Path planning method and system and transportation controller

Abstract

The invention discloses a path planning method, a path planning system and a transportation controller. The path planning method comprises the following steps: S1, acquiring a starting point and an ending point of a path to be planned; s2, determining a plurality of passing lines from the starting point to a plurality of demarcation points in the neighbor area and first estimation time consumption of moving on each passing line through a neighbor area path planning model; s3, determining second estimation time consumption from each demarcation point to an end point through a trained far and adjacent region time estimation model; and S4, determining which passing line is selected as the optimal path according to the determined first estimation time consumption corresponding to each passing line and the second estimation time consumption from the demarcation point to the end point of the passing line. According to the method, low-efficiency blind searching in the far and adjacent areas when the Dijkstra algorithm is executed can be avoided, the path planning time and the calculation cost can be reduced under the condition that the path planning accuracy is ensured, the efficiency of the transportation controller is improved, the crown block can be guided to be away from the congestion area in time, and the carrying efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of automatic handling systems, in particular to a path planning method, system and transportation controller for an overhead crane handling system in semiconductor processing. Background Art

[0002] The overhead crane handling system (OHT system) is a key device for material handling in automated factories such as semiconductor manufacturing and high-end manufacturing.

[0003] In the overhead crane handling system, there are a large number of overhead cranes running on the tracks. Therefore, reasonably planning the movement paths of each overhead crane is crucial for improving the handling efficiency.

[0004] The path planning method disclosed in the patent document with the authorization announcement number CN116705679B simply considers the distance between nodes to determine the optimal path. This method ignores the influence of other overhead cranes on the passage efficiency of the optimal path on the track, resulting in poor passage efficiency of the optimal path and affecting the handling efficiency.

[0005] The patent document with the application publication number CN119692588A discloses another path planning method. In this method, the influence of other overhead cranes on the path is considered, and the accuracy is high. However, when performing path planning in this way, it is necessary to calculate the time taken for multiple paths, which requires a large amount of real-time calculation. This causes a greater calculation cost, long calculation time and response speed. At the same time, when the overhead crane handling system is actually running, there are a large number of simultaneous path planning requirements, which greatly increases the computing power requirements for the transportation controller. Summary of the Invention

[0006] The purpose of the present invention is to solve the above problems existing in the prior art, and provide a path planning method, system and transportation controller.

[0007] The purpose of the present invention is achieved through the following technical solutions: A path planning method, comprising the following steps: S1, obtaining the starting point and the ending point of the path to be planned; S2, determining multiple passage lines from the starting point to multiple demarcation points within the neighboring area and the first estimated time taken to move on each passage line through the neighboring area path planning model; S3, determining the second estimated time taken from each demarcation point to the ending point through the trained far neighboring area time estimation model; S4, determining which passage line is selected as the optimal path according to the first estimated time taken corresponding to each determined passage line and the second estimated time taken from the demarcation point of the passage line to the ending point.

[0008] Preferably, when exploring a passing route within the neighboring area through the neighboring area path planning model, if it is determined that the explored passing route includes the end point, the passing route with the shortest estimated time from the starting point to the end point is selected as the optimal path; If it is determined that the explored passing route does not include the end point, then determine multiple passing routes from the starting point to multiple demarcation points within the neighboring area and the first estimated time-consuming for moving on each passing route.

[0009] Preferably, the neighboring area path planning model adopts an improved Dijkstra algorithm, and the weight of the edge of the improved Dijkstra algorithm is determined based on the number of overhead cranes existing on the edge, whether the edge is on the branch controlled by the area control unit, the number of overhead cranes in the contactless power supply area to which the edge belongs, and the number of available loading and unloading machine platforms between the two nodes of the edge.

[0010] Preferably, A demarcation point within the neighboring area refers to the node closest to the overhead crane after the overhead crane moves forward a predetermined length from the starting point; Or a demarcation point within the neighboring area refers to the node where the overhead crane finally locates when moving forward a predetermined number of nodes; Or a demarcation point within the neighboring area refers to the grid cell where the overhead crane locates when moving forward a predetermined number of grid cells.

