Edge-cloud collaborative battery state estimation system and operating method thereof

Abstract

An edge-cloud collaboration-based battery status estimation system according to one embodiment of the present invention may include a plurality of edge devices and a cloud server. An edge device may estimate the state of a battery by applying data collected from a multimodal sensor to a machine learning model, generate a warning signal when an abnormal sign is detected in a battery exceeding a threshold, and transmit aggregated data to a cloud server. A cloud server may diagnose the state of a battery by applying data collected from the edge device to a deep learning model, predict the performance of the battery, establish a response strategy based on the predicted performance of the battery, and generate an update package and transmit it to the edge device.

Description

Technology Field

[0001] The present invention relates to a system and method for estimating and managing the state of an electric vehicle battery in real time by utilizing edge-cloud collaboration and multimodal sensor fusion. Background Technology

[0002] Electric vehicles (EVs) are rapidly becoming a sustainable means of transportation worldwide, and batteries, a core component of these vehicles, are a critical factor in determining the vehicle's performance, driving range, and safety. Electric vehicle batteries generally utilize lithium-ion batteries, which provide high energy density and a long lifespan. A battery management system (BMS) is a key system that monitors and controls the condition of these batteries, estimating and managing the battery's state of charge (SOC), state of health (SOH), and remaining lifespan.

[0003] Electric vehicles utilize high-capacity battery packs composed of hundreds of individual battery cells, and advanced sensing technologies and data analysis techniques are required to efficiently manage these complex systems. Recent advancements in artificial intelligence (AI) and the Internet of Things (IoT) technologies have enabled more precise and predictable battery management, significantly contributing to improved performance, extended lifespan, and enhanced safety of electric vehicles. Furthermore, advancements in cloud computing and edge computing technologies allow for simultaneous real-time data processing and large-scale data analysis, enabling the efficient management of not only individual vehicles but also entire vehicle fleets.

[0004] The information described above disclosed in the background technology of this invention is intended only to enhance understanding of the background of the present invention and may therefore include information that does not constitute prior art. The problem to be solved

[0005] The objective of the present invention is to solve the problem of battery state estimation accuracy in existing systems by introducing a multimodal approach that integrates and analyzes various sensor data.

[0006] The problem that the present invention aims to solve is to resolve the trade-off between real-time response and in-depth analysis, thereby simultaneously achieving immediate safety response and long-term performance optimization.

[0007] The objective of the present invention is to overcome the limitations of generalized models by providing a customized battery management solution that takes into account the characteristics and usage patterns of each vehicle.

[0008] The problem that the present invention aims to solve is to reduce the burden of data processing and storage, and to enable long-term battery life prediction and performance optimization through large-scale data analysis.

[0009] The problem that the present invention aims to solve is to build a system that enables stable battery management even in situations where the network connection is unstable.

[0010] The problem that the present invention aims to solve is to design a flexible system structure that can rapidly adapt to new battery technologies or changes in the usage environment.

[0011] The problem that the present invention aims to solve is to support reliable decision-making by quantifying the uncertainty of battery state estimation.

[0012] The problem that the present invention aims to solve is to establish a mechanism that efficiently utilizes data from the entire vehicle fleet to improve the performance of individual vehicles and enables continuous improvement of the entire system.

[0013] However, the technical problems that the present invention aims to solve are not limited to those described above, and other unmentioned problems can be clearly understood by those skilled in the art from the description of the invention below. means of solving the problem

[0014] An edge-cloud collaboration-based battery state estimation system according to one embodiment of the present invention comprises a plurality of edge devices implemented as a battery management system (BMS) that monitors and manages the state of a battery, and a cloud server that exchanges data with the plurality of edge devices and dynamically manages the performance of the battery based on data collected from the plurality of edge devices. The edge device is configured to estimate the state of the battery by applying battery-related data collected from a multimodal sensor to a machine learning model, generate one or more of a warning signal and an action signal when an abnormal sign is detected in the battery because the battery-related data exceeds a threshold value, and transmit the aggregated data to the cloud server. The cloud server is configured to diagnose the state of the battery by applying the comparison result of the data collected from the edge device and fleet data to a first deep learning model, predict the performance of the battery by applying the data collected from the edge device to a second deep learning model, establish a response strategy based on the battery performance prediction result, and generate an edge device update package based on the aggregated data and transmit it to the edge device.

[0015] A method of operation of an edge-cloud collaboration-based battery state estimation system according to an embodiment of the present invention may include a plurality of edge devices implemented as a battery management system (BMS) that monitors and manages the state of a battery, and a cloud server that exchanges data with the plurality of edge devices and dynamically manages the performance of the battery based on data collected from the plurality of edge devices. The method may include the steps of: the edge device estimating the state of the battery by applying battery-related data collected from a multimodal sensor to a machine learning model; generating one or more of a warning signal and an action signal when an abnormal sign is detected in the battery because the battery-related data exceeds a threshold value; and transmitting aggregated data to the cloud server. The cloud server may also include the steps of: diagnosing the state of the battery by applying the comparison result of the data collected from the edge device and fleet data to a first deep learning model; predicting the performance of the battery by applying the data collected from the edge device to a second deep learning model; establishing a response strategy based on the battery performance prediction result; and generating an edge device update package based on the aggregated data and transmitting it to the edge device.

[0016] In addition to this, other methods for implementing the present invention, other systems, and computer-readable recording media storing a computer program for executing said methods may be further provided.

[0017] Other aspects, features, and advantages other than those described above will become clear from the following drawings, claims, and detailed description of the invention. Effects of the invention

[0018] While conventional technology relied on single or limited sensor data, according to the present invention, the accuracy of battery state estimation can be significantly improved through multimodal sensor fusion.

[0019] In addition, unlike conventional technology which relied on either an edge device or the cloud, the present invention enables real-time response and in-depth analysis to be realized simultaneously through edge-cloud collaboration.

[0020] In addition, while conventional technology was limited to long-term data analysis due to limitations in storage and processing capabilities, the present invention can improve the accuracy of battery life cycle prediction by introducing cloud-based big data processing technology.

[0021] In addition, while conventional technology suffered from system performance degradation when the network connection was interrupted due to high cloud dependency, the present invention strengthens the self-processing capabilities of the edge device, enabling stable battery management even in the event of network instability.

[0022] In addition, while conventional technology required a complete redesign of the entire system to respond to new battery technologies or changes in the usage environment, the present invention allows for rapid adaptation by applying a modular architecture.

