An unmanned aerial vehicle formation battery intelligent screening method and system

Abstract

The application discloses a kind of unmanned aerial vehicle formation battery intelligent screening method and system.The method is collected by on-board data recording and back transmission module in the process of unmanned aerial vehicle flight discharge data containing battery ID, and is uploaded to cloud after flight ends;When battery accesses intelligent charging dock and charges, the charging data containing battery ID is collected and uploaded to cloud;Cloud is fused with the discharge data and charging data of the same battery to form charge-discharge health record with battery ID as association key, and cross validation is carried out to identify battery aging state;Predictive model is trained using group battery historical charge-discharge data, with battery health index as input and next flight voltage drop risk probability as output;Before flight mission, recommended use battery list and disabled battery list are generated according to the risk probability of each battery and safety threshold.The application realizes cloud fusion of charge-discharge dual-source data and voltage drop risk prediction, can exclude hidden danger battery from source, and improve formation flight safety.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) formation flight control and intelligent lithium battery management technology, and in particular to an intelligent battery selection method and system for UAV formations. Background Technology

[0002] Large-scale drone formation performances have developed rapidly in recent years, with single performances often using thousands of drones. The consistency of battery performance directly affects flight safety and formation effects, and is one of the key factors restricting the industry's development. After repeated charging and discharging, lithium batteries exhibit significant individual variations in parameters such as internal resistance, capacity, and cell voltage difference. Two batteries with the same number of cycles may have vastly different actual health conditions.

[0003] Currently, the common practice of managing batteries in formation flying is to use them as soon as they are fully charged, which has significant drawbacks. On the one hand, it ignores individual differences in batteries and mixes batteries in different health conditions, making it difficult to guarantee flight safety. On the other hand, the existing management method lacks predictive means and cannot predict whether a particular battery will experience a voltage drop during the next flight. Batteries with abnormally increasing internal resistance may suddenly lose power in the later stages of flight, causing drones to fall behind or even crash. Such accidents occur frequently in formation flying performances.

[0004] Furthermore, data collection within the industry is incomplete. Data is either recorded only during charging or transmitted in real-time via 4G during flight; few systems systematically collect data from both charging and discharging stages. Due to the lack of complete charging and discharging data, cross-validation and trend prediction of battery health are impossible. Therefore, there is an urgent need for a method and system that can fuse charging and discharging data to accurately predict and automatically screen for battery voltage drop risks before flight missions. Summary of the Invention

[0005] The technical problem to be solved by this invention is: how to accurately predict the voltage drop risk of the formation batteries before each flight mission and automatically select safe and usable batteries based on cloud fusion of charging and discharging dual-source data and group historical data modeling.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution.

[0007] In a first aspect, the present invention provides a method for intelligent battery selection in drone formations, comprising the following steps: During the flight of the drone, the onboard data recording and transmission module on the drone collects discharge data including the battery ID, and does not send battery-related data to the outside during the entire flight. After the flight, the discharge data will be uploaded to the cloud-based health management platform; When the battery is connected to the smart charging dock for charging, the smart charging dock collects charging data including the battery ID and uploads the charging data to the cloud health management platform; The cloud-based health management platform uses battery ID and cycle count as association keys for time-series alignment, classifying the Nth flight record of the same battery and the next charging record with cycle count N+1 as the same charge-discharge cycle. This allows time-adjacent charging records and flight records of the same battery to form a continuous charge-discharge cycle sequence, thereby fusing the discharge data and charging data of the same battery to form a charge-discharge health profile for the battery. The discharge data and charging data are then cross-validated to identify the aging state of the battery. The cross-validation of the discharge and charge data to identify the battery's aging status includes: taking the average dynamic internal resistance for each flight segment, converting the average dynamic internal resistance to a reference temperature using a temperature coefficient, comparing the converted average dynamic internal resistance with the historical average and standard deviation of the battery's dynamic internal resistance over the most recent preset number of flights; if the converted average dynamic internal resistance exceeds the sum of the historical average and the standard deviation of a preset multiple, and exceeds the standard in two consecutive flights, it is marked as having high dynamic internal resistance; comparing the static internal resistance measured before each charge with the historical average and standard deviation of the battery's static internal resistance over the most recent preset number of charges; if the static internal resistance exceeds the sum of the historical average and the standard deviation of a preset multiple, it is marked as having high static internal resistance; the battery is determined to be aging only when both high dynamic internal resistance and high static internal resistance exist simultaneously. The cloud-based health management platform trains a prediction model using historical charge and discharge data of the group of batteries. The prediction model takes the health indicators of the batteries as input and the probability of voltage drop in the next flight as output. Before each flight mission, a recommended list of batteries to use and a list of batteries to be deactivated are generated based on the probability of voltage drop risk for each battery and the safety threshold set for the mission.

