Battery health and remaining life prediction system based on deep learning

Abstract

A deep learning-based battery health and remaining life prediction system includes: a database storing a first set of data including multiple battery data collected from an input unit; a processor preprocessing the first set of data based on a first artificial intelligence model, learning a second artificial intelligence model by inputting the preprocessed first set of data, and predicting battery state health and remaining battery life based on the learned second artificial intelligence model. The processor includes: a preprocessing unit that performs preprocessing by converting the first set of data into a spectrogram and removing noise based on the first artificial intelligence model; a learning unit that extracts features from the input preprocessed first set of data and learns the second artificial intelligence model to estimate battery state health based on the extracted features; and a prediction unit that estimates battery state health from a second set of data input by a user based on the second artificial intelligence model learned by the learning unit, and predicts the remaining battery life.

Description

Technical Field

[0001] One disclosed embodiment relates to a system for predicting battery health and remaining lifespan based on a deep learning model. Background Technology

[0002] Lithium-ion batteries play a crucial role in various applications such as electric vehicles, energy storage systems (ESS), and consumer electronics due to their high energy density and long lifespan. However, during charging and discharging, batteries accumulate physical and chemical degradation, leading to performance decline and shortened lifespan, and in severe cases, potential safety issues. Therefore, accurate assessment of battery state of health (SoH) and prediction of remaining useful life (RUL) are essential for ensuring battery reliability and stability.

[0003] Previous techniques for predicting battery health and remaining life can be broadly categorized into equivalent circuit models (ECM), physics-based models, and data-driven models. ECM models model the battery's electrical characteristics as a simple circuit to estimate the battery's state of health (SoH). This method is computationally fast and easy to implement, but its accuracy is limited because it cannot fully reflect the complex physical and chemical changes that cause battery degradation. Physics-based models mathematically model the internal chemical reactions and physical degradation mechanisms of the battery to calculate SoH and remaining life (RUL). This model provides high theoretical accuracy, but its drawback is that it requires detailed information about the battery's internal structure and operating conditions, resulting in very high computational costs. Data-driven models learn from large-scale battery data to predict SoH and RUL. This model does not rely on a fixed mathematical model and automatically extracts features from the data, offering great flexibility. However, it still has limitations in the effectiveness of feature extraction and the model's generalization ability.

[0004] Several drawbacks exist with these previous techniques. First, they cannot effectively learn or reflect the nonlinear characteristics of battery degradation and the various patterns caused by environmental factors. Second, most methods only demonstrate reliable performance in static data or limited experimental environments, lacking generalization ability in real-world operating environments. Third, many techniques rely on labeled data, and their performance degrades significantly in environments lacking large-scale, high-quality data. Finally, previous techniques struggled to estimate battery state health (SoH) or predict remaining battery life (RUL) in real time, resulting in limited optimization of battery life management and maintenance plans. Summary of the Invention

[0005] The problem to be solved

[0006] To overcome the limitations of the prior art, one disclosed embodiment relates to a system that uses a deep learning model to remove noise from battery data and predict battery state of health (SoH) and remaining useful life (RUL).

[0007] Solution to the problem

[0008] A deep learning-based battery health and remaining life prediction system according to one disclosed embodiment includes: a database storing a first set of data including multiple battery data collected from an input unit; a processor preprocessing the first set of data based on a first artificial intelligence model, learning a second artificial intelligence model by inputting the preprocessed first set of data, and predicting battery state of health (SoH) and remaining battery life (RUL) based on the learned second artificial intelligence model; the processor includes: a preprocessing unit that converts the first set of data into a spectrogram, removes noise based on the first artificial intelligence model, and performs preprocessing; a learning unit that extracts features from the input preprocessed first set of data and learns the second artificial intelligence model to estimate battery state of health based on the extracted features; and a prediction unit that estimates battery state of health from a second set of data input by a user based on the second artificial intelligence model learned by the learning unit, and predicts the remaining battery life.