[0011] Preferably, the far neighboring area time estimation model is a deep neural network, and the deep neural network includes an input layer, an embedding layer, at least one hidden layer, and an output layer.

[0012] Preferably, the training data required for training the far neighboring area time estimation model is obtained by simulating through a simulation system.

[0013] Preferably, determine the total estimated time corresponding to each passing route according to the following formula, and select the passing route with the least total estimated time as the optimal path; ; where t0 is the total estimated time, t1 is the first estimated time-consuming corresponding to a passing route; is the discount factor; t2 is the second estimated time-consuming from the demarcation point of the passing route to the end point.

[0014] Preferably, when the overhead crane approaches or moves to the demarcation point of the selected passing route each time, use the demarcation point as the starting point and execute S2 - S4.

[0015] A path planning system, comprising: A starting point and end point acquisition unit, configured to acquire the starting point and the end point of the path to be planned; A first estimation unit for determining multiple access routes from the starting point to multiple demarcation points within the neighboring area and the first estimated time consumption for moving on each access route through a neighboring area path planning model; A second estimation unit for determining the second estimated time consumption from each demarcation point to the end point through a trained far - away area time estimation model; An optimal path determination unit for determining which access route to select as the optimal path according to the first estimated time consumption corresponding to each determined access route and the second estimated time consumption from the demarcation point of the access route to the end point.

[0016] A transport controller includes a memory and a processor. The memory stores a program executable by the processor. When the program is executed, it implements the path planning method described in any one of the above.

[0017] The advantages of the technical solution of the present invention are mainly reflected in: The path planning method of the present invention considers the driving path of the overhead crane from the perspective of the global traffic state. First, it uses the classic shortest path algorithm to determine the access routes and travel times in the neighboring area close to the overhead crane, and uses a deep neural network to estimate the future congestion state and time consumption in the far - away area far from the overhead crane and directly participates in the auxiliary decision - making for determining the optimal path in the neighboring area of the overhead crane. The effective prediction of the far - away area can avoid the inefficient and blind search in the far - away area when executing the Dijkstra algorithm. It can greatly reduce the path planning time of the transport controller and reduce the computational cost while ensuring the accuracy of path planning, improve the efficiency of the transport controller, reduce the intervention times of the transport controller to a certain extent, and can timely guide the overhead crane away from the congested area to improve the handling efficiency.

[0018] The present invention adopts an improved Dijkstra algorithm. When determining the weight of the edge, it is based on elements such as the number of overhead cranes existing on the edge, whether the edge is on the branch controlled by the area control unit, the number of overhead cranes in the contactless power supply area to which the edge belongs, and the number of available loading and unloading machine platforms at the edge, etc., which can effectively ensure the accurate time estimation in the neighboring area, thus providing reliable data support for accurately planning the optimal path.

[0019] The present invention uses a simulation system to obtain the data required for model training, which can simulate various road conditions that may occur during actual operation as much as possible, so that the far - away area time estimation model can fully learn various situations on the track for subsequent accurate estimation. Brief Description of the Drawings

[0020] Figure 1 is a process schematic diagram of the path planning method of the present invention; Figure 2It is a schematic diagram of the present invention for determining the neighboring area by moving forward a certain distance; Figure 3 It is a schematic diagram of the present invention for determining the neighboring area by moving forward a predetermined number of nodes; Figure 4 It is a schematic diagram of the present invention for determining the neighboring area by moving forward a predetermined number of grid cells. The area within the dashed box in the figure is the neighboring area; Figure 5 It is a schematic diagram of the states of the passing routes determined by the neighboring area path planning model of the present invention at different times. The circles with arrows in the figure represent the overhead cranes; Figure 6 It is a schematic diagram of a side having multiple overhead cranes in an example of the present invention. The circles with arrows in the figure represent the overhead cranes; Figure 7 It is a schematic structural diagram of the far - neighboring area time estimation model of the present invention; Figure 8 It is a schematic diagram of the first path planning by the method of the present invention; Figure 9 It is a schematic diagram of the second path planning by the method of the present invention; Figure 10a It is a schematic diagram of the optimal path determined by the first path planning of the method of the present invention in a grid - shaped track; Figure 10b It is a schematic diagram of the optimal path determined by the second path planning of the method of the present invention in a grid - shaped track; Figure 10c It is a schematic diagram of the optimal path determined by the third path planning of the method of the present invention in a grid - shaped track; Figure 10d It is a schematic diagram of the optimal path determined by the fourth path planning of the method of the present invention in a grid - shaped track; Figure 10e It is a schematic diagram of the complete path determined by multiple path plannings of the method of the present invention in a grid - shaped track. Detailed implementation manners

