Automatic battery jump method and automatic jump device based on battery data detection

Abstract

The application provides a battery data detection-based automatic battery cross-connection method and automatic battery cross-connection device, which comprises the following steps: obtaining battery position information of each battery unit in a battery pack and collecting battery parameters of each battery unit; preprocessing the battery parameters and extracting static parameters and dynamic parameters; constructing a characteristic fuzzy relation matrix based on the battery characteristic parameters, and performing fuzzy comprehensive evaluation on the battery units according to the characteristic fuzzy relation matrix; inputting abnormal battery characteristic parameters of an abnormal battery unit into a preset battery fault identification model to identify a fault battery unit in the abnormal battery unit; and if the battery fault identification model identifies the fault battery unit, generating an automatic cross-connection task of the automatic battery cross-connection device, and controlling the automatic battery cross-connection device to realize automatic cross-connection processing on the fault battery unit. The application can automatically detect and cross-connect fault batteries with high efficiency.

Description

Technical Field

[0001] This invention belongs to the field of battery fault detection and processing technology, specifically relating to an automatic battery jump-connection method and device based on battery data detection. Background Technology

[0002] In power systems, smart substations use battery banks as emergency power sources for critical equipment, ensuring the system can maintain normal operation when the main power source fails. Battery banks require long-term management and maintenance. If a single battery in the bank fails, it reduces the overall capacity and output voltage of the battery bank, potentially causing it to fail to meet the normal power supply needs of substation equipment in emergencies. Therefore, it is necessary to promptly perform jumper operations on faulty individual batteries in the battery bank. By timely jumpering, the battery bank can continue to operate at near-normal performance levels, ensuring its reliability as a backup power source.

[0003] Traditional battery management relies mainly on manual inspections and operations, which proves inadequate for large-scale, distributed battery systems. Firstly, manual inspections have limited frequency and coverage, making it difficult to detect potential faults promptly. Even if a faulty individual battery is detected manually, it requires manual jump-connection, which is susceptible to human error. Summary of the Invention

[0004] This invention provides an automatic battery jump-connection method and device based on battery data detection to solve the problems of low efficiency and high safety risks when manually inspecting and jumping faulty batteries.

[0005] In a first aspect, the present invention provides an automatic battery jump-connection method based on battery data detection, applicable to a smart substation equipped with battery banks and an automatic jump-connection device. The battery banks are equipped with a smart sensor network. The automatic jump-connection device includes an insulating shell and a mobile chassis. The insulating shell is fixedly mounted on the top of the mobile chassis. The automatic jump-connection device also includes a main control module, a jump-connection protection module, and a jump-connection execution module. The main control module is connected to the mobile chassis, the jump-connection protection module, and the jump-connection execution module respectively. The jump-connection protection module is electrically connected to the jump-connection execution module. The main control module is used to receive automatic jump-connection tasks and generate automatic jump-connection instructions according to the automatic jump-connection tasks. The mobile chassis responds to the automatic jump-connection instructions and moves to the location of the faulty battery cell. The jump-connection execution module responds to the automatic jump-connection instructions and completes the automatic jump-connection process for the faulty battery cell. The jump-connection protection module responds to the automatic jump-connection instructions and protects the jump-connection execution module during the automatic jump-connection process.

[0006] The method includes the following steps:

[0007] Obtain the battery location information of each battery cell in the battery pack, and collect the battery parameters of each battery cell through a smart sensor network;

[0008] Preprocess battery parameters, extract static and dynamic parameters from the preprocessed battery parameters, and integrate the static and dynamic parameters into battery characteristic parameters;

[0009] A feature fuzzy relation matrix is ​​constructed based on battery feature parameters. Fuzzy comprehensive evaluation of the battery unit is performed based on the feature fuzzy relation matrix to obtain the battery status score of each battery unit.

[0010] Battery cells whose battery condition score is lower than a preset score threshold are marked as abnormal battery cells;

[0011] The abnormal battery characteristic parameters of the abnormal battery unit are input into the preset battery fault identification model. The faulty battery unit in the abnormal battery unit is identified by the battery fault identification model, which is built based on support vector machine.

[0012] If the battery fault identification model identifies a faulty battery cell, it generates an automatic jumper task by combining the battery location information and static parameters of the faulty battery cell. The automatic jumper task controls the automatic jumper to achieve automatic jumper processing of the faulty battery cell.

[0013] Optionally, the method may also include the following steps:

[0014] If the battery fault identification model fails to identify the faulty battery cell, then all battery characteristic parameters are converted into multi-dimensional characteristic time series data.

[0015] A battery life prediction model was constructed by combining long short-term memory networks and graph convolutional networks.

[0016] Multidimensional feature time series data is input into the battery life prediction model. Based on the graph network construction module in the battery life prediction model, a battery influence correlation graph of the battery unit is generated through a data-driven method. The graph network construction module is composed of a long short-term memory network layer and a self-attention mechanism layer. The graph nodes in the battery influence correlation graph are all battery units, and the graph node edges in the battery influence correlation graph represent the physical connection relationship or electrical coupling relationship between battery units.

[0017] High-dimensional impact features are extracted from the battery impact correlation graph through the feature extraction module in the battery life prediction model. The feature extraction module includes a graph convolutional layer, a graph domain transformation layer, a long short-term memory network layer, and a graph domain inverse transformation layer.

[0018] The high-dimensional impact features are reduced to low-dimensional impact features using the dimensionality reduction prediction module in the battery life prediction model. The fully connected layer in the dimensionality reduction prediction module generates life prediction estimates based on the low-dimensional impact features and reconstructs the low-dimensional impact features. The reconstructed low-dimensional impact features are then backpropagated to the graph network construction module, so that the graph network construction module can perform residual fusion with the subsequently input multi-dimensional feature time series data and the reconstructed low-dimensional impact features.

[0019] Identify the aging battery cell with the lowest life prediction estimate among all abnormal battery cells. If the life prediction estimate of the aging battery cell is lower than the preset lifespan threshold, mark the aging battery cell as a faulty battery cell.

[0020] An automatic jump-connection task is generated by combining the battery location information and static parameters of the faulty battery unit. The automatic jump-connection task controls the automatic jump-connection device to achieve automatic jump-connection processing of the faulty battery unit.

[0021] Optionally, preprocessing the battery parameters, extracting static and dynamic parameters from the preprocessed battery parameters, and integrating the static and dynamic parameters into battery characteristic parameters includes the following steps:

[0022] Outliers in battery parameters are processed using a median filtering method based on an adaptive window mechanism, and missing values ​​in battery parameters are filled using a multiple interpolation method to obtain the preprocessed target battery parameters.

[0023] Static and dynamic parameters are extracted from the target battery parameters according to the preset parameter selection rules. Both static and dynamic parameters are time-series data.

[0024] After performing time alignment processing on the static and dynamic parameters, for each parameter timestamp, the static and dynamic parameters belonging to the parameter timestamp are merged into a multidimensional feature vector;

[0025] The multidimensional feature vectors of all parameter timestamps are integrated into battery feature parameters.

[0026] Optionally, static parameters include open-circuit voltage data, internal resistance data, and operating temperature data, while dynamic parameters include capacity decay rate, charging curve characteristics, discharging curve characteristics, charging efficiency, self-discharge rate, coulombic efficiency, internal resistance growth rate, and cycle life decay rate.

[0027] Optionally, a feature fuzzy relation matrix is ​​constructed based on battery feature parameters. Fuzzy comprehensive evaluation is then performed on the battery cells based on this matrix to obtain the battery status score for each cell. This process includes the following steps:

[0028] The parameter weights of static and dynamic parameters are calculated by combining the analytic hierarchy process (AHP) and the entropy weight method, respectively.

[0029] The step function is used as the membership function of the static parameter, and the sigmoid function is used as the membership function of the dynamic parameter. Based on the battery characteristic parameters and combined with the parameter weights and membership functions, a feature fuzzy relation matrix is ​​constructed.

[0030] The battery cells are evaluated using a fuzzy comprehensive method based on the feature fuzzy relation matrix to obtain the battery status score for each battery cell.

[0031] In a second aspect, the present invention also provides an automatic jumper device according to the first aspect. The device includes an insulating shell and a movable chassis. The insulating shell is fixedly disposed on the top of the movable chassis. The insulating shell is divided into multiple isolation chambers. A cable sealing sleeve is provided at the cable crossing point between any two isolation chambers. Each isolation chamber is equipped with an independent grounding system. The movable chassis is used to provide basic support for the insulating shell and also to drive the insulating shell to move.

[0032] The device also includes a main control module, a power management module, a jumper protection module, and a jumper execution module. The main control module, power management module, and jumper protection module are deployed in different isolated chambers in the insulating housing according to different voltage levels. The power management module is connected to the main control module, the jumper protection module, and the jumper execution module respectively, and supplies power to the main control module, the jumper protection module, and the jumper execution module. The jumper execution module is located on the outside of the insulating housing.