[0009] The aforementioned first set of data and second set of data may include voltage, charge / discharge capacity, and current.

[0010] The feature of this embodiment is that the preprocessing unit can use the Short-Time Fourier Transform (STFT) to convert the first set of data into a spectrum.

[0011] The feature of this embodiment is that the preprocessing unit can denoise the converted spectrogram data based on a denoising convolutional neural network (DnCNN).

[0012] The feature of this embodiment is that the learning unit can input the data generated in the preprocessing unit, extract feature vectors, and learn a second artificial intelligence model based on the feature vectors in order to estimate the state of health (SoH) of the battery.

[0013] The feature of this embodiment is that the learning unit can apply Bayesian optimization during the learning process of the second artificial intelligence model to optimize hyperparameters including learning rate, dropout rate and batch size.

[0014] The feature of this embodiment is that the prediction unit can estimate the state of health (SoH) of the battery in real time from the second set of data input by the user based on the second artificial intelligence model learned in the learning unit, and predict the remaining useful life (RUL) of the battery based on the state of health estimation result.

[0015] According to another disclosed embodiment, the deep learning-based battery health and remaining life prediction method includes: storing multiple data received from an input unit into a first set of data; converting the first set of data into a spectrogram; inputting the converted spectrogram data into a first artificial intelligence model to preprocess the data and remove noise; inputting the noise-removed data into a second artificial intelligence model to learn the second artificial intelligence model in order to estimate the state of health (SoH); and based on the learned second artificial intelligence model, estimating the state of health (SoH) from a second set of data input by the user and predicting the remaining useful life (RUL).

[0016] The feature of this embodiment is that the conversion to the above-mentioned spectrum is based on the Short-Time Fourier Transform (STFT) to convert the above-mentioned first set of data into two-dimensional data including features in the time-frequency domain.

[0017] The feature of this embodiment is that the first artificial intelligence model described above is a denoising convolutional neural network that uses residual learning technology.

[0018] The aforementioned first artificial intelligence model may include at least one of the following: Denoising Convolutional Neural Network (DnCNN), Flexible and Fast Denoising Network (FFDNet), and Memory Network (MemNet).

[0019] The aforementioned second artificial intelligence model may include at least one of the following: Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Variational Autoencoder (VAE), and Transformer.

[0020] A deep learning-based battery health and remaining life prediction system according to another disclosed embodiment includes: a database storing a first set of data including multiple data received from an input unit; a processor converting the first set of data into a spectrogram, removing noise by inputting the converted spectrogram data into a first artificial intelligence model, and inputting the noise-removed first set of data into a second artificial intelligence model to estimate the state of health (SoH) and predict the remaining battery life (RUL); and an output unit that outputs the results of predicting the state of health (SoH) and remaining battery life (RUL) from the second set of data input by the user based on the learned first and second artificial intelligence models.

[0021] In the deep learning-based battery health and remaining life prediction system according to the disclosed embodiments, a decline in battery health can be detected based on the prediction results of battery state health (SoH), and detection results can be provided.

[0022] In the deep learning-based battery health and remaining life prediction system according to the disclosed embodiments, a battery maintenance plan including battery replacement time can be generated based on the predicted battery remaining life (RUL), and the generated maintenance plan is provided.

[0023] The effects of the invention

[0024] The deep learning-based battery health and remaining life prediction system according to one disclosed embodiment can remove noise from battery data and predict battery state of health (SoH) and remaining useful life (RUL) with high accuracy. This provides the advantages described below.

[0025] First, the deep learning-based battery health and remaining life prediction system according to one disclosed embodiment removes noise from battery data using a denoising convolutional neural network (DnCNN) to improve data quality. By learning complex degradation patterns through deep learning models such as convolutional neural networks (CNNs) or transformers, it can accurately predict SoH and RUL. In particular, it maintains high reliability under various battery usage environments and conditions.