[0021] The objectives, advantages and features of the present invention will be illustrated and explained by the non - restrictive description of the following preferred embodiments. These embodiments are only typical examples of applying the technical solutions of the present invention. Any technical solutions formed by equivalent replacement or equivalent transformation fall within the scope of protection required by the present invention.

[0022] In the description of the solution, it should be noted that the orientation or positional relationship indicated by terms such as "center", "upper", "lower", "left", "right", "front", "rear", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of description and simplification, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation on the present invention. In addition, the terms "first", "second", "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance. Embodiment 1

[0023] The path planning method disclosed by the present invention will be described below with reference to the accompanying drawings. The path planning method can be used in a variety of scenarios. For example, it can be used for path planning when a crane needs to move from its current position to the position of a material to be transported; or it can be used for path planning when a crane grabs a material and then moves from its current position to the position where the material to be transported is to be moved; or it can be used for path planning when a crane moves from its current position to a specified position to avoid other cranes, etc.

[0024] Correspondingly, as shown in the attached Figure 1 drawings, the path planning method includes the following steps: S1. Obtain the starting point and the ending point of the path to be planned; S2. Determine multiple passing lines from the starting point to multiple demarcation points within the neighboring area and the first estimated time consumption for moving on each passing line through the neighboring area path planning model; S3. Determine the second estimated time consumption from each demarcation point to the ending point through the trained far - away area time estimation model; S4. Determine which passing line is selected as the optimal path according to the first estimated time consumption corresponding to each determined passing line and the second estimated time consumption from the demarcation point of the passing line to the ending point.

[0025] In S1, the starting point can be the current position of the crane in each of the above - mentioned scenarios, or the ending point of the optimal path within the neighboring area determined by the following one - time path planning. The ending point is the position where the crane is to move in each of the above - mentioned scenarios.

[0026] In S2, the near-neighboring area can be determined in different ways. For example, in an overhead crane transport system, the track may be an irregular track arranged according to the equipment layout, as shown in the patent documents with application publication numbers CN119671441A and CN119692588A, and the overhead crane travels unidirectionally on the track. Of course, in another embodiment, the track may also be a grid track, as shown in the patent documents with authorization announcement numbers CN112533813B and CN108698757B. There may be multiple routes in the near-neighboring area. The far-neighboring area is the track area between the dividing point and the end point.

[0027] When the track is an irregular track, the adjacent area can be understood as the track area between the starting point and a position at a certain distance from the starting point in the forward direction. The certain distance can be, for example, 10m, 15m, 20m, 50m, etc., which can be determined according to specific needs and is not limited here. Correspondingly, a dividing point in the adjacent area refers to the node that the overhead crane is closest to after the overhead crane moves forward a predetermined distance from the starting point. Of course, it can also be the position where the overhead crane moves to. The predetermined length is the above-mentioned certain distance.

[0028] For example, as shown in the attached Figure 2 As shown, with node 1 as the starting point, it can move forward in two directions, and after moving forward in two directions for a predetermined distance, there are nodes 2 and 3. From node 2, it can continue to move toward node 4 and node 5 in two directions, and from node 3, it can continue to move forward toward node 5 and node 6 in two directions. When the overhead crane moves from node 2 to node 4 to a certain position, and the moving distance between node 1 and the certain position is equal to the predetermined length, it is determined that the distance from the certain position to node 4 is less than the distance from it to node 2, that is, the certain position is closer to node 4, and the node 4 is determined to be a demarcation point in the adjacent area. Correspondingly, the adjacent area also includes two other demarcation points, namely node 5 and node 6.