[0008] Secondly, the present invention provides a smart battery selection system for drone formations, including an airborne data recording and transmission module, a smart charging dock, a cloud-based health management platform, and a smart selection output module. The airborne data recording and transmission module is installed on the drone and is used to collect discharge data including the battery ID during the drone's flight. It does not send battery-related data to the outside during the entire flight and uploads the discharge data to the cloud health management platform after the flight ends. The smart charging dock is used to collect charging data including the battery ID when the battery is connected for charging, and upload the charging data to the cloud health management platform; The cloud-based health management platform includes a data fusion module, a cross-validation module, a feature engineering module, and a model training module; The data fusion module is used to perform time-series alignment using battery ID and cycle count as association keys, classifying the Nth flight record of the same battery and the next charging record with cycle count N+1 as the same charge-discharge cycle, so that the charging records and flight records of the same battery that are adjacent in time form a continuous charge-discharge cycle sequence, so as to fuse the discharge data and charging data of the same battery to form a charge-discharge health record including a subset of discharge sequence and a subset of charging sequence. The cross-validation module is used to cross-validate the discharge data and charging data to identify the aging state of the battery. The cross-validation includes: taking the average dynamic internal resistance of the entire flight segment for each flight, converting the average dynamic internal resistance of the entire flight segment to a reference temperature using a temperature coefficient, comparing the converted average dynamic internal resistance with the historical average and standard deviation of the dynamic internal resistance of the battery in the most recent preset number of flights, and if the converted average dynamic internal resistance exceeds the sum of the historical average and the standard deviation of a preset multiple and exceeds the standard in two consecutive flights, it is marked as high dynamic internal resistance; comparing the static internal resistance measured before each charge with the historical average and standard deviation of the static internal resistance of the battery in the most recent preset number of charges, and if the static internal resistance exceeds the sum of the historical average and the standard deviation of a preset multiple, it is marked as high static internal resistance; the battery is determined to be aging only when both high dynamic internal resistance and high static internal resistance exist simultaneously. The feature engineering module is used to calculate the average monthly growth rate of internal resistance; The model training module is used to train a GRU neural network using historical charge and discharge data of the group of batteries to obtain a prediction model. The prediction model takes the health index of the battery as input and the probability of voltage drop of the battery in the next flight as output. The intelligent filtering output module is used to generate a recommended battery list and a deactivated battery list before each flight mission, based on the voltage drop risk probability of each battery and the safety threshold set for this mission.

[0009] The beneficial effects of this invention are as follows: During the drone's flight, the onboard data recording and transmission module collects discharge data including the battery ID, ensuring that battery-related data is not transmitted externally throughout the flight and is uploaded to the cloud-based health management platform after the flight. Simultaneously, the smart charging dock collects charging data including the battery ID during charging and uploads it to the cloud-based health management platform. This ensures the reliable aggregation of flight discharge and charging data for the same battery to the cloud via two independent data sources, guaranteeing the integrity of the charging and discharging data. Furthermore, the cloud-based health management platform uses the battery ID and cycle count as the association key for time-series alignment, classifying the Nth flight record and the next charging record with cycle count N+1 of the same battery into the same charge-discharge cycle. This allows adjacent charging and flight records for the same battery to form a continuous charge-discharge cycle sequence. Even when the clocks at both ends are independent and have deviations, the discharge and charging data of the same battery can still be accurately paired, forming a complete charge-discharge health profile for each battery from charging to discharging and back to charging. This overcomes the shortcomings of single-link solutions, such as incomplete data and inability to perform trend analysis. By analyzing each... The average dynamic internal resistance of the battery across the entire flight segment is calculated and converted to a reference temperature using a temperature coefficient. This value is then compared with the historical average and standard deviation of the dynamic internal resistance over the most recent preset number of flights. Only when the dynamic internal resistance exceeds the standard in two consecutive flights is it marked as high. Simultaneously, the static internal resistance measured before each charge is compared with the historical average and standard deviation of the static internal resistance over the most recent preset number of charges, and this is marked as high. Battery aging is only determined when both high dynamic and high static internal resistance are present. This method can differentiate between the reversible temperature rise caused by high-current discharge during flight and permanent aging. This approach differentiates batteries, reducing misjudgments of their health status. By training a predictive model using historical charge and discharge data from a group of batteries, and taking battery health indicators as input and the probability of voltage drop during the next flight as output, battery management shifts from post-event detection to pre-event prediction. Before each flight mission, a recommended battery list and a deactivated battery list are generated based on the voltage drop risk probability of each battery and the safety threshold set for the mission. This automatically eliminates batteries with potential voltage drop risks from the source before flight missions, improving the safety of drone swarm flights. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0011] Figure 1This is a flowchart of the intelligent battery selection method for drone formation according to an embodiment of the present invention; Figure 2 This is a structural and hardware block diagram of the charging dock according to an embodiment of the present invention; Figure 3 This is a schematic block diagram of the intelligent battery selection system for drone formations according to an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0013] It should be noted that the descriptions involving "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Furthermore, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. The method provided by this invention can be executed by the system provided by this invention. The system includes five components: an airborne data recording and transmission module 10, an intelligent charging dock 30, a ground data aggregation node 20, a cloud-based health management platform 40, and an intelligent filtering and output module 50. Flight discharge data and charging data are collected separately through two independent links and finally fused in the cloud according to battery ID.

[0015] As attached Figure 1 - Appendix Figure 3 As shown, the method embodiment of the present invention is: a method for intelligent selection of batteries in drone formations, comprising the following steps: S1. During the flight of the UAV, the onboard data recording and transmission module 10 installed on the UAV collects discharge data including the battery ID.

[0016] In this step, the airborne data recording and transmission module 10 is integrated on each drone to collect battery discharge data during flight. The discharge data collected in this step constitutes the first data source, providing a complete record of the flight process for subsequent fusion with charging data.

[0017] S11. In a specific embodiment, during the flight of the UAV, the onboard data recording and transmission module 10 installed on the UAV collects discharge data including the battery ID; wherein, the onboard data recording and transmission module 10 collects the discharge data from the battery management system at a set sampling period through the onboard bus, adds a timestamp to the discharge data and writes it into the onboard local storage medium to form a flight log, and does not send battery-related data to the outside during the entire flight; the onboard bus is an SMBus bus or a CAN bus, the set sampling period is 1 second, and the onboard local storage medium is an SD card; the discharge data also includes at least one of the following: number of cycles, total battery pack voltage, discharge current, battery temperature, voltage of each cell, highest and lowest cell voltage difference, and DC internal resistance; the flight log is a CSV format file, named by battery ID and flight time, with each line corresponding to a sampling time.

[0018] In this embodiment, the airborne data recording and transmission module 10 communicates with the battery management system (BMS) of the intelligent battery via the flight control's SMBus or CAN bus, reading battery data at a sampling rate of 1Hz during flight. Throughout the entire flight, the airborne terminal does not send any battery-related data outwards, thereby avoiding data loss due to airborne signal interference and network congestion, ensuring the reliability of data recording during the flight phase. The airborne module adds a timestamp to the read data and writes it to the airborne SD card, forming a CSV format flight log file named according to battery ID and flight time, with each line corresponding to a sampling time.

