A cold chain warehouse robot path planning method based on AI processing

Abstract

The present invention discloses a cold chain warehouse robot path planning method based on AI processing, which belongs to the field of path planning technology and includes the following steps: S1, rasterizing the cold chain warehouse map and determining the obstacle density of each grid at the historical timestamp; S2, using random forest to determine the predicted obstacle density of each grid according to the obstacle density of each grid at the historical timestamp; S3, generating a robot's transport path for the goods to be transported according to the predicted obstacle density of each grid and the obstacle situation at the current moment. The transport path is automatically generated based on dynamic driving parameters, which realizes the automation of path planning, and considers the energy consumption cost from the current grid to the adjacent grid in path planning, which helps to select a path with lower energy consumption.

Description

Technical Field

[0001] The present invention belongs to the technical field of path planning, and particularly relates to a path planning method for a cold chain warehouse robot based on AI processing. Background Art

[0002] In the modern cold chain logistics field, the operation efficiency of the warehouse and the freshness preservation of goods are key factors in the competitiveness of enterprises. With the rapid development of artificial intelligence (AI) technology, its application in cold chain warehouse management has become an important means to improve efficiency and reduce losses. In a cold chain warehouse, due to the strict requirements for temperature control of stored items, the dynamic changes in the warehouse environment (such as door opening and closing, personnel movement, and equipment failures), and the frequent inbound and outbound of goods, the distribution of obstacles inside the warehouse is constantly changing, which poses a severe challenge to the path planning of warehouse robots.

[0003] Traditional path planning methods often rely on static maps or simple dynamic obstacle detection, and it is difficult to accurately predict and adapt to the complex and changeable obstacle situations in cold chain warehouses, resulting in low path planning efficiency and even possible damage to goods or robot failures due to collisions. Therefore, it is particularly important to develop an intelligent path planning method that can comprehensively consider historical obstacle data, real-time obstacle detection, and predict future obstacle distributions. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a path planning method for a cold chain warehouse robot based on AI processing.

[0005] The technical solution of the present invention is: a path planning method for a cold chain warehouse robot based on AI processing includes the following steps:

[0006] S1. Gridify the cold chain warehouse map and determine the obstacle density of each grid at historical timestamps;

[0007] S2. According to the obstacle density of each grid at historical timestamps, use a random forest to determine the predicted obstacle density of each grid;

[0008] S3. Generate a handling path for the robot for the goods to be handled according to the predicted obstacle density of each grid and the current obstacle situation.

[0009] Further, in S1, the obstacle density of the grid from time m to time n on the k-th day is calculated by the formula: ; where represents the duration during which there are obstacles in the grid from time m to time n on the k-th day, represents the total duration from time m to time n on the k-th day.

[0010] Further, S2 includes the following sub-steps:

[0011] S21. Construct a decision tree for each day at the historical timestamp according to the obstacle density of each grid at the historical timestamp.

[0012] S22. Combine the decision trees of all days at the historical timestamp to generate a random forest.

[0013] S23. Use the random forest to determine the predicted obstacle density of each grid in the cold chain warehouse.

[0014] The beneficial effects of the above further solution are as follows: In the present invention, constructing a decision tree for each day can finely capture the change law of the grid obstacle density every day, making the model closer to the dynamic changes of the actual warehouse environment. The decision tree model can prevent overfitting through pruning algorithms and improve the generalization ability of the model. In the cold chain warehouse environment, this helps to reduce the model deviation caused by abnormal data or special events. The random forest improves the prediction accuracy and stability by integrating the results of multiple decision trees, can more accurately predict the future obstacle density of the grid, and the random forest can handle high-dimensional data without feature selection.

[0015] Further, S21 includes the following sub-steps:

[0016] S211. Take the obstacle density of all grids on the k-th day at the historical timestamp as the root node.

[0017] S212. Calculate several partitioning features according to the obstacle density of each grid on the k-th day.

[0018] S213. Divide the obstacle density of all grids on the k-th day into several child nodes according to several partitioning features.

[0019] S214. Take the child nodes as the leaf nodes of the decision tree and use the root node to generate a decision tree for the k-th day.

