Load balancing control method of unmanned aerial vehicle battery BMS system

Abstract

The invention discloses a load balancing control method for an unmanned aerial vehicle battery BMS system, and relates to the technical field of computers, and the method comprises the steps: collecting a multi-source flight state and battery parameters in real time, and carrying out the filtering; dynamically estimating the SOC of each battery cell by fusing an open-circuit voltage method and an ampere-hour integral method, and calculating a consistency deviation; identifying a flight stage based on the lightweight decision tree, and dynamically adjusting a balance trigger threshold; when the conditions are met, single-channel equalization is carried out on the SOC extreme value battery cell through bidirectional flyback topology, and the optimal equalization current is analyzed and solved; meanwhile, closed-loop energy efficiency feedback is introduced, and the driving duty ratio is dynamically adjusted according to a copper loss and iron loss model so as to suppress loss. According to the method, quick response, efficient low-noise balance and whole machine endurance improvement under the high-dynamic working condition are achieved, the SOC consistency convergence time is remarkably shortened, and the balance energy consumption is lower than the preset proportion upper limit of the total discharge energy.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a load balancing control method for a drone battery management system (BMS). Background Technology

[0002] With the rapid development of drone technology, higher demands are being placed on the energy density, safety, and endurance of power battery systems. As a core component ensuring the safe and efficient operation of drone battery packs, the battery management system's load balancing control capability directly impacts battery life and flight reliability. Especially in high-dynamic flight missions, batteries must frequently withstand complex conditions such as high-rate charging and discharging, drastic temperature fluctuations, and voltage surges. This poses a severe challenge to the BMS's real-time response capabilities, energy efficiency management accuracy, and system lightweight design.

[0003] The load balancing control of the drone battery management system focuses on maintaining the consistency of the state of charge (SOC) of multiple cells. It aims to adjust the energy distribution of each individual cell actively or passively to suppress capacity decay and performance imbalance caused by manufacturing tolerances, aging differences, or uneven thermal environments. An ideal balancing strategy should achieve millisecond-level response and high-precision SOC synchronization while minimizing energy loss, to support the stable power supply requirements of the drone during typical flight maneuvers such as climb, hovering, and rapid descent.

[0004] However, existing BMS equalization control methods have significant shortcomings when dealing with the special operating conditions of UAVs. First, mainstream active equalization architectures rely on centralized computing and multi-module collaboration, introducing state evaluation indices, path optimization algorithms, and adaptive current regulation mechanisms. While performing well in large-scale energy storage scenarios such as electric vehicles, their high computational complexity and large hardware resource consumption make them difficult to deploy on low-power embedded platforms of airborne BMS. Second, existing strategies do not fully couple UAV flight state parameters, resulting in equalization decisions being disconnected from actual battery dynamics. This leads to lag in response to sudden voltage drops or temperature gradient changes, failing to promptly suppress the expansion of SOC deviation between cells. Furthermore, traditional equalization current control lacks detailed modeling of the conduction losses of switching devices and the efficiency of magnetic components, causing unnecessary energy waste and further compressing the already limited flight time. In addition, most solutions use fixed threshold triggering for equalization, ignoring the differentiated tolerance for battery consistency at different flight stages. This leads to over-equalization during non-critical periods, while failure due to delayed start-up during high-risk phases. Summary of the Invention

[0005] The purpose of this invention is to provide a load balancing control method for UAV battery BMS systems to solve the problems of slow response and low energy efficiency of traditional fixed threshold balancing strategies in high-dynamic flight scenarios.

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

[0007] Load balancing control methods for drone battery BMS systems include:

[0008] Step S1: Real-time acquisition of multi-source flight status and battery parameter data: The onboard sensors synchronously acquire the three-axis acceleration, attitude angle, motor load current, terminal voltage, temperature and current of each individual battery cell of the UAV, and filter the raw data based on the sliding time window to generate a denoised state vector.

[0009] Step S2, dynamically calculate the state of charge (SOC) and consistency deviation: adopt a dual-mode SOC estimation model that combines the open-circuit voltage method and the ampere-hour integral method, and combine the temperature compensation factor to correct the SOC of each cell, and calculate the difference between the maximum SOC and the minimum SOC as the consistency deviation.