[0019] In this embodiment, the specific fields and their units in the flight log file, along with their descriptions, are shown in Table 1 below.

[0020] Table 1: Description of Flight Log Fields S2. After the flight is completed, the discharge data is uploaded to the cloud health management platform 40.

[0021] In this step, after the aircraft lands and returns to its original position, the onboard terminal automatically transmits the flight log to the ground data aggregation node 20, which then uploads it to the cloud health management platform 40. No manual intervention is required, so that the discharge data is automatically and reliably aggregated to the cloud after landing and returning to its original position.

[0022] S21. In a specific embodiment, after the flight ends, the discharge data is uploaded to the cloud health management platform 40. The discharge data is first automatically transmitted wirelessly to the ground data aggregation node 20 by the airborne data recording and transmission module 10, and then uploaded to the cloud health management platform 40 by the ground data aggregation node 20. When the motor is detected to have stopped and the flight duration exceeds the minimum effective duration, the airborne wireless module is woken up and connected to the ground wireless access point. When the flight control detects that the motor has stopped, the UAV is within the wireless coverage range of the ground wireless access point, and the airborne wireless module is connected to the ground wireless access point, the discharge data is automatically pushed to the ground data aggregation node 20 via a file transfer protocol. After the transmission is completed, the airborne wireless module is automatically turned off.

[0023] In this embodiment, the minimum effective duration is 30 seconds. The airborne terminal runs a monitoring script to listen to the flight control status in real time. When it detects that the motor changes from "running" to "stopping" and the current flight exceeds 30 seconds, it immediately wakes up the Wi-Fi module, connects to the ground-based preset AP, and automatically pushes the newly generated log file in the SD card to the ground aggregation node via FTP / SFTP. After the transmission is completed, the Wi-Fi module automatically turns off.

[0024] In this embodiment, the ground station has already distributed scripts to each aircraft via APs before takeoff. A fixed correspondence exists between APs and aircraft. Log transmission after landing is the reverse process of the same link, reusing the original AP allocation without needing to re-establish a connection. Each aircraft's log is approximately 2-5KB. The ground station allocates receiving threads according to APs, and each thread receives files from its associated aircraft in parallel, writes them to a local temporary directory, and returns an ACK. Upon receiving the ACK, the aircraft automatically turns off Wi-Fi. Reception and upload are decoupled; the upload task runs independently, retrieving files from the cache according to the bandwidth limit and sending them to the cloud. Upon successful upload, the local cache is cleared. For exception handling: if a single-machine reception times out (30 seconds), the connection is disconnected and marked for retransmission; if the upload is interrupted, it resumes transmission after network recovery; if the cache is lost, the original files are retained on the onboard SD card and can be retransmitted upon next SD card readjustment.

[0025] S3. When the battery is connected to the smart charging dock 30 for charging, the smart charging dock 30 collects charging data including the battery ID and uploads the charging data to the cloud health management platform 40.

[0026] In this step, the intelligent charging dock 30 is a multi-device charging device used to charge the batteries of the formation drones. When the batteries are connected to the charging dock, the dock reads the charging data from the battery's BMS through the charging interface. There is no data interaction between the charging dock and the aircraft; the charging dock is only responsible for collecting charging data. The two links do not affect each other and can be deployed and maintained separately. The collected charging data is uploaded to the ground data aggregation node 20 or the cloud via wired network or Wi-Fi. The charging data collected in this step constitutes a second data source independent of the flight discharge data, laying the foundation for the fusion and cross-validation of charging and discharging dual-source data.

[0027] In this embodiment, the battery ID is written to and stored in the BMS non-volatile memory during battery production, ensuring its uniqueness. The ID remains fixed after battery production and will not change due to cyclic use or capacity decay. The battery parameters read by the onboard module via the SMBus / CAN bus already include this field. The charging dock also reads the same field from the BMS through its charging interface. After flight data and charging data are uploaded to the cloud respectively, the system associates the two types of data with the same file using the battery ID as the key.

[0028] In this embodiment, if a battery is determined to be disabled, charging should be prohibited. Specifically, the real-time status of each battery can be viewed in the cloud and the operator can be reminded to remove the abnormal battery, or a status light or buzzer can be set on the charging dock for real-time reminder.

[0029] S31. In a specific embodiment, when the battery is connected to the smart charging dock 30 for charging, the smart charging dock 30 collects charging data including the battery ID and uploads the charging data to the cloud health management platform 40; wherein, the charging data also includes charging current curve, voltage rise curve, voltage of each cell, battery temperature and static internal resistance; before the battery is connected and charging begins, the smart charging dock 30 applies a short-time high-current pulse to the battery using the hybrid pulse power characteristic method, while measuring the change in battery terminal voltage, and calculates the static internal resistance according to Ohm's law.

[0030] In this embodiment, the static internal resistance is the DC internal resistance (DCIR) measured by the battery under no-load conditions. Before the battery is connected and charging begins, the charging dock uses the hybrid pulse power characteristic (HPPC) method to measure the static internal resistance, that is, the static internal resistance is measured under constant temperature conditions in the hangar. The measurement conditions are constant, providing a benchmark for subsequent cross-validation with the dynamic internal resistance during flight.

[0031] S32. In a specific embodiment, after the flight ends, the discharge data is uploaded to the cloud health management platform 40; when the battery is connected to the smart charging dock 30 for charging, the smart charging dock 30 collects charging data containing the battery ID and uploads the charging data to the cloud health management platform 40; wherein, both the discharge data and the charging data are parsed, cleaned, and categorized by battery ID by the ground data aggregation node 20 before being uploaded to the cloud health management platform 40; the parsing includes splitting each record by field separator, converting the string into the corresponding data type, and verifying whether each field is within a reasonable range, and removing records with format errors or missing fields; the cleaning includes deduplicating the same file uploaded repeatedly by file name, file size, and generation time, removing data that exceeds a reasonable physical range, and completing the data by linear interpolation when the sampling interval is continuously missing for no more than a preset duration, discarding the data segment when it exceeds the preset duration, and uniformly converting the timestamp to UTC time; the categorization involves establishing a directory structure by battery ID, storing multiple flight records and charging records of the same battery ID by date.