[0029] Of course, when the track is an irregular track, the neighboring area can also be understood as the track area between the starting point and the nth node downstream of the starting point in the forward direction, and the nth node is, for example, the third node, the fourth node, etc., which is determined according to needs and is not limited here. Correspondingly, a demarcation point in the neighboring area refers to the node where the overhead travelling vehicle finally locates when it moves forward a predetermined number of nodes.

[0030] For example, as attached Figure 3As shown, taking node 1 as the starting point, there are two forward directions. The predetermined quantity is, for example, 2. Then, after the overhead crane moves in the first direction through node 2 and node 4, the final node 4 where it is located is a demarcation point within the adjacent region. Correspondingly, the adjacent region also includes another 2 demarcation points, namely node 5 and node 6.

[0031] When the track is a grid-shaped track, the adjacent region is the track region between the starting point and the grid cell where the overhead crane is located when it moves forward a predetermined number of grid cells from the starting point. Correspondingly, a demarcation point within the adjacent region refers to the grid cell where the overhead crane is located when it moves forward a predetermined number of grid cells.

[0032] As shown in the appendix Figure 4 As shown, when the overhead crane is in grid cell 11, it can move forward in three directions. Assuming the predetermined quantity is 4, the grid cells where the overhead crane is located when it moves 4 grid cells in the three directions are grid cell 33, grid cell 51, and grid cell 42 respectively. Correspondingly, the demarcation points within the adjacent region are grid cell 33, grid cell 51, and grid cell 42 respectively.

[0033] Moreover, define the above-mentioned predetermined length of forward movement, predetermined quantity of nodes, and predetermined quantity of grid cells as the hyperparameter K. The value of the hyperparameter K determines the ranges of the adjacent region and the far adjacent region, and plays a crucial role in the entire path planning.

[0034] If the value of the hyperparameter K is small, this means that the path planning model for the adjacent region only searches a very small range, and the computational cost is relatively small. However, the optimal passing route in the adjacent region will seriously depend on the second estimated time-consuming estimated by the time estimation model for the far adjacent region. Since the time estimation model for the far adjacent region cannot grasp the real-time dynamic changes of the entire track map (distribution map of the track), the smaller the value of the hyperparameter K, the longer the distance between the demarcation point and the end point in the far adjacent region, and the estimation of the time for a long distance will increase the uncertainty of the path planning. At the same time, a smaller hyperparameter K will also lead to frequent execution of the path planning for the adjacent region, increasing the frequency of path planning and also posing higher requirements on the response ability of the transportation controller (TCS).

[0035] On the contrary, if the value of the hyperparameter K is too large, the entire path planning will be more dependent on the path planning of the adjacent region. The larger the value of the hyperparameter K, the larger the range that the path planning model for the adjacent region needs to search when performing path planning, resulting in a greater computational cost. In addition, the larger the value of the hyperparameter K, the greater the change of the weight of the edge far from the starting point position over time, that is, the worse the reliability of the weight of the edge far from the starting point position.

[0036] As shown in the appendix Figure 5As shown in the figure, at time t1, a traffic line planned by the near-neighbor area path planning model is from node 1001, passing through node 1002, node 1017, node 1018, and node 1019. There is a vehicle on section 1 (1017->1018), and there is no vehicle on section 2 (1018->1019). Therefore, the edge weights of both section 1 (1017->1018) and section 2 (1018->1019) are relatively small.

[0037] At time t2, when the overhead crane travels from node 1001 to node 1002, at this time, there is an overhead crane performing unloading operations on section 1 (1017->1018) and it has not left. At the same time, there are two other overhead cranes moving to section 1 (1017->1018), and two overhead cranes moving to section 2 (1018->1019). At this time, the weights of the edges of both section 1 (1017->1018) and section 2 (1018->1019) increase. Then, the optimal path determined at time t1 may not be a better path at time t2. Therefore, the larger the value of the hyperparameter K, the greater the unreliability of the traffic line of the near-neighbor area path planning model.