[0033] The main control module is connected to the mobile chassis, the jumper protection module, and the jumper execution module. The jumper protection module is electrically connected to the jumper execution module. The main control module is used to receive automatic jumper tasks and generate automatic jumper commands based on the automatic jumper tasks. The mobile chassis responds to the automatic jumper commands and moves to the location of the faulty battery unit. The jumper execution module responds to the automatic jumper commands and completes the automatic jumper processing of the faulty battery unit. The jumper protection module responds to the automatic jumper commands and protects the jumper execution module during the automatic jumper processing.

[0034] Optionally, the main control module includes a central processing unit, a positioning unit, and a communication unit. Both the positioning unit and the communication unit are connected to the central processing unit. The positioning unit is used to acquire the real-time location information of the automatic jump-connection device, and the communication unit is used to receive the automatic jump-connection task. The central processing unit is used to parse the battery location information and static parameters of the faulty battery unit from the automatic jump-connection task. The central processing unit is used to generate automatic jump-connection instructions by combining the real-time location information, battery location information, and static parameters. The automatic jump-connection instructions include a movement instruction, a protection module configuration instruction, and a jump-connection execution instruction, which are executed sequentially. The movement instruction is generated by combining the real-time location information and battery location information and using a path planning algorithm. The protection module configuration instruction is generated based on static parameters. The communication unit is used to send the movement instruction to the mobile chassis, the protection module configuration instruction to the jump-connection protection module, and the jump-connection execution instruction to the jump-connection execution module.

[0035] Optionally, the bridging protection module includes an array switching controller, a relay array, a diode array, and a drive circuit. The relay array includes multiple solid-state relays connected in parallel, and the diode array includes multiple Schottky diodes connected in parallel. The relay array is used to reduce current surges during automatic bridging, and the diode array is used to prevent reverse current and voltage spikes during automatic bridging. The relay array and diode array form a bridging protection circuit that is electrically connected to the bridging execution module to form a bridging loop. The bridging protection circuit is electrically connected to the array switching controller through the drive circuit. The array switching controller is connected to a communication unit. The array switching controller is used to receive protection module configuration instructions and adjust the number of solid-state relays connected in parallel in the relay array and the number of Schottky diodes connected in parallel in the diode array according to the protection module configuration instructions.

[0036] Optionally, the automatic jumper also includes a jumper monitoring module. The jumper monitoring module is deployed in an isolated chamber within an insulated housing. The power management module is connected to and supplies power to the jumper monitoring module. The jumper monitoring module includes a voltage monitoring unit, a current monitoring unit, a temperature monitoring unit, and a data processing unit. The voltage monitoring unit, current monitoring unit, and temperature monitoring unit are all connected to the data processing unit, which is connected to the main control module. The voltage monitoring unit and current monitoring unit are used to monitor the jumper voltage and jumper current data of the jumper circuit during the automatic jumper process. The temperature monitoring unit is used to monitor the device temperature data of the automatic jumper. The data processing unit is used to monitor the data changes of the jumper voltage, jumper current, and device temperature data. When the jumper voltage and jumper current data change abruptly or the device temperature data exceeds a preset temperature threshold, the data processing unit is also used to send an alarm message to the main control module. The main control module is also used to generate a jumper interruption command based on the alarm message and send the jumper interruption command to the jumper execution module so that the jumper execution module interrupts the automatic jumper process for the faulty battery unit.

[0037] Optionally, the jumper execution module includes a microprocessor, a vision sensing unit, a servo motor, a linear actuator, and a robotic arm. The robotic arm is a six-degree-of-freedom articulated structure with a battery clamp at its end. The vision sensing unit, servo motor, and linear actuator are all connected to the microprocessor, which is connected to a communication unit. The microprocessor receives jumper execution instructions and generates image acquisition instructions based on these instructions. The vision sensing unit responds to the image acquisition instructions and acquires battery image information of the faulty battery unit. The microprocessor also identifies and locates the jumper interface on the faulty battery unit using image recognition technology and generates robotic arm operation instructions based on the position of the jumper interface. The servo motor and linear actuator simultaneously respond to the robotic arm operation instructions and collaboratively control the robotic arm to connect the battery clamp to the jumper interface.

[0038] The beneficial effects of this invention are:

[0039] This invention utilizes an intelligent sensor network to collect battery parameters from each cell in a battery pack in real time. Combined with preprocessing, feature extraction, and fuzzy comprehensive evaluation techniques, it achieves accurate assessment of the battery's condition. This data-driven approach significantly improves the accuracy and timeliness of fault detection, enabling the identification of potential problem cells at an early stage. Secondly, by constructing a battery fault identification model based on support vector machines, the system can intelligently distinguish between abnormal states and actual faults, reducing the probability of false alarms and missed alarms. This not only improves the accuracy of fault diagnosis but also reduces unnecessary maintenance costs. Furthermore, this invention seamlessly integrates fault identification with automatic jump-connection processing. Once a faulty cell is identified, the system automatically generates a jump-connection task and controls the automatic jump-connection device to execute it. This fully automated processing significantly improves response speed, reduces the need for manual intervention, and thus lowers the risk of human error. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the device structure of the automatic bridging device in one embodiment of this application.

[0041] Figure 2 This is a schematic diagram of the instruction transmission of the main control unit in one embodiment of this application.

[0042] Figure 3 This is a schematic diagram of the module structure of some modules in the automatic bridging device in one embodiment of this application.

[0043] Figure 4 This is a flowchart illustrating an automatic battery jump-connection method based on battery data detection in one embodiment of this application.

[0044] Figure 5 This is a schematic diagram of the battery life prediction model in one embodiment of this application.

[0045] Explanation of reference numerals in the attached figures:

[0046] 1. Insulating outer shell; 11. Isolation chamber; 2. Mobile chassis; 3. Main control module; 4. Power management module; 5. Jumper protection module; 6. Jumper execution module; 7. Jumper monitoring module; 8. Jumper circuit. Detailed Implementation

[0047] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0048] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0049] This invention discloses an automatic jumper device, with reference to Figure 1 The automatic jumper device of this invention adopts a unique partitioned isolation design concept. Its core lies in ensuring the safety and reliability of the equipment through the combined structure of the insulating shell 1 and the mobile chassis 2. Specifically, the insulating shell 1 is made of high-strength engineering plastic material (such as modified PC / ABS material), which not only has excellent insulation performance but can also withstand mechanical impacts during daily operation. The interior of the shell is carefully divided into multiple independent isolation chambers 11, a design inspired by the explosion-proof design concept of power equipment. Each isolation chamber 11 is equipped with an independent grounding system, using a copper grounding busbar with a grounding resistance of no more than 4 ohms, so that even if a fault occurs in one chamber, it will not affect other chambers. Special cable sealing sleeves are installed at the cable crossings between chambers. These sealing sleeves adopt a three-layer structure design: an outer metal protective layer, a middle silicone sealing layer, and an inner flame-retardant insulating layer, effectively preventing interference and fault propagation between different voltage levels. The mobile chassis 2 adopts an aluminum alloy frame structure and is equipped with four omnidirectional wheels, each wheel with an independent drive motor and steering motor, achieving flexible omnidirectional movement. The chassis has a load-bearing capacity of no less than 200 kg, a maximum moving speed of 1 m / s, and a positioning accuracy of ±5 mm. This structural design not only ensures the safety of equipment operation but also greatly improves its mobility and adaptability. In practical applications, such as in a standard battery room, the device can move flexibly through narrow passages, accurately locate the faulty battery, and ensure that all functional modules are always in a safe and isolated state.

[0050] like Figure 1As shown, the automatic jumper device adopts a modular design, mainly comprising four core modules: main control module 3, power management module 4, jumper protection module 5, and jumper execution module 6. The importance of this modular design lies in achieving functional decoupling and zoned management of safety levels. The power management module 4, as the energy hub of the entire system, employs a multi-level power architecture. Internally, it includes a main power supply unit (typically a 24V lithium battery pack with a capacity of not less than 100Ah) and multiple DC-DC conversion circuits. The conversion circuits can provide multiple voltage levels of output: for example, providing a stable 5V power supply to the main control module 3, a 12V operating voltage to the jumper protection module 5, and a 24V drive voltage to the jumper execution module 6. Each output is equipped with overvoltage protection, overcurrent protection, and short-circuit protection circuits, with a response time of less than 10ms. The arrangement of each module within the insulating housing 1 follows electromagnetic compatibility design principles. The main control module 3 is typically deployed in the middle-level isolation chamber 11, away from the power components that may generate strong electromagnetic interference. The power management module 4 is located on the lower level for easy heat dissipation and wiring. The jumper protection module 5 is located in the upper chamber near the jumper execution module 6 to reduce the control signal transmission distance and improve response speed. Communication between all modules utilizes fiber optic isolation technology, achieving a communication rate of up to 10Mbps and ensuring reliable signal transmission. Power cables are double-shielded, with an outer braided shielding layer and an inner aluminum foil shielding layer, providing a shielding effect better than 60dB. All cable routing is managed using dedicated cable trays and maintained at a distance of at least 15cm from signal lines.