[0026] Secondly, the deep learning-based battery health and remaining life prediction system according to one disclosed embodiment automatically extracts features from large-scale data to evaluate the battery's state in real time, significantly improving effectiveness and accuracy compared to existing manual feature extraction methods. This system also offers greater adaptability and versatility compared to constant current charge-discharge analysis or equivalent circuit models.

[0027] Third, the deep learning-based battery health and remaining life prediction system according to one disclosed embodiment can detect the initial degradation state of the battery at an early stage, improving safety and preventing dangers such as battery explosions. Furthermore, it optimizes battery replacement time and maintenance plans, reducing costs and improving resource efficiency.

[0028] Finally, the deep learning-based battery health and remaining life prediction system according to one disclosed embodiment can be easily integrated with a Battery Management System (BMS) to contribute to battery performance management and improved reliability in various application areas such as electric vehicles, energy storage systems (ESS), and consumer electronics. This can extend battery life and increase user convenience and safety. Attached Figure Description

[0029] Figure 1 A diagram used to briefly illustrate the disclosed deep learning-based battery health and remaining life prediction system.

[0030] Figure 2 This is a control block diagram of the disclosed deep learning-based battery health and remaining life prediction system.

[0031] Figure 3 This is a flowchart illustrating the process by which a processor receives aggregated data and predicts the health and remaining lifespan of a battery.

[0032] Figure 4 This diagram is used to illustrate the operation of the preprocessing unit.

[0033] Figure 5 This diagram is used to illustrate the actions of the learning and prediction departments.

[0034] Explanation of reference numerals in the attached figures

[0035] 1: A deep learning-based battery health and remaining life prediction system

[0036] 2: Communication Network

[0037] 3: Battery Management System

[0038] 4: Personal Computer

[0039] 9: Input Section

[0040] 10: Processor

[0041] 11: Ministry of Communications

[0042] 12: Database

[0043] 13: Output Department Detailed Implementation

[0044] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the technical concept of the present invention can also be embodied in other forms and is not limited to the embodiments described herein. Rather, the embodiments described herein are provided so that the disclosure is thorough and complete, and can fully convey the ideas of the present invention to those skilled in the art.

[0045] In this specification, when a structural element is mentioned as being on another structural element, it means that it can be directly formed on the other structural element, or that a third structural element can be located between them. Furthermore, in several figures, shapes and sizes are exaggerated for the purpose of effectively illustrating the technical content.

[0046] Furthermore, in various embodiments of this specification, terms such as first, second, and third are used to describe multiple structural elements, but such structural elements should not be limited to these terms. These terms are only used to distinguish one structural element from another. Therefore, what is referred to as a first structural element in one embodiment may also be referred to as a second structural element in another embodiment. The various embodiments described and illustrated herein also include complementary embodiments thereof. Furthermore, in this specification, "and / or" is used to mean including at least one of the structural elements listed above.

[0047] In this specification, singular expressions include plural expressions unless the context clearly distinguishes them. Furthermore, terms such as "comprising" or "having" should not be construed as excluding the presence or additional possibilities of one or more other features, numbers, steps, structural elements, or combinations thereof, but are used to specify the presence of the features, numbers, steps, structural elements, or combinations thereof described in the specification. Also, in this specification, the term "connected" is used to include both indirect and direct connections between multiple structural elements.

[0048] Furthermore, in the following, when it is determined that the detailed description of relevant known functions or structures in the process of explaining the present invention may unnecessarily obscure the essence of the present invention, their detailed description will be omitted.

[0049] Figure 1 A diagram used to briefly illustrate the disclosed deep learning-based battery health and remaining life prediction system.