[0038] Therefore, the hyperparameter K needs to take a suitable value. Therefore, in order to obtain a suitable hyperparameter K, in the simulation environment, the appropriate hyperparameter K can be determined by running the near-neighbor area path planning model and the far-neighbor area time estimation model with multiple different hyperparameters K for a trial run. The simulation environment, for example, generates handling instructions through the following simulation system, and inputs the starting point and ending point of the handling instructions into the near-neighbor area path planning model and the far-neighbor area time estimation model with different hyperparameters K for path planning, and simulates the overhead crane moving to the ending point according to the paths planned by the two models, and records the handling time taken by the overhead crane to move from the starting point to the ending point during the simulation process.

[0039] After each near-neighbor area path planning model corresponding to the hyperparameter K has undergone a trial run for a predetermined duration, the predetermined duration can be determined as needed and is not limited here. The average handling time and total throughput of all handling instructions for path planning using the near-neighbor area path planning model with each hyperparameter K can be calculated (that is, the total number of handling instructions executed by the near-neighbor area path planning model with a hyperparameter K within the above-mentioned predetermined duration) to measure the performance of each near-neighbor area path planning model corresponding to the hyperparameter K and the corresponding far-neighbor area time estimation model. Then, select the hyperparameter K with the optimal performance as the target value. Of course, the optimal hyperparameter K corresponding to different track maps is also different and needs to be determined through experiments in the simulation environment.

[0040] The adjacent area path planning model can explore each demarcation point in the adjacent area, the passing routes from the starting point to each demarcation point, and the first estimated time consumption for moving on each passing route through the known Dijkstra algorithm.

[0041] More preferably, the adjacent area path planning model adopts an improved Dijkstra algorithm. A main change in the improved Dijkstra algorithm is to improve the calculation method of the weight of each edge on the basis of the existing Dijkstra algorithm, that is, the weight of each edge is determined based on the number of overhead cranes existing on the edge, whether the edge is on the branch controlled by the ZCU (Zone Control Unit), the number of overhead cranes in the contactless power supply area to which the edge belongs, and the number of available loading and unloading machine platforms between the two nodes of the edge.

[0042] Specifically, an edge is the track between two nodes in the track map, and each node can be a confluence point, a bifurcation point or a set position point in the track.

[0043] When there is an overhead crane on the line corresponding to an edge, when determining the weight of this edge, it is necessary to consider the delay that the overhead cranes existing on this edge may cause to other overhead cranes that need to pass through this edge.

[0044] For example, as shown in the appendix Figure 6 If the section from node N5 to node N6 is defined as edge E56, and there are two overhead cranes on edge E56, then when calculating the weight of edge E56, the node distance D 56 or the passing time T of the overhead crane moving from node N5 to node N6 at the set vehicle speed 56 plus the additional delay T generated by each overhead crane existing on edge E56 to other overhead cranes that need to pass through this edge V That is, when calculating the weight of this edge E56, it is to use the node distance D 56 or the passing time T 56 plus 2T V .

[0045] Meanwhile, the maximum number of powered devices of a CPS (contactless power supply device) determines the total number of overhead cranes that can enter a contactless power supply area (the track area powered by a CPS power supply device). When the number of overhead cranes in a contactless power supply area reaches the upper limit warning value of this contactless power supply area, there is a risk of power supply shortage in this contactless power supply area. At this time, other overhead cranes that want to enter this contactless power supply area will be controlled to wait at the entrance of this contactless power supply area or make way. Therefore, during the path planning process, when determining the weight of each edge of this contactless power supply area, the node distance of the edge or the passing time corresponding to the edge needs to be added with the corresponding delay weight T cps 。

[0046] For example, if the number of overhead cranes in the contactless power supply area to which edge E56 in the above example belongs reaches the upper limit warning value of this contactless power supply area, then when determining the weight of edge E56, use the node distance D 56 or the passing time T 56 plus the delay weight T cps 。