[0051] In practical applications, such as when the system needs to bridge a group of faulty batteries, power management module 4 first performs a power assessment to ensure the system has sufficient energy reserves. If abnormal power consumption of a module is detected (exceeding 120% of the rated value), power management module 4 immediately activates a protection mechanism, first issuing a warning signal, and if the situation persists, cutting off the power supply to the corresponding module. This multi-layered protection mechanism ensures the system can operate safely and stably under various operating conditions. The effectiveness of this modular design is mainly reflected in three aspects:

[0052] Enhanced safety: By combining physical and electrical isolation, the spread of faults is effectively prevented;

[0053] Easy maintenance: Each module can be maintained and replaced independently, greatly reducing maintenance time and costs;

[0054] System reliability: Even if one module fails, other modules can still continue to work normally, improving the system's fault tolerance.

[0055] In one embodiment, reference is made to Figure 2The main control module 3, acting as the brain of the automatic jump-connection device, adopts a layered architecture design, comprising a central processing unit, a positioning unit, and a communication unit to form a complete intelligent decision-making system. The central processing unit uses a high-performance ARM architecture with a main frequency of no less than 1.2GHz and a memory capacity of no less than 2GB, and is equipped with a real-time operating system to ensure real-time instruction processing. The positioning unit integrates an indoor UWB positioning system and an inertial navigation module. The UWB positioning accuracy reaches ±10cm, and the inertial navigation module's gyroscope accuracy is better than 0.1° / h, and the accelerometer accuracy is better than 0.1mg. The communication unit supports Wi-Fi 6, industrial Ethernet, and 4G / 5G communication to ensure reliable data transmission. In actual operation, the communication unit first receives JSON format task data sent by the automatic jump-connection system, including faulty battery ID, location coordinates, specifications, and other information. The central processing unit then parses this data, extracting the battery's three-dimensional coordinate position information, battery model, rated voltage, and other static parameters, as well as task priority information.

[0056] Based on the analysis results, the main control module 3 generates three types of core instructions through the central processing unit: 1. Movement instructions: Path planning is performed using the A* algorithm to generate a sequence of segmented path point coordinates; 2. Protection module configuration instructions: Protection levels and thresholds are set based on battery parameters; 3. Cross-connection execution instructions configure specific operating parameters. Finally, instructions are distributed to each execution unit through different communication protocols. The mobile chassis 2 uses a real-time Ethernet protocol with a cycle of 20ms; the protection module uses a CAN bus with a rate of 500kbps; and the execution module uses the EtherCAT protocol with a synchronization cycle of 1ms. Taking a typical 48V battery cross-connection task as an example, after receiving the battery position coordinates and parameters, the main control module 3 plans the optimal path to avoid obstacles, limits the protection current to within 100A, and generates a precise robotic arm operation sequence. This intelligent control system can achieve a comprehensive positioning accuracy of ±5cm, a task response delay of no more than 200ms, and ensures operational safety through multiple protection mechanisms, adapting to the cross-connection needs of various battery models.

[0057] In one implementation, when the mobile chassis 2 receives a movement command, its built-in positioning system first activates the GPS module, combining it with RTK technology to achieve centimeter-level positioning accuracy. Simultaneously, the inertial measurement unit (IMU) begins operation, providing precise attitude information via accelerometers and gyroscopes. The chassis's obstacle avoidance sensor system, such as lidar, ultrasonic sensor arrays, or stereo vision cameras, begins scanning the surrounding environment. The electric wheel hubs begin moving according to the planned path; each hub can independently control its speed and direction, achieving precise motion control. During movement, the obstacle avoidance system continues to operate, adjusting the path in real time if an obstacle is detected. Throughout the navigation process, the device continuously communicates with the main control module 3, reporting its current position and status, and receiving possible path update commands. When the device reaches the vicinity of the target location, a precise positioning program is initiated, using a visual recognition system (such as a deep learning-based target detection algorithm) to accurately identify the location of the faulty battery, ensuring a positioning error of less than 1 centimeter.

[0058] In one embodiment, reference is made to Figure 3 The jumper protection module 5 adopts a "multiple redundancy, dynamic adjustment" design concept, achieving comprehensive protection through array-type device configuration and intelligent control strategies. The array switching controller uses a 32-bit microcontroller with a main frequency of no less than 100MHz, an integrated 12-bit ADC, a sampling rate of no less than 1MSPS, supports multiple communication interfaces, and has a response time of less than 1ms. The relay array uses a 10-channel parallel solid-state relay structure, with a single relay rated current of 100A, a switching time of less than 100μs, built-in overheat protection, a withstand voltage of no less than 1000V, and a typical on-state resistance of 0.5mΩ. The diode array uses 12 channels of parallel Schottky diodes, with a single diode rated current of 80A, a forward voltage drop of no more than 0.4V, a reverse breakdown voltage of 1000V, a reverse recovery time of less than 50ns, and an operating temperature range of -55℃ to +175℃. The drive circuit adopts an optocoupler isolation scheme, with a drive voltage of 12V, signal rise and fall times of less than 1μs, and an isolation withstand voltage of 2500VRMS.

[0059] In this embodiment, such as Figure 3 As shown, a bridging protection circuit is constructed by combining a relay array and a diode array using a hybrid series-parallel topology. Specifically, each route of the relay array consists of two solid-state relays connected in series, forming a dual control point. The diode array adopts a ladder parallel structure, with 12 Schottky diodes divided into three groups of four connected in parallel. Current balancing is achieved between the three groups through current-sharing resistors. These two arrays are connected in series in the order of "relay array - diode array - current-sharing network" to form a complete protection loop. For the connection with the bridging execution module 6, a twisted-pair shielded cable (cross-sectional area not less than 50mm²) is used. 2The connector connects to the battery clip at the end of the robotic arm via a specially designed high-voltage connector (rated voltage 1500V, rated current 800A). The connector employs a quick-locking mechanism to ensure reliable connection and ease of maintenance during operation.

[0060] When the system initiates a bridging operation, the relay array first turns on stage by stage, with a 50μs delay between each stage. This soft-start method reduces inrush current. The diode array remains on throughout the process, preventing reverse current and voltage spikes. The current sharing network ensures even current distribution in the parallel branches, avoiding localized overload. In practical applications, this bridging circuit 8 can withstand a maximum continuous operating current of 800A, with a transient overcurrent capability of 1200A (lasting no more than 100ms), and an insulation withstand voltage of 5000V. Through optimized circuit design, the system voltage drop is controlled within 0.8V, and the energy conversion efficiency exceeds 99%, meeting the requirements of high-power bridging operations.

[0061] The jumper protection module 5 dynamically adjusts the number of parallel channels according to the load: 3-4 channels are activated under light load (<200A), 5-7 channels are activated under medium load (200A-500A), and more than 8 channels are activated under heavy load (>500A). Protection strategies include current surge protection, reverse protection, and overvoltage protection, implemented through mechanisms such as relay interleaving (50μs interval) and real-time monitoring. Taking a 48V / 500Ah battery jumper as an example, the system will configure 8 parallel channels based on the expected maximum current of 600A, ensuring safety through step-by-step activation and real-time monitoring. This module can control the current rise rate to within 2000A / ms, the overcurrent response time to less than 100μs, has a reverse voltage blocking capability of 1000V, a response time to less than 50ns, a leakage current to less than 1mA, and can control the device temperature rise to within 40℃.

[0062] In this embodiment, such as Figure 3As shown, the jumper monitoring module 7 acts as the "nerve sensing system" of the entire automatic jumper device, ensuring the safety and reliability of the jumper process through multi-dimensional real-time monitoring and data analysis. This module adopts a distributed sensing architecture, including a voltage monitoring unit, a current monitoring unit, a temperature monitoring unit, and a data processing unit. The voltage monitoring unit uses high-precision differential sampling technology, employing a precision operational amplifier for signal conditioning. It achieves a 16-bit sampling accuracy, a range covering 0-1000V, a sampling rate of no less than 10kHz, and a common-mode rejection ratio greater than 100dB, effectively avoiding electromagnetic interference in industrial environments. The current monitoring unit integrates a Hall sensor and a shunt dual detection scheme. The Hall sensor has a range of 0-1000A, an accuracy better than 0.1%, and a response bandwidth of 100kHz. The shunt is made of manganese-copper alloy with a temperature coefficient of less than 20ppm / ℃, maintaining stable measurement accuracy within a temperature range of -40℃ to 85℃. The temperature monitoring unit is equipped with multiple thermocouples and PT100 temperature sensors to achieve real-time monitoring of the temperature of key components. The temperature measurement range is -50℃ to 200℃, the accuracy is better than ±0.5℃, and the sampling period is 100ms.

