Unmanned aerial vehicle battery installation and management method and system of intelligent charging cabinet

Abstract

The invention provides an unmanned aerial vehicle battery installation and management method and system for an intelligent charging cabinet, and relates to the technical field of unmanned aerial vehicles, comprising the steps that the intelligent charging cabinet selects a matched standby battery according to battery specification information, and battery replacement is completed through a visual positioning system and a mechanical arm; collecting performance parameters of the standby battery, generating a charging strategy and performing real-time adjustment; evaluating the performance of the standby battery based on a deep learning algorithm to generate a health evaluation report; and when the evaluation result is lower than a safety threshold, triggering recovery processing, and uploading the historical performance parameters to the cloud management platform to update the battery state evaluation model.

Description

Technical Field

[0001] The present invention relates to drone technology, and in particular to a method and system for installing and managing drone batteries in an intelligent charging cabinet. Background Art

[0002] The traditional battery replacement process usually requires manual operation, which is not only inefficient but also prone to safety hazards due to human error. The lack of an intelligent battery replacement system makes it impossible to achieve fast, accurate and safe battery replacement operations, affecting the continuous operation capability of drones.

[0003] Existing battery charging management methods often use fixed charging strategies and cannot be dynamically adjusted according to the real-time status of the battery. This simple charging method not only fails to fully utilize the battery performance, but may also cause the battery to be overcharged or undercharged, shorten the battery life, and even cause safety accidents.

[0004] Current battery health management systems generally lack support for deep learning algorithms and are unable to accurately evaluate and predict battery performance. This makes it difficult to detect potential battery failures in a timely manner and cannot provide a reliable basis for battery maintenance and replacement decisions. At the same time, the lack of cloud-based data analysis and model optimization mechanisms makes it difficult for battery management systems to continuously improve their performance and accuracy. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for installing and managing drone batteries in a smart charging cabinet, which can solve the problems in the prior art.

[0006] In a first aspect of an embodiment of the present invention, a method for installing and managing drone batteries in a smart charging cabinet is provided, including: the smart charging cabinet includes a battery storage area and a battery replacement area, the battery replacement area is provided with a robotic arm and a visual positioning system; the smart charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request, the visual positioning system in the battery replacement area collects the three-dimensional position information of the drone battery compartment, and controls the robotic arm to complete the removal of the battery to be replaced and the installation of the spare battery based on the three-dimensional position information; the smart charging cabinet collects the performance parameters of the spare battery in the battery storage area, the performance parameters include temperature information, power information, cycle number information and internal resistance information; the performance parameters are input into a preset The battery status assessment model generates a charging strategy, and the charging strategy includes a charging current curve, a charging time, and a temperature control curve; in the process of executing the charging strategy, the charging current curve, the charging time, and the temperature control curve are adjusted in real time based on the performance parameters collected in real time; the smart charging cabinet evaluates the performance of the backup battery in the battery storage area based on the deep learning algorithm, and generates a backup battery health assessment report including the remaining cycle life prediction, the degree of performance degradation, and the charging and discharging efficiency; when the backup battery health assessment result is lower than the safety threshold, the battery recycling process is triggered, and the historical performance parameters of the battery that triggered the recycling process are uploaded to the cloud management platform, and the cloud management platform updates the battery status assessment model by analyzing the historical performance parameters.

[0007] In an optional embodiment, the smart charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request, the visual positioning system of the battery replacement area collects the three-dimensional position information of the drone battery compartment, and the steps of controlling the robotic arm to complete the removal of the battery to be replaced and the installation of the spare battery based on the three-dimensional position information include: the smart charging cabinet receives a battery replacement request message sent by the drone, the battery replacement request message includes specification information of the battery model and capacity; queries the battery storage area database according to the specification information to obtain the storage location coordinates of the target spare battery, generates a material retrieval path planning scheme for the robotic arm based on the storage location coordinates, and the robotic arm removes the target spare battery from the battery compartment according to the material retrieval path planning scheme The battery storage area is transferred to the waiting position of the battery replacement area; a binocular vision system is used to collect the image of the UAV battery compartment, the feature points of the UAV battery compartment are identified through the SURF feature extraction algorithm, the parallax information of the feature points of the UAV battery compartment is calculated using the stereo matching algorithm, and the spatial position coordinates and attitude angles of the UAV battery compartment are calculated through triangulation based on the parallax information and camera calibration parameters; the motion trajectory of the robotic arm is planned according to the spatial position coordinates and attitude angles, and the robotic arm is controlled to move to the position of the UAV battery compartment; a strong tactile sensor is installed at the end of the robotic arm, and a spiral search method is used to approach the UAV battery compartment, the clamping force of the robotic arm is controlled to remove the battery to be replaced, and the spare battery placed in the waiting position is installed into the UAV battery compartment.

[0008] In an optional embodiment, the intelligent charging cabinet collects performance parameters of the backup battery in the battery storage area, the performance parameters including temperature information, power information, cycle number information and internal resistance information; the performance parameters are input into a preset battery status evaluation model to generate a charging strategy, the charging strategy including a charging current curve, a charging time and a temperature control curve, the steps comprising: the temperature information is collected by a platinum resistance sensor, the power information is collected by a coulomb counting method, the cycle number information is collected by detecting the number of conversions between the charging cut-off voltage and the discharging cut-off voltage, and the internal resistance information is collected by an AC impedance method; data preprocessing is performed on the performance parameters, including performing a moving average filter on the temperature information to remove mutation points, and performing a moving average filter on the power information to remove mutation points. The Kalman filtering algorithm is used to correct the accumulated error of the internal resistance information, and the high-frequency noise is eliminated by wavelet transform on the internal resistance information to obtain the pre-processed performance parameters; the pre-processed performance parameters are input into a preset battery state evaluation model, and the battery state evaluation model uses a support vector regression model, and a radial basis kernel function is used to establish a nonlinear mapping relationship, and an adaptive weight coefficient is set according to the contribution of different performance parameters to the battery life prediction, and the battery health state index, the remaining cycle life prediction value and the capacity attenuation trend are output; based on the battery health state index, the remaining cycle life prediction value and the capacity attenuation trend, a particle swarm optimization algorithm is used to generate a charging strategy, and the charging strategy includes a segmented constant current-constant voltage charging current curve, a charging time and a temperature control curve.

[0009] In an optional embodiment, the battery state assessment model uses a support vector regression model, adopts a radial basis kernel function to establish a nonlinear mapping relationship, and sets an adaptive weight coefficient according to the contribution of different performance parameters to the battery life prediction, and the steps of outputting a battery health state index, a remaining cycle life prediction value, and a capacity attenuation trend include: standardizing the performance parameters of the battery and converting the performance parameters into a standardized feature vector; constructing a support vector regression model and mapping the standardized feature vector into an output value, wherein the mapping process includes a weight vector to be optimized; constructing a loss function of the support vector regression model, wherein the loss function of the support vector regression model includes a second norm term and an adaptive adjustment term of the weight vector to be optimized; converting the gradient value of the weight vector to be optimized by the loss function of the support vector regression model into an adaptive weight coefficient through an exponential function, wherein the adaptive weight coefficient is used to characterize the effect of different performance parameters on the battery The contribution of the life prediction is that the adaptive adjustment item is composed of the product of the adaptive weight coefficient and the two norms of the weight vector to be optimized; the standardized feature vector is mapped to the high-dimensional feature space through the radial basis kernel function, and the nonlinear relationship between the standardized feature vectors is established. The penalty factor, regularization coefficient, learning rate and kernel function parameters of the support vector regression model are determined by cross-validation; the support vector regression model is trained based on the standardized feature vector to obtain the optimal weight vector, and the battery health status index is generated according to the weighted combination of the standardized feature vector mapped by the radial basis kernel function and the optimal weight vector; the remaining cycle life prediction value is calculated according to the battery health status index, and the remaining cycle life prediction value is determined by the value of the battery health status index and its change rate; based on the remaining cycle life prediction value, a capacity decay trend is generated, and the capacity decay trend uses an exponential function to describe the change law of the battery capacity with the remaining cycle life.

[0010] In an optional embodiment, during the execution of the charging strategy, the step of adjusting the charging current curve, the charging time and the temperature control curve in real time based on the real-time collected performance parameters includes: establishing an adjustment mechanism for the charging current curve based on the value of the battery health state index, the adjustment mechanism for the charging current curve including: multiplying the value of the battery health state index by the rated maximum charging current to obtain a first charging current value, comparing the first charging current value with the temperature-related current limit, and selecting the smaller value between the first charging current value and the current limit as the real-time charging current; establishing an adjustment mechanism for the charging time based on the remaining cycle life prediction value, the adjustment mechanism for the charging time converts the remaining cycle life prediction value into a charging time increment through an exponential function, and adding the charging time increment to the current moment to obtain the charging termination time; establishing an adjustment mechanism for the temperature control curve based on the capacity decay trend, the adjustment mechanism for the temperature control curve adds the reference temperature to the temperature adjustment amount related to the capacity decay to obtain the target temperature; normalizing the value of the battery health state index, the remaining cycle life prediction value and the capacity decay trend to construct a fuzzy set, performing reasoning operations based on the fuzzy rule base, and using the center of gravity method to convert the fuzzy output into the control amount of the charging current curve, the charging time and the temperature control curve.

