An AI-based method and system for obstacle avoidance and path planning inventory in a cold chain warehouse by an unmanned aerial vehicle

Abstract

The present invention discloses an AI-intelligent cold chain warehouse drone obstacle avoidance and path planning inventory method and system, belonging to the field of intelligent obstacle avoidance. The method includes: arranging different types of sensors in the cold chain warehouse to sense environmental changes in the cold chain warehouse; the edge gateway integrates different types of sensors and preprocesses the collected data, and sends the processed data to the cloud database; constructing a digital twin model of the target cold chain warehouse and an intelligent inventory path planning algorithm in the cloud database; the cloud sends the path plan output by the intelligent inventory path planning algorithm to the drone terminal and the digital twin model at the same time; the drone terminal completes the inventory task according to the path plan, while the digital twin model updates the digital twin model in real time according to the path plan and the data fed back by the drone terminal. The present invention can ensure that the drone efficiently and safely performs the inventory task in the complex cold chain warehouse environment.

Description

Technical Field

[0001] The present invention relates to the field of intelligent obstacle avoidance, and particularly to an AI-based intelligent obstacle avoidance and path planning inventory method and system for drones in cold chain warehouses. Background Art

[0002] With the rapid development of the cold chain logistics industry, higher requirements are put forward for the environmental perception ability and operation efficiency of cold chain warehousing management systems. There are some deficiencies in the environmental perception of existing cold chain warehousing management systems, the fusion effect of sensor data is poor, and the environmental perception accuracy and reliability are low. At the same time, existing systems lack intuitive environmental monitoring means and it is difficult to comprehensively grasp the real-time situation of the warehouse. In addition, in terms of inventory operations, the operation efficiency of existing systems is low and there are certain safety hazards, and they cannot adapt to the changes in dynamic environments. Summary of the Invention

[0003] One of the purposes of the present invention is to provide an AI-based intelligent obstacle avoidance and path planning inventory method for drones in cold chain warehouses to solve the problems of low efficiency and certain safety hazards in the inventory operations of cold chain warehouses in the prior art.

[0004] The present invention is realized by the following technical solutions. An AI-based intelligent obstacle avoidance and path planning inventory method for drones in cold chain warehouses includes the following steps: S100. Different types of sensors for sensing environmental changes in the cold chain warehouse are arranged in the cold chain warehouse. The edge gateway integrates different types of sensors into a unified sensor cluster, preprocesses the collected data, and sends the processed data to the cloud database; S200. A digital twin model of the target cold chain warehouse and an intelligent inventory path planning algorithm are built in the cloud database. The cloud database builds a sensor database according to the received data, maps the data in the sensor database to the digital twin model, and the intelligent inventory path planning algorithm outputs a drone obstacle avoidance-inventory path plan according to the sensor data; S300. The cloud sends the path plan output by the intelligent inventory path planning algorithm to the drone terminal and the digital twin model at the same time; S400. The drone terminal completes the inventory task according to the path plan, and the digital twin model updates the digital twin model in real time according to the path plan and the data fed back by the drone terminal.

[0005] Furthermore, different types of sensors include temperature sensors, humidity sensors, 3D lidar sensors, multi-spectral ToF sensors, rgbd camera sensors, long-wave infrared thermal imagers, millimeter-wave radar arrays, and UWB / RFID positioning tags.

[0006] Furthermore, temperature sensors, humidity sensors, multispectral ToF sensors, and RGBD camera sensors are installed at a certain distance on the shelves of the cold chain warehouse; the temperature sensors and humidity sensors are used to detect changes in temperature and humidity in the cold chain warehouse; the multispectral ToF sensor is used to accurately locate the shelf spacing; the RGBD camera sensor is used for near-field object recognition and shelf code OCR analysis.

[0007] Furthermore, 3D lidar sensors, millimeter-wave radar arrays and long-wave infrared thermal imagers are installed on the beams of the cold chain warehouse; the 3D lidar sensors are used to detect major obstacles, the millimeter-wave radar arrays are used for dynamic obstacle tracking; the long-wave infrared thermal imagers are used for temperature gradient field modeling, and can also be used to detect low-temperature risks such as hidden icicles.

[0008] Furthermore, the preprocessing includes: S110, deploying multiple ZigBee wireless temperature measurement nodes on each row of shelves, each ZigBee wireless temperature measurement node is a temperature field grid; S120, the perception level dynamic weight allocation strategy in the edge gateway builds a temperature-sensitive confidence evaluation system according to the number of temperature field grids collected by the ZigBee wireless temperature measurement node, and realizes dynamic weight adjustment.

