A method and system for balancing control of a drone battery pack

Abstract

This invention relates to the field of energy management technology for energy storage systems, and discloses a method and system for equalization control of UAV battery packs. The method includes: acquiring a voltage sequence and preprocessing it to obtain a smoothed voltage sequence, then sorting it by deviation level to obtain a priority processing unit set; calculating the equalization current based on the priority processing unit set and the smoothed voltage sequence to obtain an initial high-voltage current and an initial low-voltage current, and calculating the charging loss to obtain the total loss; optimizing the allocation based on the total loss to obtain an optimized high-voltage current and an optimized low-voltage current, and applying maximum current limiting to obtain a limited high-voltage current and a limited low-voltage current; predicting the range using a voltage range prediction model based on the limited high-voltage current and limited low-voltage current, and iteratively optimizing based on the predicted range value to obtain the final current pair; and generating current control commands based on the final current pair to control the battery pack. This method achieves voltage consistency during the charging process of the UAV battery pack.

Description

Technical Field

[0001] This invention relates to the field of energy management technology for energy storage systems, and in particular to a method and system for equalizing control of unmanned aerial vehicle (UAV) battery packs. Background Technology

[0002] Currently, the widespread application of drones in logistics, agriculture, and other fields relies on their ability to operate stably for extended periods. As the core power source, the voltage consistency of each individual cell in the battery pack directly impacts flight safety and mission efficiency. With increasing charge-discharge cycles, voltage differences gradually widen, leading to overcharging or over-discharging of some cells, decreased energy utilization, and even situations where a single battery pack's voltage drops too low, forcing the entire battery pack to prematurely stop supplying power and interrupting the mission. Therefore, a technology is needed that can identify and adjust the voltage differences of each cell in real time under dynamic operating conditions using industrial big data to ensure the overall stable and reliable performance of the battery pack.

[0003] In one existing technology, multiple individual battery cells are connected in series to form a battery pack. A constant current source charges the entire battery pack with a fixed current while monitoring the total voltage of the battery pack. When the total voltage reaches a preset threshold, the charging switches to constant voltage charging until the current drops to a set value. During this process, each series-connected cell receives the same current simultaneously, but the voltage of each individual cell is not independently regulated. Due to differences in capacity, internal resistance, etc., between individual cells, some cells will reach full charge voltage first, while other cells remain at a lower voltage. As the charging process continues, the cells that are already fully charged become overcharged. The existing technology, by using series constant current charging and only monitoring the total voltage for charging and switching, results in some cells being overcharged while others are not fully charged, further widening the voltage difference.

[0004] Therefore, existing technology cannot guarantee voltage consistency during the charging process of drone battery packs. Summary of the Invention

[0005] This invention provides a method and system for equalizing control of drone battery packs to ensure voltage consistency during the charging process of drone battery packs.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for equalizing control of a drone battery pack, comprising:

[0007] The voltage sequence of the battery cell is obtained and preprocessed to obtain a smooth voltage sequence;

[0008] Based on the smoothed voltage sequence, the deviation levels are sorted to obtain a priority processing unit set;

[0009] Based on the priority processing unit set and the smoothed voltage sequence, the equalization current of the range unit is calculated to obtain the preliminary high voltage current and the preliminary low voltage current.

[0010] The charging loss is calculated using the initial high voltage current and the initial low voltage current to obtain the total loss. The current distribution is optimized based on the total loss to obtain the optimized high voltage current and the optimized low voltage current. The optimized high voltage current and the optimized low voltage current are then subject to maximum current limiting to obtain the limited high voltage current and the limited low voltage current.

[0011] Based on the limited high-voltage current and the limited low-voltage current, the voltage range is predicted by a pre-built voltage range prediction model to obtain the predicted range value, and the limited high-voltage current and the limited low-voltage current are iteratively optimized based on the predicted range value to obtain the final current pair;

[0012] The final current is used to generate and execute current control commands.

[0013] Secondly, the present invention provides a drone battery pack equalization control system, comprising:

[0014] The data acquisition module is used to acquire the voltage sequence of the battery cells and perform preprocessing to obtain a smooth voltage sequence;

[0015] The deviation sorting module is used to sort the deviation levels according to the smoothed voltage sequence to obtain a priority processing unit set;

[0016] The current calculation module is used to perform equal current calculation of the range unit based on the priority processing unit set and the smoothed voltage sequence to obtain the preliminary high voltage current and the preliminary low voltage current.

[0017] The loss optimization module is used to calculate the charging loss using the initial high voltage current and the initial low voltage current to obtain the total loss, optimize the current distribution based on the total loss to obtain the optimized high voltage current and the optimized low voltage current, and limit the optimized high voltage current and the optimized low voltage current to obtain the limited high voltage current and the limited low voltage current.

[0018] The range optimization module is used to predict the voltage range based on the limited high voltage current and the limited low voltage current by using a pre-built voltage range prediction model to obtain the predicted range value, and to iteratively optimize the limited high voltage current and the limited low voltage current based on the predicted range value to obtain the final current pair.

[0019] The instruction output module is used to generate current control instructions to control the battery pack based on the final current pair.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] (1) This invention obtains a smooth voltage sequence by acquiring the voltage sequence of the battery cell and performing median filtering and moving average preprocessing. Then, it calculates the average voltage and absolute deviation sequence based on the smooth voltage sequence, extracts the cells that exceed the deviation threshold, and uses the bubble sort algorithm to obtain a set of priority processing cells. Since the set of priority processing cells focuses on the battery cell with the largest voltage deviation, the equalization resources are no longer evenly distributed to all cells, but the cells that need the most intervention are corrected first. This solves the problem that the existing series constant current charging treats all cells equally and the deviation cells cannot be adjusted in time, thus improving the pertinence and response speed of equalization control.

[0022] (2) This invention identifies the highest voltage unit and the lowest voltage unit in the priority processing unit set, calculates the initial high voltage current and the initial low voltage current, and then introduces the equivalent resistance value from the charging power supply to each unit to calculate the high voltage charging loss and the low voltage charging loss. With the minimum total loss as the objective function and the total equalization current value remaining unchanged as the constraint, the Lagrange multiplier method is used to optimize the current distribution, and the optimized current is subject to maximum current limit. Since the Lagrange multiplier method can analytically solve the current distribution scheme with minimum loss under the constraint of meeting the equalization requirements, the maximum current limit avoids exceeding the battery's safe charging capacity. This solves the problem of energy waste caused by not considering the path resistance difference in the equalization current distribution in the prior art and the overcharge risk caused by not considering the battery safety limit. It achieves the minimization of charging path loss under the premise of ensuring battery safety and improves the energy utilization efficiency of the equalization process.