[0050] Reference Figure 1According to one embodiment of the disclosed deep learning-based battery health and remaining life prediction system 1, it can be implemented by a computer or portable terminal that collects battery data including voltage, resistance, temperature, discharge capacity, charge capacity and current from external devices 3 and 4 via a communication network 2. The computers include, for example, laptops, desktops, laptops, tablets, and touchscreen tablets equipped with web browsers. Portable terminals, as wireless communication devices ensuring portability and mobility, may include Personal Communication Systems (PCS), Global System for Mobile Communications (GSM), Personal Digital Cellular (PDC), Personal Handyphone Systems (PHS), Personal Digital Assistants (PDAs), International Mobile Telecommunications (IMT)-2000, Code Division Multiple Access (CDMA)-2000, Wideband Code Division Multiple Access (W-CDMA), Wireless Broadband Internet (WiBro) terminals, and smartphones. All kinds of handheld wireless communication devices such as phones, and wearable devices such as watches, rings, bracelets, anklets, necklaces, glasses, contact lenses, or head-mounted devices (HMDs).

[0051] The deep learning-based battery health and remaining life prediction system 1 can collect battery data, including voltage, resistance, temperature, discharge capacity, charge capacity, and current, from the battery management system (BMS) 3 and personal computer 4.

[0052] Specifically, a Battery Management System (BMS) is a structure for measuring battery data, which can collect battery data including voltage, resistance, temperature, discharge capacity, charge capacity, and current.

[0053] The Personal Computer 4 is a structure that can collect battery data on each user's terminal, and is used to illustrate that battery data collected not only now, but also in the past can be used as learning data.

[0054] Battery data collected from various devices 3 and 4 are transmitted via communication network 2 to a deep learning-based battery health and remaining life prediction system 1.

[0055] Communication network 2 serves as a channel for receiving data from structures 3 and 4 described above. The deep learning-based battery health and remaining life prediction system 1 receives data from communication network 2. The received battery data is stored in database 12 of the deep learning-based battery health and remaining life prediction system 1 (see reference). Figure 2 The deep learning-based battery health and remaining life prediction system 1 can use a deep learning model to remove noise from the data stored in the database 12 and predict the battery's state of health (SoH) and remaining useful life (RUL). The specific actions and methods of the deep learning-based battery health and remaining life prediction system 1 in judging battery health and predicting remaining life are described later through the following different figures.

[0056] On the other hand, besides Figure 1In addition to structures 3 and 4 shown, the deep learning-based battery health and remaining life prediction system 1 can also receive battery data from various devices capable of storing data, such as smartphones, laptops, or tablets, and can also receive various types of battery data from web servers or cloud servers. Furthermore, the deep learning-based battery health and remaining life prediction system 1 can also directly collect voice data from users via peripheral devices such as Universal Serial Bus (USB), without going through the communication network 2.

[0057] Figure 2 This is a control block diagram of the disclosed deep learning-based battery health and remaining life prediction system.

[0058] Reference Figure 2 A deep learning-based battery health and remaining life prediction system 1 includes: an input unit 9 for collecting battery data; a communication unit 11 for communicating with a communication network 2; a database 12 for converting a set of data (hereinafter referred to as first set data) including multiple battery data collected from the input unit 9 or the communication unit 11 into a spectrogram, removing noise, storing a preprocessed first artificial intelligence model and a second artificial intelligence model for predicting battery state and remaining life from the preprocessed first set data; a processor 10 for preprocessing the first set data stored in the database 12, learning the artificial intelligence model, and predicting battery state and remaining life based on the learned artificial intelligence model; and an output unit 13 for outputting the result of predicting battery state and remaining life based on newly received set data (hereinafter referred to as second set data).

[0059] Specifically, the input unit 9 may include various hardware devices such as buttons or switches, foot switches, keyboards, mice, trackballs, various joysticks, handles, or sticks to receive user input.

[0060] As an example, input unit 9 can receive whether the artificial intelligence model is learned through a first set of data or whether the learned artificial intelligence model is used to detect abnormal battery degradation based on a second set of data.

[0061] The communication unit 11 may include a deep learning-based battery health and remaining life prediction system 1, which can communicate with external devices ( Figure 1 (3, 4) Various structures for communication, for example, may include at least one of a short-range communication module, a wired communication module and a wireless communication module.