[0023] However, the effects obtainable through the present invention are not limited to those described above, and other unmentioned technical effects will be clearly understood by those skilled in the art from the description of the invention below. Brief explanation of the drawing

[0024] The following drawings attached to this specification illustrate preferred embodiments of the present invention and serve to further enhance understanding of the technical concept of the present invention together with the detailed description of the invention provided below; therefore, the present invention should not be interpreted as being limited only to the matters described in such drawings. FIG. 1 illustrates the configuration of an edge-cloud collaboration-based battery state estimation system according to one embodiment of the present invention. FIG. 2 illustrates the configuration of an edge device and a cloud server included in an edge-cloud collaboration-based battery status estimation system according to an embodiment of the present invention. FIG. 3 is a flowchart illustrating a data processing method performed on an edge device according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating the operation method of an edge-cloud collaboration-based battery state estimation system according to an embodiment of the present invention. Specific details for implementing the invention

[0025] Preferred embodiments of the present invention will be described in detail below with reference to the attached drawings. Prior to this, terms and words used in this specification and claims should not be interpreted as being limited to their ordinary or dictionary meanings. Instead, based on the principle that the inventor may appropriately define the concepts of terms to best describe his invention, they should be interpreted in a meaning and concept consistent with the technical spirit of the present invention. Therefore, it should be understood that the embodiments described in this specification and the configurations illustrated in the drawings are merely some of the most preferred embodiments of the present invention and do not represent all of the technical spirit of the present invention; thus, various equivalents and modifications that can replace them may exist at the time of filing this application. Furthermore, as used in this specification, "comprise" or "include" and / or "comprising" or "including" specify the presence of the mentioned features, numbers, steps, actions, parts, elements, and / or groups thereof, and do not exclude the presence or addition of one or more other features, numbers, actions, parts, elements, and / or groups. In addition, when describing embodiments of the present invention, "may" and "may be" may include "one or more embodiments of the present invention."

[0026] Additionally, to aid in understanding the invention, the attached drawings are not drawn to actual scale, and the dimensions of some components may be exaggerated. Furthermore, the same reference numerals may be assigned to identical components in different embodiments.

[0027] The statement that two subjects of comparison are "identical" means that they are "substantially identical." Therefore, substantial identity may include deviations considered low in the industry, for example, deviations within 5%. Additionally, the statement that a parameter is uniform in a given area may mean that it is uniform from an average perspective.

[0028] Although terms such as "first," "second," etc., are used to describe various components, it goes without saying that these components are not limited by these terms. These terms are used merely to distinguish one component from another, and unless specifically stated otherwise, the first component may also be the second component.

[0029] Throughout the specification, unless specifically stated otherwise, each component may be singular or plural.

[0030] The fact that any configuration is placed on the "upper (or lower)" of a component or on the "upper (or lower)" of a component may mean not only that any configuration is placed in contact with the upper (or lower) surface of said component, but also that another configuration may be interposed between said component and any configuration placed on (or below) said component.

[0031] Furthermore, where it is stated that one component is "connected," "coupled," or "connected" to another component, it should be understood that while said components may be directly connected or connected to each other, another component may be "interposed" between each component, or that each component may be "connected," "coupled," or "connected" through another component. Additionally, when it is stated that a part is electrically coupled with another part, this includes not only cases where they are directly connected but also cases where they are connected with another component in between.

[0032] Throughout the specification, "A and / or B" means A, B, or A and B unless specifically stated otherwise. That is, "and / or" includes any combination or any combination of the enumerated items. "C to D" means C or more and D or less, unless specifically stated otherwise.

[0033] FIG. 1 illustrates the configuration of an edge-cloud collaboration-based battery state estimation system according to an embodiment of the present invention. Referring to FIG. 1, the battery state estimation system (1) may include an edge device (100), a cloud server (200), and a network (300).

[0034] In the present embodiment, the edge device (100) may include the first edge device to the Nth edge device (100-1, 100-2, 100-N). The edge device (100) may include a master BMS (battery management system) provided inside the vehicle.

[0035] The vehicle is equipped with multiple slave BMSs (not shown) and one master BMS, which can monitor and manage the status of the battery pack supplying power. The slave BMS can monitor status data, such as voltage, current, temperature, and humidity, of the battery cells included in the battery module (200-i) it manages in real time. Based on the monitoring results, the slave BMS can calculate the state of charge (SOC) and state of health (SOH) of the battery cells and manage the battery cells accordingly. Additionally, the slave BMS can perform cell balancing operations to uniformly adjust the voltage between individual battery cells within the battery pack based on the monitoring results. The master BMS can integrally manage battery status data collected from multiple slave BMSs and analyze the status of the battery system (not shown). Through this, the master BMS can establish an optimal operation strategy for the battery system and take necessary measures.

[0036] The edge device (100) can estimate the state of the battery by applying battery-related data collected from a multimodal sensor to a machine learning model. In this embodiment, the multimodal sensor can integrate various types of sensors to comprehensively monitor and analyze the state of the battery. The multimodal sensor can detect the voltage, current, temperature, humidity, etc. of the battery cell in real time and transmit them to the edge device (100). The edge device (100) can generate one or more of a warning signal and an action signal when the battery-related data exceeds a threshold and an abnormal sign is detected in the battery. The edge device (100) can compress the aggregated data and transmit it to the cloud server (200). The edge device (100) can perform an update to adjust the parameters of the cell balancing algorithm, the parameters of the machine learning model, and the threshold using an update package received from the cloud server (200). The edge device (100) can recalculate the driving range based on the results of the update.

[0037] The edge device (100) is equipped with a low-power ARM (advanced reduced instruction set computer machine) processor and a dedicated NPU (neural processing unit) to perform real-time processing and initial analysis of battery-related data locally. The edge device (100) uses QNNs (quantized neural networks) to estimate the battery status every second, and if an abnormal sign (e.g., sudden fluctuation in voltage) is detected, it can immediately transmit detailed data from the last minute to a cloud server (200). This enables fast response and real-time data processing, and allows for direct action to be taken locally if necessary.

[0038] The cloud server (200) can diagnose the battery status and predict the battery performance by applying a deep learning model to the data collected from the edge device (100). The cloud server (200) can establish a response strategy based on the battery performance prediction results. The cloud server (200) can generate an edge device update package based on the aggregated data and send it to the edge device (100). The cloud server (200) can analyze the impact of the external environment on the battery performance based on the data collected from the edge device (100). The cloud server (200) can establish a customized battery management strategy by analyzing the user's long-term habits and driving style based on the data collected from the edge device (100). The cloud server (200) can generate dashboards and analysis reports provided to the administrator based on the data collected from the edge device (100). The cloud server (200) can evaluate the possibility of reusing or recycling the battery based on the data collected from the edge device (100).

[0039] The cloud server (200) can perform large-scale data analysis and update machine learning models by utilizing a GPU (graphics processing unit) cluster and a distributed processing system (e.g., Apache Spark). The cloud server (200) receives data transmitted from the edge device (100), analyzes it together with data from the entire vehicle fleet, and thereby optimizes battery management strategies. Additionally, the cloud server (200) can periodically (e.g., daily) transmit new machine learning models to update the models of the edge device (100). This process can play an important role in continuously monitoring and optimizing the battery performance of the vehicles.