[0009] Furthermore, the dynamic weight allocation strategy includes the following steps: S121, calculating the temperature gradient and the temperature change rate according to the collected temperature data, and evaluating the dynamic change of the temperature field, wherein the temperature gradient is represented by the following formula: ,in, is the Laplace operator, is the temperature field in three-dimensional space; is the rate of change of temperature in the x direction, is the rate of change of temperature in the y direction, is the rate of change of temperature in the z direction; S123, according to the calculated temperature gradient and change rate, calculate the confidence of each area, and the confidence is calculated by the following formula: ,in, is the confidence level, is the weight coefficient used to adjust the contribution of temperature gradient to confidence, is the modulus of the temperature gradient, To adjust the weight coefficient of the temperature change rate contribution to the confidence, is the absolute value of the temperature change rate; S124, judging whether the current temperature change is drastic according to the temperature gradient and the change rate. When the temperature change is drastic, the sensor weight adjustment stage is entered. If the temperature change is not drastic, the current weight is kept unchanged; S125, in the sensor weight adjustment stage, the sensor weight is dynamically adjusted according to the confidence of each area. The weight is a time-varying weight, which is calculated by the following formula: , where is the time-varying weight of sensor i; is the confidence level of sensor i at the current moment, is the basic weight of the i-th sensor of the same type.

[0010] Furthermore, the basic weight is calculated by the following formula:

[0011] , where is the basic weight of the i-th sensor of the same type, is an exponential function; is the priority coefficient of the i-th sensor; is the influence coefficient of temperature on the performance of sensor i; N is the total number of all sensors of the same type, is the overall priority coefficient of sensors of the same type, is the influence coefficient of sensors of the same type, j represents sensors of the same class, and i represents the i-th sensor in the same class of sensors.

[0012] Furthermore, the intelligent inventory path planning algorithm is constructed based on the A* algorithm and includes the following steps: S210. Convert the environment of the cold chain warehouse into a discretized model composed of temperature-space joint voxels, and the temperature-space joint voxel is expressed as: , where is the identifier of the voxel, and a, b, and c respectively represent the indices of the voxel in the three dimensions x, y, and z of the space coordinate system. x, y, and z are space coordinates, representing the position of the voxel in three-dimensional space, and T is the temperature value in the voxel, is the gradient of the temperature value in the voxel, is the confidence level of the sensor data contained in the voxel; S220. Optimize the path of the UAV according to the relevant information provided in the discretized model, and the multi-constraint optimization objective function includes: , where π is the optimized path, is the temperature exposure term, is the time cost term, is the sensor risk term; is the weight coefficient of the time cost term, is the weight coefficient of the sensor risk term, and these two weight coefficients are used to control the influence degree of each constraint on the optimized path; S230. Replace the cost function of the A* algorithm with the multi-constraint optimization objective function, redefine the cost function f(n) of the A* algorithm so that it can comprehensively consider temperature exposure, time cost, and sensor risk, and finally generate the optimal path through the A* algorithm.

[0013] Further, the temperature exposure term can be expressed by the following formula:

[0014] ,

[0015] where k is the exponential decay factor, is the temperature value at a certain position s on the path, and T ref is the reference temperature, representing the ideal temperature state of the cold chain warehouse. The degree of temperature deviation from the reference temperature on the path will affect the magnitude of the penalty term.

[0016] Further, the time cost term is expressed by the following formula:

[0017] ,

[0018] where the total path length refers to the total distance of the UAV path planning, that is, the sum of the spatial distances between all s points on the path, and the average cruising speed of the UAV represents the average flight speed of the UAV when performing tasks.

[0019] Further, the sensor risk term can be expressed by the following formula:

[0020] ,

[0021] where represents the sensor confidence at a certain position s in the path.

[0022] Further, replacing the cost function of the A* algorithm through the multi-constraint optimization objective function includes the following steps: S231. Initialize the open list and the closed list, and the of the starting point is 0, and the heuristic function is initialized according to the estimation of the target path; S232. Starting from the current node, expand its neighboring nodes. For each neighboring node, calculate: , where is the new cost function, is the actual cost from the starting point to the current node n, and h optimized (n) is the new heuristic function. Select the node with the minimum cost for expansion according to the cost. When expanding the node, check whether the path encounters an obstacle or whether there is a temperature anomaly in the area; S233. Continuously iterate to expand the path until the target node is found. Each time it is expanded, consider the cost of the current path, and select the path with the minimum cost for expansion.

[0023] Further, the new heuristic function is expressed by the following formula:

[0024] , where , , is the weight coefficient, used to balance the importance of different factors; d(n) is the spatial distance from node n to the target node; is the estimated temperature exposure at node n, is the estimated sensor risk at node n.