[0010] Step S3, construct a flight phase identification and equalization trigger threshold mapping mechanism: based on the flight state parameters, the current flight phase is identified as takeoff, climb, hover, rapid descent or return to home using a lightweight decision tree algorithm, and the equalization start threshold is dynamically adjusted according to the preset phase-threshold mapping table, wherein the takeoff and rapid descent phase thresholds are set to the first preset ratio, the hovering phase thresholds are set to the second preset ratio, and the return to home phase thresholds are set to the third preset ratio.

[0011] Step S4, perform lightweight balancing path planning and current optimization: when the balancing triggering conditions are met, based on the cell SOC ranking results, select several cells with the highest and lowest SOC to form a balancing pair, adopt a bidirectional flyback topology, and solve the optimal balancing current analytically to maximize energy transfer efficiency, while limiting the switching frequency to no more than the preset frequency upper limit to reduce electromagnetic interference.

[0012] Step S5, implement closed-loop energy efficiency feedback and loss suppression: during the equalization process, monitor the input power and output power of the equalization circuit in real time, calculate the instantaneous conversion efficiency, and if the efficiency is lower than the preset efficiency threshold, automatically reduce the equalization current amplitude, and dynamically adjust the drive duty cycle according to the copper loss and iron loss model of the magnetic components to ensure that the overall equalization energy consumption is lower than the preset ratio upper limit of the total discharge energy.

[0013] In step S1, the sliding time window length is set to a predetermined time period, and the filtering algorithm uses a first-order low-pass digital filter with a preset cutoff frequency to effectively suppress high-frequency noise while retaining voltage change characteristics.

[0014] In step S2, the dual-mode SOC estimation model switches to the open-circuit voltage method when the battery resting time is greater than or equal to a preset time threshold. During the dynamic discharge stage, the recursive ampere-hour integral method with a forgetting factor is adopted. The forgetting factor λ is a preset constant. The temperature compensation factor is obtained by the lookup table method. For each preset temperature range change, the SOC correction amount is a preset proportional value.

[0015] The lightweight decision tree algorithm in step S3 contains only nodes of a predetermined number of layers. The input features are acceleration modulus, pitch angle change rate and motor current standard deviation. The training samples are derived from multiple real flight mission data. The inference time is less than a preset time threshold. It can run in real time on an embedded microcontroller with a preset main frequency.

[0016] In step S4, the number of equalization pairs is limited to a preset proportion upper limit of the total number of cells in the battery pack, and only one equalization pair is activated in each equalization operation to avoid control conflicts caused by multiple channels operating in parallel; the optimal equalization current Through formula Calculation, where To balance the equivalent resistance of the circuit, Both the on-resistance of the switching device and the resistance of the switching device are obtained through offline calibration.

[0017] In step S5, the copper loss of the magnetic element With iron loss From the formulas respectively and Modeling, in which This is the effective value of the winding current. The DC resistance of the winding. Let f be a material constant, B be the switching frequency, B be the magnetic flux density, α be a preset exponential constant, and D be the drive duty cycle based on the total losses. Make dynamic adjustments to meet the requirements. The minimum condition.

[0018] The equalization circuit adopts an integrated PCB layout, the magnetic components are smaller than a preset volume threshold, the switching devices are gallium nitride field-effect transistors, the on-resistance is smaller than a preset resistance threshold, and the overall power consumption of the equalization module is lower than a preset power threshold in standby mode.

[0019] The load balancing control method of the UAV battery BMS system also includes uploading the timestamp, duration, transferred power, and SOC convergence accuracy of the balancing event to the ground station after each flight mission, in order to build a battery health status assessment database and support subsequent flight strategy optimization.

[0020] The SOC convergence accuracy is defined as the standard deviation of SOC of each cell within a predetermined time period after the equalization is completed. This value is required to be less than a preset proportional threshold. If the convergence accuracy exceeds the standard in multiple consecutive tasks, the BMS self-check process will be triggered and an anomaly will be reported.

[0021] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0022] This invention breaks through the limitations of traditional fixed threshold equalization. Through flight phase identification and dynamic threshold mapping mechanism, the equalization strategy is accurately matched with the actual working conditions of the UAV. The equalization is started in advance in high-risk phases, and the response delay is less than the preset time threshold, while the average delay of traditional methods is more than the preset time threshold.