[0020] The beneficial effects of the above further solution are as follows: In the present invention, by calculating multiple partitioning features, the distribution characteristics of the grid obstacle density can be more comprehensively described, the expression ability and accuracy of the decision tree can be improved, and the obtained multiple child nodes can more accurately describe the obstacle density differences in different regions. Taking the child nodes as the leaf nodes ensures that the decision tree covers the obstacle density of all grids, enabling the model to comprehensively reflect the dynamic changes of the warehouse environment. By generating independent decision trees for each day, it can be conveniently extended to the dataset of the entire historical timestamp to construct a more comprehensive random forest model, further improving the prediction ability and robustness.

[0021] Further, in S212, the calculation formula for the partitioning feature corresponding to the k-th day from the m-th moment to the n-th moment is as follows: ; where represents the obstacle density of the r-th grid from time m to time n on the k-th day, and R represents the number of grids in the cold storage warehouse map.

[0022] The duration of each day can be appropriately divided. For example, it can be divided into 6 time periods, then the time period from time m to time n can be expressed as 0:00 - 3:00 (referring to 3:59), 4:00 - 7:00, 8:00 - 11:00, 12:00 - 15:00, 16:00 - 19:00, and 20:00 - 23:00 (referring to 23:59), and the division characteristics of each time period can be obtained.

[0023] Furthermore, in S213, all the division characteristics on the k-th day are sorted from large to small, and every two adjacent division characteristics are used as the upper and lower limits of the division interval, and all the obstacle densities belonging to the same division interval are used as a sub-node.

[0024] Furthermore, S3 includes the following sub-steps:

[0025] S31. Calculate the dynamic driving parameters of each grid by using the predicted obstacle density of each grid and whether there is an obstacle currently;

[0026] S32. Generate a handling path for the cold chain warehouse robot according to the dynamic driving parameters of each grid.

[0027] Furthermore, in S31, the calculation formula for the dynamic driving parameter of the grid at the i-th row and j-th column is:

[0028] ; where represents the energy consumption cost from the grid at the i-th row and j-th column to the grid at the (i + 1)-th row and j-th column, represents the energy consumption cost from the grid at the i-th row and j-th column to the grid at the i-th row and (j + 1)-th column, represents the energy consumption cost from the grid at the i-th row and j-th column to the grid at the (i - 1)-th row and j-th column, represents the energy consumption cost from the grid at the i-th row and j-th column to the grid at the i-th row and (j - 1)-th column, γ i,j_now represents the weight corresponding to whether there is an obstacle currently at the grid at the i-th row and j-th column, c represents a constant, P i,j represents the predicted obstacle density of the grid at the i-th row and j-th column.

[0029] The beneficial effects of the above further solutions are as follows: In the present invention, the weights of different obstacles are different, and the magnitude relationship is: weight of immovable obstacle > weight of movable obstacle > weight of no obstacle, and the weights can be set artificially. The path planning method based on dynamic driving parameters can comprehensively consider the driving cost of the robot and the obstacle situation, generate a more efficient handling path, and thus reduce the operation cost.

[0030] Further, in S32, among the rows from the starting row of the goods to be transported to the ending row of the goods to be transported, grids with the minimum dynamic driving parameters in each row are determined, and all the determined grids are connected to generate a transportation path.

[0031] The beneficial effects of the present invention are as follows: The cold chain warehouse is rasterized, and a decision tree is constructed using the obstacle density data of each grid at historical timestamps to determine a random forest, effectively integrating historical information, improving the prediction accuracy of future obstacle density, enhancing the effectiveness and reliability of planning, and enabling the robot to avoid potential high-risk areas when planning a path; The transportation path is automatically generated based on dynamic driving parameters, realizing the automation of path planning, and considering the energy consumption cost from the current grid to the adjacent grid in path planning, which helps to select a path with lower energy consumption. Description of the Drawings

[0032] Figure 1 It is a flowchart of a path planning method for a cold chain warehouse robot based on AI processing. Detailed Embodiments

[0033] The embodiments of the present invention will be further described below with reference to the drawings.

[0034] As Figure 1 shown, the present invention provides a path planning method for a cold chain warehouse robot based on AI processing, including the following steps:

[0035] S1. Rasterize the cold chain warehouse map and determine the obstacle density of each grid at historical timestamps;

[0036] S2. According to the obstacle density of each grid at historical timestamps, use a random forest to determine the predicted obstacle density of each grid;

[0037] S3. Generate a transportation path for the goods to be transported by the robot according to the predicted obstacle density of each grid and the obstacle situation at the current moment.