[0047] Furthermore, the number of available unloading machine platforms at each edge will also greatly affect the passage of other overhead cranes. Therefore, when determining the weight of an edge, it is also necessary to determine according to the number N port of available unloading machine platforms between the two nodes corresponding to the edge and the additional delay T port caused by the overhead crane taking and unloading goods at each available unloading machine platform to other overhead cranes passing through. For example, there is one available unloading machine platform at edge E56 in the above example, that is, N port =1, then when determining the weight of edge E56, use the node distance D 56 or the passing time T 56 plus N port *T port 。

[0048] In addition, the ZCU is used to control the overhead crane to queue up in sequence and pass through the crossover track section of the track one by one. When a node of an edge belongs to an upstream end or a downstream end of a crossover track section controlled by a ZCU, that is, the edge is on the branch controlled by the ZCU, it is necessary to consider the additional delay T zcu generated when the ZCU controls the overhead crane to pass through the crossover track section one by one. That is, when determining the weight of an edge, when the edge is on a branch of a crossover track section managed by a ZCU, the weight of the edge is the node distance or passing time corresponding to the edge plus the T zcu 。

[0049] And, the T V 、T cps 、T port 、T zcuIt can be set to a fixed value respectively. Of course, it can also be obtained through analysis by other system analysis methods, which is not limited herein.

[0050] In S3, the far neighbor region time estimation model can adopt a known deep neural network DNN (Deep Neural Net), as shown in the appendix Figure 7 As shown, the deep neural network is, for example, a multi-layer perceptron, which includes an input layer, an embedding layer, at least one hidden layer, and an output layer; the specific functions and working mechanisms of the input layer, the embedding layer, the hidden layer, and the output layer are known technologies and not the innovation of the present invention, so they will not be elaborated here. And, the input layer contains two neurons, and the embedding layer and the hidden layer adopt a fully connected manner, that is, each neuron in the embedding layer is connected to each neuron in the hidden layer. The hidden layer is preferably at least two. When the track map is large or the number of nodes is large, more hidden layers and neurons can be considered to be set to learn more details. In the training stage, when there is underfitting, the number of hidden layers can be tried to be increased; when there is overfitting, the number of hidden layers can be reduced or Dropout can be introduced to enhance the generalization ability of the model. The hidden layers are connected in a fully connected manner, the hidden layer and the output layer are fully connected, and the output layer has only one neuron, representing the predicted time consumption between the demarcation point and the destination. And the rectified linear unit ReLU (Rectified Linear Unit) activation function is used in the model to calculate the output.

[0051] Furthermore, when the track information is particularly large, a convolutional layer can be added between the embedding layer and the hidden layer to achieve dimensionality reduction, thereby reducing the problem of particularly large computational complexity in the training process.

[0052] To facilitate the training of the model, the present invention proposes a simulation system to improve the prediction ability of the far neighbor region time estimation model. The simulation system consists of three parts, namely a transportation controller (TCS) holding an algorithm, an upper-level simulation generation handling instruction module, and a lower-level simulation overhead crane operation simulation system.

[0053] The upper-level simulation generation handling instruction module simulates the MCS (Material Control System) and is used to generate handling instructions and send them to the TCS. The handling instructions include a starting point, an ending point, a handling priority, etc., where the starting point and the ending point are randomly configured. Of course, the MCS can also be directly used to replace the upper-level simulation generation handling instruction module to send handling instructions to the TCS.

[0054] The TCS performs path planning according to the handling instructions according to the existing path planning method and instructs the overhead crane to perform handling.

[0055] The simulation system that simulates the operation of the lower-level gantry crane uses software to simulate the operation of the gantry crane on the track, which includes a track section and a gantry crane section. Among them, the track section is in accordance with the actual Figure 1 : 1. Simulated track map, which includes nodes, FOUP (Front Opening Unified Pod) access ports, ZCU, and CPS power supply equipment, etc. The gantry crane section simulates the actual operation logic of the gantry crane. For example, it simulates the gantry crane driving along the simulated track at a preset speed, simulates the gantry crane decelerating or even stopping when encountering traffic congestion ahead, simulates the gantry crane receiving traffic control from the ZCU at the intersection track section, and receiving handling instructions from the TCS, etc.