[0025] On the other hand, the present invention provides a cold chain warehouse UAV obstacle avoidance and path planning inventory system based on AI intelligence. The system includes a processor and a memory. A computer program is stored in the memory. When the computer program is executed by the processor, the cold chain warehouse UAV obstacle avoidance and path planning inventory method based on AI intelligence as described above is implemented.

[0026] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0027] 1. By configuring a temperature-dominated perception level dynamic weight allocation strategy in the edge gateway, the present invention dynamically adjusts the weights of different types of sensors according to the dynamic changes of the temperature field, thereby optimizing the data fusion effect of the sensor cluster, improving the accuracy and reliability of environmental perception, and effectively solving the problems of poor data fusion effect of sensors, low environmental perception accuracy and reliability in a long-term low-temperature environment.

[0028] 2. By constructing a digital twin model of the target cold chain warehouse, the present invention can reflect the actual situation of the cold chain warehouse in real time. The background management personnel can intuitively monitor the environmental changes in the cold chain warehouse, solving the problems of lack of intuitive environmental monitoring means and difficulty in comprehensively grasping the real-time status of the warehouse in the prior art.

[0029] 3. The intelligent inventory path planning algorithm of the present invention can ensure that the UAV efficiently and safely performs the inventory task in a complex cold chain warehouse environment, avoids colliding with abnormal areas, and can adapt to dynamic environmental changes, thereby improving the efficiency and safety of the inventory operation, and solving the problems of low inventory operation efficiency, potential safety hazards and inability to adapt to dynamic environmental changes in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not limit the embodiments of the present invention. In the drawings:

[0031] Figure 1 is the overall method flow chart provided by Embodiment 1 of the present invention.

[0032] Figure 2 is the overall method timing diagram provided by Embodiment 1 of the present invention.

[0033] Figure 3 is the timing diagram of the dynamic weight allocation strategy provided by Embodiment 1 of the present invention.

[0034] Figure 4 This is the algorithm timing diagram for step 2 provided in Embodiment 1 of the present invention.

[0035] Figure 5 This is the overall algorithm flowchart provided in Embodiment 1 of the present invention. Detailed implementation manners

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Generally, the components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations.

[0037] Embodiment 1

[0038] This embodiment discloses a method for obstacle avoidance and path planning inventory of an unmanned aerial vehicle (UAV) in a cold chain warehouse based on AI intelligence.

[0039] In the existing cold chain warehouse inventory, due to the lack of intuitive environmental monitoring means, it is difficult to comprehensively grasp the real-time situation of the warehouse, resulting in low inventory operation efficiency, there are certain safety hazards, and it cannot adapt to the changes in the dynamic environment.

[0040] To solve the above problems, this embodiment discloses a method for obstacle avoidance and path planning inventory of an unmanned aerial vehicle (UAV) in a cold chain warehouse based on AI intelligence. This method arranges various types of sensors in the target cold chain warehouse, integrates these sensors into a unified sensor cluster through an edge gateway, and configures a perception-level dynamic weight allocation strategy dominated by temperature in the edge gateway. According to the dynamic changes of the temperature field, the weights of different types of sensors are dynamically adjusted, so as to optimize the data fusion effect of the sensor cluster and improve the accuracy and reliability of environmental perception. At the same time, by building a digital twin model of the target cold chain warehouse in the cloud, the background management personnel can intuitively monitor the environmental changes in the warehouse. In addition, this method also proposes an intelligent inventory path planning algorithm, which can ensure that the UAV efficiently and safely conducts inventory operations in a complex cold chain warehouse environment, avoid colliding with abnormal areas, and adapt to the changes in the dynamic environment.

[0041] The solution in this embodiment realizes the autonomous inventory and real-time digital twin modeling of the UAV in a low-temperature and complex environment through a hierarchical architecture.

[0042] Specifically, the solution in this application can be roughly divided into 4 levels:

[0043] A perception layer composed of multiple sensors, which is used for environmental perception of the cold chain warehouse.

[0044] Located at the decision-making layer of the cloud data center, it is constructed by a path planning algorithm. The decision-making layer, based on the sensor data from the perception layer, combines AI intelligent algorithms to achieve local obstacle avoidance and combines the shelf topology semantics to output the obstacle avoidance and inventory path plans for the drones.

[0045] Located at the digital twin layer of the cloud data center, the digital twin layer constructs a digital twin model of the target cold chain warehouse based on digital twin technology. The digital twin layer is used to, based on the data from the perception layer and the decision-making layer, a 3D dynamic map based on point cloud registration, and fuse the RFID shelf coding and temperature field data to achieve real-time reconstruction of the digital twin model, which can display the specific situation of the target cold chain warehouse in real time, facilitating monitoring by the back-end management personnel.