[0032] In this embodiment, ground data aggregation node 20 (ground station) is deployed at the takeoff and landing site to receive flight log files transmitted back by the aircraft after it returns to its position and charging data uploaded by the charging dock. The flight logs are in CSV format, and each record contains fields such as timestamp, battery ID, voltage, current, temperature, internal resistance, cell voltage, and GPS. The ground station reads the CSV file, breaks down each line according to field delimiters, converts the strings into corresponding data types (numerical, time, string, etc.), verifies whether each field is within a reasonable range, and removes records with incorrect format or missing fields. The charging data is parsed in the same way, and the fields include charging time, battery ID, charging current, voltage rise curve, cell voltage, temperature, static internal resistance, etc.

[0033] In this embodiment, during deduplication, if the same aircraft repeatedly uploads the same log file (e.g., due to network jitter causing retransmission), the ground station verifies the filename, file size, and generation time, discarding duplicate files. For outlier handling, data exceeding reasonable physical ranges is removed, such as voltages of 0V or greater than 5V / section, temperatures below -20℃ or above 80℃, and internal resistances that are negative or greater than 200mΩ, while retaining valid data rows. During data completion, in rare cases where the sampling interval exceeds 1 second (e.g., high flight control load causing frame drops), linear interpolation is used to complete the data if the consecutive missing interval does not exceed 3 seconds; otherwise, the segment is discarded. For data format standardization, timestamps are uniformly converted to UTC time, GPS coordinates are uniformly converted to degrees (decimal format), and temperatures are uniformly converted to °C.

[0034] In this embodiment, a directory structure is created based on battery ID during classification: / data / [Battery ID] / [Date] / Flight Log.csv and / data / [Battery ID] / [Date] / Charging Log.csv. Multiple flight records and charging records for the same battery ID are stored by date, facilitating direct retrieval of all historical data from the cloud by battery ID. After processing, the ground station packages the data categorized by battery ID and uploads it to the cloud server via wired broadband or 4G / 5G.

[0035] S4. The cloud-based health management platform 40 uses the battery ID as the association key to merge the discharge data and charging data of the same battery to form a charge and discharge health record of the battery, and cross-validates the discharge data and charging data to identify the aging state of the battery.

[0036] In this step, the cloud-based health management platform 40 will align the flight discharge data and charging data received from the ground aggregation node according to the battery ID, so that each battery can form a complete and cross-verifiable charge and discharge health profile. Through cross-verification, it can distinguish between reversible temperature rise effects and permanent aging, thereby reducing misjudgments of battery health status.

[0037] S41. In a specific embodiment, the cloud-based health management platform 40 uses the battery ID as the association key to merge the discharge data and charging data of the same battery to form a charge-discharge health profile for the battery, and performs cross-validation on the discharge data and charging data to identify the aging state of the battery; wherein, the charge-discharge health profile includes a subset of discharge sequences and a subset of charging sequences; the cloud-based health management platform 40 uses the battery ID and the number of cycles as the association key to perform time-series alignment, classifying the Nth flight record of the same battery and the next charging record with a cycle number of N+1 as the same charge-discharge cycle, so that the charging records and flight records of the same battery that are adjacent in time form a continuous charge-discharge cycle sequence.

[0038] In this embodiment, the discharge sequence subset includes dynamic internal resistance, voltage drop curves, etc., and the charging sequence subset includes static internal resistance, charging efficiency, etc. The timestamp of the flight log comes from the onboard terminal (GPS timing), and the timestamp of the charging data comes from the charging dock system clock (e.g., Figure 2As shown, the charging dock supports Wi-Fi or Ethernet connectivity and can synchronize absolute time via NTP. Both clocks are traceable to standard UTC time, but the onboard and charging dock timers are independently synchronized, resulting in millisecond to second-level discrepancies. To address this, the cloud retrieves all historical data by battery ID, using the `cycle_count` field as the association key: after the Nth flight, the battery cycle count is incremented by 1. In subsequent charging records, `cycle_count` equals N+1. Based on this, the cloud classifies the current flight record and the next charging record as belonging to the same charge-discharge cycle. Temporally adjacent charging and flight records for the same battery form a complete "charging → discharging → recharging" cycle sequence. Since the modeling requires overall data for each cycle (such as the average internal resistance of the current flight and the static internal resistance of the current charge), cycle-level accuracy is sufficient, and precise matching of flight end time and charging start time is not required.

[0039] In this embodiment, the voltage drop curve is a recorded sequence of changes in the total battery pack voltage over time during flight. The airborne terminal periodically reads the voltage value from the BMS register (command 0x09 in the SMBus specification, returning the battery pack terminal voltage in mV) via the SMBus / CAN bus at a sampling rate of 1Hz, adds a timestamp, and writes it to the flight log. Each record in the flight log contains the voltage value at the current moment, and the entire flight log arranged by timestamps constitutes the voltage drop curve. For example, a 10-minute flight generates approximately 600 voltage sampling records, which, when connected in chronological order, form the voltage change curve for that flight. Voltage naturally fluctuates with load changes during flight (a brief drop when pushing the throttle and recovery when decelerating), which is normal. Cloud modeling focuses on whether there are abnormal, continuous voltage drops or drops exceeding a threshold; therefore, the log records an entire time series rather than a single value.