[0040] The cloud server (200) can process and manage large-scale battery data through various technical means. To process large-scale battery data, the cloud server (200) can use a distributed storage system utilizing Hadoop Distributed File System (HDFS), a big data processing engine using Apache Spark, an Influx DB time-series database, and advanced analysis algorithms including Prophet and XGBoost.

[0041] In this embodiment, the edge-cloud collaboration structure can utilize the respective advantages of an edge device (100) and a cloud server (200) through integration. The edge device (100) enables rapid response by performing real-time data processing and initial analysis, and can immediately transmit detailed data to the cloud server (200) when an important event occurs. The cloud server (200) utilizes this data to perform large-scale data analysis and model updates, and can periodically update the edge device's model. This edge-cloud collaboration structure can provide insights that contribute to the advancement of battery technology and optimize battery life, performance, and safety by comprehensively analyzing data from the entire vehicle fleet while simultaneously securing real-time responsiveness and precise analytical capabilities.

[0042] The network (300) can support wireless communication between the edge device (100) and the cloud server (200). For efficient and secure transmission of data, the edge device (100) and the cloud server (200) may use standardized communication protocols such as MQTT (message queuing telemetry transport), AMQP (advanced message queuing protocol), or HTTPS (hypertext transfer protocol secure). Data transmission through the network (300) can be performed through various wireless network technologies such as cellular networks (e.g., LTE, 5G), satellite communication, and Wi-Fi.

[0043] FIG. 2 illustrates the configuration of an edge device and a cloud server included in an edge-cloud collaboration-based battery status estimation system according to an embodiment of the present invention. In the following description, parts that overlap with the description of FIG. 1 will be omitted.

[0044] Referring to FIG. 2, the edge device (100) may include an edge communication unit (110), an edge processor (120), and an edge memory (130). The cloud server (200) may include a cloud communication unit (210), a cloud processor (220), and a cloud memory (230).

[0045] First, the edge device (100) will be described. The edge communication unit (110) can perform the necessary functions for transmitting and receiving data with the edge device (100) and the cloud server (200). The edge communication unit (110) can exchange information with the cloud server (200) in real time using a wireless communication protocol.

[0046] The edge processor (120) can monitor the battery status by performing real-time data processing and analysis. Additionally, the edge processor (120) can immediately detect abnormal signs in the battery and transmit important data to a cloud server (200). In this embodiment, the edge processor (120) may include a data processing unit (121), a data management unit (122), and a user interface processing unit (123).

[0047] The data processing unit (121) can accurately analyze the state of the battery in real time using battery-related data collected from a multimodal sensor. The data processing unit (121) can estimate the state of charge (SOC) of the battery at preset intervals (e.g., 1 second). The data processing unit (121) can continuously monitor voltage fluctuations of the battery cell to immediately detect abnormal conditions and take necessary countermeasures. The data processing unit (121) can continuously monitor the temperature of the battery pack to prevent overheating and compensate for the safety and lifespan of the battery. The data processing unit (121) can optimize power management by quickly calculating the expected power consumption. The data processing unit (121) can rapidly process data by utilizing a pre-trained lightweight machine learning model.

[0048] The data management unit (122) can be responsible for the efficient management and security of data. The data management unit (122) can collect raw data related to batteries from multimodal sensors, sort the raw data related to batteries by a common time stamp to generate time-synchronized data, apply a statistical outlier detection algorithm to the time-synchronized data to generate outlier-removed data, smooth the outlier-removed data to generate high-frequency noise-removed data, normalize the high-frequency noise-removed data to generate normalized data, extract statistical characteristics from the normalized data to generate time-series data, combine the time-series data to generate a single input vector, and apply a data augmentation algorithm to the single input vector to generate input data for a machine learning model.

[0049] The data management unit (122) can perform a local data caching function that ensures the continuity and integrity of data by temporarily storing data when the network connection is unstable and continuing transmission after the connection is restored. The data management unit (122) can perform an emergency safety control function that ensures the safety of the battery and the user by immediately shutting down the system when overheating or overvoltage of the battery is detected. The data management unit (122) can perform a sensor health check function that maintains the reliability of the system by periodically checking whether the sensors included in the multi-model sensor are operating normally. The data management unit (122) can perform an update that adjusts the parameters of the cell balancing algorithm, the parameters of the machine learning model, and the threshold value using an update package received from the cloud server (200).

[0050] The user interface processing unit (123) can provide interaction with the user (driver) and necessary feedback. The user interface processing unit (123) can intuitively provide the driver with necessary battery information, so that the driver can easily understand the battery status and take necessary actions.

[0051] The edge processor (120) according to the present embodiment can predict battery-related data through lightweight first to fifth machine learning models. In the present embodiment, the edge processor (120) can predict the state of the battery through the first machine learning model and the second machine learning model. The edge processor (120) can predict the battery operation optimization state through the third machine learning model and the fourth machine learning model. The edge processor (120) can predict the safety and maintenance state of the battery through the fifth machine learning model. The edge processor (120) can transmit the battery-related prediction results through the first to fifth machine learning models to the cloud server (200).

[0052] The edge processor (120) can estimate the state of charge (SOC) of a battery corresponding to the battery-related data collected from the multimodal sensor by using a first machine learning model that estimates the state of charge (SOC) of a battery based on battery-related data collected from the multimodal sensor. Here, the first machine learning model may be a model trained in a supervised learning manner using training data that has the state of charge (SOC) of a battery as a label, and takes as input the battery-related data collected from the multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors.

[0053] The edge processor (120) can train an initially configured first machine learning model using labeled training data in a supervised learning manner. Here, the initially configured first machine learning model is an initial model designed to be configured as a model capable of estimating the state of charge (SOC) of a battery using battery-related data collected from a multimodal sensor, and the parameter values ​​are set to arbitrary initial values. As the initial model is trained through the aforementioned training data, the parameter values ​​are optimized so that it can be completed as a model that accurately estimates the state of charge (SOC) of a battery corresponding to the battery-related data collected from the multimodal sensor.

[0054] The edge processor (120) can predict the state of health (SOH) of a battery corresponding to the battery-related data collected from the multimodal sensor by using a second machine learning model that predicts the state of health (SOH) of a battery based on battery-related data collected from the multimodal sensor. Here, the second machine learning model may be a model trained in a supervised learning manner by training data labeled with the state of health (SOH) of a battery, and inputs battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, and a first battery performance indicator including the number of charge / discharge cycles, average DOD (depth of discharge), maximum discharge current ratio, average operating temperature, elapsed time, and previous state of health (SOH) generated using the battery-related data collected from the multimodal sensor.