[0011] In an optional embodiment, the smart charging cabinet evaluates the performance of the backup batteries in the battery storage area based on a deep learning algorithm, and generates a backup battery health assessment report including a remaining cycle life prediction, a degree of performance degradation, and a charge and discharge efficiency; when the backup battery health assessment result is lower than a safety threshold, a battery recycling process is triggered, and the historical performance parameters of the battery that triggered the recycling process are uploaded to a cloud management platform, and the cloud management platform updates the battery status assessment model by analyzing the historical performance parameters, the steps comprising: collecting battery performance data including voltage, current, temperature, state of charge, and timestamp, and constructing the battery performance data into a feature matrix; inputting the feature matrix into a multi-scale convolutional neural network, the multi-scale convolutional neural network comprising three parallel one-dimensional convolution branches, each one-dimensional convolution branch being provided with a convolution kernel of a different size, and extracting performance characteristics of different time scales through a convolution operation; splicing the performance characteristics output by the one-dimensional convolution branch, and fusing the performance characteristics through a fully connected layer, and the outputs of the fully connected layer are respectively connected to the remaining A remaining cycle life prediction branch, a performance degradation degree prediction branch and a charge and discharge efficiency prediction branch; a multi-task hybrid loss function is constructed based on the attention mechanism, the multi-task hybrid loss function includes a remaining cycle life loss term, a performance degradation degree loss term and a charge and discharge efficiency loss term, and the weight coefficient of each loss term is automatically adjusted through back propagation; the remaining cycle life prediction value, the performance degradation degree prediction value and the charge and discharge efficiency prediction value are compared with the preset safety threshold, and a backup battery health assessment report is generated, and when any prediction value exceeds the corresponding safety threshold, the battery recycling process is triggered; the historical performance data of the battery that triggers the recycling process is uploaded to the cloud management platform, the cloud management platform performs performance degradation feature analysis on the historical performance data, updates the weight coefficient of the attention mechanism based on the performance degradation feature analysis results, and adjusts the network structure parameters of the multi-scale convolutional neural network; the cloud management platform selects samples with a similarity greater than a preset similarity threshold from the historical performance data based on the maximum mean difference criterion to update the battery status assessment model.

[0012] In an optional embodiment, a multi-task hybrid loss function is constructed based on the attention mechanism, and the multi-task hybrid loss function includes a remaining cycle life loss term, a performance degradation degree loss term, and a charge and discharge efficiency loss term. The step of automatically adjusting the weight coefficient of each loss term through back propagation includes: inputting the hidden layer feature vector of the battery performance data into the temporal attention layer, calculating the attention score through the attention weight matrix and the bias term, and normalizing the attention score through the softmax function to obtain the temporal attention weight; constructing a weighted mean square error loss function based on the temporal attention weight to calculate the remaining cycle life loss term, and the weighted mean square error loss function weights the error term between the actual remaining cycle life value and the predicted value; A weighted cross entropy loss function is used to calculate the performance degradation loss term, and the weighted cross entropy loss function introduces a balance factor calculated based on the inverse of the number of samples to balance the contribution of samples with different degradation degrees; the deviation between the true value and the predicted value of the battery charge and discharge efficiency is substituted into the Gaussian kernel function for weight mapping to generate a charge and discharge efficiency loss term, and the parameters of the weight mapping are determined by the battery state of charge value; the remaining cycle life loss term, the performance degradation loss term and the charge and discharge efficiency loss term are combined into a multi-task hybrid loss function, and the weight coefficients of the remaining cycle life loss term, the performance degradation loss term and the charge and discharge efficiency loss term are automatically adjusted through back propagation, and the weight coefficients are exponentially and normalized to ensure that they are non-negative and sum to one.

[0013] According to a second aspect of an embodiment of the present invention, a drone battery installation and management system of a smart charging cabinet is provided, including: a first unit, for the smart charging cabinet including a battery storage area and a battery replacement area, the battery replacement area is provided with a mechanical arm and a visual positioning system; the smart charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request, the visual positioning system in the battery replacement area collects the three-dimensional position information of the drone battery compartment, and controls the mechanical arm to complete the removal of the battery to be replaced and the installation of the spare battery based on the three-dimensional position information; a second unit, for the smart charging cabinet to collect performance parameters of the spare battery in the battery storage area, the performance parameters including temperature information, power information, number of cycles information and internal resistance information; the performance parameters are input into the A preset battery status assessment model is input to generate a charging strategy, and the charging strategy includes a charging current curve, a charging time and a temperature control curve; in the process of executing the charging strategy, the charging current curve, the charging time and the temperature control curve are adjusted in real time based on the performance parameters collected in real time; the third unit is used for the smart charging cabinet to evaluate the performance of the backup battery in the battery storage area based on the deep learning algorithm, and generate a backup battery health assessment report including the remaining cycle life prediction, the degree of performance degradation and the charging and discharging efficiency; when the backup battery health assessment result is lower than the safety threshold, the battery recycling process is triggered, and the historical performance parameters of the battery that triggered the recycling process are uploaded to the cloud management platform, and the cloud management platform updates the battery status assessment model by analyzing the historical performance parameters.

[0014] According to a third aspect of an embodiment of the present invention, an electronic device is provided, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0016] The present invention realizes the automatic replacement of UAV batteries through the battery storage area and battery replacement area of ​​the intelligent charging cabinet, combined with a mechanical arm and a visual positioning system. The visual positioning system collects the three-dimensional position information of the UAV battery compartment, ensures the accuracy and reliability of battery replacement, and improves the efficiency and safety of the entire replacement process.

[0017] The intelligent charging cabinet generates a targeted charging strategy by collecting the performance parameters of the backup battery and combining it with the preset battery status evaluation model. During the charging process, the system can dynamically adjust the charging strategy based on the real-time collected performance parameters, ensuring the optimization of the charging process and the maximum utilization of battery performance, thus extending the battery life.

[0018] Based on deep learning algorithms, the smart charging cabinet can comprehensively evaluate the performance of the backup battery and generate a detailed health assessment report. This not only achieves predictive maintenance of battery performance, but also triggers battery recycling by setting safety thresholds, thereby improving the safety of the entire system. At the same time, by uploading the historical performance parameters of the battery to the cloud management platform, the battery status assessment model is continuously updated and optimized, further improving the intelligence level and management efficiency of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of a method for installing and managing drone batteries in a smart charging cabinet according to an embodiment of the present invention; Figure 2 This is a structural schematic diagram of a drone battery installation and management system of a smart charging cabinet according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0022] Figure 1 FIG. 1 is a flow chart of a method for installing and managing drone batteries in a smart charging cabinet according to an embodiment of the present invention. Figure 1As shown, the method includes: S1. The intelligent charging cabinet includes a battery storage area and a battery replacement area, and the battery replacement area is provided with a mechanical arm and a visual positioning system; the intelligent charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request, and the visual positioning system in the battery replacement area collects the three-dimensional position information of the drone battery compartment, and controls the mechanical arm to complete the removal of the battery to be replaced and the installation of the spare battery based on the three-dimensional position information; S2. The intelligent charging cabinet collects the performance parameters of the spare battery in the battery storage area, and the performance parameters include temperature information, power information, cycle number information and internal resistance information; the performance parameters are input into a preset battery status evaluation model to generate a charging The charging strategy includes a charging current curve, a charging time and a temperature control curve; in the process of executing the charging strategy, the charging current curve, the charging time and the temperature control curve are adjusted in real time based on the performance parameters collected in real time; S3. The intelligent charging cabinet evaluates the performance of the backup battery in the battery storage area based on the deep learning algorithm, and generates a backup battery health assessment report including the remaining cycle life prediction, the degree of performance degradation and the charging and discharging efficiency; when the backup battery health assessment result is lower than the safety threshold, the battery recycling process is triggered, and the historical performance parameters of the battery that triggered the recycling process are uploaded to the cloud management platform, and the cloud management platform updates the battery status assessment model by analyzing the historical performance parameters.

[0023] Exemplarily, the smart charging cabinet includes a battery storage area and a battery replacement area. The battery storage area is used to store spare batteries of various types, and each type of battery is stored in a separate storage compartment. The battery replacement area is equipped with a 6-degree-of-freedom robotic arm and a high-precision visual positioning system.