[0040] S42. In a specific embodiment, the cloud-based health management platform 40 uses the battery ID as the association key to merge the discharge data and charging data of the same battery to form a charge / discharge health record for the battery, and performs cross-validation on the discharge data and charging data to identify the aging state of the battery; wherein, the cross-validation includes: taking the average dynamic internal resistance of the entire flight segment for each flight, and using a temperature coefficient to convert the average dynamic internal resistance of the entire flight segment to a reference temperature, and comparing the converted average dynamic internal resistance with the dynamic internal resistance of the battery in the most recent preset number of flights. The historical mean and standard deviation of the internal resistance are compared. If the converted dynamic internal resistance mean exceeds the sum of the historical mean and the standard deviation of a preset multiple, and exceeds the standard in two consecutive flights, it is marked as high dynamic internal resistance. The static internal resistance measured before each charge is compared with the historical mean and standard deviation of the static internal resistance of the battery in the most recent preset number of charges. If the static internal resistance exceeds the sum of the historical mean and the standard deviation of a preset multiple, it is marked as high static internal resistance. The battery is determined to be aging only when both high dynamic internal resistance and high static internal resistance exist simultaneously.

[0041] In this embodiment, the high dynamic internal resistance measured during flight may be due to temperature rise caused by high current discharge. If the static internal resistance measured by the smart charging dock 30 is normal, it indicates that the increase in internal resistance is a reversible temperature rise effect rather than permanent aging. Only if the internal resistance is high under both conditions is it determined to be battery aging. Since the absolute values ​​of the dynamic internal resistance R_dyn (measured during flight, affected by temperature and current) and the static internal resistance R_static (measured at a constant temperature before charging) are different, they cannot be directly compared. Cross-validation compares the degree of deviation of each relative to its own historical mean.

[0042] In this embodiment, when determining high dynamic internal resistance, the average internal resistance R_dyn_avg of the entire flight segment is taken for each flight. The cloud maintains the internal resistance sequence of each battery for the most recent 30 flights, and calculates the mean μ_dyn and standard deviation σ_dyn. If R_dyn_avg is greater than μ_dyn + 2σ_dyn, and exceeds the standard for two consecutive flights, it is marked as "high dynamic internal resistance". When determining high static internal resistance, R_static is measured before each charge. The cloud maintains the static internal resistance sequence of the most recent 30 charges, and calculates the mean μ_static and standard deviation σ_static. If R_static is greater than μ_static + 1.5σ_static, it is marked as "high static internal resistance". The threshold of 1.5σ is taken because the static measurement conditions are constant, and the normal fluctuation is smaller than that of the dynamic.

[0043] In this embodiment, the dynamic internal resistance changes significantly with temperature. The same battery can differ by more than 30% at 35℃ and 5℃. Directly comparing historical averages can lead to misjudgment. Therefore, it is necessary to convert the measured internal resistance to the same reference temperature before comparison. The temperature coefficient is obtained as follows: For the battery cell model actually used in the fleet, several batteries with different aging levels are selected. Different temperature points (such as 0℃, 10℃, 25℃, 35℃, 45℃) are set in a constant temperature chamber. After being allowed to stand for a sufficient period of time at each temperature point, the internal resistance is measured. The temperature and internal resistance data are fitted with an exponential function to obtain the temperature coefficient α of the battery cell model. If the fleet uses multiple battery cell models at the same time, the corresponding α value is obtained by testing each model separately. The coefficient can be reused for newly purchased batteries of the same model without repeated testing. The corrected formula is: R_25℃ = R_measured × exp[α × (T_measured - 25)], where R_measured is the average internal resistance recorded in the flight log for this flight, and T_measured is the average temperature for this flight. The corrected R_25℃ is substituted into the above 2σ criterion for calculation, replacing the original R_measured. The static internal resistance of the charging dock is measured under constant temperature conditions in the hangar, with minimal temperature fluctuations, and therefore no compensation is made. Through the above cross-validation, reversible temperature rise effects and permanent aging can be effectively distinguished, reducing misjudgments.

[0044] S5. The cloud-based health management platform 40 uses historical charge and discharge data of the group of batteries to train a prediction model. The prediction model takes the health indicators of the battery as input and the probability of voltage drop of the battery in the next flight as output.

[0045] In this step, the cloud uses all the accumulated historical battery data to train a prediction model, transforming battery screening from a post-discovery process to a pre-prediction process, allowing the risk of voltage drop in each battery to be predicted before the flight mission.

[0046] S51. In a specific embodiment, the cloud-based health management platform 40 trains a prediction model using historical charge and discharge data of the group of batteries. The prediction model takes the battery's health indicators as input and the probability of voltage drop during the next flight as output. The prediction model is a GRU neural network. The training data uses each complete flight of each battery as a sample. The input features include the monthly average growth rate of internal resistance, the current number of cycles, the average temperature of the most recent preset number of flights, the average static internal resistance of the most recent preset number of charges, and the average cell voltage difference. The output label is whether a voltage drop occurred during the flight. When making a prediction, the data of the most recent preset number of flights of each battery are concatenated into an input sequence, and the probability of voltage drop is calculated by the GRU neural network.

[0047] In this embodiment, the feature engineering module calculates derived features before training, mainly the monthly average internal resistance growth rate (IRGR). The calculation formula is: IRGR = (Average internal resistance this month - Average internal resistance last month) / Average internal resistance last month / Number of months between intervals. The monthly average internal resistance growth rate reflects the accelerated degradation trend of battery internal resistance over time and is one of the important features for predicting voltage drop.

[0048] In this embodiment, the training data uses each complete flight of each battery as a sample. Input features include the monthly internal resistance growth rate (IRGR), the current cycle count, the average temperature of the last 3 flights, the average static internal resistance of the last 3 charges, and the average cell voltage difference. The output label indicates whether a voltage drop occurred during the flight, with a minimum voltage below 3.3V / cell marked as 1, and otherwise marked as 0. The training set covers no fewer than 1000 batteries, with no fewer than 10 flight records for each battery.

[0049] In this embodiment, the model uses a single-layer GRU with 64-dimensional hidden layers. The input sequence length is the continuous data of the most recent 5 flights. If a battery has fewer than 5 flights, it is filled with the most recent available data. During training, a mini-batch (32 samples) is taken each time, with cross-entropy as the loss function, using the Adam optimizer with a learning rate of 0.001, and training stops after 200 rounds.