[0062] Short-range communication modules can include various short-range communication modules that use wireless communication networks to transmit and receive signals at close range, such as Bluetooth modules, infrared communication modules, radio frequency identification (RFID) communication modules, wireless local access network (WLAN) communication modules, NFC communication modules, and Zigbee communication modules.

[0063] Wired communication modules include not only various wired communication modules such as Local Area Network (LAN) modules, Wide Area Network (WAN) modules, or Value Added Network (VAN) modules, but also various cable communication modules such as Universal Serial Bus (USB), High Definition Multimedia Interface (HDMI), Digital Visual Interface (DVI), Recommended Standard 232 (RS-232), Power Line Communication (POTS), or Plain Old Telephone Service (POTS).

[0064] In addition to Wi-Fi modules and wireless broadband modules, wireless communication modules can also include wireless communication modules that support multiple wireless communication methods, such as GSM (global system for mobile communication), CDMA (code division multiple access), WCDMA (wideband code division multiple access), UMTS (universal mobile telecommunications system), TDMA (time division multiple access), and LTE (long term evolution).

[0065] Database 12 not only stores various sets of data collected by input unit 9 or communication unit 11, but also stores a first artificial intelligence model for preprocessing the sets of data, a second artificial intelligence model for predicting battery health and remaining lifespan from the preprocessed sets of data, an artificial intelligence model for learning from the sets of data, and an artificial intelligence model after learning has ended.

[0066] Database 12 may be implemented using at least one of the following: non-volatile storage devices such as cache, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), and flash memory; volatile storage devices such as random access memory (RAM); or storage media such as hard disk drive (HDD) and read-only optical disc (CD-ROM), but is not limited thereto. Figure 2 As shown, database 12 can be a memory implemented by a chip independent of processor 10, but it can also be implemented as a single chip with processor 10 as needed.

[0067] The output unit 13 outputs data including the results estimated by the processor 10, that is, data predicting the battery's health and remaining lifespan based on the second set of data. For example, the output unit 13 can display the battery health and remaining lifespan predicted from the battery data on a screen, and output a battery maintenance plan including the battery replacement time generated based on the prediction results through a user interface.

[0068] For the aforementioned actions, the output unit 13 may include various hardware devices such as a Digital Light Processing (DLP) panel, a Plasma Display Penalty, a Liquid Crystal Display (LCD) panel, an Electroluminescence (EL) panel, an Electrophoretic Display (EPD) panel, an Electrochromic Display (ECD) panel, a Light Emitting Diode (LED) panel, or an Organic Light Emitting Diode (OLED) panel.

[0069] On the other hand, the output unit 13 may also include a graphical user interface (GUI) such as a touch pad, i.e., a software device, for user input. The touch pad is implemented by a touch screen panel (TSP) and can form a layer structure with the input unit 9.

[0070] Processor 10 controls the overall deep learning-based battery health and remaining life prediction system 1. To this end, processor 10 can execute functions for control... Figure 2 The algorithm or program that reproduces the algorithm shown in the diagram. That is, processor 10 means a hardware-built-in data processing device with physically structured circuitry to execute functions represented by code or commands included in the program. It is an example of a hardware-built-in data processing device and may include, but is not limited to, microprocessors, central processing units (CPUs), processor cores, multiprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), graphics processing units (GPUs), etc. Processor 10 may also be implemented by more than one chip.

[0071] In terms of software, the processor 10 can be divided into a preprocessing unit 110 that converts the first set of data into a spectrogram and removes noise based on residual learning, a learning unit 120 that extracts features based on the data generated in the preprocessing unit and learns an artificial intelligence model to predict the state and remaining life of the battery, and a prediction unit 130 that predicts the state and remaining life of the battery from the second set of data input by the user based on the artificial intelligence model learned by the learning unit.

[0072] The specific operation of the processor 10, which is distinguished by software, will be described later with reference to the following additional diagrams.