[0024] When receiving a drone battery replacement request, the smart charging cabinet first parses the battery specification information contained in the request, such as battery model, capacity, etc. Then it selects a matching spare battery from the battery storage area based on the specification information.

[0025] Next, the visual positioning system in the battery replacement area collects the 3D position information of the drone battery compartment. The visual positioning system includes a depth camera, which can accurately locate the 3D coordinates and attitude angle of the battery compartment by acquiring the point cloud data of the battery compartment and combining it with a pre-trained deep learning model.

[0026] Based on the acquired three-dimensional position information, the smart charging cabinet controls the robotic arm to remove the battery to be replaced and install the spare battery. The end of the robotic arm is equipped with a special fixture that can adapt to the gripping of batteries of different models. The motion trajectory of the robotic arm is planned in real time through the inverse kinematics algorithm to ensure that the battery replacement operation is completed safely and smoothly.

[0027] In terms of battery management, the smart charging cabinet collects the performance parameters of the backup batteries in the battery storage area, including temperature information, power information, cycle number information and internal resistance information. Temperature information is collected by a thermocouple sensor with a sampling frequency of 1Hz. Power information is measured by a high-precision coulomb meter with an accuracy of up to 0.1%. Cycle number information is recorded in the battery management system. Internal resistance information is measured by the AC impedance method with a measurement frequency of 1kHz.

[0028] The collected performance parameters are input into the preset battery status evaluation model to generate a personalized charging strategy. The charging strategy includes the charging current curve, charging time and temperature control curve. For example, for a new battery, the charging current curve may be: 0.5C constant current charging to 4.1V, then 0.2C constant current charging to 4.2V, and finally constant voltage charging to a current less than 0.05C. The charging time is about 2 hours. The temperature control curve is set to: linearly adjust the cooling fan speed in the range of 25-35°C.

[0029] In the process of executing the charging strategy, the smart charging cabinet adjusts the charging current curve, charging time and temperature control curve in real time based on the real-time collected performance parameters. For example, when an abnormal increase in battery temperature is detected, the charging current will be reduced and the cooling fan speed will be increased. When an abnormal increase in internal resistance is detected, the charging process will be terminated in advance.

[0030] In addition, the smart charging cabinet evaluates the performance of the backup batteries in the battery storage area based on a deep learning algorithm. The deep learning model takes historical performance parameters as input and outputs a backup battery health assessment report that includes the remaining cycle life prediction, performance degradation degree, and charge and discharge efficiency. For example, for a certain type of battery that has been used 200 times, the model may predict that its remaining cycle life is 300 times, the capacity retention rate is 92%, and the charge and discharge efficiency is 95%.

[0031] When the backup battery health assessment result is lower than the safety threshold, such as the remaining cycle life is less than 50 times or the capacity retention rate is less than 80%, the smart charging cabinet will trigger the battery recycling process. At the same time, the historical performance parameters of the battery that triggered the recycling process will be uploaded to the cloud management platform.

[0032] The cloud management platform aggregates battery performance data from multiple smart charging cabinets and continuously optimizes and updates the battery status assessment model through big data analysis and machine learning algorithms. The updated model will be regularly distributed to each smart charging cabinet to improve the accuracy and efficiency of battery management.

[0033] The present invention realizes the automated replacement and charging management of UAV batteries, greatly improving the efficiency and safety of UAV operations. Through visual positioning and robotic arm operation, battery replacement can be completed accurately and quickly, reducing the errors and risks of manual operation; personalized charging strategies and real-time adjustment mechanisms are adopted to extend the battery life and improve charging efficiency and safety. The charging parameters are dynamically adjusted according to the real-time status of the battery to avoid the occurrence of adverse conditions such as overcharging and over-discharging; a battery health assessment and prediction system based on deep learning is established to realize intelligent management of the entire life cycle of the battery. By timely discovering and processing batteries with degraded performance, the reliability of UAV operations is guaranteed, while the utilization efficiency of battery resources is optimized.

[0034] In an optional embodiment, the smart charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request, the visual positioning system of the battery replacement area collects the three-dimensional position information of the drone battery compartment, and the steps of controlling the robotic arm to complete the removal of the battery to be replaced and the installation of the spare battery based on the three-dimensional position information include: the smart charging cabinet receives a battery replacement request message sent by the drone, the battery replacement request message includes specification information of the battery model and capacity; queries the battery storage area database according to the specification information to obtain the storage location coordinates of the target spare battery, generates a material retrieval path planning scheme for the robotic arm based on the storage location coordinates, and the robotic arm removes the target spare battery from the battery compartment according to the material retrieval path planning scheme The battery storage area is transferred to the waiting position of the battery replacement area; a binocular vision system is used to collect the image of the UAV battery compartment, the feature points of the UAV battery compartment are identified through the SURF feature extraction algorithm, the parallax information of the feature points of the UAV battery compartment is calculated using the stereo matching algorithm, and the spatial position coordinates and attitude angles of the UAV battery compartment are calculated through triangulation based on the parallax information and camera calibration parameters; the motion trajectory of the robotic arm is planned according to the spatial position coordinates and attitude angles, and the robotic arm is controlled to move to the position of the UAV battery compartment; a strong tactile sensor is installed at the end of the robotic arm, and a spiral search method is used to approach the UAV battery compartment, the clamping force of the robotic arm is controlled to remove the battery to be replaced, and the spare battery placed in the waiting position is installed into the UAV battery compartment.

[0035] For example, the smart charging cabinet receives a battery replacement request message sent by the drone. The message contains the specification information of the battery model and capacity, such as "TB55-7660mAh-22.8V". The smart charging cabinet parses the message and extracts the battery model "TB55" and capacity "7660mAh" information.

[0036] The smart charging cabinet queries the battery storage area database based on the extracted specification information. The database pre-stores the storage location coordinates of batteries of different models and capacities. For example, the storage location coordinates of the target backup battery TB55 are queried as (x=100mm, y=200mm, z=300mm).

[0037] Based on the queried storage location coordinates, the smart charging cabinet generates a path planning scheme for the robot arm to pick up materials. The path planning uses the A* algorithm, taking into account factors such as obstacle avoidance, to generate the optimal path from the current position of the robot arm to the target battery storage location.

[0038] The robot arm actuator moves to the target battery storage location according to the planned path. The gripper installed at the end of the robot arm grabs the target backup battery and transfers it to the preset waiting position in the battery replacement area along the planned path, such as (x=500mm, y=600mm, z=200mm).

[0039] The battery replacement area of ​​the smart charging cabinet is equipped with a binocular vision system, which includes two parallel-mounted industrial cameras. The binocular cameras collect stereo image pairs of the drone battery compartment.

[0040] The SURF feature extraction algorithm (accelerated robust feature extraction algorithm) is used for the collected stereo images. The SURF algorithm detects and describes the feature points in the image, such as the edge of the battery compartment, screw holes and other significant features. Each feature point contains information such as position, scale, and direction.

[0041] The disparity information of feature points is calculated using a stereo matching algorithm. First, the SURF feature points of the left and right images are matched to establish a corresponding relationship. Then the pixel coordinate difference of the matched feature points is calculated to obtain the disparity value.

[0042] Based on the disparity information and camera calibration parameters, the three-dimensional coordinates of the feature points are calculated by triangulation. Camera calibration parameters include focal length, principal point coordinates, baseline length, etc. The depth information of the feature points is calculated based on the disparity value, focal length, and baseline length. Combined with the pixel coordinates, the three-dimensional coordinates of the feature points in the camera coordinate system are obtained.

[0043] The PnP algorithm is used to estimate the posture of the drone battery compartment. The coordinates of the three-dimensional feature points are aligned with the corresponding points of the pre-established battery compartment CAD model, and the transformation relationship from the camera coordinate system to the battery compartment coordinate system is solved to obtain the spatial position and posture angle of the battery compartment.

[0044] According to the calculated spatial position and attitude angle of the battery compartment, the motion trajectory of the robot arm is planned. A quintic polynomial interpolation method is used to generate a smooth trajectory to ensure that the end of the robot arm can accurately approach the battery compartment position.

[0045] The robot arm is controlled to move along the planned trajectory to the vicinity of the drone battery compartment. A six-dimensional force / torque sensor is installed at the end of the robot arm to accurately sense the contact force.

[0046] The spiral search method is used to make the end of the robot arm approach the battery compartment accurately. With the current position as the center, a small range search is carried out along the spiral trajectory until the force sensor detects the contact force.

[0047] According to the feedback from the force sensor, the robot gripper is controlled to apply appropriate force to remove the battery to be replaced. The gripper grabs the battery and moves it out of the battery compartment along the preset trajectory.

[0048] The robot arm turns to the waiting position and grabs the spare battery. It moves the spare battery to the battery compartment along the planned trajectory and accurately positions it through force feedback to complete the battery installation.