[0063] The data processing unit adopts a dual-core processor architecture with a main frequency of 1GHz, equipped with a hardware floating-point arithmetic unit, and uses a real-time multi-tasking operating system to ensure the real-time performance and reliability of data processing. In practical applications, when the device performs jump-connection processing on a group of faulty batteries, the monitoring module will collect and analyze various parameters in the jump-connection circuit 8 in real time: voltage surge detection uses a sliding window algorithm with a window length of 100ms, triggering an alarm when the voltage change rate exceeds a set threshold (e.g., 10V / ms); current monitoring uses oversampling technology, identifying current harmonics through 1024-point FFT analysis, and issuing a warning when the total harmonic distortion exceeds 5%; temperature monitoring uses a fuzzy control strategy, dynamically adjusting the warning threshold according to the temperature change trend. All collected data is transmitted to the main control module 3 in real time via the CAN bus at a communication rate of 500kbps, ensuring that the main control module 3 can respond to abnormal situations in a timely manner. When an abnormality is detected, the main control module 3 will complete a safety interruption of the jump-connection operation within 50ms, including disconnecting the relay, retracting the robotic arm, and a series of other actions. This comprehensive monitoring and rapid response mechanism ensures the safety of jumper operations and effectively prevents equipment damage and safety accidents caused by abnormal electrical parameters. Through extensive practical application data analysis, the monitoring system achieves a reliability of 99.99%, a false alarm rate of less than 0.1%, and a near-zero false alarm rate, providing a strong guarantee for the safe operation of automatic jumper devices.

[0064] In this embodiment, such as Figure 3As shown, the bridging execution module 6 serves as the execution terminal of the automatic bridging device, integrating intelligent visual perception and a precision mechanical control system. This module adopts a six-degree-of-freedom articulated robotic arm structure, achieving precise bridging operations through the coordinated work of a microprocessor, a vision perception unit, a servo motor, and a linear driver. The microprocessor uses an industrial-grade ARM architecture with a main frequency of 1.5GHz, equipped with a real-time operating system, and integrates a machine vision algorithm library and a motion control module. The vision perception unit is equipped with a 2-megapixel industrial camera with a frame rate of 60fps and a matching LED structured light source, enabling stable operation in complex lighting environments. Image processing employs a deep learning object detection algorithm, supporting the recognition of multiple bridging interfaces with an accuracy rate of 98% and an average recognition time of less than 100ms. The robotic arm adopts a modular design, with each joint equipped with a high-precision harmonic reducer (accuracy better than 1 arcminute) and a 17-bit absolute encoder, achieving a repeatability accuracy of ±0.02mm. The end-effector battery clamp uses an adaptive clamping mechanism, equipped with a torque sensor and an anti-slip clamping surface, allowing the clamping force to be continuously adjusted within the range of 0-500N. The servo motor is a permanent magnet synchronous motor with a rated power of 200W and a maximum speed of 3000rpm, equipped with a 20-bit incremental encoder. The linear driver uses ball screw transmission with a stroke of 300mm, a maximum thrust of 1000N, and a positioning accuracy of 0.01mm.

[0065] In actual operation, upon receiving the jumper command, the vision system first acquires a battery image and performs depth analysis to extract the 3D position information of the jumper interface. The system then plans the optimal robotic arm trajectory, considering obstacle avoidance, speed planning, and posture optimization. Through the coordinated control of servo motors and linear actuators, the system guides the robotic arm to precisely approach the target. The end effector adaptively adjusts the clamping force based on force feedback information to ensure a reliable connection. The average time for the entire jumper process is no more than 30 seconds, with a positioning success rate of 99.5%. The closed-loop force-position control strategy effectively avoids potential mechanical damage to the battery during operation. This execution module can adapt to various battery jumper interfaces, covering a spherical space with a radius of 800mm, and has a load capacity of up to 10kg, meeting the needs of most field applications.

[0066] This invention also discloses an automatic battery jump-connection method based on battery data detection, which is applied to smart substations where battery banks are deployed. Figures 1 to 3 The automatic jumper shown has a smart sensor network installed on the battery pack. Figure 4 This is a flowchart illustrating an automatic battery jump-connection method based on battery data detection in one embodiment. It should be understood that, although... Figure 4The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 4 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 4 As shown, the automatic battery jump-connection method based on battery data detection disclosed in this invention specifically includes the following steps:

[0067] S101. Obtain the battery location information of each battery cell in the battery pack, and collect the battery parameters of each battery cell through a smart sensor network.

[0068] The system utilizes a smart sensor network deployed within the battery pack to acquire key information about each battery cell. This sensor network consists of distributed nodes, each equipped with a high-precision position sensor and a multi-parameter acquisition module. The position sensors employ UWB positioning technology, achieving centimeter-level positioning accuracy in indoor environments, and calculate the three-dimensional coordinates (x, y, z) of each battery cell using the TDOA (Time Difference of Arrival) algorithm. The battery parameter acquisition module integrates a voltage sampling unit (sampling accuracy 0.1mV, range 0-100V), a current sampling unit (accuracy 0.1A, range ±1000A), a temperature sensing unit (accuracy 0.1℃, range -20℃ to 80℃), and an internal resistance measurement unit (measurement range 0.1mΩ-1000mΩ). All sensor data is transmitted in real-time to a data concentrator via a wireless network (using a mesh network topology), with a transmission cycle of 100ms. Taking a 48V battery system as an example, containing 24 2V battery cells, the location information of each cell can be represented as Pi(xi,yi,zi), where i is the cell number (1-24). The deployment of this sensor network significantly improves the automation of data acquisition, increasing acquisition efficiency by 95% compared to manual methods, and achieving a data accuracy of 99.9%, providing a reliable data foundation for subsequent analysis. The acquisition of physical parameters takes into account the influence of electromagnetic interference, employing digital filtering technology to eliminate power frequency interference, achieving a signal-to-noise ratio better than 60dB.

[0069] S102. Preprocess battery parameters, extract static and dynamic parameters from the preprocessed battery parameters, and integrate the static and dynamic parameters into battery characteristic parameters.

[0070] The process involves preprocessing and feature extraction of the collected battery parameters to construct a battery feature parameter system. The preprocessing stage begins with outlier detection and processing of the raw data. The preprocessed data is divided into two categories: static parameters and dynamic parameters. For static parameters, open-circuit voltage, internal resistance, and operating temperature are selected because these parameters directly reflect the battery's instantaneous state: open-circuit voltage reflects the battery's state of charge and electrochemical activity; internal resistance characterizes the battery's power characteristics and internal structural integrity; and operating temperature directly relates to the electrochemical reaction rate and safety. Dynamic parameters are selected based on the battery aging mechanism and performance degradation characteristics: capacity decay rate reflects a decrease in the utilization rate of active materials; charge-discharge curve characteristics reflect the degree of electrochemical polarization; charging efficiency and coulombic efficiency characterize energy conversion performance; self-discharge rate reflects the degree of side reactions; and the internal resistance growth rate and cycle life decay rate directly indicate the aging state. Specifically, static parameters include:

[0071] Open circuit voltage data (VOC): Measured with the battery in an open circuit state, with a measurement accuracy of 0.1mV.

[0072] Internal resistance data (R): Measured using the AC impedance method, frequency range 0.1Hz-1kHz.

[0073] Operating temperature data (T): The temperature at multiple points on the battery surface was measured using a thermocouple array.

[0074] Dynamic parameters include:

[0075] Capacity decay rate (Cd): Where Cn is the current capacity and C0 is the initial capacity.

[0076] Charging curve characteristics include charging platform voltage (Vc), charging time (tc), and charging voltage rise rate (δVc / δt).

[0077] Discharge curve characteristics include discharge plateau voltage (Vd), discharge time (td), and discharge voltage drop rate (δVd / δt).

[0078] Charging efficiency (ηc): Where Eout is the released energy and Ein is the input energy.

[0079] Self-discharge rate (Sr): Monthly capacity loss percentage.

[0080] Coulomb efficiency (ηQ): The ratio of output charge to input charge.

[0081] Internal resistance growth rate (Rr): Where Rn is the current internal resistance and R0 is the initial internal resistance.

[0082] Cycle life decay rate (Lr): The ratio of the current number of iterations to the rated number of iterations.

[0083] Integrate static and dynamic parameters into a feature vector:

[0084] F=[VOC, R, T, Cd, Vc, tc, δVc / δt, Vd, td, δVd / δt, ηc, Sr, ηQ, Rr, Lr].

[0085] Through this systematic parameter preprocessing and feature extraction method, the feature vector can comprehensively reflect the static characteristics and dynamic performance degradation trend of the battery, providing a quantitative basis for subsequent fuzzy evaluation. This method achieves a feature extraction accuracy of 98% and improves data reliability by 85%.

[0086] S103. Construct a feature fuzzy relation matrix based on battery feature parameters, and perform fuzzy comprehensive evaluation of the battery unit according to the feature fuzzy relation matrix to obtain the battery status score of each battery unit.