[0073] On the other hand, the disclosed deep learning-based battery health and remaining life prediction system 1, in addition to the above-mentioned structure, may also include various other structures. Figure 2 The various structures can be modified or combined in various ways according to the required actions.

[0074] Figure 3 This is a flowchart illustrating the process by which the processor 10 of the disclosed deep learning-based battery health and remaining life prediction system 1 receives aggregated data, converts it into a spectrogram, removes noise based on residual learning, performs preprocessing, and predicts battery state and remaining life from the preprocessed data based on high-level feature learning.

[0075] To avoid repeating the following explanation, please refer to... Figure 3 and Figure 4 This describes the process by which the deep learning-based battery health and remaining life prediction system 1 receives and preprocesses the set of data.

[0076] Reference Figure 3 A deep learning-based battery health and remaining life prediction system 1 receives a set of data (100).

[0077] The aggregated data (hereinafter referred to as the first aggregated data) is data received from the input unit 9 or the communication unit 11, and may include voltage, charge / discharge capacity, and current. The first aggregated data may be data measured by various sensors that measure voltage, charge / discharge capacity, and current in real time, and is used by the learning unit 120 to learn an artificial intelligence model.

[0078] Specifically, the first set of data can be collected from data measured in real time by measurement devices such as a Battery Management System (BMS). The BMS utilizes various sensors to measure battery data such as voltage, resistance, temperature, discharge capacity, charge capacity, and current. Voltage is measured by a voltage sensor, resistance by an impedance analyzer or constant current circuit, temperature by a thermocouple, resistance temperature detector (RTD), or negative temperature coefficient thermistor, discharge and charge capacity by a current sensor and data logger, and current by a Hall effect current sensor or shunt resistor.

[0079] Reference Figure 3 The deep learning-based battery health and remaining life prediction system 1 preprocesses the received set of data (101).

[0080] Reference Figure 4 The deep learning-based battery health and remaining life prediction system 1 preprocesses the data by converting voltage, charge / discharge capacity and current data in the first set of data into a spectrum (200), removing noise (300).

[0081] The deep learning-based battery health and remaining life prediction system 1 according to the disclosed embodiments can apply Short-Time Fourier Transform (STFT) to convert the voltage, charge / discharge capacity, and current data included in the first set of data into a time-frequency domain spectrum. Thus, by representing the temporal characteristics of the battery data as frequency components, important features in the deep learning model can be effectively learned.

[0082] Specifically, the first set of data received by the input unit 9 or the communication unit 11 is divided into predetermined time windows. These time windows serve as the basic unit for converting the regional characteristics of the data into frequency regions, and a Fourier Transform is performed on each time window. The window size is a crucial parameter for adjusting the resolution and time discriminability of the converted frequency components, and is optimized based on the characteristics of the battery data. A spectrogram is generated by calculating the STFT result. This spectrogram is represented as a two-dimensional matrix, with the horizontal axis representing time and the vertical axis representing frequency. Color or amplitude represents the signal strength at a specific time and frequency. The generated spectrogram reflects the characteristic patterns of the battery data under various operating conditions (charge / discharge rate, temperature changes, etc.), and is subsequently utilized in preprocessing steps, including noise reduction.

[0083] The deep learning-based battery health and remaining life prediction system 1 according to the disclosed embodiments can utilize DnCNN (Denoising Convolutional Neural Network) to remove noise from data converted into a spectrogram. The denoising step using DnCNN improves the signal quality of the battery data, subsequently yielding highly accurate results during the deep learning model learning and prediction process.

[0084] DnCNN uses residual learning techniques to learn and remove noise components from the input data. In more detail, DnCNN is fed into spectrogram data and extracts the main features of the battery data from each layer through multiple convolutional layers and Rectified Linear Unit (ReLU) activation functions. As mentioned above, DnCNN operates by predicting residual components and removing them from the input data. This effectively removes anomalous signals such as Gaussian noise or spike noise from the battery data.