[0046] The execution layer is used to receive the inventory path plan sent by the path planning algorithm, operate the drones to complete the path inventory operation, and send the information collected in real time to the digital twin layer to achieve real-time data update and reconstruction.

[0047] This application innovatively integrates a variety of advanced technologies, optimizing and improving the existing cold chain warehousing management system from multiple aspects such as environmental perception, environmental monitoring, and operation planning, and is expected to significantly improve the overall level of cold chain warehousing management.

[0048] Figure 1 The overall method flow chart in this embodiment is shown. It can be seen from the figure that this embodiment includes the following steps:

[0049] Step 1: Arrange various types of sensors in the target cold chain warehouse to sense the environmental changes in the cold chain warehouse. Integrate various types of sensors into a unified sensor cluster through an edge gateway, and preprocess the data collected by the sensors in the edge gateway, and send the processed data to the cloud database.

[0050] Specifically, in this embodiment, various types of sensors include temperature sensors, humidity sensors, 3D lidar sensors, multi-spectral ToF sensors, rgbd camera sensors, long-wave infrared thermal imagers, millimeter-wave radar arrays, and UWB / RFID positioning tags.

[0051] Among them, the temperature sensors, humidity sensors, multi-spectral ToF sensors, and rgbd camera sensors are installed at certain intervals on the shelves of the cold chain warehouse. The temperature sensors and humidity sensors are used to detect the changes in temperature and humidity in the cold chain warehouse. The multi-spectral ToF sensors are used for accurate positioning of the shelf spacing. The rgbd camera sensors are used for near-field object recognition and OCR parsing of shelf coding.

[0052] The 3D lidar sensor, millimeter-wave radar array, and long-wave infrared thermal imager are installed on the beam of the cold chain warehouse. The 3D lidar sensor is used to detect major obstacles (non-moving or slowly moving obstacles), and the millimeter-wave radar array is used for dynamic obstacle tracking. The long-wave infrared thermal imager is used for temperature gradient field modeling and can also be used to detect low-temperature risks such as hidden icicles.

[0053] It should be noted that considering the importance of the temperature in the cold chain warehouse, in this embodiment, special processing can be carried out on the temperature sensor during the layout, and the data collected by other sensors can be preprocessed with the data of the temperature sensor as the core through the edge gateway. Specifically, the preprocessing of the data collected by the sensors in the edge gateway can include:

[0054] By deploying multiple ZigBee wireless temperature measurement nodes on each column of shelves, each ZigBee wireless temperature measurement node is a temperature field grid, and all ZigBee wireless temperature measurement nodes together constitute a temperature field grid covering the entire cold chain warehouse. The specific size of this grid can be set according to the actual situation. For better illustration of the solution of this embodiment, a grid layout of 1m×1m×1m is adopted.

[0055] In the edge gateway, a perception-level dynamic weight allocation strategy based on temperature dominance is configured. This strategy calculates the temperature gradient and rate of change of temperature based on the temperature field grid data collected by the ZigBee wireless temperature measurement nodes, and realizes dynamic weight adjustment by constructing a temperature-sensitive confidence evaluation system. The dynamic weight allocation strategy takes the temperature data in the temperature field collected by the ZigBee wireless temperature measurement nodes as the basic data for calculation, and adjusts the weights of different sensors based on the dynamic changes of the temperature field, so as to achieve more efficient and accurate environmental perception. Its core is to dynamically adjust the trust and weight of different types of sensors according to the temperature change situation (such as the gradient and rate of change of the temperature field), so as to optimize the data fusion effect of the sensor cluster.

[0056] Specifically, the dynamic weight allocation strategy includes the following contents:

[0057] 1) According to the collected temperature data, calculate the temperature gradient and rate of change of temperature, so as to evaluate the dynamic change situation of the temperature field.

[0058] Specifically, the temperature gradient represents the direction and rate of temperature change, and it is the rate of change of temperature in space. For a temperature field T(x, y, z) in a three-dimensional space, its temperature gradient can be expressed as:

[0059] ,

[0060] Among them, is the Laplace operator, which is used to represent that the temperature gradient is the rate of change of the temperature field, indicating the direction and rate of temperature change; is the temperature field in three-dimensional space; is the rate of change of temperature in the x direction, is the rate of change of temperature in the y direction, is the rate of change of temperature in the z direction.

[0061] It should be noted that according to different measurement devices and data sources, the calculation methods of the temperature gradient are different. For example, the discrete difference method, the gradient estimation method or the weighted average method can be used to calculate the temperature gradient.

[0062] In this embodiment, the temperature data collected by the ZigBee wireless temperature measurement node is usually discrete data. In this case, the temperature gradient can be approximately calculated by the finite difference method.