[0023] (3) This invention obtains the range prediction value by inputting the limited high voltage current and limited low voltage current into a pre-constructed voltage range prediction model. When the range prediction value exceeds the balance tolerance threshold, the total equalization current value is iteratively reduced and the current pair is recalculated until the range prediction value meets the tolerance requirements. Then, the charging command is executed according to the final current pair and the equalized voltage range is collected. Successful samples and failure samples are stored in the successful sample set and the failure sample set, respectively. The positive error value is calculated based on the successful sample set and the negative error value is calculated based on the failure sample set. The total loss value is obtained by weighted summation and the weight parameters of the voltage range prediction model are updated by gradient descent. Since successful samples and failure samples are assigned different weights, the model pays more attention to the sample features that cause equalization failure during the training process. The iterative optimization mechanism ensures that the effect of the current scheme is evaluated by prediction and adjusted in advance before each charging. This solves the problem that the equalization control strategy in the prior art cannot adaptively optimize the model according to the actual effect and cannot predict the equalization result in advance. It realizes the continuous evolution of the voltage range prediction model and the dynamic optimization of the equalization current strategy, and ensures the stable and reliable voltage consistency during the charging process of the UAV battery pack in the long term. Attached Figure Description

[0024] Figure 1This is a schematic flowchart of the UAV battery pack equalization control method provided in the first embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of the structure of the drone battery pack equalization control system provided in the second embodiment of the present invention. Detailed Implementation

[0026] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Reference Figure 1 The first embodiment of the present invention provides a method for equalizing control of a drone battery pack, comprising the following steps:

[0028] S11, Obtain the voltage sequence of the battery cell and perform preprocessing to obtain a smooth voltage sequence;

[0029] S12, based on the smoothed voltage sequence, sort by deviation level to obtain a priority processing unit set;

[0030] S13, based on the priority processing unit set and the smoothed voltage sequence, perform equalization current calculation of the range unit to obtain the preliminary high voltage current and the preliminary low voltage current.

[0031] S14, calculate the charging loss using the initial high voltage current and the initial low voltage current to obtain the total loss, optimize the current distribution based on the total loss to obtain the optimized high voltage current and the optimized low voltage current, and apply maximum current limiting to the optimized high voltage current and the optimized low voltage current to obtain the limited high voltage current and the limited low voltage current.

[0032] S15, based on the limited high voltage current and the limited low voltage current, the voltage range is predicted by a pre-built voltage range prediction model to obtain the range prediction value, and the limited high voltage current and the limited low voltage current are iteratively optimized based on the range prediction value to obtain the final current pair;

[0033] S16, Based on the final current pair, generate a current control command to control the battery pack.

[0034] In step S11, the voltage sequence of the battery cell is obtained and preprocessed to obtain a smoothed voltage sequence, including:

[0035] Obtain the voltage sequence of the battery cells;

[0036] Based on the voltage sequence, noise is removed using a median filtering algorithm to obtain a denoised voltage sequence;

[0037] Based on the denoised voltage sequence, the data is smoothed using a moving average method to obtain a smoothed voltage sequence.

[0038] Specifically, the battery management system first collects the terminal voltage data of each battery cell in the drone's battery pack during the charging process in real time. These terminal voltage data are arranged in chronological order to form an original voltage sequence. The acquisition frequency of the original voltage sequence is consistent with the sampling period of the battery management system, which is preset to a fixed time interval, such as once every 0.1 seconds. Due to electromagnetic interference in the charging environment and measurement errors of the sensors themselves, the original voltage sequence contains high-frequency noise components. This noise can interfere with the accurate judgment of subsequent voltage deviations, so the original voltage sequence needs to be denoised. The denoising process uses a median filtering algorithm, setting a sliding window of odd length. The window length is preset to five consecutive sampling points based on the sampling frequency and noise characteristics. The window slides point by point on the original voltage sequence. For the five voltage values ​​within the window, they are arranged in ascending order of value, and the third value in the middle position after the arrangement is taken as the denoised output voltage value corresponding to the center position of the window. The sliding window starts from the beginning of the original voltage sequence and moves one sampling point to the right each time until the entire original voltage sequence is traversed, thus obtaining a denoised voltage sequence of the same length as the original voltage sequence.

[0039] Although the denoised voltage sequence obtained after median filtering eliminates isolated impulse noise, small fluctuations caused by random volatility still exist. These fluctuations can lead to unnecessary misjudgments when calculating the voltage deviation in subsequent steps. To further smooth the data, a moving average method is used to process the denoised voltage sequence. The specific implementation process of the moving average method is as follows: a fixed-length averaging window is set, the length of which is predetermined based on the voltage response time of the battery cell and the update cycle of the control system, consisting of three consecutive sampling points. This averaging window is slid point by point across the denoised voltage sequence. For the three denoised voltage values ​​within the window, their arithmetic mean is calculated, i.e., the sum of the three values ​​is divided by three. The result is used as the smoothed voltage value corresponding to the end position of the window. The sliding window starts from the beginning of the denoised voltage sequence and moves one sampling point to the right each time until it covers the entire denoised voltage sequence, ultimately generating a smoothed voltage sequence with the same length as the original voltage sequence. This smoothed voltage sequence is the output variable of step S11, used in subsequent steps to calculate the average voltage value and the absolute deviation sequence.

[0040] This step, through the cascaded preprocessing of median filtering and moving average, effectively suppresses high-frequency noise and random fluctuations in the original voltage sequence, enabling the smoothed voltage sequence to truly reflect the voltage change trend of each battery cell. This improves the accuracy of subsequent deviation level sorting and priority processing unit set identification, laying a data foundation for achieving precise battery pack equalization control.

[0041] In step S12, the deviation levels are sorted according to the smoothed voltage sequence to obtain a priority processing unit set, including:

[0042] The average voltage value of the battery cell is calculated based on the smoothed voltage sequence.

[0043] Calculate the absolute value of the difference between the smoothed voltage sequence and the average voltage value to obtain the absolute deviation sequence;

[0044] Extract the battery cells that exceed a preset deviation threshold from the absolute deviation sequence to obtain a set of cells to be processed;

[0045] The set of units to be processed is arranged in descending order of the absolute deviation sequence to obtain the set of units to be processed first.

[0046] Specifically, the average voltage of all battery cells is first calculated based on the smoothed voltage sequence. This is done by summing the smoothed voltage values ​​of all cells in the sequence, then dividing the sum by the total number of cells to obtain the average voltage. Next, the absolute deviation of each cell is calculated. The absolute deviation is the absolute value of the difference between the smoothed voltage and the average voltage for that cell; it is calculated by subtracting the average voltage from the smoothed voltage. If the difference is negative, the opposite value is taken. This yields an absolute deviation sequence that corresponds one-to-one with the smoothed voltage sequence. The absolute deviation sequence reflects the degree to which each cell's voltage deviates from the average level; a larger value indicates a more severe voltage deviation.