[0049] The present invention realizes the fully automatic replacement of UAV batteries, improving efficiency and safety. Through visual positioning and force feedback control, the accuracy and reliability of the battery replacement process are guaranteed; the binocular vision system and SURF feature extraction algorithm are adopted to accurately obtain the three-dimensional position information of the battery compartment, which can meet the battery replacement needs of different models of UAVs; the end of the mechanical arm is equipped with a force tactile sensor, which uses a spiral search method to accurately approach the battery compartment, effectively solving the position error problem and improving the success rate of battery disassembly and assembly.

[0050] In an optional embodiment, the intelligent charging cabinet collects performance parameters of the backup battery in the battery storage area, the performance parameters including temperature information, power information, cycle number information and internal resistance information; the performance parameters are input into a preset battery status evaluation model to generate a charging strategy, the charging strategy including a charging current curve, a charging time and a temperature control curve, the steps comprising: the temperature information is collected by a platinum resistance sensor, the power information is collected by a coulomb counting method, the cycle number information is collected by detecting the number of conversions between the charging cut-off voltage and the discharging cut-off voltage, and the internal resistance information is collected by an AC impedance method; data preprocessing is performed on the performance parameters, including performing a moving average filter on the temperature information to remove mutation points, and performing a moving average filter on the power information to remove mutation points. The Kalman filtering algorithm is used to correct the accumulated error of the internal resistance information, and the high-frequency noise is eliminated by wavelet transform on the internal resistance information to obtain the pre-processed performance parameters; the pre-processed performance parameters are input into a preset battery state evaluation model, and the battery state evaluation model uses a support vector regression model, and a radial basis kernel function is used to establish a nonlinear mapping relationship, and an adaptive weight coefficient is set according to the contribution of different performance parameters to the battery life prediction, and the battery health state index, the remaining cycle life prediction value and the capacity attenuation trend are output; based on the battery health state index, the remaining cycle life prediction value and the capacity attenuation trend, a particle swarm optimization algorithm is used to generate a charging strategy, and the charging strategy includes a segmented constant current-constant voltage charging current curve, a charging time and a temperature control curve.

[0051] Exemplarily, the intelligent charging cabinet collects performance parameters of the backup battery in the battery storage area, including temperature information, power information, cycle number information and internal resistance information. Specifically, the temperature information is collected by a platinum resistance sensor, the power information is collected by a coulomb counting method, the cycle number information is collected by detecting the number of conversions between the charging cut-off voltage and the discharging cut-off voltage, and the internal resistance information is collected by an AC impedance method.

[0052] When collecting temperature information, multiple platinum resistance sensors are evenly arranged on the battery surface, and temperature data is collected every 10 seconds. After collecting 30 times in succession, the average value is taken as the temperature information at that moment. For example, the 30 temperature data (unit: ℃) collected at one time are: 25.3, 25.5, 25.4, 25.6, 25.3, 25.5, 25.4, 25.7, 25.5, 25.6, 25.4, 25.5, 25.3, 25.6, 25.4, 25.5, 25.6, 25.4, 25.5, 25.3, 25.6, 25.4, 25.5, 25.7, 25.5, 25.6, 25.4, 25.5, 25.3, 25.6, and the average value is 25.5℃.

[0053] When collecting power information, a high-precision current sensor is used to monitor the battery charging and discharging current in real time, sampling once per second, and integrating the accumulated current to obtain the power change. For example, during the charging process of a battery within 1 hour, the current value (unit: A) is recorded every 10 minutes, which are: 2.5, 2.3, 2.1, 1.9, 1.7, 1.5. Based on the integration of the current over time, it can be calculated that the power charged in 1 hour is about 2.0Ah.

[0054] When collecting cycle information, the charge cut-off voltage is set to 4.2V and the discharge cut-off voltage is set to 3.0V. Each time the battery voltage rises from 3.0V to 4.2V and then drops to 3.0V, it is recorded as a complete cycle. For example, after a battery has been used for a long time, its voltage change curve shows that it has completed 500 charge and discharge cycles.

[0055] When collecting internal resistance information, a small-amplitude sinusoidal AC signal (frequency range 1Hz-1kHz) is applied to the battery, the amplitude ratio and phase difference of voltage and current are measured, and the internal resistance value is obtained through complex operations. For example, for a 18650 lithium battery, the voltage amplitude is 10mV, the current amplitude is 5mA, and the phase difference is 30° at a frequency of 1kHz. The internal resistance at this frequency is calculated to be approximately 1.73mΩ.

[0056] The collected performance parameters are preprocessed. The temperature information is filtered by 5-point moving average to remove mutation points, and the temperature curve after filtering is smoother. The Kalman filter algorithm is used to correct the cumulative error of the power information to improve the accuracy of power estimation. The internal resistance information is transformed by 4 layers of db4 wavelet to eliminate high-frequency noise and retain the effective signal.

[0057] For example, the original temperature data sequence (unit: ℃) collected at a certain time is: 25.5, 25.4, 25.6, 28.3, 25.5, 25.7, 25.6, 25.5, 25.4, 25.6. After 5-point moving average filtering, the results are: 25.5, 25.5, 26.1, 26.1, 26.1, 25.6, 25.6, 25.5, 25.5, 25.5, which effectively removes the outlier value of 28.3℃.

[0058] For power information, suppose the actual remaining capacity of a battery is 3000mAh, but due to the current integration cumulative error, the calculated remaining capacity is 3200mAh. Using the Kalman filter algorithm, combined with historical data and measurement noise, the estimated value can be corrected to 3050mAh, which is closer to the actual value.

[0059] For the internal resistance information, assuming that the original internal resistance data sequence (unit: mΩ) measured at a frequency of 1 kHz is: 1.73, 1.75, 1.72, 1.78, 1.71, 1.76, 1.74, 1.77, 1.73, 1.75, after wavelet transform denoising, the results are: 1.74, 1.74, 1.73, 1.75, 1.73, 1.75, 1.74, 1.75, 1.74, 1.74, which effectively eliminates high-frequency fluctuations.

[0060] The preprocessed performance parameters are input into the preset battery status evaluation model. The model uses the support vector regression method and the radial basis kernel function to establish a nonlinear mapping relationship. According to the contribution of different performance parameters to the battery life prediction, the adaptive weight coefficients are set: temperature information weight 0.2, power information weight 0.3, cycle number information weight 0.3, internal resistance information weight 0.2.

[0061] The model input is the pre-processed temperature, power, number of cycles and internal resistance data, and outputs the battery health index (value 0-1, 1 means brand new), remaining cycle life prediction value (unit: times) and capacity decay trend (unit: % / 100 cycles). For example, the input parameters of a 18650 lithium battery are: average temperature 25.5℃, remaining capacity 2900mAh, 500 cycles, internal resistance 1.74mΩ at 1kHz frequency, and the model output results are: health index 0.85, remaining cycle life prediction value 450 times, capacity decay trend 2.5% / 100 cycles.

[0062] Based on the battery status evaluation results, the particle swarm optimization algorithm is used to generate a charging strategy, including a segmented constant current-constant voltage charging current curve, charging time, and temperature control curve. Each particle in the particle swarm represents a set of candidate charging parameters, including charging current, charging cut-off voltage, charging time, and target temperature at each stage. Through iterative optimization, the optimal parameter combination that can maximize battery life while ensuring charging efficiency is sought.

[0063] For example, for the above battery, the generated charging strategy is: the first stage is constant current charging, the current is 1.5A, and it lasts for 60 minutes; the second stage is constant current charging, the current is 1.0A, and it lasts for 30 minutes; the third stage is constant voltage charging, the voltage is 4.2V, and the cut-off current is 0.1A or the charging time reaches 90 minutes. The temperature control target is to maintain 25-30℃, and the cooling system is activated when the temperature exceeds 30℃. The total charging time does not exceed 180 minutes.

[0064] The present invention uses a variety of sensors to accurately collect battery performance parameters, combined with data preprocessing technology, to improve the accuracy and reliability of battery status assessment; adopts support vector regression model and adaptive weight coefficient to achieve accurate prediction of battery health status and remaining life, providing a basis for formulating reasonable charging strategy; based on the charging strategy generated by particle swarm optimization algorithm, it can dynamically adjust charging parameters according to the actual state of the battery, while ensuring charging efficiency while maximizing the battery service life, thereby improving the management level and economic benefits of smart charging cabinets.