[0063] For example, assuming that we have the values of the temperature field T(x, y, z) at discrete points in space, then the temperature gradient can be calculated by the following formula:

[0064] The rate of change in the x direction is:

[0065] ,

[0066] The rate of change in the y direction is:

[0067] ,

[0068] The rate of change in the z direction is:

[0069] ,

[0070] where , and are the distances between adjacent measurement points in the space of different coordinates in the space coordinate system, respectively.

[0071] It should be noted that: the temperature gradient describes the rate and direction of the temperature change in the cold chain warehouse in space. In practical applications, according to different scenarios and sensor layout methods, different methods can be used for the temperature gradient. The finite difference method disclosed in this embodiment is only for explanation and cannot be considered as a limitation to the invention in the present invention.

[0072] 2) According to the calculated temperature gradient and rate of change, calculate the confidence of each area, providing a basis for dynamically adjusting the sensor weights. Specifically, the confidence can be calculated by the following formula:

[0073] ,

[0074] Among them, is the confidence level, is the weight coefficient used to adjust the contribution of the temperature gradient to the confidence level, is the modulus of the temperature gradient, is the weight coefficient used to adjust the contribution of the temperature change rate to the confidence level, is the absolute value of the temperature change rate.

[0075] 3) At the same time, judge whether the current temperature change is drastic according to the temperature gradient and the change rate. When the temperature change is drastic, enter the stage of adjusting the sensor weights; if the temperature change is not drastic, keep the current weights unchanged.

[0076] 4) In the stage of adjusting the sensor weights, the weights of the sensors are dynamically adjusted according to the confidence levels calculated for each region.

[0077] First, the basic weight of the sensor is calculated by the following formula:

[0078] ,

[0079] Among them, is the basic weight of the i-th sensor of the same type, is the exponential function; is the priority coefficient of the i-th sensor, which is used to reflect the importance of different sensors in the system. This coefficient can be adjusted according to the actual situation and is obtained based on the subjective evaluation of the priority of this type of sensor in the target cold chain warehouse; is the influence coefficient of temperature on the performance of sensor i. N is the total number of all sensors of this type, is the overall priority coefficient of this type of sensor, is the influence coefficient of this type of sensor. j represents the sensors of the same class, and i represents the i-th sensor among the sensors of the same class.

[0080] It should be noted that the above formula starts from the idea of weighted average and dynamically adjusts the weights of each sensor through the exponential function and normalization. The denominator part is the exponential weighted sum of all sensors of the same type for normalization processing. In this formula, the exponential function (exp) is used to ensure that the weights are always non-negative. Since the exponential function grows rapidly with the increase of the input value, this makes the weight distribution very sensitive to the differences in the input values, so as to highlight those sensors with better performance. In the formula, is used to amplify the differences in the influence of temperature on sensor performance. is used to represent the influence coefficient of temperature on the performance of the i-th type of sensor. This coefficient reflects the performance changes of the sensor under different temperature conditions. When constructing When considering, the temperature gradient and the rate of temperature change need to be taken into account. The temperature gradient is used in the construction to characterize the influence of the rate of temperature change in space on this type of sensor. In places with a large temperature gradient, a certain type of sensor may require higher response capabilities. The rate of temperature change is used in the construction to characterize the influence of the rate of temperature change over time on this type of sensor. In places with a fast rate of change, the performance of a certain type of sensor may be more affected. For example: A possible construction method is: , where a i and b i are coefficients related to the performance of this type of sensor affected by the temperature gradient and the rate of temperature change.

[0081] After obtaining the basic weight of the sensor through the above formula, the confidence level is combined with the basic weight to obtain the time-varying weight, thereby realizing the dynamic adjustment of the weight of each sensor. The time-varying weight is calculated through the following formula:

[0082] ,

[0083] where, is the time-varying weight of sensor i; is the confidence level of sensor i at the current moment, indicating the data credibility of this sensor in an area or environment with large temperature changes.

[0084] It should be noted that through the time-varying weight, the contribution of the data collected by the sensor can be dynamically adjusted according to the real-time temperature change situation. With a higher confidence level, the weight of the corresponding sensor increases; with a lower confidence level, the weight of the corresponding sensor decreases. Since the weights of different sensors are dynamically adjusted, it can better reflect the changes in the current environment and improve the accuracy of the overall sensor data fusion. The sensor data with a high confidence level is given a higher weight, enhancing the monitoring effect on key areas. In areas with gentle temperature changes, the weight of the sensor is low, which can effectively reduce the influence of data errors caused by sensor noise on the entire system, thereby optimizing the data fusion effect. That is to say, the time-varying weight realizes the real-time response to the temperature change situation by dynamically adjusting the contribution of the sensor, and optimizes the data fusion effect of the sensor cluster. It enables the perception layer composed of multiple sensors to more flexibly and accurately reflect and respond to environmental changes, thereby improving the performance and reliability of the overall system.