[0047] To identify battery cells requiring equalization, a pre-defined deviation threshold is set. This threshold is determined by collecting absolute deviation sequence data recorded multiple times during the drone's battery pack's historical charging process. All historical absolute deviation values ​​are arranged in ascending order, and the 85th percentile value in the sequence is used as the deviation threshold. If no historical absolute deviation sequence data is available, the 85th percentile of the absolute deviation sequence measured under standard charging conditions for this battery pack model is used as the initial deviation threshold. This percentile is obtained through factory testing of the same battery pack model. This percentile selection ensures that 15% of larger deviations are prioritized, while avoiding an excessively low threshold that would overload the system by processing too many cells simultaneously. Each absolute deviation value in the absolute deviation sequence is compared to this threshold, and battery cells with absolute deviations exceeding the threshold, along with their corresponding absolute deviation values, are extracted. These battery cells constitute the set of cells to be processed.

[0048] The battery cells in the set to be processed are arranged in descending order of their absolute deviations using a bubble sort algorithm. The specific process of bubble sort is as follows: The battery cells in the set to be processed are arranged into a list according to their absolute deviation values. Starting from the first cell in the list, the absolute deviation values ​​of adjacent cells are compared. If the absolute deviation value of the preceding cell is smaller than that of the following cell, the two cells are swapped. After one round of comparisons and swaps from beginning to end, the largest absolute deviation value is moved to the end of the list. This process is repeated, reducing the range of comparisons by one cell in each round, until there are no more positions to swap in the entire list. After sorting, the battery cells in the list are arranged in descending order of their absolute deviation values, forming a priority processing set.

[0049] This step uses deviation threshold screening and bubble sorting in descending order to ensure that the battery cells with the most serious voltage deviations are processed first. This allows resources to be concentrated and balanced to correct the cells with the largest deviations within a limited charging time, effectively preventing individual cells from deviating too much from the charging efficiency of the entire battery pack or causing safety risks.

[0050] In step S13, based on the priority processing unit set and the smoothed voltage sequence, the equalization current of the range unit is calculated to obtain the preliminary high-voltage current and the preliminary low-voltage current, including:

[0051] Extract the smoothed voltage sequence from the priority processing unit set, identify the highest voltage value and the highest voltage unit, the lowest voltage value and the lowest voltage unit, and subtract the lowest voltage value from the highest voltage value to obtain the voltage range;

[0052] Multiply the preset equalization current base value by the voltage range and then divide by the preset range reference value to obtain the total equalization current value.

[0053] The high voltage deviation is obtained by subtracting the preset target voltage value from the highest voltage value, and the low voltage deviation is obtained by subtracting the target voltage value from the lowest voltage value.

[0054] The absolute values ​​of the high-voltage deviation and the low-voltage deviation are added together to obtain the deviation sum. The total equalization current value is multiplied by the absolute value of the high-voltage deviation and then divided by the deviation sum to obtain the preliminary high-voltage current. The total equalization current value is multiplied by the absolute value of the low-voltage deviation and then divided by the deviation sum to obtain the preliminary low-voltage current.

[0055] Specifically, the priority processing unit set is already sorted according to voltage deviation from largest to smallest, with the first battery unit being the one with the most severe deviation. However, to calculate the equalization current, it is necessary to extract the voltage extreme value units from the priority processing unit set. The specific process is as follows: traverse the unit number of each battery unit in the priority processing unit set, and search for the corresponding smoothed voltage value in the smoothed voltage sequence based on the unit number. Among all the found smoothed voltage values, identify the largest value as the highest voltage value, and the battery unit corresponding to this value is the highest voltage unit; simultaneously, identify the smallest value as the lowest voltage value, and the battery unit corresponding to this value is the lowest voltage unit. Subtracting the lowest voltage value from the highest voltage value yields the voltage range, which reflects the severity of voltage inconsistency in the current battery pack.

[0056] The total balancing current is calculated based on the voltage range. A baseline balancing current value and a range reference value are pre-set. These parameters are determined as follows: the baseline balancing current value is 20% of the maximum allowable charging current of the drone battery pack under standard charging conditions. This value is obtained by statistically analyzing historical charging data where the voltage range is within a safe range, showing the average balancing current. The range reference value is 5% of the rated voltage calibrated at the factory for the battery pack. This value is determined by analyzing historical fault records showing a significant increase in the probability of overcharging or over-discharging events after the voltage range exceeds this value. If no historical charging data is available, the baseline balancing current value is 15% of the maximum allowable charging current of the battery pack, and the range reference value is 3% of the rated voltage. These two initial values ​​are calibrated through factory testing of the same battery pack model. The total balancing current value is calculated by multiplying the baseline balancing current value by the voltage range, and then dividing the product by the range reference value. The total balancing current value increases with the increase in voltage range; that is, the more severe the voltage inconsistency, the greater the total balancing current required.

[0057] To distribute the total equalization current value to the highest and lowest voltage cells, high-voltage deviation and low-voltage deviation need to be calculated. A target voltage value is preset, which is the rated voltage that each cell should reach when the battery pack is fully charged, provided by the battery manufacturer. Subtracting the highest voltage value from the target voltage value yields the high-voltage deviation, representing the amount of charging the highest voltage cell needs to reach the target voltage. Since the highest voltage value may already exceed the target voltage value, the high-voltage deviation is negative, indicating that discharge equalization is needed for that cell. Subtracting the lowest voltage value from the target voltage value yields the low-voltage deviation, representing the amount of charging the lowest voltage cell needs to reach the target voltage. The low-voltage deviation is positive, indicating that charging compensation is needed for that cell. The absolute values ​​of the high-voltage and low-voltage deviations are added together to obtain the deviation sum. The initial high-voltage current is calculated by multiplying the total equalization current value by the absolute value of the high-voltage deviation and then dividing by the deviation sum. The initial low-voltage current is calculated by multiplying the total equalization current value by the absolute value of the low-voltage deviation and then dividing by the deviation sum. If the sum of the absolute values ​​of the deviations is zero, that is, all unit voltages are equal to the target voltage, then the initial high voltage current and the initial low voltage current are directly set to zero, and no balancing is required. Through this method of allocating according to the deviation ratio, the unit with the larger deviation obtains a larger balancing current, so that the highest voltage unit and the lowest voltage unit move towards the target voltage value synchronously during the balancing process.