[0087] The process involves constructing a feature fuzzy relation matrix based on extracted battery feature parameters and performing fuzzy comprehensive evaluation, employing a three-layer fuzzy evaluation system. The division of the evaluation factor set follows a two-dimensional evaluation system of "static features - dynamic features." Static features reflect the instantaneous state of the battery, offering timeliness and intuitiveness; dynamic features embody the performance change trend of the battery, providing predictability and comprehensiveness. This division ensures both real-time evaluation and forward-looking assessment. First, an evaluation factor set U = {U1, U2, U3} is established, where U1 is the static feature subset, containing open-circuit voltage, internal resistance data, and operating temperature data; U2 is the dynamic feature subset I, containing capacity decay rate, charge / discharge curve characteristics, and charging efficiency; and U3 is the dynamic feature subset II, containing self-discharge rate, coulombic efficiency, internal resistance growth rate, and cycle life decay rate. The evaluation grade set V = {Excellent, Good, Average, Poor, Faulty}. The weight allocation was determined using the Analytic Hierarchy Process (AHP) and the entropy weight method. Then, first-level and second-level fuzzy comprehensive evaluation calculations were performed separately to obtain the final evaluation result vector, representing the membership degree of the battery state to each level. Finally, the battery state score was calculated based on the membership degree.

[0088] S104. Mark battery cells whose battery status score is lower than the preset score threshold as abnormal battery cells.

[0089] This step primarily involves identifying and marking abnormal battery units by comparing the battery status score with a preset scoring threshold. In this embodiment, the preset scoring threshold is determined using statistical methods combined with expert experience, specifically comprising three levels of thresholds: a warning threshold (75 points), a pre-warning threshold (65 points), and a fault threshold (55 points). The theoretical basis for setting these thresholds is as follows:

[0090] The warning threshold (75 points) is determined based on a large amount of historical data statistics, using the 3σ principle, and the calculation formula is as follows:

[0091] T warning =μ-2σ

[0092] Where μ is the mean score of a normal battery (approximately 90 points), and σ is the standard deviation (approximately 7.5 points). When the score falls below this threshold, it indicates that battery performance begins to decline significantly.

[0093] The warning threshold (65 points) is determined based on the analysis of the battery performance inflection point:

[0094] T alert =μ-3σ

[0095] At this point, the battery is already in a phase of accelerated degradation and requires close monitoring.

[0096] The fault threshold (55 points) corresponds to the minimum availability standard for the battery:

[0097] T fault =μ-4σ

[0098] Batteries below this threshold can no longer meet normal usage requirements.

[0099] The anomaly marking process can be implemented by establishing a hierarchical marking matrix M:

[0100]

[0101] Where i is the battery pack number and j is the unit number.

[0102] Record three consecutive rating results and construct a time-series feature vector:

[0103] V t =[Score t Score t-1 Score t-2 ]

[0104] Calculate the rate of change in scores:

[0105]

[0106] A battery cell is marked as abnormal if any of the following conditions are met:

[0107] 1. Current score is below the fault threshold: Score<55.

[0108] 2. The score is below the warning threshold and the rate of decline exceeds the threshold: Score < 65 and δ_t < -2 points / day.

[0109] 3. Three consecutive scores are all below the warning threshold: min(V_t)<75.

[0110] S105. Input the abnormal battery characteristic parameters of the abnormal battery unit into the preset battery fault identification model, and identify the faulty battery unit in the abnormal battery unit through the battery fault identification model.

[0111] Among them, a battery fault identification model based on support vector machines (SVM) is constructed to accurately identify faulty battery cells among abnormal battery cells. The SVM model was chosen based on the following considerations: it has strong small-sample learning and generalization capabilities, can effectively handle high-dimensional nonlinear classification problems, and has relatively low computational complexity. The construction process of the fault identification model is as follows:

[0112] 1. Eigenvector construction:

[0113] The input features include two main categories: static parameters and dynamic parameters, totaling 11 indicators:

[0114] X = [VOC, R, T, Cd, Vc] c urve, Vd c urve, η c Sr, η Q ,Rr,Lr]

[0115] Feature standardization:

[0116]

[0117] 2. Construction of training sample set:

[0118] 1000 typical fault samples were collected, including: capacity decay faults: 300 samples; increased internal resistance faults: 250 samples; increased self-discharge faults: 200 samples; deteriorated charge / discharge performance faults: 150 samples; and other types of faults: 100 samples. Sample labels: y = {1: fault, 0: non-fault}

[0119] 3. Kernel function selection:

[0120] Radial basis function (RBF) is used:

[0121]

[0122] Kernel parameter optimization uses a grid search method: the penalty parameter C ranges from [2^-5, 2^15]; the kernel parameter γ ranges from [2^-15, 2^3].

[0123] 4. SVM model training:

[0124] Objective function:

[0125]

[0126] Constraints:

[0127]

[0128] ξ i ≥0, i=1,2,...,n

[0129] The optimal separating hyperplane is obtained by solving the problem.

[0130]

[0131] 5. Model optimization strategies: PSO algorithm is used to optimize SVM parameters; feature importance analysis and feature selection are performed; sample balancing is performed (SMOTE algorithm); ensemble learning is used to improve performance (Bagging method).

[0132] 6. Fault Identification Process: Input abnormal battery feature vector X_test; Preprocessing: standardization, dimensionality reduction; Model prediction:

[0133] y pred =f(X) test )

[0134] Confidence calculation:

[0135]

[0136] S106. If the battery fault identification model identifies a faulty battery cell, it generates an automatic jumper task for the automatic jumper device by combining the battery location information and static parameters of the faulty battery cell. The automatic jumper task controls the automatic jumper device to achieve automatic jumper processing of the faulty battery cell.

[0137] This step utilizes an automatic jumper to automatically jumper the faulty battery cell. Based on the fault identification result and battery location information, an automatic jumper task T = [ID, Position, Spec, Interface, Priority] is generated, which is received by the main control module and converted into a standard execution command format: Task = {header, parameters, timestamp}. The main control module sends motion commands through the mobile chassis control interface, using an improved A* algorithm to plan the optimal path (obstacle avoidance accuracy ±5cm), and controls the mobile chassis to carry the entire device to the faulty battery location. During the movement, the power management module monitors the power supply status of each part of the system in real time to ensure stable operation (voltage fluctuation <±2%). Upon reaching the designated location, the main control module activates the jumper execution module, which calculates the robotic arm's trajectory based on an inverse kinematics algorithm (positioning accuracy ±1mm) and establishes a communication link with the jumper protection module. The jumper protection module activates a protection circuit constructed from a relay array and a diode array, setting protection thresholds (overcurrent protection 800A, overvoltage protection 50V). The robotic arm of the bridging module, guided by vision, precisely approaches the faulty battery interface (alignment accuracy <0.5mm) and completes the mechanical connection. Throughout the process, all modules are deployed within an insulated housing (insulation withstand voltage >5000V) and exchange data in real time via a CAN bus (communication latency <10ms). The bridging operation employs a soft-start method, with the current ramp-up rate controlled within 100A / s. Electrical and mechanical parameters are monitored throughout the process, and a protection mechanism is automatically triggered in case of abnormalities. This solution achieves precise positioning of the faulty battery (error <10mm), safe bridging (success rate >99%), and efficient processing (average time <3min), and possesses a comprehensive safety protection mechanism (fault response time <1ms).

[0138] In one embodiment, the automatic battery jump-connection method based on battery data detection further includes the following steps:

[0139] If the battery fault identification model fails to identify the faulty battery cell, then all battery characteristic parameters are converted into multi-dimensional characteristic time series data.

[0140] A battery life prediction model was constructed by combining long short-term memory networks and graph convolutional networks.

[0141] Multidimensional feature time series data is input into the battery life prediction model. Based on the graph network construction module in the battery life prediction model, a battery influence correlation graph of the battery unit is generated through a data-driven method. The graph network construction module is composed of a long short-term memory network layer and a self-attention mechanism layer. The graph nodes in the battery influence correlation graph are all battery units, and the graph node edges in the battery influence correlation graph represent the physical connection relationship or electrical coupling relationship between battery units.

[0142] High-dimensional impact features are extracted from the battery impact correlation graph through the feature extraction module in the battery life prediction model. The feature extraction module includes a graph convolutional layer, a graph domain transformation layer, a long short-term memory network layer, and a graph domain inverse transformation layer.

[0143] The high-dimensional impact features are reduced to low-dimensional impact features using the dimensionality reduction prediction module in the battery life prediction model. The fully connected layer in the dimensionality reduction prediction module generates life prediction estimates based on the low-dimensional impact features and reconstructs the low-dimensional impact features. The reconstructed low-dimensional impact features are then backpropagated to the graph network construction module, so that the graph network construction module can perform residual fusion with the subsequently input multi-dimensional feature time series data and the reconstructed low-dimensional impact features.

[0144] Identify the aging battery cell with the lowest life prediction estimate among all abnormal battery cells. If the life prediction estimate of the aging battery cell is lower than the preset lifespan threshold, mark the aging battery cell as a faulty battery cell.

[0145] An automatic jump-connection task is generated by combining the battery location information and static parameters of the faulty battery unit. The automatic jump-connection task controls the automatic jump-connection device to achieve automatic jump-connection processing of the faulty battery unit.

