier. The outlier detection unit 212 can detect outliers based on machine learning, such as DBSCAN and One-Class SVM, that finds values that deviate from the distribution according to the characteristics of the data.

The outlier deletion and replacement unit 214 removes the outlier and, if the outlier is determined to be important, replaces it with a specific value rather than deleting it. Specifically, the outlier deletion and replacement unit 214 may replace the outlier with the most frequent value, specific value, average value, etc. In addition, the outlier deletion and replacement unit 2124 replaces outliers with specific values using linear interpolation, cubic interpolation, bilinear interpolation, and bicubic interpolation. can do. Additionally, the outlier deletion and replacement unit 214 may replace the outlier with a specific value using a hot-deck imputation method that randomly selects one of data having similar data characteristics.

The missing value preprocessor 220 detects missing values in each feature of the collected battery data, and removes the missing values or replaces the missing values with a preset second value. To this end, the missing value preprocessing unit 220 may include a missing value detection unit 222 and a missing value deletion and replacement unit 224.

The missing value detection unit 222 determines values that do not arrive at the actual signal time for each feature or are not in the desired format as missing values (special characters, spaces, etc.).

The missing value deletion and replacement unit 224 removes the missing value and, if the missing value is determined to be important, replaces it with a specific value rather than deleting it. Since the method by which the missing value deletion and replacement unit 224 replaces missing values with a specific value is the same as the method by which the outlier value deletion and replacement unit 214 replaces an outlier with a specific value, a description thereof will be omitted.

On the other hand, when learning an artificial intelligence model, reflecting many features requires a corresponding amount of data. Additionally, if there is little data, overfitting problems may occur. To solve this problem, appropriate features can be selected and used in the artificial intelligence model to increase the reliability of the artificial intelligence model. To this end, the correlation coefficient reporting unit 230 calculates the correlation between data. Specifically, the correlation coefficient reporting unit 230 corresponds to each feature and one of the remaining capacity (State of Charge), capacity life (State of Health), output life (State of Power), and balance life (State of Balance). The correlation between them can be calculated.

Additionally, the correlation coefficient reporting unit 230 may determine that a feature has no correlation as the correlation coefficient converges to 0, and that a feature has high correlation as the correlation coefficient approaches 1 or -1.

Meanwhile, depending on the communication environment and device status, noise may be mixed into the data. Therefore, a noise reduction unit is needed to smooth the data. The noise reduction unit 240 according to the present invention recovers data mixed with noise. Specifically, the noise reduction unit 240 may use a deep neural network-based noise reduction (Denoising Autoencoder) that outputs noise-reduced data when noise-mixed data is input. Additionally, the noise reduction unit 240 may use a filter-based noise reduction (moving average filter) that reduces noise by averaging data from both sides of the surrounding area. Figure 5 is a diagram illustrating an example of noise reduction results according to an embodiment of the present invention.

Below, the specific configuration of the battery state prediction system according to the present invention will be described.

Referring to FIG. 3, hardware capable of realizing the function of the battery state prediction system will be described. Figure 3 is a block diagram showing an example of hardware capable of realizing the function of the battery state prediction system according to the embodiment of the present invention.

The functions of the battery state prediction system can be realized using, for example, the hardware resources shown in FIG. 3. In other words, the functions of the battery state prediction system are realized by controlling the hardware shown in FIG. 3 using a computer program.

As shown in FIG. 3 , this hardware mainly includes a CPU 302, a read only memory (ROM) 304, a RAM 306, a host bus 308, and a bridge 310. In addition, this hardware includes an external bus 312, an interface 314, an input unit 316, an output unit 318, a storage unit 320, a drive 322, a connection port 324, and a communication unit 326. has

The CPU 302 functions, for example, as an arithmetic processing unit or a control unit, based on various programs recorded in the ROM 304, RAM 306, storage unit 320, or removable recording medium 328. Controls all or part of the operation of each component. The ROM 304 is an example of a storage device that stores programs read by the CPU 302 and data used in calculations. In the RAM 306, for example, a program read by the CPU 302 and various parameters that change when executing the program are temporarily or permanently stored.

These elements are connected to each other, for example, through a host bus 308 capable of high-speed data transfer. Meanwhile, the host bus 308 is connected to the external bus 312, which has a relatively low data transfer speed, through a bridge 310, for example. Additionally, as the input unit 316, for example, a mouse, keyboard, touch panel, touch pad, button, switch, lever, etc. are used. Additionally, as the input unit 316, a remote controller capable of transmitting a control signal using infrared or other radio waves can be used.