[0058] It should be noted that when the highest voltage exceeds the target voltage, the high-voltage deviation may be negative, indicating that the high-voltage unit needs to be discharged for equalization; the low-voltage deviation is positive, indicating that the low-voltage unit needs to be charged for equalization. After taking the absolute value, both are positive, and the calculated initial high-voltage current and initial low-voltage current are current amplitudes. In actual control, the initial high-voltage current is applied to the highest voltage unit in the discharge direction (negative value), and the initial low-voltage current is applied to the lowest voltage unit in the charging direction (positive value). By distributing the current according to the absolute value of the deviation, the unit that deviates more from the target receives a larger equalization current, thereby simultaneously narrowing the distance between the high and low voltage units and the target voltage, effectively reducing the voltage range difference.

[0059] The output variables of this step are the initial high-voltage current and the initial low-voltage current, which are used in subsequent steps to calculate charging losses and optimize current distribution. This solution process achieves a quantitative correlation between the balancing current and the degree of voltage deviation, ensuring that the balancing resources are accurately applied to the extreme value units, and providing a basic control quantity for reducing voltage range differences.

[0060] In step S14, the charging loss is calculated using the initial high-voltage current and the initial low-voltage current to obtain the total loss. Current allocation is optimized based on the total loss to obtain an optimized high-voltage current and an optimized low-voltage current. The optimized high-voltage current and the optimized low-voltage current are then subject to maximum current limiting to obtain a limited high-voltage current and a limited low-voltage current. This includes:

[0061] Obtain the equivalent resistance value from the charging power source to the highest voltage unit and the equivalent resistance value from the charging power source node to the lowest voltage unit to obtain the high voltage path resistance value and the low voltage path resistance value.

[0062] The square of the initial high-voltage current is multiplied by the high-voltage path resistance value to obtain the high-voltage charging loss, and the square of the initial low-voltage current is multiplied by the low-voltage path resistance value to obtain the low-voltage charging loss. The sum of the high-voltage charging loss and the low-voltage charging loss is calculated as the total loss.

[0063] Using the total loss as the objective function and the total equilibrium current value remaining unchanged as the constraint, the initial high voltage current and the initial low voltage current are optimized by using the Lagrange multiplier method to obtain the optimized high voltage current and the optimized low voltage current.

[0064] When the optimized high voltage current does not exceed the preset maximum allowable charging current of the battery, the optimized high voltage current is output as a limited high voltage current; when the optimized high voltage current exceeds the preset maximum allowable charging current of the battery, the maximum allowable charging current of the battery is output as a limited high voltage current.

[0065] When the optimized low-voltage current does not exceed the maximum allowable charging current of the battery, the optimized low-voltage current is output as a limited low-voltage current; when the optimized low-voltage current exceeds the maximum allowable charging current of the battery, the maximum allowable charging current of the battery is output as a limited low-voltage current.

[0066] Specifically, the high-voltage path resistance and low-voltage path resistance values ​​are read from the circuit parameter table stored within the battery management system. The high-voltage path resistance refers to the total equivalent resistance of the wires, connectors, and protection circuits from the charging power output node to the positive terminal of the highest voltage unit. The low-voltage path resistance refers to the total equivalent resistance from the charging power output node to the positive terminal of the lowest voltage unit. These two resistance values ​​are measured by a DC resistance tester after the battery pack is assembled and stored in the non-volatile memory of the battery management system, and are directly read before each charge. First, the charging loss corresponding to the initial high-voltage current is calculated: multiply the initial high-voltage current by itself to get the square of the initial high-voltage current, then multiply this square by the high-voltage path resistance value; the product is the high-voltage charging loss. The charging loss corresponding to the initial low-voltage current is calculated by multiplying the square of the initial low-voltage current by the low-voltage path resistance value; the product is the low-voltage charging loss. It should be noted that both high-voltage and low-voltage charging losses follow Joule's law, meaning that the power loss is proportional to the square of the current. The total loss is obtained by adding the high-voltage and low-voltage charging losses together.

[0067] With minimizing total loss as the optimization objective and maintaining the total equilibrium current value as the constraint, the Lagrange multiplier method is used to optimize the allocation of the initial high-voltage and low-voltage currents. The specific implementation process of the Lagrange multiplier method is as follows: First, a Lagrange function is constructed, which consists of the objective function plus the constraint multiplied by an unknown multiplier. The objective function is an expression for the total loss with respect to the high-voltage and low-voltage currents, and the constraint requires that the sum of the high-voltage and low-voltage currents equals the total equilibrium current value. Then, partial derivatives are calculated with respect to the high-voltage current, low-voltage current, and Lagrange multiplier, respectively. Setting these three partial derivatives equal to zero yields a system of three equations. Solve this system of equations: From the equation where the partial derivative of the high-voltage current is zero, solve for the relationship between the high-voltage current and the Lagrange multiplier; from the equation where the partial derivative of the low-voltage current is zero, solve for the relationship between the low-voltage current and the Lagrange multiplier. Substitute these two relationships into the constraint equations to obtain the values ​​of the Lagrange multipliers. Then substitute the Lagrange multipliers back into the relationships between the high-voltage and low-voltage currents to obtain the optimized high-voltage and low-voltage currents. This optimization process minimizes the total loss while maintaining a constant total equilibrium current, thereby improving the energy utilization efficiency during the equilibrium process.

[0068] After obtaining the optimized high-voltage current and optimized low-voltage current, both need to be limited to their maximum current to prevent exceeding the battery cell's safe charging capacity. The maximum allowable charging current of the battery is a preset threshold, determined by collecting temperature rise data and cycle life decay data of the battery cell at different charging rates during charge-discharge cycle tests. A safety boundary is set at a battery surface temperature not exceeding 45 degrees Celsius and cycle life decay not exceeding 20%. The maximum charging current value that meets this boundary condition is selected as the battery's maximum allowable charging current. This value is provided by the battery manufacturer and written into the battery management system. The optimized high-voltage current is compared with the battery's maximum allowable charging current. If the optimized high-voltage current is less than or equal to the battery's maximum allowable charging current, the optimized high-voltage current is directly used as the limited high-voltage current output; if the optimized high-voltage current is greater than the battery's maximum allowable charging current, the battery's maximum allowable charging current is used as the limited high-voltage current output. Similarly, the optimized low-voltage current is compared with the maximum allowable charging current of the battery. If the optimized low-voltage current is less than or equal to the maximum allowable charging current of the battery, the optimized low-voltage current is used as the limited low-voltage current output; if the optimized low-voltage current is greater than the maximum allowable charging current of the battery, the maximum allowable charging current of the battery is used as the limited low-voltage current output.