[0038] In the embodiment of the present invention, in S1, the obstacle density of the grid from time m to time n on the k-th day is calculated by the formula: ; where represents the duration of the presence of obstacles in the grid from time m to time n on the k-th day, represents the total duration from time m to time n on the k-th day.

[0039] In the embodiment of the present invention, S2 includes the following sub-steps:

[0040] S21. According to the obstacle density of each grid at historical timestamps, construct a decision tree for each day of the historical timestamps;

[0041] S22. Combine the decision trees for all days of the historical timestamp to generate a random forest;

[0042] S23. Use the random forest to determine the predicted obstacle density of each grid in the cold chain warehouse.

[0043] In the present invention, by constructing a decision tree for each day, the changing pattern of the grid obstacle density for each day can be precisely captured, making the model closer to the dynamic changes of the actual warehouse environment. The decision tree model can prevent overfitting through pruning algorithms and improve the generalization ability of the model. In the cold chain warehouse environment, this helps to reduce the model bias caused by abnormal data or special events. The random forest improves the prediction accuracy and stability by integrating the results of multiple decision trees, can more accurately predict the obstacle density of future grids, and the random forest can handle high-dimensional data without the need for feature selection.

[0044] In the embodiment of the present invention, S21 includes the following sub-steps:

[0045] S211. Use the obstacle density of all grids on the k-th day in the historical timestamp as the root node;

[0046] S212. Calculate several partitioning features according to the obstacle density of each grid on the k-th day;

[0047] S213. Divide the obstacle density of all grids on the k-th day into several sub-nodes according to several partitioning features;

[0048] S214. Use the sub-nodes as the leaf nodes of the decision tree and generate a decision tree for the k-th day using the root node.

[0049] In the present invention, by calculating multiple partitioning features, the distribution characteristics of the grid obstacle density can be more comprehensively described, the expression ability and accuracy of the decision tree can be improved, the obtained multiple sub-nodes can more precisely describe the obstacle density differences in different regions, using the sub-nodes as leaf nodes ensures that the decision tree covers the obstacle density of all grids, enabling the model to comprehensively reflect the dynamic changes of the warehouse environment. By generating independent decision trees for each day, it can be conveniently extended to the dataset of the entire historical timestamp to construct a more comprehensive random forest model, further improving the prediction ability and robustness.

[0050] In the embodiment of the present invention, in S212, the calculation formula for the partitioning feature corresponding to the k-th day from time m to time n is: ; where

[0051] ; in the formula, represents the obstacle density of the r-th grid from time m to time n on the k-th day, and R represents the number of grids in the cold storage warehouse map.

[0052] The daily duration can be appropriately divided. For example, it can be divided into 6 time periods. Then, the time period from m to n can be expressed as 0:00 - 3:00 (referring to 3:59), 4:00 - 7:00, 8:00 - 11:00, 12:00 - 15:00, 16:00 - 19:00, and 20:00 - 23:00 (referring to 23:59), and the division characteristics of each time period are obtained.

[0053] In the embodiment of the present invention, in S213, all the division characteristics on the k-th day are sorted from large to small. Every two adjacent division characteristics are used as the upper and lower limits of the division interval, and all the obstacle densities belonging to the same division interval are used as a sub-node.

[0054] In the embodiment of the present invention, S3 includes the following sub-steps:

[0055] S31. Calculate the dynamic driving parameters of each grid by using the predicted obstacle density of each grid and whether there is an obstacle currently.

[0056] S32. Generate a handling path for the cold chain warehouse robot according to the dynamic driving parameters of each grid.

[0057] In the embodiment of the present invention, in S31, the calculation formula for the dynamic driving parameter of the grid at the i-th row and j-th column is:

[0058] ; In the formula, represents the energy consumption cost from the grid at the i-th row and j-th column to the grid at the (i + 1)-th row and j-th column, represents the energy consumption cost from the grid at the i-th row and j-th column to the grid at the i-th row and (j + 1)-th column, represents the energy consumption cost from the grid at the i-th row and j-th column to the grid at the (i - 1)-th row and j-th column, represents the energy consumption cost from the grid at the i-th row and j-th column to the grid at the i-th row and (j - 1)-th column, γ i,j_now represents the weight corresponding to whether there is an obstacle currently in the grid at the i-th row and j-th column, c represents a constant, P i,j represents the predicted obstacle density of the grid at the i-th row and j-th column.