[0146] In this implementation, the first step is to perform a time-series transformation on all battery characteristic parameters. This process can be understood as creating a data table that changes over time. Specifically, a 30-day sliding time window is used, with data recorded daily and the window shifting forward by one day each time. For each battery cell, 11 key parameters are collected: open-circuit voltage VOC (volts), internal resistance R (milliohms), and operating temperature T (degrees Celsius) as static parameters; and capacity decay rate Cd (percentage), charging curve characteristics Vc (slope and inflection point of the voltage-time curve), discharging curve characteristics Vd (plateau characteristics of the voltage-time curve), charging efficiency ηc (percentage), self-discharge rate Sr (percentage / month), coulombic efficiency ηQ (percentage), internal resistance growth rate Rr (percentage / month), and cycle life decay rate Lr (percentage / thousand cycles) as dynamic parameters. These raw data are standardized and then denoised using db4 wavelet transform to remove measurement noise and environmental interference. Finally, a [30×11]-dimensional feature matrix is ​​obtained, which retains the time-varying characteristics of the parameters and reflects the interrelationships between them.

[0147] Reference Figure 5The battery life prediction model adopts a hybrid architecture of LSTM and GCN. This structural design fully considers the temporal characteristics and spatial correlations of battery performance degradation. The LSTM network is responsible for processing temporal features and includes an input layer, two hidden layers, and an output layer. The input layer receives 11 feature parameters, and each hidden layer contains 128 neurons, using the tanh activation function. The core of LSTM lies in its special gating mechanism: the forget gate ft determines which historical information is discarded, the input gate it controls the amount of new information input, and the output gate ot determines how much information is output. The mathematical expressions for these three gates are:

[0148] f t =σ(W f ·[h t-1 x t ]+b f )

[0149] i t =σ(W i ·[h t-1 x t ]+b i )

[0150] o t =σ(W o ·[h t-1 x t ]+b o )

[0151] The GCN network is used to handle the spatial correlation between battery cells. The entire network is trained using the Adam optimizer with an initial learning rate of 0.001, decaying by 50% every 50 epochs. The loss function combines mean squared error and L2 regularization.

[0152]

[0153] The graph network building module generates a battery influence correlation graph using a data-driven approach. This process first utilizes an LSTM layer to process temporal features. The LSTM layer contains 128 hidden units, and the state update of each unit follows the following rules:

[0154]

[0155] h t =o t ⊙tanh(c t )

[0156] Where ⊙ denotes element-wise multiplication. The processed features are input into the self-attention mechanism layer, which calculates the association weights between different battery cells. The self-attention calculation process is as follows: First, the input features X are transformed through three different linear transformations to obtain the query matrix Q, the key matrix K, and the value matrix V, all with dimensions [n×d], where n is the number of battery cells and d is the feature dimension. Then, the attention score is calculated: scaling factor here Used to prevent gradient vanishing.

[0157] The construction of the correlation graph comprehensively considers two types of relationships: physical connections (such as series-parallel topologies) and electrical coupling relationships (such as uneven current distribution and temperature field coupling). For physical connections, the adjacency matrix is ​​directly constructed based on the circuit topology; for electrical coupling, the edge weights are determined based on parameter correlation coefficients.

[0158]

[0159] In the final interconnection graph G = (V, E), the node set V contains all battery cells, the edge set E represents the interconnection relationships between cells, and the edge weights combine the physical connection strength and the degree of electrical coupling: W ij =α·W phy +(1-α)·W ele α is the balance coefficient (usually taken as 0.6). This graph structure representation method not only intuitively reflects the interaction mechanism inside the battery pack, but also supports subsequent graph neural network operations. Practice shows that the method achieves an association recognition accuracy of 96%, effectively capturing more than 90% of significant influence relationships, with a space complexity of O(n^2) and a time complexity of O(n^2d).

[0160] The feature extraction module extracts high-dimensional impact features from the battery impact correlation graph using a four-layer network structure. First, the graph convolutional layer updates the features of each node using the following formula:

[0161] H (l+1) =σ(D -1 / 2 AD -1 / 2 H (l) W (l) )

[0162] Where A is the adjacency matrix (representing the connection relationships between battery cells), D is the degree matrix (representing the number of connections per battery cell), H is the node feature matrix, and W is the learnable weight matrix. A two-layer graph convolution is used; the first layer outputs 64-dimensional features, and the second layer outputs 32-dimensional features, with ReLU as the activation function. Next, a graph domain transformation layer maps the features to the frequency domain, using Chebyshev multinomial expansion:

[0163]

[0164] Where Λ is the eigenvalue of the graph Laplacian matrix, and K is 5. This transformation can capture global features and periodic patterns in the graph structure. The transformed features are input into an LSTM layer containing 64 hidden units to extract temporal dependencies. Finally, the features are mapped back to the spatial domain through a graph domain inverse transformation layer, preserving the physical interpretability of the features. The inverse transformation uses the inverse of the Fast Fourier Transform (FFT): x = F -1 X, where X represents the frequency domain feature and x represents the spatial domain feature. The entire feature extraction process can be represented as: H out =LSTM(FFT) -1 (GCN(FFT(H in This multi-level feature extraction method ensures accurate extraction of local features (spatial accuracy > 95%) and effectively captures global features (temporal correlation recognition rate > 92%). The module output is a 256-dimensional high-dimensional feature vector containing complete performance characteristics of the battery cell.

[0165] The dimensionality reduction prediction module processes and predicts high-dimensional influence features. First, a variational autoencoder is used to compress the 256-dimensional high-dimensional features into 32-dimensional low-dimensional features. The encoding process uses a three-layer fully connected network:

[0166] Z = f enc (H)=W3·ReLU(W2·ReLU(W1·H+b1)+b2)+b3

[0167] Where H represents the input high-dimensional features, and Z represents the generated low-dimensional features. To ensure the effectiveness of feature compression, KL divergence loss is introduced: L KL =D KL (q(z|x)||p(z)), where q(z|x) is the posterior distribution and p(z) is the prior distribution (standard normal distribution). Then, a lifetime prediction estimate is generated through a three-layer fully connected network:

[0168] Life pred =W6·ReLU(W5·ReLU(W4·Z+b4)+b5)+b6

[0169] Simultaneously, the low-dimensional features are reconstructed, and the structure of the reconstructed network is symmetrical to that of the encoding network.

[0170] Z recon =f dec (Z)=W9·ReLU(W8·ReLU(W7·Z+b7)+b8)+b9

[0171] The reconstruction loss uses mean squared error: The reconstructed low-dimensional features are backpropagated to the graph network building module through residual connections to achieve feature fusion. Where α is the adaptive fusion coefficient, initially set to 0.5, and dynamically adjusted during the training process: The total loss function for the entire module is: L total =λ1L pred +λ2L KL +λ3L recon , where λ1=0.5, λ2=0.3, λ3=0.2.

[0172] The lifespan prediction results of all abnormal battery cells are evaluated and screened. First, a multi-level evaluation standard system is established: for each battery cell, its comprehensive health index is calculated.

[0173]

[0174] Life pred,i To predict remaining lifespan, Life nom For the rated cycle life (typically 2000 cycles), Cap i For the current capacity, Cap nom R is the rated capacity. i R is the current internal resistance. nom As the initial internal resistance, the weighting coefficients are w1 = 0.5, w2 = 0.3, and w3 = 0.2. The preset service life threshold uses a piecewise function.

[0175]

[0176] Then all abnormal units are sorted by predicted lifetime: Life sort =sort({Life pred,i}), select the minimum value as the aging unit: Life min =min({Life pred,i If the condition is met: Life min <T life If the condition is met, the unit is marked as a faulty unit. The marking results are stored in a structured data format, and the replacement priority is calculated according to the following formula: This judgment method boasts high accuracy (fault warning accuracy reaches 97%) and anticipatory nature (it can provide early warning of potential faults 8-12 days in advance), with a false judgment rate of less than 2%. Furthermore, through dynamic evaluation of the health index, it effectively avoids false judgments caused by instantaneous fluctuations (improving anti-interference capability by 70%). In addition, the method has high computational efficiency (single evaluation time <100ms) and low storage overhead (data volume per unit <1KB), making it suitable for online real-time monitoring applications.

[0177] In one embodiment, preprocessing the battery parameters, extracting static and dynamic parameters from the preprocessed battery parameters, and integrating the static and dynamic parameters into battery characteristic parameters includes the following steps:

[0178] Outliers in battery parameters are processed using a median filtering method based on an adaptive window mechanism, and missing values ​​in battery parameters are filled using a multiple interpolation method to obtain the preprocessed target battery parameters.

[0179] Static and dynamic parameters are extracted from the target battery parameters according to the preset parameter selection rules. Both static and dynamic parameters are time-series data.

[0180] After performing time alignment processing on the static and dynamic parameters, for each parameter timestamp, the static and dynamic parameters belonging to the parameter timestamp are merged into a multidimensional feature vector;

[0181] The multidimensional feature vectors of all parameter timestamps are integrated into battery feature parameters.