[0069] This step, through loss optimization and current limiting, ensures that the equalization process operates with minimal energy loss within the battery's safety boundaries, avoiding unnecessary heat generation and energy waste.

[0070] In step S15, based on the limited high-voltage current and the limited low-voltage current, a voltage range prediction value is obtained by using a pre-built voltage range prediction model. Then, based on the predicted range value, the limited high-voltage current and the limited low-voltage current are iteratively optimized to obtain the final current pair, including:

[0071] Based on the limited high-voltage current and the limited low-voltage current, the voltage range is predicted using a pre-built voltage range prediction model to obtain the predicted range value.

[0072] When the predicted range value does not exceed the preset balance tolerance threshold, the limited high voltage current and the limited low voltage current are output as the final current pair.

[0073] When the predicted range value exceeds the preset balance tolerance threshold, the total equalization current value is reduced by a preset ratio and the limited high voltage current and the limited low voltage current are recalculated until the predicted range value does not exceed the preset balance tolerance threshold. Then, the limited high voltage current and the limited low voltage current are output as the final current pair.

[0074] Specifically, the first step is to construct a voltage range prediction model. This model employs a three-layer neural network structure. The input layer contains two nodes, corresponding to limiting high-voltage current and limiting low-voltage current, respectively. The hidden layer contains five nodes, each using a linear rectified function as its activation function. This function outputs the input value when the input value is greater than zero and outputs zero when the input value is less than or equal to zero. The output layer contains one node, corresponding to the predicted range value. The output layer does not use an activation function to maintain a linear output.

[0075] The model's training data comes from samples collected during historical charging processes. Each sample contains a set of limiting high-voltage current values, limiting low-voltage current values, and the equalized voltage range measured after charging. The data collection process is as follows: in each previous charging equalization control, the limiting high-voltage current and limiting low-voltage current output in step S14 are recorded, and after charging, the equalized voltage range is calculated according to the method in step S16. These three values ​​are stored as a training sample in the historical database. When the number of samples in the historical database reaches more than 500, model training begins.

[0076] It should be noted that when there are no training samples in the historical database, a pre-trained default model is used as the voltage range prediction model. This default model is obtained by selecting the same model of UAV battery pack and conducting at least one hundred charge-discharge tests under standard charging conditions. The limiting high voltage current, limiting low voltage current, and equalized voltage range are collected during each charging process. The default model parameters are obtained by pre-training using the same neural network structure and training method. The default model is stored in the non-volatile memory of the battery management system. When the number of historical samples reaches five hundred, the model is continuously optimized using online collected actual operating samples.

[0077] The training process employs gradient descent, with the loss function defined as the square of the difference between the predicted range output by the model and the actual voltage range after equilibrium. First, all weight parameters are initialized to random numbers between -0.5 and +0.5. Then, 32 samples are randomly selected from the historical database each time as a training batch. For each sample, forward propagation is performed to calculate the predicted range, followed by the calculation of the loss value. Next, backpropagation is used to calculate the gradient of each weight parameter. Finally, the weight parameters are updated according to a preset learning rate (0.001). This training batch process is repeated for a preset number of 1000 iterations, with a preset loss threshold of 0.001. Training is stopped when the number of training iterations exceeds 1000 or the calculated loss value is less than 0.001. Simultaneously, training is also terminated early if the average loss value of 50 consecutive batches no longer decreases to prevent overfitting. The resulting model is the usable voltage range prediction model.

[0078] When using this model, the limited high-voltage current and limited low-voltage current are used as two values ​​in the input layer. They are then passed through the weighted summation of each node in the hidden layer and activated by the linear rectification function, and then through the weighted summation of the output layer to finally obtain a value as the range prediction value.

[0079] After obtaining the predicted voltage range value, it needs to be compared with the preset balance tolerance threshold. The balance tolerance threshold is determined by collecting the actual measured voltage range data of the drone battery pack after equalization during historical charging processes. All historical equalization voltage range values ​​are arranged in ascending order, and the value at the 90th percentile of the arrangement is taken as the balance tolerance threshold. This threshold represents the voltage consistency level that 90% of successful equalization cases can achieve. Exceeding this threshold indicates that the equalization effect does not meet the requirements. When the predicted voltage range value is less than or equal to the balance tolerance threshold, it is considered that the current limited high-voltage current and limited low-voltage current can enable the battery pack to achieve acceptable voltage consistency after charging. Therefore, the limited high-voltage current and limited low-voltage current are directly used as the final current pair output. If there is no historical equalization voltage range data, the balance tolerance threshold is taken as the maximum allowable voltage range value specified by the battery pack manufacturer at the time of manufacture.

[0080] When the predicted range value exceeds the balance tolerance threshold, it indicates that the voltage range is still too large after charging according to the current current scheme. The total equalization current needs to be reduced to slow the charging rate, thus allowing more time for voltage equalization. At this point, iterative optimization is performed: the total equalization current value is multiplied by a preset proportional coefficient, which is 0.9, meaning the total equalization current value is reduced by 10% each time. Specifically, the reduced total equalization current value replaces the original total equalization current value, keeping the high-voltage and low-voltage deviations unchanged. The initial current is recalculated according to the calculation method for the initial high-voltage and initial low-voltage currents in step S13. Then, loss optimization and limiting are performed using the Lagrange multiplier method in step S14, resulting in a new set of limited high-voltage and limited low-voltage currents. This new set of currents is input into the voltage range prediction model again to obtain a new predicted range value. The new predicted range value is compared with the balance tolerance threshold again. If it still exceeds the threshold, the current total equalization current value is multiplied by 0.9 again, and the above recalculation and prediction process is repeated. Each iteration reduces the total equalization current by 10%, thereby decreasing the charging current intensity and allowing the battery pack to charge under gentler conditions, which is beneficial for the convergence of voltage range. When the predicted range value obtained in a certain iteration is less than or equal to the balance tolerance threshold, the iteration stops, and the limited high-voltage current and limited low-voltage current obtained in that iteration are used as the final current pair output.

[0081] This step, through forward evaluation of the voltage range prediction model and iterative adjustment of the total equalization current value, ensures that the final charging scheme can control the voltage range within an acceptable range as expected, thus avoiding equalization failure due to excessive current.

[0082] In step S16, a current control command is generated based on the final current pair to control the battery pack.