[0055] The edge processor (120) can train an initially configured second machine learning model using labeled training data in a supervised learning manner. Here, the initially configured second machine learning model is an initial model designed to be configured as a model capable of predicting the state of health (SOH) of a battery using battery-related data collected from a multimodal sensor, and the parameter values ​​are set to arbitrary initial values. As the initial model is trained using the aforementioned training data, the parameter values ​​are optimized so that it can be completed as a model that accurately predicts the state of health (SOH) of a battery corresponding to the battery-related data collected from the multimodal sensor.

[0056] The edge processor (120) can predict the remaining driving distance of a vehicle corresponding to the battery-related data and driving environment data collected from the multimodal sensor by using a third machine learning model that predicts the remaining driving distance of the vehicle based on battery-related data and driving environment data collected from the multimodal sensor. Here, the third machine learning model may be a model trained in a supervised learning manner by training data labeled as driving distance, which takes as input the battery-related data collected from the multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a second battery performance indicator including the state of charge (SOC) and state of health (SOH) of the battery generated using the battery-related data collected from the multimodal sensor, and driving environment data including the average energy consumption rate, driving speed, terrain information, and whether the air conditioner is used.

[0057] The edge processor (120) can train an initially configured third machine learning model using labeled training data in a supervised learning manner. Here, the initially configured third machine learning model is an initial model designed to be configured as a model capable of predicting the remaining driving distance of a vehicle using battery-related data and driving environment data collected from a multimodal sensor, and the parameter values ​​are set to arbitrary initial values. As the initial model is trained using the aforementioned training data, the parameter values ​​are optimized so that it can be completed as a model capable of accurately predicting the remaining driving distance of a vehicle using battery-related data and driving environment data collected from a multimodal sensor.

[0058] The edge processor (120) can predict the optimal charging profile of a battery corresponding to the battery-related data and battery charging control data collected from the multimodal sensor by using a fourth machine learning model that predicts the optimal charging profile of the battery based on battery-related data and battery charging control data collected from the multimodal sensor. Here, the fourth machine learning model may be a model trained in a supervised learning manner by training data that takes as input the battery-related data collected from the multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a fourth battery performance indicator including the initial state of charge (SOC) and the current state of health (SOH) of the battery using the battery-related data collected from the multimodal sensor, and battery charging control data including the target state of charge (SOC), available charging power, and desired charging completion time, and has the hourly charging current profile and hourly charging voltage profile as labels.

[0059] The edge processor (120) can train an initially configured fourth machine learning model using labeled training data in a supervised learning manner. Here, the initially configured fourth machine learning model is an initial model designed to be configured as a model capable of predicting the optimal charging profile of a battery using battery-related data and battery charging control data collected from a multimodal sensor, and the parameter values ​​are set to arbitrary initial values. As the initial model is trained through the aforementioned training data, the parameter values ​​are optimized so that it can be completed as a model capable of accurately predicting the optimal charging profile of a battery using battery-related data and battery charging control data collected from a multimodal sensor.

[0060] The edge processor (120) can detect abnormalities in the battery corresponding to the battery-related data and battery operation data collected from the multimodal sensor by using a fifth machine learning model that detects abnormalities in the battery based on battery-related data and battery operation characteristic data collected from the multimodal sensor. Here, the fifth machine learning model may be a model trained in a supervised learning manner by training data that takes as input battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a third battery performance indicator including a voltage deviation between battery cells, a temperature deviation between battery cells, a charge / discharge efficiency, and a rate of change in internal resistance generated using the battery-related data collected from the multimodal sensor, and battery operation characteristic data including the self-discharge rate of the battery, the frequency of voltage drop of the battery, and the frequency of overheating of the battery, and labels whether the abnormal state includes one of normal or abnormal, and an abnormal type including one of cell imbalance, overheating, capacity degradation, internal short circuit, and poor connection.

[0061] The edge processor (120) can train an initially configured fifth machine learning model using labeled training data in a supervised learning manner. Here, the initially configured fifth machine learning model is an initial model designed to be configured as a model capable of detecting battery abnormalities using battery-related data and battery operation characteristic data collected from a multimodal sensor, and the parameter values ​​are set to arbitrary initial values. As the initial model is trained using the aforementioned training data, the parameter values ​​are optimized so that it can be completed as a model capable of accurately detecting battery abnormalities.

[0062] The edge processor (120) can process instructions of a computer program by performing basic arithmetic, logic, and input / output operations. Meanwhile, the edge processor (120) can perform at least some of the data analysis, processing, and result information generation for performing the above-described operations using at least one of machine learning, neural network, or deep learning algorithms as a rule-based or artificial intelligence (AI) algorithm. Examples of neural networks may include models such as CNN (convolutional neural network), DNN (deep neural network), and RNN (recurrent neural network).

[0063] For example, the edge processor (120) may be implemented as an array of multiple logic gates, or as a combination of a general-purpose microprocessor and memory storing a program that can be executed on the microprocessor. For example, the processor may include a general-purpose processor, a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a controller, a microcontroller, a state machine, etc. In some environments, the edge processor (120) may include an application-specific integrated circuit (ASIC), a programmable logic device (PLD), a field programmable gate array (FPGA), etc. For example, the edge processor (120) may refer to a combination of processing devices such as a combination of a digital signal processor (DSP) and a microprocessor, a combination of multiple microprocessors, a combination of one or more microprocessors combined with a digital signal processor (DSP) core, or any other combination of such configurations.

[0064] The edge memory (130) is operably connected to the edge processor (120) and can store at least one code associated with an operation performed by the edge processor (120).

[0065] Additionally, the edge memory (130) can perform the function of temporarily or permanently storing data processed by the edge processor (120). Here, the edge memory (130) may include a magnetic storage medium or a flash storage medium, but the scope of the present invention is not limited thereto. Such edge memory (130) may include internal memory and / or external memory, and may include volatile memory such as DRAM, SRAM, or SDRAM, non-volatile memory such as OTPROM, PROM, EPROM, EEPROM, mask ROM, flash ROM, NAND flash memory, or NOR flash memory, flash drives such as SSD, CF card, SD card, Micro-SD card, Mini-SD card, xD card, or Memory Stick, or storage devices such as HDD.

[0066] Next, the cloud server (200) will be described. The cloud communication unit (210) can perform the necessary functions for transmitting and receiving data with the cloud server (200) and the edge device (100). The cloud communication unit (210) can exchange information with the edge device (100) in real time using a wireless communication protocol.

[0067] The cloud processor (220) can utilize data collected from the edge device (100) to comprehensively analyze the performance and status of the battery, establish a long-term maintenance plan, manage software updates, and provide a customized battery management solution by analyzing user and environment data. In this embodiment, the cloud processor (220) may include a predictive analysis unit (221), a system management unit (222), a user-environment analysis unit (223), and an information provision unit (224).