[0182] In this implementation, outliers and missing values ​​in battery parameters are handled by first employing median filtering based on an adaptive window to process outliers. The adaptive window mechanism dynamically adjusts the window size according to the degree of data fluctuation: W s =max(3,ceil(2σ) t )), where σ t Let x be the standard deviation for the current time period. For each data point in the time series x(t), calculate its median within the window: x med (t) = median(x(tk:t+k)). Then, outliers are determined based on the 3σ criterion: if |x(t)-x med (t)|>3σ t If x(t) is not found, then x(t) is marked as an outlier. For the detected outliers, a locally weighted regression is used for correction. The weighting coefficients use a Gaussian kernel function: Bandwidth parameter h is adaptively adjusted: h = 0.9σ t ·n -0.2 For missing values, a multiple imputation method is used. First, a Markov chain Monte Carlo (MCMC) model is constructed to estimate the conditional distribution of the missing data. Then, multiple (usually 5) random sampling imputations are performed, with each imputation based on the result of the previous one. The final imputed value is a weighted average of the multiple imputations.

[0183] The preprocessed battery parameters are classified and extracted based on preset parameter selection rules. Parameter selection follows the "three highs and one low" principle: high signal-to-noise ratio (SNR > 20dB), high stability (coefficient of variation < 0.1), high diagnostic value (information gain > 0.3), and low acquisition cost. A standardized scoring function is established for each type of parameter. The sampling period for static parameters is typically 5 minutes, and the calculation period for dynamic parameters is 30 minutes. All parameters are stored as time series. To ensure the temporal continuity of the data, linear interpolation is performed on data points exceeding the sampling period. Time alignment processing adopts a multi-level sliding window mechanism. First, all parameters are initially aligned according to the minimum sampling period (5 minutes). For each time window W(t), the time labels of static and dynamic parameters are mapped to a unified time grid. For parameters with different sampling frequencies, a piecewise linear interpolation method is used for time point matching. To ensure the accuracy of interpolation, a time constraint is introduced: when the time interval between adjacent sampling points exceeds a preset threshold τ (usually 30 minutes), local polynomial regression is used for filling. The aligned parameters are merged into a multidimensional feature vector at each timestamp t.

[0184] The multidimensional feature vectors of all parameter timestamps are integrated into complete battery feature parameters, organized using a hierarchical matrix structure. First, the basic feature matrix is ​​constructed: F base = [X(t1), X(t2), ..., X(t)] n )] T , where n is the total number of timestamps. To capture the correlation between parameters, the feature correlation matrix is ​​calculated:

[0185] Based on the correlation matrix, a dynamic weight adjustment mechanism is introduced:

[0186] Where λ is the weight adjustment parameter (default value 1.5), |C i | represents the average correlation between the i-th parameter and other parameters, and the feature vectors are weighted and integrated. To preserve temporal features, a time difference matrix is ​​constructed:

[0187] ΔF=[X(t2)-X(t1),X(t3)-X(t2),...,X(t n )-X(t n-1 )]

[0188] Simultaneously, moving average statistical characteristics are calculated, including mean μ(t), standard deviation σ(t), skewness s(t), and kurtosis k(t), with a window size of 12 time points (1 hour). The final characteristic parameters consist of the following components:

[0189] F total ={F base,ΔF,[μ(t),σ(t),s(t),k(t)],C,w(t)}

[0190] To improve data usability, the feature parameters are standardized: The integrated feature parameters have the characteristics of multiple scales (time scale, parameter scale) and multiple levels (original features, statistical features, correlation features).

[0191] In one implementation, a feature fuzzy relation matrix is ​​constructed based on battery feature parameters, and fuzzy comprehensive evaluation is performed on the battery cells according to the feature fuzzy relation matrix to obtain the battery state score of each battery cell. The steps include the following:

[0192] The parameter weights of static and dynamic parameters are calculated by combining the analytic hierarchy process (AHP) and the entropy weight method, respectively.

[0193] The step function is used as the membership function of the static parameter, and the sigmoid function is used as the membership function of the dynamic parameter. Based on the battery characteristic parameters and combined with the parameter weights and membership functions, a feature fuzzy relation matrix is ​​constructed.

[0194] The battery cells are evaluated using a fuzzy comprehensive method based on the feature fuzzy relation matrix to obtain the battery status score for each battery cell.

[0195] In this embodiment, the principle of parameter selection is first explained. The static parameters selected are open-circuit voltage, internal resistance, and operating temperature because these parameters directly reflect the instantaneous state of the battery: open-circuit voltage reflects the battery's state of charge and electrochemical activity; internal resistance characterizes the battery's power characteristics and internal structural integrity; and operating temperature directly relates to the electrochemical reaction rate and safety. The dynamic parameters are selected based on the battery aging mechanism and performance degradation characteristics: capacity decay rate reflects a decrease in the utilization rate of active materials; charge-discharge curve characteristics reflect the degree of electrochemical polarization; charging efficiency and coulombic efficiency characterize energy conversion performance; self-discharge rate reflects the degree of side reactions; and the internal resistance growth rate and cycle life decay rate directly indicate the aging state.

[0196] The parameter weights are determined using a combination of the Analytic Hierarchy Process (AHP) and the entropy weight method. The AHP calculation process is as follows:

[0197] Construct the judgment matrix A using the 1-9 scaling method:

[0198] For static parameters:

[0199] Calculate the eigenvalues ​​and eigenvectors: A1x = λx;

[0200] Solve for the maximum eigenvalue λmax and the corresponding eigenvector, and then normalize to obtain the weight vector:

[0201] W1=[0.539, 0.297, 0.164];

[0202] For dynamic parameters, construct an 8×8 judgment matrix:

[0203]

[0204] Similarly, the weight vector is obtained by solving for it:

[0205] W2 = [0.251, 0.146, 0.146, 0.127, 0.082, 0.127, 0.060, 0.060] Entropy weight method calculation process:

[0206] Standardize the raw data: Calculate the information entropy of the j-th indicator:

[0207] in:

[0208] Calculate the weights: The entropy weight vectors W1 entropy and W2 entropy are obtained.

[0209] Combined weight calculation:

[0210] The AHP weights and entropy weights are combined using the product method: The final static parameter combination weights are: W1 = [0.45, 0.35, 0.20].

[0211] Dynamic parameter combination weights:

[0212] W2=[0.23, 0.15, 0.15, 0.12, 0.10, 0.12, 0.07, 0.06]

[0213] Next, we construct the membership function. The static parameter membership function (step function) is as follows:

[0214] Open circuit voltage VOC:

[0215]

[0216] Internal resistance R:

[0217]

[0218] Operating temperature T:

[0219]

[0220] The dynamic parameter membership function (S-shaped function) is as follows:

[0221] For the capacity decay rate Cd:

[0222]

[0223] Where a = 5%, b = 7.5%, and c = 10%.

[0224] Other dynamic parameters employ similar S-shaped functions, with parameter values ​​determined based on specific indicators. Next, a fuzzy relation matrix is ​​constructed:

[0225] Static parameter fuzzy relation matrix R1:

[0226]

[0227] Dynamic parameter fuzzy relation matrix R2:

[0228]

[0229] Next, fuzzy comprehensive evaluation is performed, including static parameter evaluation:

[0230] Dynamic parameter evaluation:

[0231] Overall assessment: Where W = [0.4, 0.6] is the weight vector of static and dynamic features.

[0232] Finally, the status score is calculated: Where v = [100, 80, 60, 40, 20] is the rating value vector.