[0023] This invention dynamically adjusts the driving parameters based on a physical model of copper and iron losses, so that the equalization conversion efficiency is stabilized above a preset efficiency threshold, improving the preset efficiency gain compared to the unmodeled scheme, and reducing the preset energy consumption ratio of a single equalization. By analytically solving for the optimal equalization current, it avoids additional heat loss caused by overcurrent, while limiting the switching frequency, taking into account both efficiency and electromagnetic compatibility, and extending the overall battery life within a preset battery life gain range.

[0024] This invention employs a limited equalization pair quantity and a single-channel activation mechanism to eliminate multi-channel interference, shortening the SOC consistency convergence time to within a preset time threshold and accelerating it by a preset factor. The introduction of an energy efficiency feedback loop and a post-task data upload mechanism not only ensures the quality of a single equalization but also provides data support for the entire battery lifecycle management, improving the long-term stability of SOC estimation error to within a preset error range. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall technical solution architecture of the load balancing control method for the UAV battery BMS system proposed in this invention.

[0026] Figure 2 This is a schematic diagram of the core principle framework of the flight phase identification and dynamic equilibrium threshold mapping mechanism proposed in this invention. Detailed Implementation

[0027] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0028] Example 1

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0030] Step S1 collects multi-source flight status and battery parameter data in real time. The onboard sensors synchronously acquire the UAV's three-axis acceleration, attitude angle, motor load current, terminal voltage, temperature and current of each individual battery cell, and filter the raw data based on a sliding time window to generate a denoised state vector.

[0031] Specifically, in step S1, the inertial measurement unit (IMU) deployed inside the UAV continuously outputs triaxial acceleration signals at a sampling frequency of 200Hz. Its measurement range is ±16g, and its zero-bias stability is better than 0.05mg. Simultaneously, the high-precision gyroscope module synchronously outputs pitch angle θ, roll angle φ, and yaw angle ψ, with an angular resolution of no less than 0.01°. The motor driver incorporates a Hall current sensor to monitor the phase current of the four brushless DC motors in real time. to The sampling rate is 1kHz, and the accuracy is ±1% of full scale. The battery pack consists of 12 series-connected lithium polymer cells, each equipped with an independent voltage detection channel and an NTC temperature sensor. The voltage sampling resolution is 1mV, and the temperature sampling accuracy is ±0.5℃. All battery parameters are transmitted to the main control BMS chip via an isolated SPI bus at a rate of 500Hz. The raw data stream is buffered through a 50ms sliding time window, which is implemented using a circular queue structure with a maximum storage capacity of 10 sampling points. The filtering algorithm uses a first-order low-pass digital filter, and its difference equation is expressed as: ,in , The sampling period is This is the time constant. The cutoff frequency is set to 30Hz, corresponding to the time constant. The system ensures effective suppression of high-frequency noise introduced by the switching power supply while preserving the rapid voltage drop characteristics of the battery terminal voltage under sudden motor load. The high-frequency noise is mainly concentrated above 100Hz, and the typical response time of the rapid voltage drop characteristic is on the order of 10ms. After filtering, the system constructs an 18-dimensional state vector. This vector serves as the unified input data source for subsequent SOC estimation and flight phase identification, and its elements are as follows: .

[0032] Step S2 dynamically calculates the state of charge (SOC) and consistency deviation. A dual-mode SOC estimation model combining the open-circuit voltage method and the ampere-hour integral method is adopted. The SOC of each cell is corrected by combining the temperature compensation factor, and the difference between the maximum SOC and the minimum SOC is calculated as the consistency deviation.