[0085] Denoising-preprocessed data can maintain the spatiotemporal characteristics of battery data while avoiding interference from unnecessary signal components in deep learning models, thus providing high reliability and accuracy in predicting battery state of health (SoH) and remaining useful life (RUL).

[0086] To avoid repeating the following explanations, please refer to... Figure 3 and Figure 5 This paper describes the process by which a deep learning-based battery health and remaining life prediction system 1 learns features from preprocessed data to predict battery status and remaining life.

[0087] Reference Figure 3 The deep learning-based battery health and remaining life prediction system 1 learns features (103) by inputting preprocessed data and predicts battery status and remaining life (104).

[0088] Reference Figure 5 A deep learning-based battery health and remaining life prediction system 1 learns an artificial intelligence model based on preprocessed data (400) and predicts battery status and remaining life in the learned artificial intelligence model (500).

[0089] According to the disclosed embodiments, the deep learning-based battery health and remaining life prediction system 1 learns a second artificial intelligence model (400) by inputting noise-removed spectrogram data after a preprocessing step.

[0090] Specifically, the second artificial intelligence model is a predictive model for battery state of health (SoH) and remaining useful life (RUL) learned based on the preprocessed first set of data. It can be at least one of convolutional neural network (CNN), recurrent neural network (RNN), variational autoencoder (VAE), and Transformer.

[0091] During the learning process of the second artificial intelligence model, the loss function can use mean squared error (MSE) or cross-entropy loss. Hyperparameters including learning rate, batch size, and dropout rate can be optimized through Bayesian optimization.

[0092] According to the disclosed embodiments, the deep learning-based battery health and remaining life prediction system 1 predicts the battery state and remaining life based on a second artificial intelligence model that has completed learning (500).

[0093] Specifically, the second AI model, after learning, estimates the battery state in real time based on the second set of user-input data, and calculates the remaining battery life (RUL) based on the predicted battery state health (SoH). The RUL prediction is achieved by reflecting the battery's degradation state and operating conditions, and provides high reliability in various operating environments through precise analysis of complex degradation patterns.

[0094] According to the disclosed embodiments, the deep learning-based battery health and remaining life prediction system 1 outputs the results of predicting battery state and remaining life based on the artificial intelligence model at the end of learning (105).

[0095] In the disclosed embodiments, the output unit 13 may visually represent the predicted state of battery health (SoH) and remaining battery life (RUL) in the processor 10 based on the user interface (UI).

[0096] Specifically, the output unit 13 can display the battery status in real time, expressing the battery state health (SoH) as a percentage (%) value and the remaining battery life (RUL) as the estimated remaining usage time or charge / discharge cycles (Cycle Count). Furthermore, when an abnormal battery condition is detected, the output unit 13 can also provide notifications including alarm information and maintenance recommendations.

[0097] The output unit 13 can share predicted battery state health (SoH) and battery remaining life (RUL) data through communication with external systems (such as battery management systems, BMS) or remote servers.

[0098] Finally, the output unit 13 can improve the effectiveness of battery life management by providing users with battery replacement schedules or maintenance plans.

[0099] The operation of this output unit 13 is just one example; there can be many variations.

[0100] The disclosed deep learning-based battery health and remaining life prediction system 1, by removing noise from the data received from the input unit and learning complex degradation patterns, accurately predicts battery state of health (SoH) and remaining useful life (RUL), which can optimize battery replacement time and maintenance plans, contributing to extending battery life and reducing operating costs.

Claims

1. A deep learning-based battery health and remaining life prediction system, characterized in that, include: A database that stores a first set of data including multiple battery data collected from the input unit; The processor preprocesses the first set of data based on a first artificial intelligence model, learns a second artificial intelligence model by inputting the preprocessed first set of data, and predicts the battery's state health and remaining battery life based on the learned second artificial intelligence model. The processors mentioned above include: The preprocessing unit converts the aforementioned first set of data into a spectrogram and removes noise based on the aforementioned first artificial intelligence model, thus performing preprocessing. The learning unit extracts features from the preprocessed first set of data and learns the aforementioned second artificial intelligence model to estimate battery state health based on the extracted features; and The prediction unit estimates the battery's state health from the second set of user-input data based on the second artificial intelligence model learned by the aforementioned learning unit, and predicts the battery's remaining lifespan.