[0233] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0234] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A method for automatic battery jump-connection based on battery data detection, characterized in that, This system is applied to intelligent substations equipped with battery banks and automatic jump-connection devices. The battery banks are equipped with an intelligent sensor network. The automatic jump-connection device includes an insulating housing and a mobile chassis. The insulating housing is fixedly mounted on the top of the mobile chassis. The automatic jump-connection device also includes a main control module, a jump-connection protection module, and a jump-connection execution module. The main control module is connected to the mobile chassis, the jump-connection protection module, and the jump-connection execution module. The jump-connection protection module is electrically connected to the jump-connection execution module. The main control module receives automatic jump-connection tasks and generates automatic jump-connection instructions based on these tasks. The mobile chassis responds to the automatic jump-connection instructions and moves to the location of the faulty battery cell. The jump-connection execution module responds to the automatic jump-connection instructions and completes the automatic jump-connection process for the faulty battery cell. The jump-connection protection module responds to the automatic jump-connection instructions and protects the jump-connection execution module during the automatic jump-connection process. The method includes the following steps: Obtain the battery location information of each battery cell in the battery pack, and collect the battery parameters of each battery cell through a smart sensor network; Preprocess battery parameters, extract static and dynamic parameters from the preprocessed battery parameters, and integrate the static and dynamic parameters into battery characteristic parameters; A feature fuzzy relation matrix is ​​constructed based on battery feature parameters. Fuzzy comprehensive evaluation of the battery unit is performed based on the feature fuzzy relation matrix to obtain the battery status score of each battery unit. Battery cells whose battery condition score is lower than a preset score threshold are marked as abnormal battery cells; The abnormal battery characteristic parameters of the abnormal battery unit are input into the preset battery fault identification model. The faulty battery unit in the abnormal battery unit is identified by the battery fault identification model, which is built based on support vector machine. If the battery fault identification model identifies a faulty battery cell, it generates an automatic jumper task by combining the battery location information and static parameters of the faulty battery cell. The automatic jumper task controls the automatic jumper to achieve automatic jumper processing of the faulty battery cell. If the battery fault identification model fails to identify the faulty battery cell, then all battery characteristic parameters are converted into multi-dimensional characteristic time series data. A battery life prediction model was constructed by combining long short-term memory networks and graph convolutional networks. Multidimensional feature time series data is input into the battery life prediction model. Based on the graph network construction module in the battery life prediction model, a battery influence correlation graph of the battery unit is generated through a data-driven method. The graph network construction module is composed of a long short-term memory network layer and a self-attention mechanism layer. The graph nodes in the battery influence correlation graph are all battery units, and the graph node edges in the battery influence correlation graph represent the physical connection relationship or electrical coupling relationship between battery units. High-dimensional impact features are extracted from the battery impact correlation graph through the feature extraction module in the battery life prediction model. The feature extraction module includes a graph convolutional layer, a graph domain transformation layer, a long short-term memory network layer, and a graph domain inverse transformation layer. The high-dimensional impact features are reduced to low-dimensional impact features using the dimensionality reduction prediction module in the battery life prediction model. The fully connected layer in the dimensionality reduction prediction module generates life prediction estimates based on the low-dimensional impact features and reconstructs the low-dimensional impact features. The reconstructed low-dimensional impact features are then backpropagated to the graph network construction module, so that the graph network construction module can perform residual fusion with the subsequently input multi-dimensional feature time series data and the reconstructed low-dimensional impact features. Identify the aging battery cell with the lowest life prediction estimate among all abnormal battery cells. If the life prediction estimate of the aging battery cell is lower than the preset lifespan threshold, mark the aging battery cell as a faulty battery cell. An automatic jump-connection task is generated by combining the battery location information and static parameters of the faulty battery unit. The automatic jump-connection task controls the automatic jump-connection device to achieve automatic jump-connection processing of the faulty battery unit.

2. The automatic battery jump-connection method based on battery data detection according to claim 1, characterized in that, Preprocessing battery parameters involves extracting static and dynamic parameters from the preprocessed parameters and integrating them into battery characteristic parameters. This process includes the following steps: Outliers in battery parameters are processed using a median filtering method based on an adaptive window mechanism, and missing values ​​in battery parameters are filled using a multiple interpolation method to obtain the preprocessed target battery parameters. Static and dynamic parameters are extracted from the target battery parameters according to the preset parameter selection rules. Both static and dynamic parameters are time-series data. After performing time alignment processing on the static and dynamic parameters, for each parameter timestamp, the static and dynamic parameters belonging to the parameter timestamp are merged into a multidimensional feature vector; The multidimensional feature vectors of all parameter timestamps are integrated into battery feature parameters.

3. The automatic battery jump-connection method based on battery data detection according to claim 2, characterized in that, Static parameters include open-circuit voltage data, internal resistance data, and operating temperature data, while dynamic parameters include capacity decay rate, charging curve characteristics, discharging curve characteristics, charging efficiency, self-discharge rate, coulombic efficiency, internal resistance growth rate, and cycle life decay rate.

4. The automatic battery jump-connection method based on battery data detection according to claim 3, characterized in that, The process of constructing a feature fuzzy relation matrix based on battery feature parameters, and then performing fuzzy comprehensive evaluation on the battery cells based on the feature fuzzy relation matrix to obtain the battery status score for each battery cell includes the following steps: The parameter weights of static and dynamic parameters are calculated by combining the analytic hierarchy process (AHP) and the entropy weight method, respectively. The step function is used as the membership function of the static parameter, and the sigmoid function is used as the membership function of the dynamic parameter. Based on the battery characteristic parameters and combined with the parameter weights and membership functions, a feature fuzzy relation matrix is ​​constructed. The battery cells are evaluated using a fuzzy comprehensive method based on the feature fuzzy relation matrix to obtain the battery status score for each battery cell.

5. An automatic jumper device according to any one of claims 1 to 4, characterized in that, The device includes an insulating shell and a mobile chassis. The insulating shell is fixedly mounted on the top of the mobile chassis. The interior of the insulating shell is divided into multiple isolation chambers. A cable sealing sleeve is installed at the cable crossing point between any two isolation chambers. Each isolation chamber is equipped with an independent grounding system. The mobile chassis serves as a foundation support for the insulating shell and also drives the insulating shell to move. The device also includes a main control module, a power management module, a jumper protection module, and a jumper execution module. The main control module, power management module, and jumper protection module are deployed in different isolated chambers in the insulating housing according to different voltage levels. The power management module is connected to the main control module, the jumper protection module, and the jumper execution module respectively, and supplies power to the main control module, the jumper protection module, and the jumper execution module. The jumper execution module is located on the outside of the insulating housing. The main control module is connected to the mobile chassis, the jumper protection module, and the jumper execution module. The jumper protection module is electrically connected to the jumper execution module. The main control module is used to receive automatic jumper tasks and generate automatic jumper commands based on the automatic jumper tasks. The mobile chassis responds to the automatic jumper commands and moves to the location of the faulty battery unit. The jumper execution module responds to the automatic jumper commands and completes the automatic jumper processing of the faulty battery unit. The jumper protection module responds to the automatic jumper commands and protects the jumper execution module during the automatic jumper processing.

6. The automatic bridging device according to claim 5, characterized in that, The main control module includes a central processing unit, a positioning unit, and a communication unit. Both the positioning unit and the communication unit are connected to the central processing unit. The positioning unit is used to acquire the real-time location information of the automatic jump-connection device, and the communication unit is used to receive the automatic jump-connection task. The central processing unit is used to parse the battery location information and static parameters of the faulty battery unit from the automatic jump-connection task. The central processing unit is used to generate automatic jump-connection instructions by combining the real-time location information, battery location information, and static parameters. The automatic jump-connection instructions include a movement instruction, a protection module configuration instruction, and a jump-connection execution instruction, which are executed sequentially. The movement instruction is generated by combining the real-time location information and battery location information and using a path planning algorithm. The protection module configuration instruction is generated based on static parameters. The communication unit is used to send the movement instruction to the mobile chassis, the protection module configuration instruction to the jump-connection protection module, and the jump-connection execution instruction to the jump-connection execution module.

7. The automatic bridging device according to claim 6, characterized in that, The jumper protection module includes an array switching controller, a relay array, a diode array, and a drive circuit. The relay array includes multiple solid-state relays connected in parallel, and the diode array includes multiple Schottky diodes connected in parallel. The relay array is used to reduce current surges during automatic jumper processing, and the diode array is used to prevent reverse current and voltage spikes during automatic jumper processing. The relay array and diode array together form a jumper protection circuit that is electrically connected to the jumper execution module to form a jumper loop. The jumper protection circuit is electrically connected to the array switching controller through the drive circuit. The array switching controller is connected to a communication unit. The array switching controller is used to receive configuration instructions from the protection module and adjust the number of solid-state relays connected in parallel in the relay array and the number of Schottky diodes connected in parallel in the diode array according to the configuration instructions.

8. The automatic bridging device according to claim 7, characterized in that, The automatic jumper also includes a jumper monitoring module, which is deployed in an isolated chamber within an insulated housing. A power management module is connected to and supplies power to the jumper monitoring module. The jumper monitoring module includes a voltage monitoring unit, a current monitoring unit, a temperature monitoring unit, and a data processing unit. All three units are connected to the data processing unit, which is in turn connected to the main control module. The voltage and current monitoring units monitor the jumper voltage and current data during the automatic jumper process. The temperature monitoring unit monitors the device temperature data of the automatic jumper. The data processing unit monitors changes in the jumper voltage, current, and temperature data. When the jumper voltage and current data change abruptly, or the device temperature exceeds a preset temperature threshold, the data processing unit sends an alarm message to the main control module. The main control module generates a jumper interruption command based on the alarm message and sends it to the jumper execution module, causing the jumper execution module to interrupt the automatic jumper process for the faulty battery unit.

9. The automatic bridging device according to claim 6, characterized in that, The jumper execution module includes a microprocessor, a vision sensing unit, a servo motor, a linear actuator, and a robotic arm. The robotic arm is a six-degree-of-freedom articulated structure with a battery clamp at its end. The vision sensing unit, servo motor, and linear actuator are all connected to the microprocessor, which is also connected to a communication unit. The microprocessor receives jumper execution commands and generates image acquisition commands based on these commands. The vision sensing unit responds to the image acquisition commands and acquires battery image information of the faulty battery unit. The microprocessor also identifies and locates the jumper interface on the faulty battery unit using image recognition technology and generates robotic arm operation commands based on the position of the jumper interface. The servo motor and linear actuator simultaneously respond to the robotic arm operation commands and collaboratively control the robotic arm to connect the battery clamp to the jumper interface.

Images