[0085] Step 2: Build a digital twin model of the target cold chain warehouse and an intelligent inventory path planning algorithm in the cloud database.

[0086] The cloud database receives the sensor data sent by the edge gateway, constructs a sensor database based on the received data, and maps the data in the sensor database to the digital twin model, enabling the back-end management personnel to intuitively monitor the environmental changes in the cold chain warehouse.

[0087] The intelligent inventory path planning algorithm outputs a UAV obstacle avoidance-inventory path plan based on the sensor data.

[0088] Specifically, in this embodiment, the intelligent inventory path planning algorithm is a multi-level and dynamic path planning algorithm based on the A* algorithm, aiming to ensure that the UAV can efficiently and safely perform the inventory task in the complex cold chain warehouse environment.

[0089] By integrating information from different sensors, a comprehensive perception of the environment is provided. In particular, the temperature and humidity changes in the warehouse are monitored in real time through the temperature and humidity data of the sensors, ensuring that the environmental requirements of the cold chain warehouse are considered during path planning. And a preliminary path is generated through global path planning and the path is locally optimized. Through dynamic path adjustment and spiral maneuvering, it is ensured that the UAV can adapt to dynamic environmental changes and avoid colliding with abnormal areas. The specific contents are as follows:

[0090] 1) First, the environment of the cold chain warehouse is discretized, and the environment of the cold chain warehouse is transformed into a discretized model composed of a temperature-space voxel. A voxel is a pixel unit in three-dimensional space, which represents a small cube in space. In this model, each voxel contains information related to the warehouse environment, especially the temperature and humidity sensor data. In this embodiment, the voxel can be expressed as:

[0091] ,

[0092] Among them, is the identifier of the voxel, and a, b, and c respectively represent the indices of the voxel in the x, y, and z dimensions of the space coordinate system. Through these indices, the position of each voxel in the cold chain warehouse can be identified. Each voxel represents a discrete spatial area. In this embodiment, a 1m resolution is used to divide the entire space. A 1m resolution means that the side length of each voxel is 1 meter, or each voxel occupies a cube space of 1 meter × 1 meter × 1 meter. x, y, and z are spatial coordinates, representing the position of the voxel in three-dimensional space, and T is the temperature value in the voxel, with the unit of degree Celsius (°C). One of the core requirements of the cold chain warehouse is temperature control. Therefore, each voxel has a temperature value to represent the temperature at that spatial position. The temperature data is also very important for the UAV inventory path planning because the UAV needs to focus on inspecting areas with extreme temperature changes to ensure the safety of the goods. is the confidence level of the sensor data contained in the voxel (i.e., the data obtained from the confidence level calculation in step 1).

[0093] 2) Based on the discretization model, according to the relevant information provided in the discretization model, the path of the UAV is optimized through a multi-constraint optimization objective function.

[0094] Specifically, the multi-constraint optimization objective function includes:

[0095] ,

[0096] where π is the optimized path, is the temperature exposure term, is the time cost term, is the sensor risk term.

[0097] It should be noted that in this embodiment, the temperature exposure term can be expressed by the following formula:

[0098] ,

[0099] where k is the exponential decay factor, is the temperature value at a certain position s on the path, T ref is the reference temperature, representing the ideal temperature state of the cold chain warehouse. The degree of temperature deviation from the reference temperature on the path will affect the size of the penalty term.

[0100] It should be noted that in this formula is the exponential term, which is used to indicate that when the temperature deviates from the reference temperature, it means that the temperature exposure on the path will increase, and this place needs to be carefully inspected to ensure the safety of the cold chain goods. is the temperature gradient term. Areas with large temperature changes usually cause problems for the goods in the cold storage, so the UAV should try to pass through these areas as little as possible.

[0101] The time cost term can be expressed by the following formula:

[0102] ,

[0103] where the total path length refers to the total distance of the UAV path planning, that is, the sum of the spatial distances between all s points on the path. The average cruising speed of the UAV represents the average flight speed of the UAV when performing tasks.

[0104] The sensor risk term can be expressed by the following formula:

[0105] ,

[0106] where represents the sensor confidence at a certain position s in the path.

[0107] 3) Use a multi-constraint optimization objective function to replace the cost function of the A* algorithm. By redefining the cost function f(n) of the A* algorithm, it can comprehensively consider temperature exposure, time cost, and sensor risk. Then, use the A* algorithm to find an optimal path that can not only ensure the safety of the goods but also efficiently complete the task.