[0083] Specifically, firstly, a current control command is generated based on the final current pair, which includes a final high-voltage current and a final low-voltage current. The final high-voltage current is used to control the equalization discharge branch connected to the highest voltage unit, and the final low-voltage current is used to control the equalization charging branch connected to the lowest voltage unit. The specific format of the current control command is a digital signal containing a target current value and an execution duration. This signal is sent to the bidirectional DC-DC converters in the charging network through the controller area network bus of the battery management system. Each bidirectional DC-DC converter adjusts the magnitude of its output current according to the received command, applying a discharge current to the highest voltage unit and a charging current to the lowest voltage unit, thereby dynamically adjusting the voltage of each battery unit and reducing the voltage difference. When the highest voltage unit reaches the set target voltage value and both the final high-voltage current and the final low-voltage current drop below 0.05 amperes, the charging process is considered complete, thus achieving precise physical control of the UAV battery pack charging process.

[0084] In one embodiment, after generating a current control command to control the battery pack based on the final current pair, the method further includes:

[0085] The current control command is sent to the charging network;

[0086] Obtain the equalized voltage sequence of the battery cells from the start of charging to the completion of charging, identify the highest and lowest voltage values, and subtract the lowest voltage value from the highest voltage value to obtain the equalized voltage range;

[0087] When the equalized voltage range does not exceed the balance tolerance threshold, the final current pair, the predicted range value, and the equalized voltage range are stored in the successful sample set.

[0088] When the equalized voltage range exceeds the balance tolerance threshold, the final current pair, the predicted range value, and the equalized voltage range are stored in the failure sample set.

[0089] The voltage range prediction model is inversely optimized based on the successful sample set and the failed sample set.

[0090] During charging, the battery management system continuously collects the terminal voltage data of each battery cell at a fixed sampling period, consistent with the sampling period in step S11. From the start of charging until the end of charging, all voltage data collected during this period are stored in chronological order as an equalization voltage sequence. The criterion for determining when charging is complete is that the highest voltage cell reaches the target voltage value, and simultaneously, both the final high-voltage current and the final low-voltage current drop to a preset low-current threshold close to zero. This preset low-current threshold is 0.05A, obtained by statistically analyzing the average value of the current naturally dropping to near zero at the end of charging for this type of battery pack, and stored in the battery management system.

[0091] After charging is complete, the highest and lowest voltage values ​​occurring throughout the entire charging process are extracted from the equalization voltage sequence. Specifically, the voltage values ​​of all battery cells at all sampling times in the equalization voltage sequence are iterated through. The maximum value in the entire sequence is identified as the global highest voltage value, and the minimum value is identified as the global lowest voltage value. The difference between the global highest and global lowest voltage values ​​is the equalization voltage range. This equalization voltage range is a quantitative indicator for evaluating the actual effect of the equalization control, reflecting the final improvement in voltage inconsistency throughout the entire charging process. The equalization voltage range is compared with the balance tolerance threshold. The method for determining the balance tolerance threshold is given in step S15, which involves collecting historical equalization voltage range data and taking the 90th percentile value. If the equalization voltage range is less than or equal to the balance tolerance threshold, it indicates that the equalization control has successfully maintained the battery pack voltage consistency within an acceptable range. At this point, the final high-voltage current and final low-voltage current in the final current pair, the predicted range value obtained in step S15, and the equalization voltage range calculated in this step are considered as a successful sample and stored in the successful sample set. The successful sample set is stored in the non-volatile memory of the battery management system for subsequent forward optimization of the voltage range prediction model. If the voltage range after equalization is greater than the equalization tolerance threshold, it indicates that the equalization control has failed to meet the expected voltage consistency requirements. In this case, the final current pair, the predicted range value, and the equalized voltage range are treated as a single failure sample and stored in the failure sample set. The failure sample set is also stored in non-volatile memory for subsequent backward optimization of the voltage range prediction model.

[0092] This step distinguishes between successful and failed samples and stores them separately. Subsequent model optimization can then specifically enhance the learning of the characteristics of failed samples, thereby gradually improving the predictive model's ability to identify adverse operating conditions and enabling the equilibrium control system to continuously improve itself in actual use.

[0093] In one embodiment, the step of performing reverse optimization of the voltage range prediction model based on the successful sample set and the failed sample set includes:

[0094] Based on the successful sample set, the square of the difference between the predicted range value and the equalized voltage range is calculated, and then the average value is obtained to get the positive error value.

[0095] Based on the set of failed samples, the square of the difference between the predicted range value and the equalized voltage range is calculated, and then the average value is obtained to get the reverse error value.

[0096] Based on the positive error value and the negative error value, the positive error value is multiplied by a preset positive weight to obtain a weighted positive error, and the negative error value is multiplied by a preset negative weight to obtain a weighted negative error. The weighted positive error and the weighted negative error are then added together to obtain the total loss value.

[0097] Based on the total loss value, the weight parameters of the voltage range prediction model are updated using the gradient descent method.

[0098] Specifically, each sample in the successful sample set contains three data points: the final high-voltage current and the final low-voltage current in the final current pair, the predicted range value obtained in step S15, and the equalized voltage range obtained from the actual measurement after charging. Each sample in the failed sample set contains the same type of data. To optimize the voltage range prediction model, the error values ​​of the successful and failed sample sets need to be calculated separately, and the two are weighted and combined into a total loss value. Then, the weight parameters of the model are updated using the gradient descent method.

[0099] First, calculate the positive error value, which reflects the accuracy of the voltage range prediction model on successful cases. Take the predicted range value and the equalized voltage range for each sample in the successful sample set, calculate the difference between them, and then multiply this difference by itself to get the square of the difference. Sum the squared differences for all samples in the successful sample set, and then divide the sum by the total number of samples in the successful sample set. The average value obtained is the positive error value. The smaller the positive error value, the more accurate the model's prediction is under successful equalization conditions.

[0100] Next, the back-inference error value is calculated. This error value reflects the degree of prediction deviation of the voltage range prediction model in failed cases. For each sample in the failed sample set, the predicted range and the equalized voltage range are taken. The difference between the two is calculated, and then multiplied by itself to obtain the square of the difference. The squares of the differences for all samples in the failed sample set are summed, and then the sum is divided by the total number of samples in the failed sample set. The average value obtained is the back-inference error value. The smaller the back-inference error value, the smaller the prediction deviation of the model under the operating conditions that lead to equalization failure. Since failed samples correspond to situations where the equalized voltage range exceeds the balance tolerance threshold, these samples have a higher warning significance for model optimization and therefore need to be assigned greater weight.