[0065] In an optional embodiment, the battery state assessment model uses a support vector regression model, adopts a radial basis kernel function to establish a nonlinear mapping relationship, and sets an adaptive weight coefficient according to the contribution of different performance parameters to the battery life prediction, and the steps of outputting a battery health state index, a remaining cycle life prediction value, and a capacity attenuation trend include: standardizing the performance parameters of the battery and converting the performance parameters into a standardized feature vector; constructing a support vector regression model and mapping the standardized feature vector into an output value, wherein the mapping process includes a weight vector to be optimized; constructing a loss function of the support vector regression model, wherein the loss function of the support vector regression model includes a second norm term and an adaptive adjustment term of the weight vector to be optimized; converting the gradient value of the weight vector to be optimized by the loss function of the support vector regression model into an adaptive weight coefficient through an exponential function, wherein the adaptive weight coefficient is used to characterize the effect of different performance parameters on the battery The contribution of the life prediction is that the adaptive adjustment item is composed of the product of the adaptive weight coefficient and the two norms of the weight vector to be optimized; the standardized feature vector is mapped to the high-dimensional feature space through the radial basis kernel function, and the nonlinear relationship between the standardized feature vectors is established. The penalty factor, regularization coefficient, learning rate and kernel function parameters of the support vector regression model are determined by cross-validation; the support vector regression model is trained based on the standardized feature vector to obtain the optimal weight vector, and the battery health status index is generated according to the weighted combination of the standardized feature vector mapped by the radial basis kernel function and the optimal weight vector; the remaining cycle life prediction value is calculated according to the battery health status index, and the remaining cycle life prediction value is determined by the value of the battery health status index and its change rate; based on the remaining cycle life prediction value, a capacity decay trend is generated, and the capacity decay trend uses an exponential function to describe the change law of the battery capacity with the remaining cycle life.

[0066] For example, the design of the battery state assessment model adopts the support vector regression method, establishes a nonlinear mapping relationship through the radial basis kernel function, and introduces an adaptive weight coefficient to reflect the contribution of different performance parameters to the battery life prediction. The model output includes the battery health state index, the remaining cycle life prediction value and the capacity decay trend.

[0067] First, the collected battery performance parameters are standardized. Taking the temperature parameter as an example, the collected temperature data is subtracted from the sample mean and divided by the standard deviation to make the data distributed in the interval [-1, 1]. The same method is used to standardize the power, cycle number and internal resistance information to obtain a standardized feature vector. In actual applications, the temperature data of a certain model of lithium battery is in the range of 20-45℃, and is mapped to the interval [-0.8, 0.9] after standardization.

[0068] In the process of constructing the support vector regression model, the standardized feature vector is used as input and the output value is obtained through the mapping function. The mapping process introduces the weight vector to be optimized as an adjustable parameter. The loss function of the model contains two terms: the bi-norm term of the weight vector is used to prevent overfitting, and the adaptive adjustment term is used to balance the importance of different features. In the specific implementation, the weight vector is initialized to a uniformly distributed random value.

[0069] After the loss function calculates the gradient of the weight vector, the gradient value is converted into an adaptive weight coefficient through an exponential function. The base of the exponential function used in the conversion is the natural logarithm e, and the exponent is the normalized gradient value. The obtained adaptive weight coefficient reflects the importance of different performance parameters to life prediction. Practice shows that the weight coefficient of the temperature parameter is usually between 0.3-0.4, while the weight coefficient of the internal resistance parameter is between 0.2-0.3.

[0070] The standardized feature vector is mapped to the high-dimensional feature space through the radial basis kernel function to establish the nonlinear relationship between the features. The model parameters are determined by the five-fold cross-validation method: the penalty factor ranges from 1 to 100, the regularization coefficient is between 0.01 and 0.1, the learning rate is set to 0.001-0.01, and the kernel function parameter is selected from 0.1 to 1.

[0071] The optimal weight vector is obtained by training the model based on the standardized feature vector. The mapped feature vector is weightedly combined with the optimal weight vector to generate a battery health index with a value range of 0-1. The health index of a certain model of battery is about 0.85 after 500 cycles. The remaining cycle life is calculated based on the value of the health index and its rate of change. Every decrease of 0.1 in the index corresponds to a reduction of about 200 cycles in life. Finally, an exponential function is used to fit the decay trend of battery capacity with the number of cycles. The battery life is determined to end when the capacity decays from the initial value to 80%.

[0072] The present invention dynamically adjusts the importance of different performance parameters through adaptive weight coefficients, thereby improving the sensitivity of the evaluation model to battery performance degradation; adopts radial basis kernel function to realize nonlinear feature mapping, better capturing the complex relationship between battery performance parameters; optimizes model parameters based on cross-validation, thereby enhancing the reliability and generalization ability of the evaluation results; and the prediction results can be directly used to guide charging strategy optimization and battery management decisions.

[0073] In an optional embodiment, during the execution of the charging strategy, the step of adjusting the charging current curve, the charging time and the temperature control curve in real time based on the real-time collected performance parameters includes: establishing an adjustment mechanism for the charging current curve based on the value of the battery health state index, the adjustment mechanism for the charging current curve including: multiplying the value of the battery health state index by the rated maximum charging current to obtain a first charging current value, comparing the first charging current value with the temperature-related current limit, and selecting the smaller value between the first charging current value and the current limit as the real-time charging current; establishing an adjustment mechanism for the charging time based on the remaining cycle life prediction value, the adjustment mechanism for the charging time converts the remaining cycle life prediction value into a charging time increment through an exponential function, and adding the charging time increment to the current moment to obtain the charging termination time; establishing an adjustment mechanism for the temperature control curve based on the capacity decay trend, the adjustment mechanism for the temperature control curve adds the reference temperature to the temperature adjustment amount related to the capacity decay to obtain the target temperature; normalizing the value of the battery health state index, the remaining cycle life prediction value and the capacity decay trend to construct a fuzzy set, performing reasoning operations based on the fuzzy rule base, and using the center of gravity method to convert the fuzzy output into the control amount of the charging current curve, the charging time and the temperature control curve.

[0074] For example, in the process of executing the charging strategy, it is first necessary to collect the performance parameters of the battery in real time, including the battery health status index, the remaining cycle life prediction value and the capacity decay trend, etc. Based on these parameters collected in real time, the charging current curve, charging time and temperature control curve are adjusted in real time.

[0075] First, a regulation mechanism for the charging current curve is established based on the value of the battery health index. The mechanism multiplies the value of the battery health index by the rated maximum charging current to obtain a first charging current value. For example, if the battery health index is 0.9 and the rated maximum charging current is 10A, the first charging current value is 9A. The first charging current value is then compared with the temperature-related current limit, and the smaller value of the two is selected as the real-time charging current. If the temperature-related current limit is 8A, the final real-time charging current is 8A.

[0076] Secondly, a charging time adjustment mechanism is established based on the remaining cycle life prediction value. This mechanism converts the remaining cycle life prediction value into a charging time increment through an exponential function. For example, if the remaining cycle life prediction value is 500 times, the charging time increment obtained by exponential function conversion is 10 minutes. The charging time increment is added to the current time to obtain the charging termination time. If the current time is 14:00, the charging termination time is 14:10.

[0077] Again, a temperature control curve adjustment mechanism is established based on the capacity decay trend. This mechanism adds the reference temperature to the temperature adjustment amount related to capacity decay to obtain the target temperature. For example, if the reference temperature is 25°C and the temperature adjustment amount related to capacity decay is 2°C, the final target temperature is 27°C.

[0078] Finally, the values ​​of the battery health status index, the remaining cycle life prediction value, and the capacity decay trend are normalized to construct a fuzzy set. For example, the battery health status index is normalized to 0.9, the remaining cycle life prediction value is normalized to 0.8, and the capacity decay trend is normalized to 0.7. The inference operation is performed based on the pre-set fuzzy rule base to obtain the fuzzy output. The center of gravity method is used to convert the fuzzy output into the control quantity of the charging current curve, charging time, and temperature control curve.

[0079] In practical applications, the real-time performance parameters of the battery need to be collected first. The battery management system (BMS) can be used to collect basic parameters such as battery voltage, current, temperature, etc., and then the battery health status index, remaining cycle life prediction value and capacity decay trend can be calculated through algorithms.

[0080] The battery health index can be calculated based on the battery's internal resistance, capacity, self-discharge rate and other parameters. For example, if the rated internal resistance of the battery is 100mΩ, the measured internal resistance is 110mΩ, the rated capacity is 100Ah, the measured capacity is 95Ah, and the self-discharge rate is within the normal range, the battery health index can be calculated to be about 0.95.

[0081] The remaining cycle life prediction value can be predicted based on the battery's cumulative charge and discharge times, deep cycle times, shallow cycle times and other historical data, combined with the battery's usage environment and working status. For example, assuming that the battery has undergone 500 charge and discharge cycles, including 200 deep cycles and 300 shallow cycles, and the current usage environment is mild and the working status is stable, the remaining cycle life can be predicted to be 1,000 times.

[0082] The capacity attenuation trend can be obtained by measuring the battery capacity regularly and fitting the capacity change curve over time or cycle number. For example, assuming that the capacity of the battery after 100, 300 and 500 times is 98Ah, 96Ah and 95Ah respectively, a capacity attenuation curve can be fitted to predict the future capacity attenuation trend.