[0033] Specifically, in step S2, the SOC estimation engine runs on an ARM Cortex-M7 core microcontroller with a main frequency of 120MHz. Its internal Flash memory stores a pre-calibrated OCV-SOC lookup table, which was obtained through a 0.05C rate charge-discharge experiment under constant temperature conditions. This lookup table contains 101 discrete points from SOC 0% to 100% in 1% steps. When any cell is idle for a certain period of time... At time s, the system determines that it has entered steady-state conditions and switches to the open-circuit voltage (OCV) method: read the terminal voltage of the cell. The initial value is obtained by looking up the OCV-SOC table through linear interpolation. .like If s is a variable, then the recursive ampere-hour integral method with a forgetting factor is used, and its iterative formula is:

[0034]

[0035] Where η is the coulomb efficiency, which is 0.99 by default; Let be the battery current in the kth sampling period, where a positive current value represents discharge; 2ms The rated capacity of the battery cell is expressed in Ah. This is for temperature compensation. The forgetting factor λ is set to 0.995, and the historical current integration error is exponentially decayed and weighted to prevent long-term cumulative drift. The temperature compensation factor is obtained through a two-dimensional lookup table method: based on the cell temperature... Using the current SOC as an index, the system retrieves the correction amount from a pre-stored compensation matrix. This matrix is ​​constructed based on the Arrhenius equation fitted to experimental data, with a SOC correction of 0.8% for every 10°C temperature change. After completing the SOC correction for each cell, the system iterates through all 12 SOC values ​​and extracts the maximum value. and minimum value Calculate the consistency deviation This deviation serves as the core criterion for determining the equilibrium trigger.

[0036] Step S3 constructs a flight phase identification and equalization trigger threshold mapping mechanism. Based on the flight state parameters, a lightweight decision tree algorithm is used to identify the current flight phase as takeoff, climb, hover, rapid descent, or return. The equalization start threshold is dynamically adjusted according to a preset phase-threshold mapping table. The takeoff and rapid descent phase thresholds are set to a first preset ratio, the hovering phase thresholds are set to a second preset ratio, and the return phase thresholds are set to a third preset ratio.

[0037] Specifically, in step S3, the decision tree model depth is limited to 3 layers, containing 7 nodes including 1 root node, 2 internal nodes, and 4 leaf nodes. Its training process is completed using the CART algorithm on a dataset containing 500 real flight missions. Input feature vector It consists of three parts: acceleration magnitude Pitch angle change rate The standard deviation of the current of the four motors was calculated using the five-point central difference method. The root node split threshold is set to... g: If If 'g', it is determined to be a high-dynamic phase, i.e., takeoff or rapid descent; otherwise, proceed to the next judgment. For the high-dynamic branch, based on... Distinguishing: If 15° / s, then it is classified as takeoff; if -20° / s is classified as a rapid descent. For low dynamic branches... g, according to Judgment: If If A, it is determined to be hovering; otherwise, the remaining battery power is further checked. The system checks if the condition is met; if so, it initiates a return to base; otherwise, it initiates a climb. The entire inference process requires only 3 floating-point comparisons and 2 array indexing operations, taking less than 80μs, which meets real-time requirements at a 120MHz clock frequency. The phase identification results are mapped to a balanced trigger threshold table: thresholds corresponding to takeoff and rapid descent phases. % % corresponding to the hovering phase % corresponding to the return phase %; the hovering threshold is used during the climb phase. The system compares ΔSOC with the current phase threshold in real time, when... When the equalization trigger flag is set.

[0038] Step S4 performs lightweight balancing path planning and current optimization. When the balancing triggering conditions are met, based on the cell SOC ranking results, several cells with the highest and lowest SOC are selected to form a balancing pair. A bidirectional flyback topology is adopted, and the optimal balancing current is solved analytically to maximize energy transfer efficiency. At the same time, the switching frequency is limited to not exceed the preset frequency upper limit to reduce electromagnetic interference.

[0039] Specifically, in step S4, the balanced path planning module first performs a fast sort on the SOC values ​​of the 12 cells using a heap sort algorithm, with a time complexity of [missing information]. Generate an ascending sequence list Equilibrium on quantity The limit is no more than 25% of the total number of battery cells, that is System selection That is, the highest SOC and That is, the lowest SOC constitutes the first equilibrium pair, if Then continue selecting and And so on. Each equalization operation activates only one equalization pair, while the remaining equalization channels remain in a high-impedance state to avoid control conflicts and magnetic coupling interference caused by multiple channels operating in parallel. The equalization circuit adopts an integrated bidirectional flyback topology, including one high-frequency transformer, two gallium nitride (GaN) field-effect transistors, and two synchronous rectifiers. and output filter capacitor. Switching frequency. Fixed at 500kHz, below the preset frequency upper limit of 600kHz, to suppress conducted and radiated electromagnetic interference. Optimal equalization current. Calculated analytically:

[0040]

[0041] in, To balance the equivalent resistance of the circuit, including the transformer winding resistance, PCB trace resistance and contact resistance, the rated value is 25mΩ; The on-resistance of the GaN switching device is calibrated to 8mΩ. Both values ​​were obtained through an offline calibration process: a known test current was applied under a constant temperature of 25°C. A. Measure the voltage drop ΔV in the equalization circuit and calculate... After deducting the theoretical trace resistance, the separation is... and . The calculation results are limited to ensure they do not exceed the maximum allowable equalization current of the battery cell, 3A. The drive signal generation module then... Value setting PWM duty cycle ,in A, This ensures that the magnetic core remains unsaturated.

[0042] In the load balancing control method of the aforementioned UAV battery BMS system, step S5 implements closed-loop energy efficiency feedback and loss suppression. During the balancing process, the input power and output power of the balancing circuit are monitored in real time, and the instantaneous conversion efficiency is calculated. If the efficiency is lower than a preset efficiency threshold, the balancing current amplitude is automatically reduced, and the drive duty cycle is dynamically adjusted based on the copper loss and iron loss model of the magnetic components to ensure that the overall balancing energy consumption is lower than the preset upper limit of the total discharge energy. Specifically, in step S5, the power monitoring unit synchronously collects the input current of the balancing circuit through a high-bandwidth current probe with a bandwidth of 10MHz and a differential voltage amplifier. With voltage and output current With voltage The sampling rate is 1MHz. Instantaneous input power. Output power Conversion efficiency Preset efficiency threshold ,like Then the equalization current controller will Multiply by the attenuation coefficient The driving parameters are recalculated. Simultaneously, the system undergoes refined control based on the magnetic component loss model: copper loss... With iron loss They are modeled as follows:

[0043]

[0044] in The effective value of the primary winding current of the transformer is calculated by integrating the real-time current waveform. The DC resistance of the winding is 18mΩ. The constant of the PC95 ferrite core material. f is the switching frequency of 500kHz; B is the magnetic flux density, calculated using the volt-second product formula. calculate, The number of turns on the original side. Where is the effective cross-sectional area of ​​the magnetic core; α is taken as 2.3. Total loss. The driving duty cycle D is based on the minimum condition. Dynamic adjustments are made. The optimal duty cycle derived from this condition is... satisfy: .because Both B and D are linear functions of D. The system is solved online using a first-order Taylor expansion. The PWM register is updated with a 50μs cycle to balance energy consumption. Accumulated and combined with total discharge energy Comparison, requirements 1.5%. If this limit is exceeded, the load balancing will be forcibly terminated and the event log will be recorded.

[0045] After each flight mission, the system executes a data upload process: uploading the timestamp, duration, transferred charge, and SOC convergence accuracy of the current equilibrium event to the ground station. The timestamp is in UTC format, and the transferred charge is recorded via... The SOC convergence accuracy is obtained through integration. It is defined as the standard deviation of the SOC of each cell within 60 seconds after the equalization process ends. ,Require < 0.8%. If in 3 consecutive tasks This triggers the BMS self-test process: sequentially detecting voltage sampling channel offset, temperature sensor drift, and equalization. An abnormal conduction voltage drop was detected, and fault code 0x2A was reported to the flight control computer via the CAN bus. The equalization circuit adopts a 4-layer FR-4 PCB integrated layout, and the magnetic component, namely the EE13 magnetic core, has a volume of [missing information]. The volume is less than the preset volume threshold of 150 mm³. The switching device is a gallium nitride field-effect transistor with a conduction resistance of [missing information]. The resistance is mΩ, which is less than the preset resistance threshold of 10mΩ. The overall equalization module consumes less than 5mW in standby mode and is powered by the BMS main power rail through LDO regulation.