[0050] In this embodiment, before a new flight mission, the data from the last 5 flights of each battery are spliced ​​into an input sequence, and the model calculates the voltage drop probability P (a floating-point number between 0 and 1) in the forward calculation. This probability value is the direct basis for battery screening in the next step.

[0051] S6. Before each flight mission, generate a recommended battery list and a deactivated battery list based on the voltage drop risk probability of each battery and the safety threshold set for this mission.

[0052] In this step, before each flight mission, maintenance personnel initiate a screening request on the client. The system reads the latest data of all available batteries, inputs it into the model to obtain the voltage drop probability P of each battery, and automatically generates a list of recommended batteries and a list of deactivated batteries based on the safety threshold set for this mission, thereby eliminating potential batteries from the source before each flight mission.

[0053] In this embodiment, the threshold is determined based on the following: Regarding flight duration, 8-minute and 15-minute flight times correspond to different requirements. The flight time can be divided into 30-second segments, and different P thresholds can be trained for each segment. The threshold corresponding to the duration is selected. Regarding formation position, the core position (key point of the main pattern) is most affected by power failure, so the P threshold is reduced, while the tolerance for edge positions is slightly higher. Regarding safety level, it is divided into three levels: strict, standard, and lenient. The division of the three levels is shown in Table 2 below.

[0054] Table 2: Safety Level Classification Table. In this embodiment, the data integrity of each battery is checked before screening. Batteries that do not meet the conditions are marked as "insufficient data" and are not included in this screening. The judgment rules are as follows: the flight record must include at least 3 complete flights (excluding abnormal flights such as takeoff failure and emergency landing), and the most recent valid flight must be within 30 days; for charging records, there must be a corresponding charging record after the most recent flight, including a complete charging curve and static internal resistance; only batteries with complete data in both categories are included in the prediction screening after the flight record and charging record are verified to be consistent by battery ID. The anomaly exclusion rules are as follows: short takeoffs and landings with a flight duration of less than 30 seconds are not included in the average calculation; if the GPS coordinates are not within the takeoff and landing area after the flight ends, it is considered as not returning to its original position, and the data of this flight is not included in the statistics; if the charging curve is interrupted (incomplete curve), the charging record is considered invalid.

[0055] In this embodiment, the deactivated list includes the reason for deactivated status (such as "internal resistance growth exceeds threshold" or "capacity decay exceeds limit"), which helps maintenance personnel determine the subsequent handling method; the recommended list is sorted from low to high risk. The screening results are displayed through the user terminal 60 and can also be sent to the smart charging dock 30, which indicates to the operator which batteries are available and which are deactivated via indicator lights or screen.

[0056] In a specific embodiment, such as Figure 3 As shown, the present invention provides an intelligent battery screening system for drone formations, including an airborne data recording and transmission module 10, an intelligent charging dock 30, a cloud-based health management platform 40, and an intelligent screening output module 50.

[0057] The airborne data recording and transmission module 10 is installed on the UAV and is used to collect discharge data including the battery ID during the flight of the UAV, and upload the discharge data to the cloud health management platform 40 after the flight ends.

[0058] The smart charging dock 30 is used to collect charging data including the battery ID when the battery is connected for charging, and upload the charging data to the cloud health management platform 40.

[0059] The cloud-based health management platform 40 is used to merge the discharge data and charging data of the same battery with the battery ID as the association key to form the battery's charge and discharge health record. The discharge data and charging data are cross-validated to identify the battery's aging state. The platform also uses the historical charge and discharge data of the group of batteries to train a prediction model. The prediction model takes the battery's health indicators as input and the probability of the battery experiencing a voltage drop in the next flight as output.

[0060] The intelligent screening output module 50 is used to generate a recommended battery list and a deactivated battery list before each flight mission, based on the voltage drop risk probability of each battery and the safety threshold set for this mission.

[0061] In a specific embodiment, the system further includes a ground data aggregation node 20. The ground data aggregation node 20 is used to receive the discharge data and the charging data, parse and clean the discharge data and the charging data, and upload them to the cloud health management platform 40 after classifying them according to the battery ID.

[0062] In a specific embodiment, the airborne data recording and transmission module 10 includes a data acquisition unit, a log storage unit, and a wireless transmission unit. The data acquisition unit is used to collect discharge data, including the battery ID, from the battery management system via an SMBus bus or CAN bus at a sampling period of 1 second. The discharge data also includes at least one of the following: cycle count, total battery pack voltage, discharge current, battery temperature, individual cell voltage, highest and lowest cell voltage difference, and DC internal resistance. The log storage unit is used to add timestamps to the discharge data and write it to an SD card in CSV format to form a flight log. The flight log is named according to the battery ID and flight time, with each line corresponding to a sampling time, and no battery-related data is transmitted externally during the entire flight. The wireless backhaul unit is used to wake up the onboard Wi-Fi module and connect to the ground wireless access point when the flight ends and the motor is detected to have changed from running to stopping and the flight duration exceeds the minimum effective duration. When the flight controller detects that the motor has stopped, the UAV is within the wireless coverage range of the ground wireless access point, and the onboard Wi-Fi module has been connected to the ground wireless access point, the newly generated flight log in the SD card is automatically pushed to the ground data aggregation node 20 through the file transfer protocol. After the transmission is completed, the onboard Wi-Fi module is automatically turned off.

[0063] In a specific embodiment, the smart charging dock 30 includes a charging data acquisition unit and a static internal resistance measurement unit. The charging data acquisition unit is used to acquire charging data, including the battery ID, through the charging interface when the battery is connected for charging. The charging data also includes a charging current curve, a voltage rise curve, the voltage of each cell, the battery temperature, and the static internal resistance. The charging data is then uploaded to the ground data aggregation node 20 via a wired network or Wi-Fi. The static internal resistance measurement unit is used to apply a short-duration high-current pulse to the battery using a hybrid pulse power characteristic method before the battery is connected and charging begins, while simultaneously measuring the change in battery terminal voltage and calculating the static internal resistance according to Ohm's law.