[0083] Based on these real-time collected and calculated parameters, the charging strategy can be adjusted in real time. The first is to adjust the charging current curve. Multiply the battery health index by the rated maximum charging current to get a preliminary charging current value. For example, if the rated maximum charging current is 10A and the battery health index is 0.95, the preliminary charging current value is 9.5A. Then consider the temperature factor. Assuming that the current temperature is 30°C and the corresponding temperature-related current limit is 9A, the final real-time charging current is 9A.

[0084] The adjustment of charging time is based on the remaining cycle life prediction value. The remaining cycle life prediction value is converted into a charging time increment through an exponential function. For example, if the remaining cycle life prediction value is 1000 times, an exponential function can be set so that when the prediction value is 1000 times, the corresponding charging time increment is 5 minutes. This charging time increment is added to the original charging end time to obtain a new charging end time.

[0085] The temperature control curve is adjusted based on the capacity decay trend. First, set a reference temperature, such as 25°C, and then determine a temperature adjustment amount based on the capacity decay trend. For example, if the capacity decay trend is slow, the temperature adjustment amount can be set to 1°C; if the capacity decay trend is fast, the temperature adjustment amount can be set to 2°C. Add the reference temperature and the temperature adjustment amount to get the final target temperature.

[0086] Finally, the three parameters of battery health index, remaining cycle life prediction value and capacity decay trend are normalized to construct a fuzzy set. For example, the battery health index can be divided by 1, the remaining cycle life prediction value can be divided by 2000, and the capacity decay trend can be mapped to a value between 0 and 1. Then, reasoning operations are performed based on the pre-set fuzzy rule base. Fuzzy rules can include "If the battery is in good health and has a large remaining cycle life, a larger charging current can be used" and so on. After obtaining the fuzzy output through fuzzy reasoning, the centroid method is used to convert it into a specific control quantity, which is used to adjust the charging current curve, charging time and temperature control curve.

[0087] The present invention improves the safety and reliability of the charging process. By adjusting the charging current, charging time and temperature control parameters in real time, it can effectively avoid safety hazards such as overcharging and overheating, and extend the service life of the battery. It optimizes the charging efficiency and dynamically adjusts the charging strategy based on the real-time status of the battery, which can not only ensure the charging speed but also take into account the battery life, thus achieving a balance between charging efficiency and battery life.

[0088] In an optional embodiment, the smart charging cabinet evaluates the performance of the backup batteries in the battery storage area based on a deep learning algorithm, and generates a backup battery health assessment report including a remaining cycle life prediction, a degree of performance degradation, and a charge and discharge efficiency; when the backup battery health assessment result is lower than a safety threshold, a battery recycling process is triggered, and the historical performance parameters of the battery that triggered the recycling process are uploaded to a cloud management platform, and the cloud management platform updates the battery status assessment model by analyzing the historical performance parameters, the steps comprising: collecting battery performance data including voltage, current, temperature, state of charge, and timestamp, and constructing the battery performance data into a feature matrix; inputting the feature matrix into a multi-scale convolutional neural network, the multi-scale convolutional neural network comprising three parallel one-dimensional convolution branches, each one-dimensional convolution branch being provided with a convolution kernel of a different size, and extracting performance characteristics of different time scales through a convolution operation; splicing the performance characteristics output by the one-dimensional convolution branch, and fusing the performance characteristics through a fully connected layer, and the outputs of the fully connected layer are respectively connected to the remaining A remaining cycle life prediction branch, a performance degradation degree prediction branch and a charge and discharge efficiency prediction branch; a multi-task hybrid loss function is constructed based on the attention mechanism, the multi-task hybrid loss function includes a remaining cycle life loss term, a performance degradation degree loss term and a charge and discharge efficiency loss term, and the weight coefficient of each loss term is automatically adjusted through back propagation; the remaining cycle life prediction value, the performance degradation degree prediction value and the charge and discharge efficiency prediction value are compared with the preset safety threshold, and a backup battery health assessment report is generated, and when any prediction value exceeds the corresponding safety threshold, the battery recycling process is triggered; the historical performance data of the battery that triggers the recycling process is uploaded to the cloud management platform, the cloud management platform performs performance degradation feature analysis on the historical performance data, updates the weight coefficient of the attention mechanism based on the performance degradation feature analysis results, and adjusts the network structure parameters of the multi-scale convolutional neural network; the cloud management platform selects samples with a similarity greater than a preset similarity threshold from the historical performance data based on the maximum mean difference criterion to update the battery status assessment model.

[0089] For example, first, the smart charging cabinet collects the performance data of the backup battery in the storage area, including voltage, current, temperature, state of charge and corresponding timestamp. Taking a certain type of lithium battery as an example, the collection interval is 1 minute, and the collection is continued for 24 hours, and 1440 sets of data are obtained. These data are constructed into a feature matrix with a matrix dimension of 1440×5.

[0090] Next, the feature matrix is ​​input into the designed multi-scale convolutional neural network. The network contains three parallel one-dimensional convolution branches with convolution kernel sizes of 3, 5, and 7, respectively, to extract performance characteristics at different time scales. Taking the first convolution branch as an example, 32 one-dimensional convolution kernels of size 3 and a step size of 1 are used to perform convolution operations on the input matrix, and the feature map size is 1438×32. The other two branches are processed similarly, and feature maps of 1436×32 and 1434×32 are obtained respectively.

[0091] Then the feature graphs output by the three branches are spliced ​​in the time dimension to obtain a 4308×32 fusion feature. The fusion feature is reduced in dimension through two fully connected layers (the number of nodes is 256 and 128 respectively) to obtain a 128-dimensional comprehensive feature vector. This feature vector is connected to three prediction branches: the remaining cycle life prediction branch (1 output node), the performance degradation degree prediction branch (3 output nodes, corresponding to the three levels of mild, moderate and severe) and the charge and discharge efficiency prediction branch (1 output node).

[0092] In the training phase, a multi-task hybrid loss function is constructed. The remaining cycle life prediction uses the mean square error loss, the performance degradation prediction uses the cross entropy loss, and the charge and discharge efficiency prediction uses the mean absolute error loss. The attention mechanism is introduced to adaptively adjust the weights of the three loss items. The initial weights are all 1 / 3. As the training progresses, the model automatically adjusts the importance of each loss item.

[0093] After training is completed, the real-time performance data of the battery is input into the model to obtain the predicted values ​​of the remaining cycle life, performance degradation degree and charge and discharge efficiency. The predicted value is compared with the preset safety threshold to generate a health assessment report. Taking a certain brand of battery as an example, the safety threshold is set as: the remaining cycle life is not less than 200 times, the performance degradation degree does not exceed "moderate", and the charge and discharge efficiency is not less than 85%. When any indicator exceeds the threshold, the recycling process is triggered.

[0094] The historical performance data of the battery that triggers the recycling is uploaded to the cloud management platform. The platform analyzes the performance degradation characteristics in the data, such as the capacity decay rate, the internal resistance growth trend, etc. Based on the analysis results, the attention mechanism weight coefficient is updated, such as increasing the weight of the remaining cycle life prediction. At the same time, the network structure is adjusted, such as increasing the number of convolution kernels or changing the number of nodes in the fully connected layer.

[0095] Finally, the cloud platform selects samples with high similarity to the current battery performance from historical data based on the maximum mean difference criterion. Specifically, the similarity value is obtained by calculating the mean difference between the current battery and the historical samples in various performance indicators, and samples with similarity values ​​greater than the preset threshold are selected. These samples are used to update and fine-tune the battery status assessment model to improve the model's adaptability to new models or abnormal batteries.

[0096] The present invention uses a multi-scale convolutional network and attention mechanism to achieve accurate assessment of the multi-dimensional health status of the battery, thereby improving the comprehensiveness and accuracy of the assessment. The maximum mean difference criterion is introduced to select similar samples to update the model, thereby enhancing the model's generalization ability for new types of batteries and improving the robustness of the assessment.

[0097] In an optional embodiment, a multi-task hybrid loss function is constructed based on the attention mechanism, and the multi-task hybrid loss function includes a remaining cycle life loss term, a performance degradation degree loss term, and a charge and discharge efficiency loss term. The step of automatically adjusting the weight coefficient of each loss term through back propagation includes: inputting the hidden layer feature vector of the battery performance data into the temporal attention layer, calculating the attention score through the attention weight matrix and the bias term, and normalizing the attention score through the softmax function to obtain the temporal attention weight; constructing a weighted mean square error loss function based on the temporal attention weight to calculate the remaining cycle life loss term, and the weighted mean square error loss function weights the error term between the actual remaining cycle life value and the predicted value; A weighted cross entropy loss function is used to calculate the performance degradation loss term, and the weighted cross entropy loss function introduces a balance factor calculated based on the inverse of the number of samples to balance the contribution of samples with different degradation degrees; the deviation between the true value and the predicted value of the battery charge and discharge efficiency is substituted into the Gaussian kernel function for weight mapping to generate a charge and discharge efficiency loss term, and the parameters of the weight mapping are determined by the battery state of charge value; the remaining cycle life loss term, the performance degradation loss term and the charge and discharge efficiency loss term are combined into a multi-task hybrid loss function, and the weight coefficients of the remaining cycle life loss term, the performance degradation loss term and the charge and discharge efficiency loss term are automatically adjusted through back propagation, and the weight coefficients are exponentially and normalized to ensure that they are non-negative and sum to one.