As the output unit 318, for example, a display device such as a Cathode Ray Tube (CRT), Liquid Crystal Display (LCD), Plasma Display Panel (PDP), or Electro-Luminescence Display (ELD) may be used. Additionally, as the output unit 318, an audio output device such as a speaker or headphone, or a printer may be used.

The storage unit 320 is a device for storing various data. As the storage unit 320, for example, a magnetic storage device such as an HDD is used. Additionally, as the storage unit 320, a semiconductor storage device such as a solid state drive (SSD) or RAM disk, an optical storage device, or a magneto-optical storage device may be used.

The drive 322 is a device that reads information recorded on the removable recording medium 328, which is a removable recording medium, or records information on the removable recording medium 328. As the removable recording medium 328, for example, a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory is used. Additionally, a program defining the operation of the battery state prediction system may be stored in the removable recording medium 328.

The connection port 324 is for connecting an external connection device 330, such as a USB (Universal Serial Bus) port, IEEE 1394 port, SCSI (Small Computer System Interface), RS-232C port, or optical audio terminal. It's a port. As the external connection device 330, a printer or the like is used, for example.

The communication unit 326 is a communication device for connecting to the network 332. As the communication unit 326, for example, a communication circuit for a wired or wireless LAN, a communication circuit for a WUSB (Wireless USB), a communication circuit for a mobile phone network, etc. may be used. The network 332 is, for example, a network connected by wire or wireless.

Above, the hardware of the battery state prediction system has been described. In addition, the hardware described above is an example, and modifications such as omitting some elements or adding new elements are possible.

Figure 4 is a flowchart showing a battery data preprocessing method for battery condition diagnosis according to an embodiment of the present invention.

Referring to FIG. 4, the battery state prediction system collects battery data including at least one of voltage, current, and temperature from the battery management system (S410).

The battery status prediction system selects variables (i.e. features) to be input into the pre-trained artificial intelligence model (S420). In other words, when diagnosing battery status, the input values (data sensed from the battery, such as voltage, current, and temperature) are not always fixed, and the parameter values to be input to the learned artificial intelligence model must be selected. For example, when voltage = V, current = I, and temperature = T, the battery status prediction system is one of ① V, I, T ② V, I ③ V, T ④ I, T ⑤ V ⑥ I ⑦ T You can select . Additionally, the battery health prediction system can select any combination of data measured from the battery, including maximum, minimum, and average.

The battery status prediction system is a preprocessing method for selected variables, including outlier preprocessing to remove outliers or replace outliers with preset first values, missing value preprocessing to remove missing values or replace missing values with preset second values, and correlation between data. At least one of the correlation coefficient reporting process that calculates the relationship and the noise reduction process that recovers data mixed with noise is selected (S430).

The battery state prediction system performs selected preprocessing on the selected variables (S440).

The battery status prediction system inputs the preprocessed variables into the learned artificial intelligence model and determines whether they are suitable for the learned artificial intelligence model (S450). For example, if an error occurs when the battery status prediction system inputs the preprocessed variable into the learned artificial intelligence model, it may be determined that it is not suitable for the learned artificial intelligence model. Additionally, the battery state prediction system may determine that the learned artificial intelligence model is not suitable if the time for which the result is predicted exceeds a preset time. Additionally, the battery status prediction system may determine that the result is not suitable for the learned artificial intelligence model if the result exceeds the preset capacity.

If the battery status prediction system determines that the preprocessed variables are suitable for the learned artificial intelligence model, it normalizes and standardizes the preprocessed variables and outputs them as refined data (S460). If the battery state prediction system determines that the preprocessed variables are not suitable for the learned artificial intelligence model, it returns to step S420 and reselects variables to be input into the pre-trained artificial intelligence model.

The present invention can be used in all industries where lithium-ion batteries are applied, such as mobile devices such as electric vehicles, electric scooters, electric ships, forklifts, golf carts, and UAVs (Unmanned Aerial Vehicles), as well as UPS (Uninterruptible Power Supply) and energy storage systems. You can.

The above-described method may be implemented through various means. For example, embodiments of the present invention may be implemented by hardware, firmware, software, or a combination thereof.

In the case of hardware implementation, the method according to embodiments of the present invention uses one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), and Programmable Logic Devices (PLDs). , can be implemented by FPGAs (Field Programmable Gate Arrays), processors, controllers, microcontrollers, and microprocessors.

In the case of implementation by firmware or software, the method according to embodiments of the present invention may be implemented in the form of a module, procedure, or function that performs the functions or operations described above. Software code can be stored in a memory unit and run by a processor. The memory unit is located inside or outside the processor and can exchange data with the processor through various known means.