[0046] To verify the effectiveness of the above method, a specific application example is constructed as follows: A hexacopter logistics drone performs urban delivery tasks. The battery pack is configured with 12S2P lithium polymer batteries, with a nominal voltage of 44.4V and a capacity of 16000mAh. The flight mission is divided into five phases: 0-30 seconds for ground launch, 30-45 seconds for vertical takeoff, 45-180 seconds for high-altitude hovering, 180-195 seconds for emergency obstacle avoidance and rapid descent, and 195-300 seconds for low-battery return-to-home. During the vertical takeoff phase, the IMU detects… , The decision tree identifies it as "takeoff," and the equilibrium threshold is set to 3.5%. At this point, the cell SOC distribution is... , This triggers a balancing process. The system selects the cells with the highest SOC (cell #3, 82.1%) and the lowest SOC (cell #9, 78.3%) to form a balancing pair, and calculates... The equilibrium period lasted 12 seconds, transferring 13.8mAh of energy, and the ΔSOC dropped to 2.9% at the end. During the emergency obstacle avoidance and rapid descent phase, , The reading was identified as a "sharp drop," but the threshold remained at 3.5%. At this point, due to the large current discharge causing a widening of the cell temperature difference, the SOC distribution was... , This triggers the equilibrium again. However, real-time monitoring shows The system automatically reduced the current to 1.47A. After 15 seconds of equalization, ΔSOC converged to 3.2%. The accuracy requirements were met. The total energy consumption during the equalization process was 218mAh, accounting for 1.7% of the total discharge energy of 12800mAh, slightly exceeding the 1.5% upper limit. However, during the return phase, the cumulative energy consumption was successfully controlled within the threshold by reducing the equalization current. After the mission was completed, all equalization data was uploaded to the cloud battery health database via the 4G module for updating the cell aging model.

[0047] Example 2

[0048] In another embodiment, the lightweight decision tree algorithm in step S3 is replaced by a support vector machine (SVM) classifier, whose kernel function uses a radial basis function and a penalty parameter. kernel coefficient The training sample was expanded to 1000 flight missions, covering extreme environments such as high altitude, high temperature, and strong wind. The input feature was expanded to include the rate of change of air pressure at altitude. GPS velocity vector magnitude This generates a 5-dimensional feature vector. Although the SVM inference time increases to 150μs, it still meets the 200Hz control cycle requirement. The stage-threshold mapping mechanism is correspondingly expanded: a new "strong wind anti-interference" stage is added, with a threshold set to 4.0%; the threshold for the "high temperature cruise" stage is set to 4.5%. The remaining steps are consistent with Example 1. This scheme reduces the equilibrium false trigger rate to 0.3% under complex weather conditions, but increases memory usage to 8KB, making it suitable for high-performance BMS platforms.

[0049] Example 3

[0050] In another embodiment, the bidirectional flyback topology in step S4 is replaced with a capacitor storage type flying capacitor balancing circuit. This circuit includes 11 flying capacitors (one between each adjacent cell) and 22 MOSFET switches. The balancing path planning uses a greedy algorithm: each time, the adjacent cell pair with the largest SOC difference is selected for energy transfer. The optimal balancing current calculation formula is modified as follows:

[0051]

[0052] in The equivalent series resistance of the flying capacitor is 15mΩ. The on-resistance of the MOSFET is 12mΩ. Since there are no magnetic components, the iron loss model in step S5 is removed, retaining only the copper loss model. This solution reduces single-cycle equalization energy consumption to 1.2% of total discharge energy on a 12S battery pack, but extends the equalization convergence time to 45 seconds. It is suitable for long-endurance reconnaissance UAVs that are extremely sensitive to weight but do not require high equalization speed.