[0098] Exemplarily, the battery performance data is first passed through a feature extraction network to obtain a hidden layer feature vector, and the vector dimension is set to 128 dimensions. The feature vector is input into the temporal attention layer, which contains a learnable weight matrix and bias term. The weight matrix is ​​initialized to a random Gaussian distribution with a mean of 0 and a standard deviation of 0.01, and the bias term is initialized to a zero vector. The attention score is calculated by the weight matrix and the bias term. The attention score at a certain moment reflects the importance of the feature at that moment to the prediction task. The attention score is normalized by the softmax function (normalized exponential function) to obtain a temporal attention weight with a sum of 1. Practice shows that for cycle life prediction tasks, the temporal weight of the rapid decay stage of battery performance is usually between 0.15-0.25.

[0099] In the calculation of the remaining cycle life loss term, the weighted mean square error loss function is used. Taking a certain model of lithium battery as an example, the actual remaining cycle life is 800 times, the model prediction value is 750 times, and the prediction error of 50 times is substituted into the loss function. The weight of the error term is determined by the time series attention weight, so that the model pays more attention to the prediction accuracy of key time series points. For the performance degradation degree loss term, the weighted cross entropy loss function is used. The battery performance degradation degree is divided into four levels: normal, slightly degraded, moderately degraded and severely degraded, with sample numbers of 1000, 800, 400 and 200 respectively. The balance factor is calculated based on the inverse of the sample number, and the values ​​are 0.001, 0.00125, 0.0025 and 0.005, respectively, to balance the contribution of samples with different degradation degrees.

[0100] The calculation of the charge and discharge efficiency loss term uses the Gaussian kernel function for weight mapping. Taking the charge and discharge efficiency under a certain working condition as an example, the true value is 92%, the predicted value is 90%, and the 2% deviation is substituted into the Gaussian kernel function. The parameters of the kernel function are dynamically determined by the battery state of charge value. When the state of charge is in the range of 20%-80%, the variance parameter of the kernel function is small, making the model more sensitive to the prediction deviation in this interval.

[0101] The three loss terms are combined into a multi-task hybrid loss function, and the initial weight coefficients are all set to one-third. During the model training process, the weight coefficients of each loss term are automatically adjusted through the back-propagation algorithm. The weight coefficients are indexed and normalized to ensure that the weights are non-negative and sum to 1. After 1,000 rounds of iterative training, the weight coefficients of the remaining cycle life loss term, the performance degradation loss term, and the charge and discharge efficiency loss term converge to about 0.4, 0.35, and 0.25, respectively.

[0102] The present invention captures the long-term dependencies of battery performance data through a temporal attention mechanism, thereby improving the ability to extract key temporal features; introduces a sample balance factor to solve the sample imbalance problem in the classification of performance degradation levels; and based on a back-propagation weight adaptive adjustment mechanism, achieves a dynamic balance of multi-task learning objectives and improves the comprehensive prediction performance of the model.

[0103] Figure 2 FIG. 1 is a schematic diagram of the structure of the drone battery installation and management system of the smart charging cabinet according to an embodiment of the present invention. Figure 2As shown, the system includes: a first unit, which is used for the intelligent charging cabinet to include a battery storage area and a battery replacement area, and the battery replacement area is provided with a mechanical arm and a visual positioning system; the intelligent charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request, and the visual positioning system in the battery replacement area collects the three-dimensional position information of the drone battery compartment, and controls the mechanical arm to complete the removal of the battery to be replaced and the installation of the spare battery based on the three-dimensional position information; the second unit is used for the intelligent charging cabinet to collect performance parameters of the spare battery in the battery storage area, and the performance parameters include temperature information, power information, cycle number information and internal resistance information; the performance parameters are input into a preset battery status evaluation model to generate A charging strategy is formed, and the charging strategy includes a charging current curve, a charging time and a temperature control curve; in the process of executing the charging strategy, the charging current curve, the charging time and the temperature control curve are adjusted in real time based on the performance parameters collected in real time; the third unit is used for the smart charging cabinet to evaluate the performance of the backup battery in the battery storage area based on the deep learning algorithm, and generate a backup battery health assessment report including the remaining cycle life prediction, the degree of performance degradation and the charging and discharging efficiency; when the backup battery health assessment result is lower than the safety threshold, the battery recycling process is triggered, and the historical performance parameters of the battery that triggered the recycling process are uploaded to the cloud management platform, and the cloud management platform updates the battery status assessment model by analyzing the historical performance parameters.

[0104] According to a third aspect of an embodiment of the present invention, an electronic device is provided, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0105] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0106] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The method for installing and managing drone batteries in a smart charging cabinet is characterized in that: include: The intelligent charging cabinet includes a battery storage area and a battery replacement area, and the battery replacement area is provided with a robotic arm and a visual positioning system; The intelligent charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request. The visual positioning system in the battery replacement area collects the three-dimensional position information of the drone battery compartment and controls the mechanical arm based on the three-dimensional position information to complete the removal of the battery to be replaced and the installation of the spare battery. The intelligent charging cabinet collects performance parameters of the backup battery in the battery storage area, wherein the performance parameters include temperature information, power information, cycle number information, and internal resistance information; the performance parameters are input into a preset battery status evaluation model to generate a charging strategy, wherein the charging strategy includes a charging current curve, a charging time, and a temperature control curve; In the process of executing the charging strategy, the charging current curve, charging time and temperature control curve are adjusted in real time based on the real-time collected performance parameters; The smart charging cabinet evaluates the performance of the backup batteries in the battery storage area based on a deep learning algorithm, and generates a backup battery health assessment report that includes remaining cycle life prediction, performance degradation degree, and charge and discharge efficiency; when the backup battery health assessment result is lower than the safety threshold, the battery recycling process is triggered, and the historical performance parameters of the battery that triggered the recycling process are uploaded to the cloud management platform, and the cloud management platform updates the battery status assessment model by analyzing the historical performance parameters.

2. The method according to claim 1, characterized in that The smart charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request, and the visual positioning system in the battery replacement area collects the three-dimensional position information of the drone battery compartment. The steps of controlling the mechanical arm to complete the removal of the battery to be replaced and the installation of the spare battery based on the three-dimensional position information include: The smart charging cabinet receives a battery replacement request message sent by the drone, wherein the battery replacement request message includes specification information of the battery model and capacity; queries the battery storage area database according to the specification information to obtain the storage location coordinates of the target backup battery, generates a material retrieval path planning scheme for the manipulator based on the storage location coordinates, and the manipulator transfers the target backup battery from the battery storage area to the waiting position in the battery replacement area according to the material retrieval path planning scheme; A binocular vision system is used to collect images of the drone battery compartment, and feature points of the drone battery compartment are identified through a SURF feature extraction algorithm. The disparity information of the feature points of the drone battery compartment is calculated using a stereo matching algorithm. Based on the disparity information and camera calibration parameters, the spatial position coordinates and attitude angle of the drone battery compartment are calculated through triangulation. The motion trajectory of the robotic arm is planned according to the spatial position coordinates and attitude angles, and the robotic arm is controlled to move to the position of the drone battery compartment; a strong tactile sensor is installed at the end of the robotic arm, and a spiral search method is used to approach the drone battery compartment, and the clamping force of the robotic arm is controlled to remove the battery to be replaced, and the spare battery placed in the waiting position is installed into the drone battery compartment.

3. The method according to claim 1, characterized in that The smart charging cabinet collects performance parameters of the backup battery in the battery storage area, wherein the performance parameters include temperature information, power information, cycle number information, and internal resistance information; and the steps of inputting the performance parameters into a preset battery state evaluation model to generate a charging strategy, wherein the charging strategy includes a charging current curve, a charging time, and a temperature control curve include: The temperature information is collected by a platinum resistance sensor, the power information is collected by a coulomb counting method, the cycle number information is collected by detecting the number of conversions between the charge cut-off voltage and the discharge cut-off voltage, and the internal resistance information is collected by an AC impedance method; data preprocessing is performed on the performance parameters, including performing a moving average filter on the temperature information to remove mutation points, using a Kalman filter algorithm to correct the accumulated error of the power information, and eliminating high-frequency noise on the internal resistance information by wavelet transform, to obtain the preprocessed performance parameters; The preprocessed performance parameters are input into a preset battery state evaluation model, which uses a support vector regression model, adopts a radial basis kernel function to establish a nonlinear mapping relationship, and sets an adaptive weight coefficient according to the contribution of different performance parameters to battery life prediction, and outputs a battery health state index, a remaining cycle life prediction value, and a capacity decay trend; Based on the battery health status index, the remaining cycle life prediction value and the capacity decay trend, a particle swarm optimization algorithm is used to generate a charging strategy, which includes a segmented constant current-constant voltage charging current curve, a charging time and a temperature control curve.