In the above, embodiments disclosed in the present specification have been described with reference to the attached drawings. As such, the embodiments shown in each drawing should not be construed as limited, but may be combined with each other by those skilled in the art who are familiar with the contents of the present specification, and when combined, some components may be omitted.

Here, terms or words used in this specification and claims should not be construed as limited to their usual or dictionary meanings, but should be interpreted as meanings and concepts consistent with the technical ideas disclosed in this specification.

Therefore, the embodiments described in this specification and the configurations shown in the drawings are only embodiments disclosed in this specification, and do not represent all of the technical ideas disclosed in this specification, and therefore, various equivalents may be substituted for them at the time of filing this application. It should be understood that variations and variations may exist.

110: data collection unit 112: sampling cycle conversion unit
120: data preprocessing unit 122: data analysis unit
124: Data size conversion unit 130: Battery state prediction unit
210: Outlier pre-processing unit 212: Outlier detection unit
214: Outlier deletion and replacement unit 220: Missing value preprocessing unit
222: Missing value detection unit 224: Missing value deletion and replacement unit
230: Correlation coefficient reporting unit 240: Noise reduction unit

Claims (8)

For battery data including at least one of voltage, current, and temperature, outlier preprocessing to remove outliers or replace the outliers with a preset first value, and missing values to remove missing values or replace the missing values with a preset second value. a data analysis unit that performs at least one of preprocessing, correlation coefficient reporting processing to calculate correlation between data, and noise reduction processing to recover data mixed with noise; and
a data size conversion unit that normalizes and normalizes the battery data preprocessed by the data analysis unit and inputs it into a pre-trained artificial intelligence model that predicts the state of the battery;
A battery data preprocessing system for battery condition diagnosis including.
According to paragraph 1,
The data analysis unit sorts each feature of the battery data in ascending order, divides each feature into 4 equal parts, and divides the difference between the 75% point and 25% point among the divided values into an interquartile range. Define, the value obtained by multiplying the interquartile range by 1.5 plus the 75% point value is determined as the maximum value, and the value obtained by multiplying the interquartile range by 1.5 plus the 25% point value is determined as the minimum value, A battery data preprocessing system for battery status diagnosis, characterized in that a value exceeding the maximum value and the minimum value is determined as the outlier.
According to paragraph 1,
A battery data preprocessing system for battery status diagnosis, wherein the data analysis unit detects the outlier based on machine learning, which finds values that deviate from the distribution according to the characteristics of the data.
According to paragraph 1,
The data analysis unit preprocesses battery data for battery status diagnosis, characterized in that among the features included in the battery data, values that do not come in at the time when the actual signal should come in or that are not in the desired form are determined as the missing values. system.
According to paragraph 1,
The data analysis unit analyzes each feature of the battery data and any one of remaining capacity (State of Charge), capacity life (State of Health), output life (State of Power), and balance life (State of Balance). A battery data preprocessing system for battery status diagnosis, characterized in that it calculates the correlation.
According to clause 5,
The data analysis unit determines that as the correlation coefficient approaches 0, the feature is uncorrelated, and as the correlation coefficient approaches 1 or -1, the data analysis unit determines that the feature is correlated.
A battery data preprocessing system includes, for battery data including at least one of voltage, current, and temperature, outlier preprocessing to remove outliers or replace the outliers with a preset first value, remove missing values, or replace the missing values with a preset first value. 2. Performing at least one preprocessing of missing value preprocessing to replace with a value, correlation coefficient reporting processing to calculate correlation between data, and noise reduction processing to recover data mixed with noise; and
The battery data pre-processing system normalizes and normalizes the pre-processed battery data and inputs it into a pre-trained artificial intelligence model that predicts the state of the battery;
Battery data preprocessing method for battery condition diagnosis including.
a data collection unit that collects battery data including at least one of voltage, current, and temperature from the battery management system, and converts the sampling period of the collected battery data;
For battery data whose sampling cycle has been converted by the data collection unit, outlier preprocessing to remove outliers or replace the outliers with a preset first value, missing values to remove missing values or replace the missing values with a preset second value A data preprocessor that performs at least one of preprocessing, correlation coefficient reporting processing to calculate correlation between data, and noise reduction processing to recover data mixed with noise, and normalizes and normalizes the preprocessed battery data; and
a battery state prediction unit that predicts the state of the battery by inputting battery data normalized and standardized by the data preprocessor into a pre-trained artificial intelligence model;
A battery state prediction system including.