[0101] Next, the total loss value is calculated. Positive and negative weights are pre-defined. These weights are determined by statistically analyzing the number of successful and failed samples in the past 100 charging events. The success rate is set as the number of successful samples divided by 100, and the failure rate as the number of failed samples divided by 100. The reciprocal of the success rate is taken as the positive baseline value, and the reciprocal of the failure rate is taken as the negative baseline value. The sum of the positive and negative baseline values ​​is calculated. The positive baseline value is divided by this sum and multiplied by 2 to obtain the positive weight. The negative baseline value is divided by this sum and multiplied by 2 to obtain the negative weight. For example, if there were 90 successful charging events and 10 failed charging events historically, the success rate is 0.9 and the failure rate is 0.1. The positive baseline value is 1.11, and the negative baseline value is 10, totaling 11.11. The positive weight is approximately 1.11 divided by 11.11 multiplied by 2, which is approximately 0.2. The negative weight is approximately 10 divided by 11.11 multiplied by 2, which is approximately 1.8. The weight values ​​are rounded to one decimal place. This weighting method, based on historical statistics, ensures that failed samples contribute significantly more to the total loss than successful samples. The positive error is multiplied by the positive weight to obtain the weighted positive error; the negative error is multiplied by the negative weight to obtain the weighted negative error. The total loss is obtained by adding the weighted positive error and the weighted negative error together.

[0102] After obtaining the total loss value, the weight parameters of the voltage range prediction model are updated using gradient descent. The voltage range prediction model is a three-layer neural network structure, with hidden layers and output layers containing multiple weight parameters. The specific implementation process of gradient descent is as follows: First, the partial derivative of the total loss value with respect to each weight parameter in the output layer is calculated, i.e., the rate of change of the total loss value relative to that weight. This can be calculated layer by layer using the chain rule in the backpropagation algorithm. Starting from the output layer, the gradient of the output layer activation value with respect to the total loss value is calculated. Then, based on the relationship between the output layer weights and activation values, the gradient value of each output layer weight is obtained. The gradient is then passed to the hidden layer, the error term of each node in the hidden layer is calculated, and the gradient value of each hidden layer weight is obtained based on the relationship between the hidden layer weights and node outputs. After obtaining the gradient data of all weights, each weight parameter is updated according to the preset learning rate. The update method is to subtract the learning rate multiplied by the gradient value corresponding to the weight from the current weight value. The learning rate is kept consistent with the learning rate used during the initial training of the model in step S15, and is set to 0.001. After completing one weight update, this optimization process ends.

[0103] This step introduces successful and failed samples into the loss function with different weights. In subsequent predictions, the model will pay more attention to the current combination characteristics that lead to equilibration failure, thereby gradually improving the ability to identify and avoid failure conditions and enabling the equilibration control strategy to continuously evolve in actual operation.

[0104] Reference Figure 2 The second embodiment of the present invention provides a drone battery pack equalization control system, comprising:

[0105] The data acquisition module is used to acquire the voltage sequence of the battery cells and perform preprocessing to obtain a smooth voltage sequence;

[0106] The deviation sorting module is used to sort the deviation levels according to the smoothed voltage sequence to obtain a priority processing unit set;

[0107] The current calculation module is used to perform equal current calculation of the range unit based on the priority processing unit set and the smoothed voltage sequence to obtain the preliminary high voltage current and the preliminary low voltage current.

[0108] The loss optimization module is used to calculate the charging loss using the initial high voltage current and the initial low voltage current to obtain the total loss, optimize the current distribution based on the total loss to obtain the optimized high voltage current and the optimized low voltage current, and limit the optimized high voltage current and the optimized low voltage current to obtain the limited high voltage current and the limited low voltage current.

[0109] The range optimization module is used to predict the voltage range based on the limited high voltage current and the limited low voltage current by using a pre-built voltage range prediction model to obtain the predicted range value, and to iteratively optimize the limited high voltage current and the limited low voltage current based on the predicted range value to obtain the final current pair.

[0110] The instruction output module is used to generate current control instructions to control the battery pack based on the final current pair.

[0111] It should be noted that the UAV battery pack equalization control system provided in this embodiment of the invention is used to execute all the process steps of the UAV battery pack equalization control method in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0112] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0113] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for balancing and controlling a UAV battery pack, characterized in that, include: The voltage sequence of the battery cell is obtained and preprocessed to obtain a smooth voltage sequence; Based on the smoothed voltage sequence, the deviation levels are sorted to obtain a priority processing unit set; Based on the priority processing unit set and the smoothed voltage sequence, the equalization current of the range unit is calculated to obtain the preliminary high voltage current and the preliminary low voltage current. The charging loss is calculated using the initial high voltage current and the initial low voltage current to obtain the total loss. The current distribution is optimized based on the total loss to obtain the optimized high voltage current and the optimized low voltage current. The optimized high voltage current and the optimized low voltage current are then subject to maximum current limiting to obtain the limited high voltage current and the limited low voltage current. Based on the limited high-voltage current and the limited low-voltage current, the voltage range is predicted by a pre-built voltage range prediction model to obtain the predicted range value, and the limited high-voltage current and the limited low-voltage current are iteratively optimized based on the predicted range value to obtain the final current pair; Based on the final current pair, a current control command is generated to control the battery pack.

2. The UAV battery pack equalization control method according to claim 1, characterized in that, The preprocessing to obtain a smooth voltage sequence includes: Based on the voltage sequence, noise is removed using a median filtering algorithm to obtain a denoised voltage sequence; Based on the denoised voltage sequence, the data is smoothed using a moving average method to obtain a smoothed voltage sequence.

3. The UAV battery pack equalization control method according to claim 1, characterized in that, The step of sorting the deviation levels according to the smoothed voltage sequence to obtain a priority processing unit set includes: The average voltage value of the battery cell is calculated based on the smoothed voltage sequence. Calculate the absolute value of the difference between the smoothed voltage sequence and the average voltage value to obtain the absolute deviation sequence; Extract the battery cells that exceed a preset deviation threshold from the absolute deviation sequence to obtain a set of cells to be processed; The set of units to be processed is arranged in descending order of the absolute deviation sequence to obtain the set of units to be processed first.

4. The UAV battery pack equalization control method according to claim 1, characterized in that, The step of performing equalization current calculation of the range unit based on the priority processing unit set and the smoothed voltage sequence to obtain the preliminary high-voltage current and preliminary low-voltage current includes: Extract the smoothed voltage sequence from the priority processing unit set, identify the highest voltage value and the highest voltage unit, the lowest voltage value and the lowest voltage unit, and subtract the lowest voltage value from the highest voltage value to obtain the voltage range; Multiply the preset equalization current base value by the voltage range and then divide by the preset range reference value to obtain the total equalization current value. The high voltage deviation is obtained by subtracting the preset target voltage value from the highest voltage value, and the low voltage deviation is obtained by subtracting the target voltage value from the lowest voltage value. The absolute values ​​of the high-voltage deviation and the low-voltage deviation are added together to obtain the deviation sum. The total equalization current value is multiplied by the absolute value of the high-voltage deviation and then divided by the deviation sum to obtain the preliminary high-voltage current. The total equalization current value is multiplied by the absolute value of the low-voltage deviation and then divided by the deviation sum to obtain the preliminary low-voltage current.