4. The method according to claim 3, characterized in that The battery state assessment model uses a support vector regression model, adopts a radial basis kernel function to establish a nonlinear mapping relationship, and sets an adaptive weight coefficient according to the contribution of different performance parameters to the battery life prediction. The steps of outputting a battery health state index, a remaining cycle life prediction value, and a capacity decay trend include: Standardizing the performance parameters of the battery and converting the performance parameters into standardized feature vectors; A support vector regression model is constructed, and the standardized feature vector is mapped to an output value, wherein the mapping process includes a weight vector to be optimized; a loss function of the support vector regression model is constructed, wherein the loss function of the support vector regression model includes a two-norm term and an adaptive adjustment term of the weight vector to be optimized; a gradient value of the weight vector to be optimized of the loss function of the support vector regression model is converted into an adaptive weight coefficient by an exponential function, wherein the adaptive weight coefficient is used to characterize the contribution of different performance parameters to the battery life prediction, and the adaptive adjustment term is composed of the adaptive weight coefficient and the two-norm product of the weight vector to be optimized; The standardized feature vectors are mapped to a high-dimensional feature space through a radial basis kernel function, a nonlinear relationship between the standardized feature vectors is established, and a penalty factor, a regularization coefficient, a learning rate and kernel function parameters of a support vector regression model are determined through cross-validation; A support vector regression model is trained based on the standardized feature vector to obtain an optimal weight vector, and a battery health status index is generated according to a weighted combination of the standardized feature vector mapped by a radial basis kernel function and the optimal weight vector; a remaining cycle life prediction value is calculated according to the battery health status index, and the remaining cycle life prediction value is determined by the value of the battery health status index and its rate of change; a capacity decay trend is generated based on the remaining cycle life prediction value, and the capacity decay trend uses an exponential function to describe the change law of the battery capacity with the remaining cycle life.

5. The method according to claim 1, characterized in that In the process of executing the charging strategy, the steps of adjusting the charging current curve, charging time and temperature control curve in real time based on the real-time collected performance parameters include: Establishing a charging current curve adjustment mechanism based on the value of the battery health state index, the charging current curve adjustment mechanism comprising: multiplying the value of the battery health state index by the rated maximum charging current to obtain a first charging current value, comparing the first charging current value with a temperature-related current limit, and selecting a smaller value between the first charging current value and the current limit as the real-time charging current; Establishing a charging time adjustment mechanism based on the remaining cycle life prediction value, wherein the charging time adjustment mechanism converts the remaining cycle life prediction value into a charging time increment through an exponential function, and adds the charging time increment to the current time to obtain a charging termination time; Establishing a temperature control curve adjustment mechanism based on the capacity decay trend, wherein the temperature control curve adjustment mechanism adds a reference temperature to a temperature adjustment amount related to the capacity decay to obtain a target temperature; The numerical value of the battery health status index, the remaining cycle life prediction value and the capacity decay trend are normalized to construct a fuzzy set. The reasoning operation is performed based on the fuzzy rule base, and the center of gravity method is used to convert the fuzzy output into the control quantity of the charging current curve, charging time and temperature control curve.

6. The method according to claim 1, characterized in that The smart charging cabinet evaluates the performance of the backup batteries in the battery storage area based on a deep learning algorithm, and generates a backup battery health assessment report including a remaining cycle life prediction, a performance degradation degree, and a charge and discharge efficiency; when the backup battery health assessment result is lower than a safety threshold, a battery recycling process is triggered, and the historical performance parameters of the battery that triggers the recycling process are uploaded to a cloud management platform. The cloud management platform updates the battery status assessment model by analyzing the historical performance parameters, including the following steps: Collecting battery performance data including voltage, current, temperature, state of charge and timestamp, and constructing the battery performance data into a feature matrix; Inputting the feature matrix into a multi-scale convolutional neural network, the multi-scale convolutional neural network comprises three parallel one-dimensional convolution branches, each one-dimensional convolution branch is provided with a convolution kernel of a different size, and the performance characteristics of different time scales are extracted through convolution operation; The performance features output by the one-dimensional convolution branch are concatenated, and the performance features are fused through a fully connected layer, and the output of the fully connected layer is respectively connected to the remaining cycle life prediction branch, the performance degradation degree prediction branch, and the charge and discharge efficiency prediction branch; A multi-task hybrid loss function is constructed based on the attention mechanism, wherein the multi-task hybrid loss function includes a remaining cycle life loss term, a performance degradation degree loss term, and a charge and discharge efficiency loss term, and the weight coefficient of each loss term is automatically adjusted through back propagation; Compare the remaining cycle life prediction value, performance degradation prediction value and charge and discharge efficiency prediction value with the preset safety threshold value to generate a backup battery health assessment report. When any prediction value exceeds the corresponding safety threshold value, the battery recycling process is triggered; Uploading the historical performance data of the battery that triggers the recycling process to the cloud management platform, the cloud management platform performs a performance degradation feature analysis on the historical performance data, updates the weight coefficient of the attention mechanism based on the performance degradation feature analysis results, and adjusts the network structure parameters of the multi-scale convolutional neural network; The cloud management platform selects samples with a similarity greater than a preset similarity threshold from historical performance data based on a maximum mean difference criterion to update a battery status evaluation model.

7. The method according to claim 6, characterized in that A multi-task hybrid loss function is constructed based on the attention mechanism, wherein the multi-task hybrid loss function includes a remaining cycle life loss term, a performance degradation degree loss term, and a charge and discharge efficiency loss term. The steps of automatically adjusting the weight coefficient of each loss term through back propagation include: Input the hidden feature vector of the battery performance data into the temporal attention layer, calculate the attention score through the attention weight matrix and the bias term, and normalize the attention score through the softmax function to obtain the temporal attention weight; A weighted mean square error loss function is constructed based on the temporal attention weight to calculate the remaining cycle life loss term, and the weighted mean square error loss function weights the error term between the true remaining cycle life value and the predicted value; a weighted cross entropy loss function is used to calculate the performance degradation degree loss term, and the weighted cross entropy loss function introduces a balance factor calculated based on the inverse of the number of samples to balance the contribution of samples with different degradation degrees; the deviation between the true value and the predicted value of the battery charge and discharge efficiency is substituted into the Gaussian kernel function for weight mapping to generate a charge and discharge efficiency loss term, and the parameters of the weight mapping are determined by the battery state of charge value; The remaining cycle life loss term, performance degradation degree loss term and charge and discharge efficiency loss term are combined into a multi-task hybrid loss function, and the weight coefficients of the remaining cycle life loss term, the performance degradation degree loss term and the charge and discharge efficiency loss term are automatically adjusted through back propagation. The weight coefficients are exponentially and normalized to ensure that they are non-negative and sum to one.

8. A drone battery installation and management system of a smart charging cabinet, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used for the intelligent charging cabinet including a battery storage area and a battery replacement area, wherein the battery replacement area is provided with a robotic arm and a visual positioning system; The intelligent charging cabinet selects a matching spare battery from the battery storage area according to the battery specification information in the drone battery replacement request. The visual positioning system in the battery replacement area collects the three-dimensional position information of the drone battery compartment and controls the mechanical arm based on the three-dimensional position information to complete the removal of the battery to be replaced and the installation of the spare battery. The second unit is used for the intelligent charging cabinet to collect performance parameters of the backup battery in the battery storage area, wherein the performance parameters include temperature information, power information, cycle number information and internal resistance information; the performance parameters are input into a preset battery state evaluation model to generate a charging strategy, wherein the charging strategy includes a charging current curve, a charging time and a temperature control curve; In the process of executing the charging strategy, the charging current curve, charging time and temperature control curve are adjusted in real time based on the real-time collected performance parameters; The third unit is used for evaluating the performance of the backup batteries in the battery storage area of ​​the smart charging cabinet based on a deep learning algorithm, and generating a backup battery health assessment report including the remaining cycle life prediction, performance degradation degree and charging and discharging efficiency; when the backup battery health assessment result is lower than the safety threshold, the battery recycling process is triggered, and the historical performance parameters of the battery that triggered the recycling process are uploaded to the cloud management platform, and the cloud management platform updates the battery status assessment model by analyzing the historical performance parameters.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images