[0068] The prediction analysis unit (221) can predict the long-term health status and performance degradation of the battery by analyzing large-scale battery-related data collected from the edge device (100). A deep learning model may be used to predict the long-term health status and performance degradation of the battery. The prediction analysis unit (221) can diagnose the state of the battery by applying the comparison results of the data collected from the edge device (100) and fleet data to the first deep learning model. The prediction analysis unit (221) can predict the performance of the battery by applying the data collected from the edge device (100) to the second deep learning model.

[0069] The predictive analysis unit (221) can perform an anomaly pattern mining function that automatically detects anomaly patterns from data collected from the edge device (100). Through the anomaly pattern mining function, data such as voltage, current, and temperature can be analyzed in real time and abnormal behavior of the battery cell can be detected. Through the anomaly pattern mining function, data accumulated during the battery's life cycle can be analyzed and signs of battery performance degradation can be identified in advance. The anomaly patterns mined from the battery data can be used as training data for a deep learning model to improve the accuracy of the model and contribute to the development of an algorithm that can predict the state of the battery more accurately.

[0070] The system management unit (222) can remotely manage the software and settings of the edge device (100) and, when necessary, perform updates via OTA (over-the-air) to maintain system stability and provide the latest features.

[0071] The user-environment analysis unit (223) can analyze the impact of external environmental changes on battery performance and generate a battery management strategy that takes this into account. Here, external environmental changes may include temperature, humidity, atmospheric pressure, driving conditions (gradient, driving speed, road conditions, etc.), climate change, etc. The user-environment analysis unit (223) can analyze long-term user (driver) data to provide a customized battery management solution that fits the individual's driving habits and driving style.

[0072] The information provider (224) can convert complex data into an easily understandable format to generate a dashboard for administrators and a detailed analysis report. For example, the information provider (224) can generate and provide a graphic that monitors basic parameters such as the battery's voltage, current, and temperature in real time. The information provider (224) can provide energy consumption patterns by time of day, day of the week, and season as a visual graph. The information provider (224) can display a warning signal on the dashboard indicating an abnormal condition of the battery or the need for imminent maintenance. The information provider (224) can generate a battery life prediction report that predicts when replacement or maintenance is needed through a life prediction analysis based on the battery's usage history and performance data. When battery performance degradation is observed, the information provider (224) can generate and provide a detailed report analyzing the cause (e.g., overload operation, temperature change, increased internal resistance, etc.). The information provider (224) can generate and provide a report evaluating the reusability and recycling conditions of the used battery.

[0073] The cloud processor (220) can provide powerful computing resources capable of performing advanced computations and large-scale data processing than the edge processor (120). The cloud processor (220) typically includes a high-performance central processing unit (CPU), multiple graphics processing units (GPUs), or a dedicated artificial intelligence (AI) processing unit, and can perform extensive analysis, including complex deep learning algorithms, in real time.

[0074] The cloud memory (230) is operably connected to the cloud processor (220) and can store at least one code in association with an operation performed by the cloud processor (220).

[0075] Additionally, the cloud memory (230) can perform the function of temporarily or permanently storing data processed by the cloud processor (220).

[0076] Cloud memory (230) provides sufficient capacity for storing and processing large datasets and can generally use high-speed accessable dynamic random access memory (DRAM), a solid state drive (SSD) suitable for big data processing, or a more economical large-capacity hard disk drive (HDD). Cloud memory (230) may also include data replication and backup functions to ensure high availability and data integrity. This configuration enables the cloud server (200) to effectively support continuously expanding data demands and complex computational requirements.

[0077] The cloud processor (220) and cloud memory (230) can process data collected from edge devices (100) using a distributed processing system, such as a framework like Apache Spark. This allows the cloud server (200) to complement the local processing capabilities of the edge devices (100), increase the precision of data analysis and insight generation, and improve the efficiency and performance of the entire system.

[0078] FIG. 3 is a flowchart illustrating a data processing method performed in an edge device according to an embodiment of the present invention. In the following description, parts that overlap with the description of FIG. 1 and FIG. 2 will be omitted. The data processing method according to the present embodiment may be performed in an edge processor (120) included in an edge device (100). The data processed in the present embodiment may be used as input data for a machine learning model.

[0079] Referring to FIG. 3, in step S301, the edge processor (120) can collect raw data related to the battery from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, a plurality of temperature sensors, and a plurality of humidity sensors. The raw data can be sampled at a 100 Hz period. The voltage sensors measure the voltage of the battery cell, and the voltage data of the battery cell can be recorded along with a time stamp. The current sensors can measure the current indicating the charging and discharging status of the battery. The temperature sensors can measure the temperature of the battery cell and the battery module. The humidity sensors can measure the relative humidity inside the battery pack.

[0080] In step S303, the edge processor (120) can generate time-synchronized data by aligning raw battery-related data to a common time stamp. The edge processor (120) can align data collected from each sensor to a common time stamp. This allows all sensor data to share the same time reference, thereby reducing temporal errors during data analysis. The edge processor (120) can, for example, combine data from a voltage sensor with data from a temperature sensor, a current sensor, and a humidity sensor to form a single data row of all sensor measurements corresponding to each time unit (ms).

[0081] In step S305, the edge processor (120) can generate data from which outliers have been removed by applying a statistical outlier detection algorithm to time-synchronized data. The edge processor (120) can detect and remove outliers within the data set by utilizing a statistical outlier detection algorithm, for example, the Z-score method. The Z-score method calculates a standard score for each data point, and if this score exceeds a set threshold (e.g., |z|>3), the data point is considered an outlier and can be excluded from the data set. This can minimize the impact of extreme values ​​on the results while preserving the statistical characteristics of the data.

[0082] In step S307, the edge processor (120) can smooth the data from which outliers have been removed to generate data from which high-frequency noise has been removed. The edge processor (120) can remove high-frequency noise by applying a moving average filter. The edge processor (120) can group consecutive data points into a preset window size (e.g., 5) and calculate the average value within each group. These calculated average values ​​can be generated as new data points that reduce noise in the original data and increase the smoothness of the data. Data from which high-frequency noise has been removed by the edge processor (120) can provide higher reliability and accuracy when used in subsequent analysis or modeling steps.

[0083] In step S309, the edge processor (120) can normalize the data from which high-frequency noise has been removed to generate normalized data. The edge processor (120) can normalize the data from which high-frequency noise has been removed to a range between 0 and 1 by applying a Min-Max scaling method. Through this process, the data values ​​are standardized so that the relative size difference between the data collected from various sensors can be adjusted.

[0084] In step S311, the edge processor (120) can extract statistical characteristics from the normalized data. The edge processor (120) can perform the role of extracting statistical characteristics from the normalized data. Here, the statistical characteristics may include the mean, standard deviation, maximum value, and minimum value.