5. The UAV battery pack equalization control method according to claim 4, characterized in that, The process of calculating charging losses using the initial high-voltage current and the initial low-voltage current to obtain total losses, optimizing current allocation based on the total losses to obtain optimized high-voltage current and optimized low-voltage current, and applying maximum current limiting to the optimized high-voltage current and optimized low-voltage current to obtain limited high-voltage current and limited low-voltage current includes: Obtain the equivalent resistance value from the charging power source to the highest voltage unit and the equivalent resistance value from the charging power source node to the lowest voltage unit to obtain the high voltage path resistance value and the low voltage path resistance value. The square of the initial high-voltage current is multiplied by the high-voltage path resistance value to obtain the high-voltage charging loss, and the square of the initial low-voltage current is multiplied by the low-voltage path resistance value to obtain the low-voltage charging loss. The sum of the high-voltage charging loss and the low-voltage charging loss is calculated as the total loss. Using the total loss as the objective function and the total equilibrium current value remaining unchanged as the constraint, the initial high voltage current and the initial low voltage current are optimized by using the Lagrange multiplier method to obtain the optimized high voltage current and the optimized low voltage current. When the optimized high voltage current does not exceed the preset maximum allowable charging current of the battery, the optimized high voltage current is output as a limited high voltage current; when the optimized high voltage current exceeds the preset maximum allowable charging current of the battery, the maximum allowable charging current of the battery is output as a limited high voltage current. When the optimized low-voltage current does not exceed the maximum allowable charging current of the battery, the optimized low-voltage current is output as a limited low-voltage current; when the optimized low-voltage current exceeds the maximum allowable charging current of the battery, the maximum allowable charging current of the battery is output as a limited low-voltage current.

6. The UAV battery pack equalization control method according to claim 4, characterized in that, The step of predicting the voltage range using a pre-built voltage range prediction model based on the limited high-voltage current and the limited low-voltage current to obtain a predicted range value, and then iteratively optimizing the limited high-voltage current and the limited low-voltage current based on the predicted range value to obtain the final current pair, includes: Based on the limited high-voltage current and the limited low-voltage current, the voltage range is predicted using a pre-built voltage range prediction model to obtain the predicted range value. When the predicted range value does not exceed the preset balance tolerance threshold, the limited high voltage current and the limited low voltage current are output as the final current pair. When the predicted range value exceeds the preset balance tolerance threshold, the total equalization current value is reduced by a preset ratio and the limited high voltage current and the limited low voltage current are recalculated until the predicted range value does not exceed the preset balance tolerance threshold. Then, the limited high voltage current and the limited low voltage current are output as the final current pair.

7. The UAV battery pack equalization control method according to claim 6, characterized in that, The construction process of the voltage range prediction model includes: Obtain historical limiting high-voltage current, historical limiting low-voltage current, and historical range values; The historical limited high-voltage current and the historical limited low-voltage current are input into the input layer of the initially constructed neural network model for training, and the range prediction value output by the output layer of the neural network model is obtained. Substitute the predicted range value and the historical range value into the loss function to calculate the loss value; The gradient of the output layer of the neural network model is calculated based on the loss value, and the gradient is passed forward layer by layer through the chain rule to calculate the gradient of the parameters of each layer and obtain the gradient data. Based on the gradient data and the preset learning rate, the parameters of each layer of the neural network model are updated using the gradient descent method. The parameters of each layer are iteratively updated until the number of training iterations of the neural network model is greater than a preset number of iterations, or the loss value of the neural network model is less than a preset loss threshold. At this point, the training is considered complete, and a trained voltage range prediction model is obtained.

8. The UAV battery pack equalization control method according to claim 6, characterized in that, After generating a current control command to control the battery pack based on the final current pair, the method further includes: The current control command is sent to the charging network; Obtain the equalized voltage sequence of the battery cells from the start of charging to the completion of charging, identify the highest and lowest voltage values, and subtract the lowest voltage value from the highest voltage value to obtain the equalized voltage range; When the equalized voltage range does not exceed the balance tolerance threshold, the final current pair, the predicted range value, and the equalized voltage range are stored in the successful sample set. When the equalized voltage range exceeds the balance tolerance threshold, the final current pair, the predicted range value, and the equalized voltage range are stored in the failure sample set. The voltage range prediction model is inversely optimized based on the successful sample set and the failed sample set.

9. The UAV battery pack equalization control method according to claim 8, characterized in that, The step of performing reverse optimization of the voltage range prediction model based on the successful sample set and the failed sample set includes: Based on the successful sample set, the square of the difference between the predicted range value and the equalized voltage range is calculated, and then the average value is obtained to get the positive error value. Based on the set of failed samples, the square of the difference between the predicted range value and the equalized voltage range is calculated, and then the average value is obtained to get the reverse error value. Based on the positive error value and the negative error value, the positive error value is multiplied by a preset positive weight to obtain a weighted positive error, and the negative error value is multiplied by a preset negative weight to obtain a weighted negative error. The weighted positive error and the weighted negative error are then added together to obtain the total loss value. Based on the total loss value, the weight parameters of the voltage range prediction model are updated using the gradient descent method.

10. A battery pack balancing control system for unmanned aerial vehicles (UAVs), characterized in that, include: The data acquisition module is used to acquire the voltage sequence of the battery cells and perform preprocessing to obtain a smooth voltage sequence; The deviation sorting module is used to sort the deviation levels according to the smoothed voltage sequence to obtain a priority processing unit set; The current calculation module is used to perform equal current calculation of the range unit based on the priority processing unit set and the smoothed voltage sequence to obtain the preliminary high voltage current and the preliminary low voltage current. The loss optimization module is used to calculate the charging loss using the initial high voltage current and the initial low voltage current to obtain the total loss, optimize the current distribution based on the total loss to obtain the optimized high voltage current and the optimized low voltage current, and limit the optimized high voltage current and the optimized low voltage current to obtain the limited high voltage current and the limited low voltage current. The range optimization module is used to predict the voltage range based on the limited high voltage current and the limited low voltage current by using a pre-built voltage range prediction model to obtain the predicted range value, and to iteratively optimize the limited high voltage current and the limited low voltage current based on the predicted range value to obtain the final current pair. The instruction output module is used to generate current control instructions to control the battery pack based on the final current pair.

Images