[0108] ,

[0109] Among them, is the new cost function, is the actual cost from the starting point to the current node n, which can include the flight distance of the drone, time cost, etc. For the A* algorithm, it can usually be represented by the length of the flight path. h optimized (n) is the new heuristic function, which is used to estimate the cost from the current node to the target node. This heuristic function not only considers the spatial distance but also comprehensively considers multiple constraints such as temperature exposure, time cost, and sensor risk.

[0110] Specifically, in this embodiment, the heuristic function needs to comprehensively consider:

[0111] Temperature exposure term: The time that may be exposed to different temperature regions on the path. To avoid high or low temperature regions, the potential risk of temperature exposure on the path can be considered in the heuristic function. For example, the temperature gradient on the possible path can be weighted to avoid regions with abnormal temperatures.

[0112] Time cost term: This part can be calculated by estimating the flight time from the current node to the target node. If there is a long flight time on the path, the value of the heuristic function will increase.

[0113] Sensor risk term: The risk of the sensor (such as unreliable data or failure) can affect the path selection, so the sensor confidence needs to be considered. If the quality of the sensor data in some regions is poor, the path selection needs to be adjusted by increasing the sensor risk.

[0114] To sum up, in this embodiment, the form of the heuristic function can be:

[0115] ,

[0116] Among them, , , are weight coefficients, which are used to balance the importance of different factors; d(n) is the spatial distance from node n to the target node; is the estimated temperature exposure at node n, is the estimated sensor risk at node n.

[0117] The path search process of the A* algorithm after replacing the cost function of the A* algorithm with a multi-constraint optimization objective function is as follows:

[0118] Initialize the open list and the closed list. For the starting point, = 0, and the heuristic function is initialized based on the estimation of the target path.

[0119] Starting from the current node, expand its neighboring nodes. For each neighboring node, calculate:

[0120] , and select the node with the minimum cost for expansion based on the cost.

[0121] When expanding the node, check whether the path encounters obstacles (such as shelves, walls, etc.) or whether there are temperature anomalies in certain areas.

[0122] Continuously iterate to expand the path until the target node is found. Each time of expansion, consider the cost of the current path , and select the path with the minimum cost for expansion.

[0123] The A* algorithm will terminate when the target node is found and return an optimal path from the starting point to the target node. In this optimal path, the cost of each edge optimizes the temperature exposure, time cost, and sensor risk as much as possible.

[0124] Step 3: The cloud simultaneously sends the path plan output by the intelligent inventory path planning algorithm to the drone terminal and the digital twin model.

[0125] Step 4: The drone terminal completes the inventory task according to the path plan, while the digital twin model updates the digital twin model in real time according to the path plan and the data feedback from the drone terminal.

[0126] The specific implementation manners described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above is only the specific implementation manners of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An AI - based cold - chain warehouse drone obstacle avoidance and path planning inventory method, characterized in that, The path planning inventory method comprises: S100: Arrange different types of sensors in the cold chain warehouse for sensing environmental changes in the cold chain warehouse. The edge gateway integrates different types of sensors into a unified sensor cluster, pre-processes the collected data, and sends the processed data to the cloud database; S200, builds a digital twin model of the target cold chain warehouse in the cloud database, and an intelligent inventory path planning algorithm, The cloud database builds a sensor database based on the received data and maps the data in the sensor database to the digital twin model. The intelligent inventory path planning algorithm outputs the obstacle avoidance-inventory path plan for the drone based on the sensor data; S300, the cloud sends the path plan output by the intelligent inventory path planning algorithm to the drone terminal and the digital twin model at the same time; S400 and the drone terminal complete the inventory task according to the path plan, while the digital twin model updates the digital twin model in real time based on the path plan and the data fed back by the drone terminal; The pre-processing comprises: S110, deploying multiple ZigBee wireless temperature measurement nodes on each row of shelves, each ZigBee wireless temperature measurement node being a temperature field grid; S120, the perception level dynamic weight allocation strategy in the edge gateway builds a temperature-sensitive confidence evaluation system according to the number of temperature field grids collected by the ZigBee wireless temperature measurement node to achieve dynamic weight adjustment; The intelligent inventory path planning algorithm is constructed based on the A* algorithm and includes the following steps: S210, converting the environment of the cold chain warehouse into a discretized model consisting of temperature-space joint voxels, wherein the temperature-space joint voxels are represented as: ， Among them, is the identifier of the voxel. a, b, and c respectively represent the indices of the voxel in the three dimensions x, y, and z in the spatial coordinate system. x, y, and z are spatial coordinates, indicating the position of the voxel in three-dimensional space. T is the temperature value in the voxel, is the gradient of the temperature value in the voxel, is the confidence level of the sensor data contained in the voxel; S220. Optimize the path of the UAV according to the relevant information provided in the discretization model and in combination with a multi-constraint optimization objective function, wherein the multi-constraint optimization objective function includes: ， where π is the optimized path, is the temperature exposure term, is the time cost term, is the sensor risk term; is the weight coefficient of the time cost term, is the weight coefficient of the sensor risk term, and these two weight coefficients are used to control the influence degree of each constraint on the optimized path; S230, replacing the cost function of the A* algorithm with a multi-constraint optimization objective function, redefining the cost function f (n) of the A* algorithm so that it can comprehensively consider temperature exposure, time cost and sensor risk, and finally generating an optimal path through the A* algorithm.