[0059] In the present invention, the weights of different obstacles are different, and the magnitude relationship is: the weight of an immovable obstacle > the weight of a movable obstacle > the weight of no obstacle, and the weights can be set artificially. The path planning method based on the dynamic driving parameters can comprehensively consider the driving cost of the robot and the obstacle situation, generate a more efficient handling path, and thus reduce the operation cost.

[0060] Those of ordinary skill in the art will realize that the embodiments described herein are provided to assist the reader in understanding the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention based on these technical revelations disclosed in the present invention, and these deformations and combinations are still within the scope of protection of the present invention.

Claims

1. A cold chain warehouse robot path planning method based on AI processing, characterized in that: The following steps are involved: S1. Rasterize the cold chain warehouse map and determine the obstacle density of each grid at the historical timestamp; S2, based on the obstacle density of each grid at the historical timestamp, using random forest to determine the predicted obstacle density of each grid; S3, generating a robot transport path for the goods to be transported according to the predicted obstacle density of each grid and the obstacle situation at the current moment; In S1, the obstacle density of the grid from time m to time n on the kth day The calculation formula is: ; In the formula, It indicates the duration of time when there is an obstacle in the grid from time m to time n on the kth day. Represents the total duration from time m to time n on the kth day; The S3 comprises the following sub-steps: S31, calculating the dynamic driving parameters of each grid using the predicted obstacle density of each grid and whether there is an obstacle at present; S32, generating a transport path for the cold chain warehouse robot according to the dynamic driving parameters of each grid; In the above S31, the calculation formula of the dynamic driving parameter of the grid at the position of the i-th row and the j-th column is: ; In the formula, represents the energy cost of traveling from the grid of row i and column j to the grid of row i+1 and column j, represents the energy cost of traveling from the grid at row i and column j to the grid at row i and column j+1, represents the energy cost of traveling from the grid in row i and column j to the grid in row i-1 and column j, represents the energy cost of traveling from the grid in the i-th row and j-th column to the grid in the i-th row and j-1-th column, γ i,j_now Indicates the weight corresponding to whether there is an obstacle in the grid of row i and column j. c represents a constant. P i,j represents the predicted obstacle density of the grid in row i and column j; In S32, in the row from the starting point of the goods to be transported to the row from the end point of the goods to be transported, the grid with the minimum dynamic driving parameter in each row is determined, and all the determined grids are connected to generate a transport path.

2. The cold chain warehouse robot path planning method based on AI processing according to claim 1 is characterized in that: The S2 comprises the following sub-steps: S21, constructing a decision tree for each day of the historical timestamp according to the obstacle density of each grid at the historical timestamp; S22, combining the decision trees of all days of the historical timestamp to generate a random forest; S23. Use random forest to determine the predicted obstacle density of each grid in the cold chain warehouse.

3. The cold chain warehouse robot path planning method based on AI processing according to claim 2 is characterized in that: The S21 comprises the following sub-steps: S211, taking the obstacle density of all grids on the kth day in the historical timestamp as the root node; S212, calculating a number of division features according to the obstacle density of each grid on the kth day; S213, dividing the obstacle density of all grids on the kth day into a number of sub-nodes according to a number of division features; S214, using the child nodes as leaf nodes of the decision tree, and using the root node to generate a decision tree for the kth day.

4. The cold chain warehouse robot path planning method based on AI processing according to claim 3 is characterized in that: In S212, the corresponding division features from time m to time n on the kth day The calculation formula is: ; In the formula, It represents the obstacle density of the rth grid from time m to time n on the kth day, and R represents the number of grids in the cold storage warehouse map.

5. The cold chain warehouse robot path planning method based on AI processing according to claim 3 is characterized in that: In S213, all the partitioning features in the kth day are sorted from large to small, every two adjacent partitioning features are used as the upper and lower limits of the partitioning interval, and all obstacle densities belonging to the same partitioning interval are used as a child node.

Images