[0085] The S313 edge processor (120) can generate time series data from statistical characteristics. The edge processor (120) can generate time series data with a sequence length of, for example, 100, for input to a long short-term memory (LSTM) model as a machine learning model. In this embodiment, each sequence may include a time stamp and the mean, standard deviation, etc. of each feature.

[0086] In step S315, the edge processor (120) can combine time series data to generate a single input vector. The edge processor (120) can sort characteristic values ​​representing voltage, current, temperature, and humidity according to time stamps, and combine the characteristic values ​​with maximum and minimum values ​​to form a single integrated vector.

[0087] In step S317, the edge processor (120) can augment data by applying a data augmentation algorithm to a single input vector. The edge processor (120) can apply data augmentation techniques to improve the generalization performance of a machine learning model. The edge processor (120) can enhance the robustness of the model and improve prediction accuracy by adding small variations to specific data. The edge processor (120) can artificially increase the diversity of the data by adding Gaussian noise to a single input vector.

[0088] In step S319, the edge processor (120) can organize batches for the data augmentation results and input them into a machine learning model. The edge processor (120) can organize batches for training the machine learning model. Each batch contains multiple data sequences, and each sequence can be organized into a tensor form containing various features. This tensor has a specific number of sequence lengths, and this sequence may contain various features. In addition, labels to be predicted can be assigned to each batch to evaluate the performance of the machine learning model during the training process. This configuration can help the machine learning model compare actual values ​​with predicted values ​​to minimize learning errors, and consequently contribute to increasing the learning efficiency and accuracy of the model.

[0089] FIG. 4 is a flowchart illustrating the operation method of an edge-cloud collaboration-based battery state estimation system according to an embodiment of the present invention. In the following description, parts that overlap with the description of FIG. 1 to 3 will be omitted. The operation method of the battery state estimation system according to the present embodiment may be performed on an edge device (100) and a cloud server (200). This may mean that it is performed on an edge processor (120) included in the edge device (100) and a cloud processor (220) included in the cloud server (200).

[0090] Referring to FIG. 4, in step S401, the edge device (100) can collect and preprocess battery-related data from a multimodal sensor. In one embodiment, when a driver starts the vehicle, the multimodal sensor initiates sensing of various types of data, including voltage, current, temperature, and humidity, and the edge processor (120) can collect battery-related data from the multimodal sensor. The voltage sensor measures the voltage of each battery cell, and the current sensor can monitor the charging and discharging current of the battery pack. The temperature sensor checks the temperature of the battery cell and module, and the humidity sensor can measure the humidity inside the battery pack. The edge device (100) can perform preprocessing on the collected battery-related data as illustrated in FIG. 3.

[0091] In step S403, the edge device (100) can estimate the battery state based on preprocessed battery-related data. The edge device (100) can estimate the battery state in real time based on the preprocessed data by utilizing a lightweight LSTM model (e.g., a first machine learning model). The edge device (100) can estimate the initial battery state of charge (SOC). The edge device (100) can immediately display the estimated battery state on the driver's dashboard so that the driver can easily check the current battery state. The information displayed on the dashboard also includes the driving range, which can provide the driver with important driving and charging planning information.

[0092] In step S405, the edge device (100) can detect and respond to signs of battery anomalies based on preprocessed battery-related data. The edge device (100) can detect signs of battery anomalies by utilizing the aforementioned fifth machine learning model. The edge device (100) can detect sudden signs of anomalies in a specific battery cell, for example, a voltage drop. A rule-based algorithm stored in the edge device (100) can quickly process these signs of anomalies and isolate the cell. A warning message can be displayed on the driver's dashboard so that the driver can immediately recognize the vehicle's battery status and take necessary measures.

[0093] In step S407, the edge device (100) can compress the recently accumulated data (e.g., over 15 minutes) using a compression algorithm. In step S409, the edge device (100) can transmit the data compressed in step S407 to the cloud server (200). In this embodiment, the edge device (100) can encrypt the accumulated data using the AES-256 algorithm. The encryption process may be essential to ensure the confidentiality and integrity of the data. Subsequently, the edge device (100) can securely transmit the encrypted data to the cloud server (200) using the MQT (message queuing telemetry transport) protocol. In this embodiment, data in which anomalies are detected is processed with priority and can be transmitted first to the cloud server (200) via a priority queue.

[0094] In step S411, the cloud server (200) can perform in-depth analysis on the data collected from the edge device (100). The cloud server (200) can decode the data collected from the edge device (100). The cloud server (200) can perform distributed processing based on Apache Spark on the data collected from the edge device (100). The cloud server (200) can analyze the voltage drop pattern of a specific battery cell by comparing the data collected from the edge device (100) with similar case data of the entire vehicle fleet. After data analysis, the cloud server (200) can perform a precise diagnosis using a first deep learning model that combines a transformer and a convolutional neural network (CNN). The cloud server (200) can generate an analysis result of the voltage drop pattern of a battery cell (e.g., 85% probability of microcracks inside the cell) by using a first deep learning model that can generate an analysis result of the voltage drop pattern of a battery cell corresponding to the comparison result of the data collected from the edge device (100) and the similar case data of the entire vehicle fleet as input. Here, the first deep learning model may be a model trained in a supervised learning manner by training data that takes the comparison result of the data collected from the edge device (100) and the similar case data of the entire vehicle fleet as input and labels the analysis result of the voltage drop pattern of the battery cell.

[0095] In step S413, the cloud server (200) continuously monitors the battery status based on data collected from the edge device (100) and can predict battery performance degradation through in-depth analysis of the data. The cloud server (200) can predict battery performance degradation (e.g., battery performance decreasing by 20% within the next 3 months) by combining the vehicle's past 6 months of operational data and current status data. The cloud server (200) can predict battery performance corresponding to the data collected from the edge device (100) (past 6 months of operational data and current status data) by using a second deep learning model that predicts battery performance using the data collected from the edge device (100) (past 6 months of operational data and current status data). Here, the second deep learning model may be a model trained in a supervised learning manner using training data labeled with battery performance, with the data collected from the edge device (100) including past 6 months of operational data and current status data as input. The cloud server (200) can establish an optimal response strategy based on these prediction results. The cloud server (200) can establish a response strategy that, for example, minimizes performance degradation by adjusting cell balancing in the short term, and, for example, recommends visiting a repair shop within two weeks in the medium term. Additionally, the cloud server (200) can establish a response strategy that, for example, recommends replacing the battery module in the long term.

[0096] In steps S415 and S417, the cloud server (200) can generate an update package to be applied to the edge device (100) to take necessary measures based on the analysis results to date, and transmit the update package to the edge device (100). In this embodiment, the update package may include parameters of a cell balancing algorithm, parameters of a machine learning model, and content for adjusting thresholds. The cloud server (200) can transmit an update package of a preset size (e.g., 500 KB) to the edge device (100) using a differential update method. By ensuring that only the parts that have changed compared to the software previously installed on the edge device (100) are updated, data usage can be minimized and the update speed can be increased. In one embodiment, the cloud server (200) can transmit the update package to the edge device (100) via a wireless communication method. For example, the cloud server (200) can transmit the update package to the edge device (100) via OTA (over-the-air).