[0053] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A load balancing control method for a UAV battery BMS system, characterized in that, Includes the following steps: Step S1: Collect multi-source flight status and battery parameter data. Obtain the UAV's three-axis acceleration, attitude angle, motor load current, terminal voltage, temperature and current of each individual battery cell through airborne sensors. Filter the raw data based on a sliding time window to generate a denoised state vector. Step S2: Dynamically calculate the state of charge (SOC) and consistency deviation. Use a dual-mode SOC estimation model that combines the open-circuit voltage method and the ampere-hour integral method. Combine the temperature compensation factor to correct the SOC of each cell and calculate the difference between the maximum and minimum SOC as the consistency deviation. Step S3: Construct a flight phase identification and equalization trigger threshold mapping mechanism. Use a lightweight decision tree algorithm to identify the current flight phase as takeoff, climb, hover, rapid descent or return. Dynamically adjust the equalization start threshold according to the preset phase-threshold mapping table. The takeoff and rapid descent phase thresholds are set to the first preset ratio, the hovering phase thresholds are set to the second preset ratio, and the return phase thresholds are set to the third preset ratio. Step S4: Perform lightweight balancing path planning and current optimization. When the balancing triggering conditions are met, select several cells with the highest and lowest SOC to form a balancing pair. Use a bidirectional flyback topology and solve the optimal balancing current analytically to maximize energy transfer efficiency. At the same time, limit the switching frequency to no more than the preset frequency limit to reduce electromagnetic interference. Step S5: Implement closed-loop energy efficiency feedback and loss suppression. During the equalization process, monitor the input power and output power of the equalization circuit in real time, calculate the instantaneous conversion efficiency, and if the efficiency is lower than the preset efficiency threshold, automatically reduce the equalization current amplitude and dynamically adjust the drive duty cycle according to the copper loss and iron loss model of the magnetic components to ensure that the overall equalization energy consumption is lower than the preset upper limit of the total discharge energy.

2. The load balancing control method for a UAV battery BMS system according to claim 1, characterized in that, The sliding time window is implemented using a circular queue structure, and the filtering algorithm is a first-order low-pass digital filter with a cutoff frequency set to 30Hz, which preserves voltage change characteristics and suppresses high-frequency noise.

3. The load balancing control method for a UAV battery BMS system according to claim 1, characterized in that, The dual-mode SOC estimation model switches to the open-circuit voltage method when the cell resting time is greater than or equal to 30 seconds. During the dynamic discharge stage, a recursive ampere-hour integral method with a forgetting factor is used, with the forgetting factor set to 0.

995. The temperature compensation factor is obtained through a two-dimensional lookup table method, and the SOC correction amount is 0.8% for every 10°C change in temperature.

4. The load balancing control method for a UAV battery BMS system according to claim 1, characterized in that, The lightweight decision tree algorithm is limited to a depth of 3 layers. The input features include acceleration magnitude, pitch angle change rate and motor current standard deviation. The inference process is completed through floating-point comparison and array indexing, taking less than 80 microseconds. It can run in real time on an embedded microcontroller with a main frequency of 120MHz.

5. The load balancing control method for a UAV battery BMS system according to claim 1, characterized in that, The number of equalization pairs does not exceed 25% of the total number of cells in the battery pack. Each equalization operation activates only one equalization pair, while the remaining equalization channels remain in a high-resistance state to avoid control conflicts and magnetic coupling interference caused by multiple channels running in parallel.

6. The load balancing control method for a UAV battery BMS system according to claim 1, characterized in that, The optimal equalization current is calculated based on the equivalent resistance of the equalization circuit and the on-resistance of the switching device. Both the equivalent resistance of the equalization circuit and the on-resistance of the switching device are obtained through offline calibration, and the equalization current is limited to not exceed the maximum allowable equalization current of the battery cell.

7. The load balancing control method for a UAV battery BMS system according to claim 1, characterized in that, The equalization circuit adopts an integrated PCB layout, the magnetic components are less than 150 cubic millimeters in size, the switching devices are gallium nitride field-effect transistors, the on-resistance is less than 10 milliohms, and the overall equalization module consumes less than 5 milliwatts in standby mode.

8. The load balancing control method for a UAV battery BMS system according to claim 1, characterized in that, After each flight mission, the timestamp, duration, transferred charge, and SOC convergence accuracy of the equalization event are uploaded to the ground station to build a battery health status assessment database, supporting subsequent flight strategy optimization.

9. The load balancing control method for a UAV battery BMS system according to claim 8, characterized in that, The SOC convergence accuracy is defined as the standard deviation of the SOC of each cell within 60 seconds after the equalization is completed. The SOC convergence accuracy is required to be less than 0.8%. If the convergence accuracy exceeds the standard in three consecutive tasks, the BMS self-check process will be triggered and an anomaly will be reported.

Images