2. The deep learning-based battery health and remaining life prediction system according to claim 1, characterized in that, The aforementioned first set of data and second set of data include voltage, charge / discharge capacity, and current.

3. The deep learning-based battery health and remaining life prediction system according to claim 1, characterized in that, The aforementioned preprocessing unit uses short-time Fourier transform to convert the first set of data into a spectrum.

4. The deep learning-based battery health and remaining life prediction system according to claim 3, characterized in that, The preprocessing unit denoises the converted spectrogram data based on a denoising convolutional neural network.

5. The deep learning-based battery health and remaining life prediction system according to claim 1, characterized in that, The learning unit inputs the data generated in the preprocessing unit, extracts feature vectors, and learns a second artificial intelligence model based on the feature vectors to estimate the battery's state health.

6. The deep learning-based battery health and remaining life prediction system according to claim 5, characterized in that, The learning unit applies Bayesian optimization during the learning process of the second artificial intelligence model to optimize hyperparameters including learning rate, dropout rate, and batch size.

7. The deep learning-based battery health and remaining life prediction system according to claim 1, characterized in that, The prediction unit estimates the battery health status in real time from the second set of user-input data based on the second artificial intelligence model learned in the learning unit, and predicts the remaining battery life based on the battery health status estimation results.

8. A method for predicting battery health and remaining life based on deep learning, characterized in that, include: The multiple data received from the input unit are stored as the first set of data. Convert the aforementioned first set of data into a spectrum graph. The transformed spectrogram data is input into the first artificial intelligence model to preprocess the data and remove noise. The noise-removed data is then input into the second artificial intelligence model for learning, in order to estimate the battery's state of health. Based on the above learning, the second artificial intelligence model estimates the battery health status from the second set of user-input data and predicts the remaining battery life.

9. The method for predicting battery health and remaining life based on deep learning according to claim 8, characterized in that, The conversion to the above spectrum diagram is based on the short-time Fourier transform, which converts the first set of data into two-dimensional data including features in the time-frequency domain.

10. The method for predicting battery health and remaining life based on deep learning according to claim 8, characterized in that, The first artificial intelligence model mentioned above is a denoised convolutional neural network that uses residual learning techniques.

11. The method for predicting battery health and remaining life based on deep learning according to claim 10, characterized in that, The aforementioned first artificial intelligence model includes at least one of the following: a denoising convolutional neural network, a flexible and fast denoising network, and a memory network.

12. The method for predicting battery health and remaining life based on deep learning according to claim 8, characterized in that, The aforementioned second artificial intelligence model includes at least one of convolutional neural networks, recurrent neural networks, variational autoencoders, and Transformers.

13. A deep learning-based battery health and remaining life prediction system, characterized in that, include: A database that stores a first set of data, including multiple data received from the input unit; The processor converts the aforementioned first set of data into a spectrogram, removes noise by inputting the converted spectrogram data into a first artificial intelligence model, and inputs the noise-removed first set of data into a second artificial intelligence model to estimate the battery's state health and predict its remaining lifespan. as well as The output section, based on the first and second artificial intelligence models learned from the learning process, outputs results predicting the battery's state of health and remaining battery life from the second set of data input by the user.

14. The deep learning-based battery health and remaining life prediction system according to claim 13, characterized in that, Based on the above prediction results of battery health status, the system detects a decline in battery health and provides detection results.

15. The deep learning-based battery health and remaining life prediction system according to claim 13, characterized in that, Based on the predicted battery life, a battery maintenance plan including battery replacement time is generated and provided.

Images