[0097] In steps S419 and S421, the edge device (100) can complete the application of the update and immediately activate the newly introduced cell balancing algorithm. In this process, the edge device (100) can redistribute the load on cells surrounding a specific cell to more efficiently optimize the energy usage of the entire battery pack. After the update, the edge device (100) can recalculate the driving range using a third machine learning model and provide the driver with updated driving range information. Additionally, the edge device (100) can display a message on the driver's dashboard recommending a visit to a repair shop based on the battery status and recent updates.

[0098] In step S423, the edge device (100) and the cloud server (200) can perform continuous monitoring and learning. The edge device (100) can continuously collect and analyze various data related to the battery even while the vehicle is in operation. This allows for real-time monitoring of the battery's condition and immediate reflection of necessary information. The edge device (100) can continuously fine-tune the installed lightweight machine learning model using online learning techniques. This may be a process to improve the model's performance through new data and ensure real-time response. The cloud server (200) can perform in-depth analysis using the data received from the edge device (100). This in-depth analysis can play an important role in predicting the long-term performance and health status of the battery. The data processed by the cloud server (200) can be used to periodically update a global model for the entire vehicle fleet.

[0099] Although the present invention has been described above by limited embodiments and drawings, the present invention is not limited thereto, and it is obvious that various modifications and variations are possible within the scope of the technical spirit of the present invention and the equivalent scope of the claims described below by those skilled in the art to which the present invention belongs. Explanation of the symbols