2. The method for obstacle avoidance and path planning inventory of an AI-intelligent cold chain warehouse drone according to claim 1, wherein The different types of sensors include temperature sensors, humidity sensors, 3D lidar sensors, multispectral ToF sensors, RGB-D camera sensors, long-wave infrared thermal imager sensors, millimeter-wave radar array sensors, and UWB / RFID positioning tags.

3. The method for obstacle avoidance and path planning inventory of drones in a cold chain warehouse based on AI intelligence according to claim 1, characterized in that, The dynamic weight allocation strategy includes the following steps: S121. Calculate the temperature gradient and the temperature change rate according to the collected temperature data to evaluate the dynamic change of the temperature field. The temperature gradient is expressed by the following formula: ， wherein, is the Laplace operator, is the temperature field in three-dimensional space; is the rate of change of temperature in the x direction, is the rate of change of temperature in the y direction, is the rate of change of temperature in the z direction; S123. Calculate the confidence of each area according to the calculated temperature gradient and change rate. The confidence is calculated by the following formula: ， Among them, is the confidence level, is the weight coefficient used to adjust the contribution of the temperature gradient to the confidence level, is the modulus of the temperature gradient, is the weight coefficient used to adjust the contribution of the temperature change rate to the confidence level, is the absolute value of the temperature change rate; S124: Determine whether the current temperature change is drastic based on the temperature gradient and the rate of change. If the temperature change is drastic, enter the sensor weight adjustment stage. If the temperature change is not drastic, keep the current weight unchanged; S125. During the stage of adjusting the sensor weights, the weights of the sensors are dynamically adjusted according to the confidence of each area. The weights are time-varying weights and are calculated by the following formula: ， Among them, is the time-varying weight of sensor i; is the confidence of sensor i at the current moment, is the basic weight of the i-th sensor of the same type.

4. The method for inventory taking of obstacle avoidance and path planning of drones in a cold chain warehouse based on AI intelligence according to claim 3, wherein The basic weights are calculated by the following formula: ， Among them, is the basic weight of the i-th sensor of the same type, is an exponential function; is the priority coefficient of the i-th sensor; is the influence coefficient of temperature on the performance of sensor i; N is the total number of all sensors of the same type, is the overall priority coefficient of sensors of the same type, is the influence coefficient of sensors of the same type, j represents sensors of the same type, and i represents the i-th sensor among sensors of the same type.

5. The inventory method for obstacle avoidance and path planning of UAVs in a cold chain warehouse based on AI intelligence according to claim 1, characterized in that The steps of replacing the cost function of the A* algorithm with the multi-constraint optimization objective function are as follows: S231. Initialize the open list and the closed list. For the starting point, = 0, and the heuristic function is initialized based on the estimate of the target path; S232. Starting from the current node, expand its neighboring nodes. For each neighboring node, calculate: ， Among them, is the new cost function, is the actual cost from the starting point to the current node n, and h optimized (n) is the new heuristic function, Select the node with the minimum cost for expansion according to the cost. When expanding the node, check whether the path encounters an obstacle or whether there is a temperature anomaly in the area; S233. Continuously iterate and expand the path until the target node is found. When expanding each time, consider the cost of the current path and select the path with the minimum cost for expansion.

6. The method for obstacle avoidance and path planning inventory of drones in a cold chain warehouse based on AI intelligence according to claim 5, wherein, The new heuristic function is represented by the following formula: ， Among them, , , are weight coefficients used to balance the importance of different factors; d(n) is the spatial distance from node n to the target node; is the estimated temperature exposure at node n, is the estimated sensor risk at node n.

7. A cold chain warehouse UAV obstacle avoidance and path planning inventory system based on AI intelligence, characterized in that, The AI intelligence-based cold chain warehouse UAV obstacle avoidance and path planning inventory system includes: A processor; A memory storing a computer program, which when executed by the processor, implements the AI intelligence-based cold chain warehouse UAV obstacle avoidance and path planning inventory method according to any one of claims 1 to 6.

Images