[0064] In a specific embodiment, the ground data aggregation node 20 includes a parsing unit, a cleaning unit, and a classification and uploading unit. The parsing unit is used to break down each record according to field delimiters, convert strings into corresponding data types, and verify whether each field is within a reasonable range, removing records with format errors or missing fields. The cleaning unit is used to deduplicatize repeatedly uploaded files based on filename, file size, and generation time, removing data exceeding a reasonable physical range, and completing the data segment using linear interpolation when the sampling interval is continuously missing for no more than 3 seconds, discarding the segment if it exceeds 3 seconds, and uniformly converting timestamps to UTC time. The classification and uploading unit is used to establish a directory structure based on battery ID, storing multiple flight records and charging records of the same battery ID by date, and uploading them to the cloud health management platform 40 via wired broadband or 4G / 5G.

[0065] In a specific embodiment, the cloud-based health management platform 40 includes a data fusion module, a cross-validation module, a feature engineering module, and a model training module. The data fusion module is used to perform time-series alignment using battery ID and cycle count as association keys, classifying the Nth flight record of the same battery and the next charging record with cycle count N+1 as the same charge-discharge cycle, and forming a charge-discharge health profile containing a subset of discharge sequences and a subset of charging sequences. The cross-validation module is used to take the average dynamic internal resistance for each flight segment, convert the average dynamic internal resistance for the entire flight segment to a reference temperature using a temperature coefficient, and compare it with the historical average and standard deviation of the dynamic internal resistance of the battery in the most recent 30 flights. If it exceeds the sum of the historical average and twice the standard deviation, and exceeds the standard in two consecutive flights, it is marked as high dynamic internal resistance. For the static internal resistance measured before each charge, it is compared with the historical average and standard deviation of the static internal resistance of the battery in the most recent 30 charges. If it exceeds the sum of the historical average and 1.5 times the standard deviation, it is marked as high static internal resistance. The battery is determined to be aging only when both high dynamic internal resistance and high static internal resistance exist simultaneously. The feature engineering module is used to calculate the monthly average growth rate of internal resistance. The calculation formula is (average internal resistance this month - average internal resistance last month) / average internal resistance last month / number of months between intervals. The model training module is used to train the GRU neural network using historical charge and discharge data of the group of batteries. The training data is based on each complete flight of each battery as a sample. The input features include the monthly average growth rate of internal resistance, the current number of cycles, the average temperature of the last 3 flights, the average static internal resistance of the last 3 charges, and the average cell voltage difference. The output label is whether a voltage drop occurred during the flight. During prediction, the data of the last 5 flights of each battery are concatenated into an input sequence, and the probability of voltage drop risk is calculated by the forward calculation of the GRU neural network.

[0066] In a specific embodiment, the intelligent screening output module 50 is also used to check the data integrity of each battery before screening, marking batteries with fewer than 3 complete flights, the most recent valid flight more than 30 days ago, or missing corresponding charging records as having insufficient data and not participating in the screening; and determining the voltage drop risk probability threshold based on the flight duration, formation position, and safety level of this mission, classifying batteries into three categories: recommended, usable, and deactivated; the deactivated list includes the reasons for deactivation, the recommended list is sorted from low to high risk, and the screening results are displayed through the user terminal 60 and can be sent to the intelligent charging dock 30, which indicates the usable and deactivated batteries to the operator through indicator lights or screen.

[0067] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for intelligent battery selection in drone formations, characterized in that, Includes the following steps: During the flight of the drone, the onboard data recording and transmission module on the drone collects discharge data including the battery ID, and does not send battery-related data to the outside during the entire flight. After the flight, the discharge data will be uploaded to the cloud-based health management platform; When the battery is connected to the smart charging dock for charging, the smart charging dock collects charging data including the battery ID and uploads the charging data to the cloud health management platform; The cloud-based health management platform uses battery ID and cycle count as association keys for time-series alignment, classifying the Nth flight record of the same battery and the next charging record with cycle count N+1 as the same charge-discharge cycle. This allows time-adjacent charging records and flight records of the same battery to form a continuous charge-discharge cycle sequence, thereby fusing the discharge data and charging data of the same battery to form a charge-discharge health profile for the battery. The discharge data and charging data are then cross-validated to identify the aging state of the battery. The cross-validation of the discharge and charge data to identify the battery's aging status includes: taking the average dynamic internal resistance for each flight segment, converting the average dynamic internal resistance to a reference temperature using a temperature coefficient, comparing the converted average dynamic internal resistance with the historical average and standard deviation of the battery's dynamic internal resistance over the most recent preset number of flights; if the converted average dynamic internal resistance exceeds the sum of the historical average and the standard deviation of a preset multiple, and exceeds the standard in two consecutive flights, it is marked as having high dynamic internal resistance; comparing the static internal resistance measured before each charge with the historical average and standard deviation of the battery's static internal resistance over the most recent preset number of charges; if the static internal resistance exceeds the sum of the historical average and the standard deviation of a preset multiple, it is marked as having high static internal resistance; the battery is determined to be aging only when both high dynamic internal resistance and high static internal resistance exist simultaneously. The cloud-based health management platform trains a prediction model using historical charge and discharge data of the group of batteries. The prediction model takes the health indicators of the batteries as input and the probability of voltage drop in the next flight as output. Before each flight mission, a recommended list of batteries to use and a list of batteries to be deactivated are generated based on the probability of voltage drop risk for each battery and the safety threshold set for the mission.

2. The intelligent battery selection method for UAV formations according to claim 1, characterized in that, The prediction model is a GRU neural network; The cloud-based health management platform trains a prediction model using historical charge and discharge data of the group of batteries. The prediction model takes battery health indicators as input and the probability of a voltage drop during the next flight as output, and includes: The training data uses each complete flight of each battery as a sample. The input features include the monthly average growth rate of internal resistance, the current number of cycles, the average temperature of the most recent preset number of flights, the average static internal resistance of the most recent preset number of charges, and the average cell voltage difference. The output label is whether a voltage drop occurred during the flight. When making predictions, the data from the most recent preset number of flights for each battery are concatenated into an input sequence, and the voltage drop risk probability is calculated by the GRU neural network.