[0056] The data required for training the time estimation model of the far neighbor area is obtained through the simulation system. Specifically, during the simulation process, the starting point, ending point, road condition information before path planning, and the travel time calculated when the gantry crane reaches the ending point from the starting point of each handling instruction are recorded and used as a data sample. The road condition information includes edge information features and load information features. The edge information feature refers to the number of gantry cranes on each edge, which reflects the current congestion situation on each edge. The load information feature reflects the recent or future congestion situation related to the workload, and is specifically divided into three types: pending load (the number of handling tasks that have not been assigned to the target gantry crane), pick-up load (the number of handling tasks that the assigned gantry crane is on the way to pick up goods), and unloading load (the number of handling tasks that the assigned gantry crane has picked up the goods and is performing the unloading journey task to the designated destination).

[0057] After a period of simulation, sufficient data samples are obtained as training data, and the training data is sent to the time estimation model of the far neighbor area for training. The starting point, ending point, and road condition information in each training sample are sent to the input layer, and the travel time is used as the data in the output layer.

[0058] Through a period of training, the deep neural network can fully learn the relationship between various road condition information and travel time; when the time estimation model of the far neighbor area tends to be stable, it is considered that the model has converged, and the training process can be stopped. The specific training process of the time estimation model of the far neighbor area is a known technology and is not an innovation of the present invention, so it will not be elaborated here. When updating the module parameters during training, preferably choose Adam as the optimizer to achieve adaptive adjustment of the learning rate and prevent overfitting problems.

[0059] After the far - adjacent area time estimation model is stabilized, the travel time generated by the simulation system can be further used for testing the far - adjacent area time estimation model. For example, the travel time from the starting point to the ending point obtained by simulation can be compared with the estimated time between the starting point and the ending point estimated by the far - adjacent area time estimation model for testing. During the testing process, the structural parameters in the far - adjacent area time estimation model can still be adjusted. For example, the number of hidden layers and the number of neurons in each hidden layer can be adjusted, and the adjusted model is further trained. Finally, a DNN model with strong learning ability and stability is obtained as the far - adjacent area time estimation model.

[0060] The effectiveness of the far - adjacent area time estimation model is affected by whether various scenarios appear during training. For scenarios that do not appear during training, the decision - making in practice may be relatively poor. However, through the simulation system of the present invention, various situations encountered during the operation of the overhead crane can be rehearsed, so that the far - adjacent area time estimation model can learn the congestion levels under various road conditions, thereby improving the prediction accuracy of the far - adjacent area time estimation model.

[0061] Of course, it is also possible to collect the data required for training the far - adjacent area time estimation model from the data recorded in the stored historical handling diaries, which will not be elaborated here.

[0062] Moreover, in order to better adapt to the changes in the actual road conditions, it is necessary to continuously improve the far - adjacent area time estimation model so that it can learn the new rail transit states.

[0063] Therefore, the present invention adopts an incremental training method, that is, during the actual operation process, the model parameters are continuously updated with small - batch data in real - time operation. For example, at regular intervals or when the handling density is low, recent historical handling records are randomly selected to update the model parameters.

[0064] When determining the second estimated time consumption through the trained far - adjacent area time estimation model, the determined current road conditions information (edge information features, load information features) of each demarcation point, the ending point, and the path planning can be input into the trained far - adjacent area time estimation model, and the far - adjacent area time estimation model predicts the second estimated time consumption from each demarcation point to the ending point. Since the far - adjacent area time estimation model only needs to estimate time and does not need to perform path planning, it can greatly reduce unnecessary path exploration and calculation consumption.

[0065] In S4, the total estimated time consumption corresponding to each passing line is determined according to the following formula, and the passing line with the least total estimated time consumption is selected as the optimal path; ; where, t0 is the total estimated time consumption, and t1 is the first estimated time consumption corresponding to a passing line; is a discount factor. The discount factor is set because the time estimation model for the far neighbor region is inherently a prediction of future congestion in areas far from the starting point, which has a certain degree of uncertainty. Therefore, a discount factor needs to be added; t2 is the second estimated time consumption from the demarcation point of the traffic route to the end point.