[0100] 100: Edge device 200: Cloud Server 300: Network

Claims

Claim 1 A battery state estimation system based on edge-cloud collaboration, comprising: a plurality of edge devices implemented as a battery management system (BMS) that monitors and manages the state of a battery; and a cloud server that exchanges data with the plurality of edge devices and dynamically manages the performance of the battery based on data collected from the plurality of edge devices, wherein the edge device is configured to estimate the state of the battery by applying battery-related data collected from a multimodal sensor to a machine learning model, generate one or more of a warning signal and an action signal when an abnormal sign is detected in the battery because the battery-related data exceeds a threshold value, and transmit aggregated data to the cloud server; and wherein the cloud server is configured to diagnose the state of the battery by applying the comparison result of the data collected from the edge device and fleet data to a first deep learning model, predict the performance of the battery by applying the data collected from the edge device to a second deep learning model, establish a response strategy based on the battery performance prediction result, and generate an edge device update package based on the aggregated data and transmit it to the edge device. Claim 2 A battery state estimation system according to claim 1, wherein the edge device is further configured to perform an update that adjusts the parameters of a cell balancing algorithm, the parameters of a machine learning model, and the threshold value using the update package received from the cloud server, and to recalculate the driving range of the battery based on the result of performing the update. Claim 3 A battery state estimation system according to claim 1, wherein the edge device collects raw battery-related data from the multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, a plurality of temperature sensors, and a plurality of humidity sensors before applying the battery-related data to a machine learning model; generates time-synchronized data by aligning the raw battery-related data with a common time stamp; generates data with outliers removed by applying a statistical outlier detection algorithm to the time-synchronized data; generates data with high-frequency noise removed by smoothing the data with outliers removed; generates normalized data by normalizing the data with high-frequency noise removed; generates time-series data by extracting statistical characteristics from the normalized data; generates a single input vector by combining the time-series data; and generates input data for the machine learning model by applying a data augmentation algorithm to the single input vector. Claim 4 In claim 1, the edge device is further configured to estimate the state of charge (SOC) of the battery corresponding to the battery-related data collected from the multimodal sensor by using a first machine learning model that estimates the state of charge (SOC) of the battery based on battery-related data collected from the multimodal sensor before transmitting the aggregated data to the cloud server, and the first machine learning model is a model trained in a supervised learning manner by training data having the state of charge (SOC) of the battery as a label, and battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors as input. Claim 5 In claim 1, the edge device is further configured to predict the state of health (SOH) of the battery corresponding to the battery-related data collected from the multimodal sensor using a second machine learning model that predicts the state of health (SOH) of the battery based on battery-related data collected from the multimodal sensor before transmitting the aggregated data to the cloud server, and the second machine learning model is a model trained in a supervised learning manner by training data labeled as the state of health (SOH) of the battery, and inputs battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, and a first battery performance indicator including the number of charge / discharge cycles, average DOD (depth of discharge), maximum discharge current ratio, average operating temperature, elapsed time, and previous state of health (SOH) generated using the battery-related data collected from the multimodal sensor. Claim 6 In claim 1, the edge device is further configured to predict the remaining driving distance of the vehicle corresponding to the battery-related data and driving environment data collected from the multimodal sensor using a third machine learning model that predicts the remaining driving distance of the vehicle based on battery-related data and driving environment data collected from the multimodal sensor before transmitting the aggregated data to the cloud server, and the third machine learning model is a model trained in a supervised learning manner by training data labeled as driving distance, which takes as input battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a second battery performance indicator including the state of charge (SOC) and state of health (SOH) of the battery generated using the battery-related data collected from the multimodal sensor, and driving environment data including an average energy consumption rate, driving speed, terrain information, and whether an air conditioner is used. Claim 7 In claim 1, the edge device is further configured to predict an optimal charging profile of the battery corresponding to the battery-related data and battery charging control data collected from the multimodal sensor using a fourth machine learning model that predicts an optimal charging profile of the battery based on battery-related data and battery charging control data collected from the multimodal sensor before transmitting the aggregated data to the cloud server, and the fourth machine learning model is a model trained in a supervised learning manner by training data having hourly charging current profiles and hourly charging voltage profiles as labels, and inputs battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a fourth battery performance indicator including the initial state of charge (SOC) and the current state of health (SOH) of the battery using the battery-related data collected from the multimodal sensor, and battery charging control data including a target state of charge (SOC), available charging power, and desired charging completion time. Claim 8 In claim 1, the edge device is further configured to detect an abnormality of the battery corresponding to the battery-related data and battery operation data collected from the multimodal sensor by using a fifth machine learning model that detects an abnormality of the battery based on battery-related data and battery operation characteristic data collected from the multimodal sensor before transmitting the aggregated data to the cloud server, and the fifth machine learning model is a model trained in a supervised learning manner by training data having a label of an abnormal state including one of normal or abnormal, and an abnormal type including one of cell imbalance, overheating, capacity degradation, internal short circuit, and connection failure, and having inputs battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a third battery performance indicator including a voltage deviation between battery cells, a temperature deviation between battery cells, charge / discharge efficiency, and a rate of change in internal resistance generated using the battery-related data collected from the multimodal sensor, and battery operation characteristic data including a self-discharge rate of the battery, a voltage drop frequency of the battery, and an overheating frequency of the battery. Claim 9 A battery state estimation system according to claim 1, wherein the cloud server is further configured to generate an edge device update package and transmit it to the edge device, and then, based on data collected from the edge device, analyze the impact of the external environment on the performance of the battery, analyze the user's long-term habits and driving style to establish a customized battery management strategy, generate a dashboard and analysis report provided to an administrator, and evaluate the reusability or recycling potential of the battery. Claim 10 A method of operation of a battery state estimation system comprising a plurality of edge devices implemented as a battery management system (BMS) that monitors and manages the state of a battery, and a cloud server that exchanges data with the plurality of edge devices and dynamically manages the performance of the battery based on data collected from the plurality of edge devices, the method comprising: a step in which the edge device applies battery-related data collected from a multimodal sensor to a machine learning model to estimate the state of the battery, generates one or more of a warning signal and an action signal when an abnormal sign is detected in the battery because the battery-related data exceeds a threshold value, and transmits the aggregated data to the cloud server; and a step in which the cloud server applies the comparison result of the data collected from the edge device and fleet data to a first deep learning model to diagnose the state of the battery, applies the data collected from the edge device to a second deep learning model to predict the performance of the battery, establishes a response strategy based on the battery performance prediction result, and generates an edge device update package based on the aggregated data and transmits it to the edge device. Claim 11 A method of operation of a battery state estimation system according to claim 10, further comprising the step of the edge device performing an update to adjust the parameters of a cell balancing algorithm, the parameters of a machine learning model, and the threshold value using the update package received from the cloud server, and recalculating the driving range of the battery based on the result of performing the update. Claim 12 A method of operation of a battery state estimation system according to claim 10, comprising the steps of: collecting raw battery-related data from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, a plurality of temperature sensors, and a plurality of humidity sensors before the edge device applies the battery-related data to a machine learning model; aligning the raw battery-related data by a common time stamp to generate time-synchronized data; applying a statistical outlier detection algorithm to the time-synchronized data to generate outlier-removed data; smoothing the outlier-removed data to generate high-frequency noise-removed data; normalizing the high-frequency noise-removed data to generate normalized data; extracting statistical characteristics from the normalized data to generate time-series data; combining the time-series data to generate a single input vector; and applying a data augmentation algorithm to the single input vector to generate input data for the machine learning model. Claim 13 A method of operation of a battery state estimation system according to claim 10, further comprising the step of estimating the state of charge (SOC) of a battery corresponding to the battery-related data collected from the multimodal sensor using a first machine learning model that estimates the state of charge (SOC) of a battery based on battery-related data collected from the multimodal sensor before the edge device transmits the aggregated data to the cloud server, wherein the first machine learning model is a model trained in a supervised learning manner by training data having the state of charge (SOC) of a battery as a label, and having battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors as input. Claim 14 A method of operation of a battery state estimation system according to claim 10, further comprising the step of predicting the state of health (SOH) of the battery corresponding to the battery-related data collected from the multimodal sensor using a second machine learning model that predicts the state of health (SOH) of the battery based on battery-related data collected from the multimodal sensor before the edge device transmits the aggregated data to the cloud server, wherein the second machine learning model is a model trained in a supervised learning manner by training data labeled as the state of health (SOH) of the battery, and inputs battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, and a first battery performance indicator including the number of charge / discharge cycles, average DOD (depth of discharge), maximum discharge current ratio, average operating temperature, elapsed time, and previous state of health (SOH) generated using the battery-related data collected from the multimodal sensor. Claim 15 A method of operation of a battery state estimation system according to claim 10, wherein the edge device further includes the step of predicting the remaining driving distance of the vehicle corresponding to the battery-related data and driving environment data collected from the multimodal sensor using a third machine learning model that predicts the remaining driving distance of the vehicle based on battery-related data and driving environment data collected from the multimodal sensor before transmitting the aggregated data to the cloud server, wherein the third machine learning model is a model trained in a supervised learning manner by training data having a driving range as a label, and inputs battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a second battery performance indicator including the state of charge (SOC) and state of health (SOH) of the battery generated using the battery-related data collected from the multimodal sensor, and driving environment data including an average energy consumption rate, driving speed, terrain information, and whether an air conditioner is used. Claim 16 A method of operation of a battery state estimation system according to claim 10, further comprising the step of predicting an optimal charging profile of a battery corresponding to the battery-related data and battery charging control data collected from the multimodal sensor using a fourth machine learning model that predicts an optimal charging profile of a battery based on battery-related data and battery charging control data collected from the multimodal sensor before the edge device transmits the aggregated data to the cloud server, wherein the fourth machine learning model is a model trained in a supervised learning manner by training data having hourly charging current profiles and hourly charging voltage profiles as labels, and inputs battery-related data collected from a multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a fourth battery performance indicator including the initial state of charge (SOC) and the current state of health (SOH) of the battery using the battery-related data collected from the multimodal sensor, and battery charging control data including a target state of charge (SOC), available charging power, and desired charging completion time. Claim 17 A method of operation of a battery state estimation system according to claim 10, further comprising the step of detecting an abnormality of a battery corresponding to the battery-related data and battery operation data collected from the multimodal sensor using a fifth machine learning model that detects an abnormality of a battery based on battery-related data and battery operation characteristic data collected from the multimodal sensor before the edge device transmits the aggregated data to the cloud server, wherein the fifth machine learning model is a model trained in a supervised learning manner by training data having a label of an abnormal state including one of normal or abnormal, and an abnormal type including one of cell imbalance, overheating, capacity degradation, internal short circuit, and connection failure, and having inputs battery-related data collected from the multimodal sensor including a plurality of voltage sensors, a plurality of current sensors, and a plurality of temperature sensors, a third battery performance indicator including a voltage deviation between battery cells, a temperature deviation between battery cells, charge / discharge efficiency, and a rate of change in internal resistance generated using the battery-related data collected from the multimodal sensor, and battery operation characteristic data including the self-discharge rate of the battery, the frequency of voltage drop of the battery, and the frequency of overheating of the battery. Claim 18 A method of operation of a battery state estimation system according to claim 10, further comprising, after the step of transmitting to the edge device, the step of the cloud server analyzing the impact of the external environment on the performance of the battery based on data collected from the edge device, establishing a customized battery management strategy by analyzing the user's long-term habits and driving style, generating a dashboard and analysis report provided to an administrator, and evaluating the reusability or recycling potential of the battery. Claim 19 A computer-readable recording medium storing a computer program for executing the method of claim 10 using a computer.

Images