3. The intelligent battery selection method for UAV formations according to claim 1, characterized in that, After the flight, uploading the discharge data to the cloud-based health management platform includes: The discharge data is first automatically transmitted wirelessly to the ground data aggregation node by the airborne data recording and transmission module, and then uploaded to the cloud health management platform by the ground data aggregation node. When the motor is detected to have stopped and the flight duration exceeds the minimum effective duration, the onboard wireless module is activated and connected to the ground wireless access point. When the flight controller detects that the motor has stopped, the UAV is within the wireless coverage area of ​​the ground wireless access point, and the onboard wireless module is connected to the ground wireless access point, the discharge data is automatically pushed to the ground data aggregation node via the file transfer protocol. After the transmission is completed, the onboard wireless module is automatically turned off.

4. The intelligent battery selection method for UAV formations according to claim 1, characterized in that, The airborne data recording and transmission module collects the discharge data from the battery management system according to a set sampling period via the airborne bus, adds a timestamp to the discharge data and writes it into the airborne local storage medium to form a flight log. The airborne bus is an SMBus bus or a CAN bus, the sampling period is set to 1 second, and the airborne local storage medium is an SD card; The discharge data also includes at least one of the following: number of cycles, total battery pack voltage, discharge current, battery temperature, voltage of each cell, voltage difference between the highest and lowest cells, and DC internal resistance. The flight log is a CSV file, named by battery ID and flight time, with each line corresponding to a sampling time.

5. The intelligent battery selection method for UAV formations according to claim 1, characterized in that, The charging data also includes charging current curves, voltage rise curves, voltage of each cell, battery temperature, and static internal resistance. Before the battery is connected to the smart charging dock and charging begins, the smart charging dock applies a short-duration high-current pulse to the battery using the hybrid pulse power characteristic method, while simultaneously measuring the change in battery terminal voltage and calculating the static internal resistance based on Ohm's law.

6. The intelligent battery selection method for UAV formations according to claim 1, characterized in that, Both the discharge data and the charging data are analyzed and cleaned by the ground data aggregation node, and then uploaded to the cloud health management platform after being classified by battery ID. The parsing process includes splitting each record by field delimiter, converting the string into the corresponding data type, verifying whether each field is within a reasonable range, and removing records with format errors or missing fields. The cleaning process includes deduplicating the same file uploaded repeatedly by filename, file size and generation time, removing data that exceeds a reasonable physical range, and completing the data segment by linear interpolation when the sampling interval is continuously missing for no more than a preset time, discarding the data segment when it exceeds the preset time, and uniformly converting the timestamp to UTC time. The classification involves establishing a directory structure based on battery ID, and storing multiple flight records and charging records for the same battery ID by date.

7. A smart battery selection system for drone formations, characterized in that, Includes an onboard data recording and transmission module, a smart charging dock, a cloud-based health management platform, and a smart screening and output module; The airborne data recording and transmission module is installed on the drone and is used to collect discharge data including the battery ID during the drone's flight. It does not send battery-related data to the outside during the entire flight and uploads the discharge data to the cloud health management platform after the flight ends. The smart charging dock is used to collect charging data including the battery ID when the battery is connected for charging, and upload the charging data to the cloud health management platform; The cloud-based health management platform includes a data fusion module, a cross-validation module, a feature engineering module, and a model training module; The data fusion module is used to perform time-series alignment using battery ID and cycle count as association keys, classifying the Nth flight record of the same battery and the next charging record with cycle count N+1 as the same charge-discharge cycle, so that the charging records and flight records of the same battery that are adjacent in time form a continuous charge-discharge cycle sequence, so as to fuse the discharge data and charging data of the same battery to form a charge-discharge health record including a subset of discharge sequence and a subset of charging sequence. The cross-validation module is used to cross-validate the discharge data and charging data to identify the aging state of the battery. The cross-validation includes: taking the average dynamic internal resistance of the entire flight segment for each flight, converting the average dynamic internal resistance of the entire flight segment to a reference temperature using a temperature coefficient, comparing the converted average dynamic internal resistance with the historical average and standard deviation of the dynamic internal resistance of the battery in the most recent preset number of flights, and if the converted average dynamic internal resistance exceeds the sum of the historical average and the standard deviation of a preset multiple and exceeds the standard in two consecutive flights, it is marked as high dynamic internal resistance; comparing the static internal resistance measured before each charge with the historical average and standard deviation of the static internal resistance of the battery in the most recent preset number of charges, and if the static internal resistance exceeds the sum of the historical average and the standard deviation of a preset multiple, it is marked as high static internal resistance; the battery is determined to be aging only when both high dynamic internal resistance and high static internal resistance exist simultaneously. The feature engineering module is used to calculate the average monthly growth rate of internal resistance; The model training module is used to train a GRU neural network using historical charge and discharge data of the group of batteries to obtain a prediction model. The prediction model takes the health index of the battery as input and the probability of voltage drop of the battery in the next flight as output. The intelligent filtering output module is used to generate a recommended battery list and a deactivated battery list before each flight mission, based on the voltage drop risk probability of each battery and the safety threshold set for this mission.

8. The intelligent battery selection system for drone formations according to claim 7, characterized in that, It also includes ground data aggregation nodes; The ground data aggregation node is used to receive the discharge data and the charging data, parse and clean the discharge data and the charging data, and upload them to the cloud health management platform after classifying them by battery ID. The intelligent screening output module is also used to check the data integrity of each battery before screening. Batteries with fewer than a preset number of flight records, more than a preset number of days for the most recent valid flight, or missing corresponding charging records are marked as having insufficient data and will not participate in the screening. The module also determines the voltage drop risk probability threshold based on the flight duration, formation position, and safety level of this mission, and classifies the batteries into three categories: recommended, usable, and deactivated.

Images