[0066] Among them, the discount factor can be determined according to different situations. When the accuracy of the time estimation model for the far neighbor region is relatively high, it can be considered ≥1; when the time estimation model for the far neighbor region is in its initial stage or the prediction is not very stable, it can be considered . More preferably, the discount factor can be determined in the following way: after training the time estimation model for the far neighbor region, the historical data or the data obtained by simulating the simulation system is re - input into the determined near neighbor region path planning model and the time estimation model for the far neighbor region, and the discount factor is determined by means of grid search. For example, within the range of 0.5 to 1.5, with a step size of 0.1 for grid search, the specific process of grid search is a known technology and will not be elaborated here.

[0067] As shown in the appendix Figure 1 As shown, in actual path planning, there will be cases where the distance between the starting point or the demarcation point and the end point is very close. Especially after multiple path planning, the distance from the demarcation point (starting point) of the latest selected traffic route to the end point is already very close. Therefore, in S2, when exploring traffic routes in the near neighbor region through the near neighbor region path planning model, if it is determined that the explored traffic route includes the end point, the traffic route with the shortest estimated time consumption from the starting point to the end point is selected as the optimal path, and the path planning can be ended.

[0068] If it is determined that the explored traffic route does not include the end point, the multiple traffic routes from the starting point to multiple demarcation points in the near neighbor region and the first estimated time consumption for moving on each traffic route are determined, and S3 is executed.

[0069] As shown in the appendix Figure 1 As shown, after one - time path planning, the overhead crane moves according to the selected optimal path. When the overhead crane approaches or moves to the demarcation point of the selected optimal path each time, the demarcation point is used as the starting point, and the near neighbor region is determined again with the same hyperparameter K as above, and S2 - S4 are executed. That is, each time it approaches or moves to the demarcation point of the optimal path, path planning is carried out again with the demarcation point as the starting point, so as to continuously plan new optimal paths.

[0070] The following uses a specific example to illustrate how to determine the optimal path in each path planning: As shown in the appendix Figure 8As shown, a handling task starts from node 1 and ends at node 99. With the hyperparameter K set to move forward 20 meters to determine the neighboring area, each time path planning is performed, the node closest to the position where the overhead crane moves forward 20 meters from the starting point is used as the demarcation point. The demarcation points reached after 20 meters from node 1 are node 4, node 5, and node 6 respectively. The passing routes determined by the neighboring area path planning model, the first estimated time consumption corresponding to each passing route, the second estimated time consumption determined by the far - neighboring area time estimation model, and the total estimated time are shown in Table 1, where the discount factor a is taken as 1.

[0071] Table 1

[0072] As can be seen from Table 1, when node 1 is used as the starting point, a total of four passing routes are explored, namely passing route 1 (1→2→4), passing route 2 (1→2→5), passing route 3 (1→3→5), and passing route 4 (1→3→6). The first estimated time consumptions corresponding to the 4 passing routes calculated by the neighboring area path planning model are 7s, 8s, 9s, and 10s respectively. The second estimated time consumptions predicted by the far - neighboring area time estimation model for node 4, node 5, and node 6 to the end point 99 are 50s, 40s, and 60s respectively.

[0073] Then, for the first path planning, the total estimated time consumptions corresponding to the four passing routes in the neighboring area are as follows: the total estimated time consumption corresponding to passing route 1 is 57 seconds, the total estimated time consumption corresponding to passing route 2 is 48 seconds, the total estimated time consumption corresponding to passing route 3 is 49 seconds, and the total estimated time consumption corresponding to passing route 4 is 70 seconds. Therefore, when determining the optimal path for the first path planning, passing route 2 will be selected as the optimal path.

[0074] During the process of the overhead crane traveling along passing route 2, when the overhead crane is about to move to node 5 or has moved to node 5, the transportation controller needs to plan the next - stage passing route for the overhead crane again. As shown in the appendix Figure 9 As shown, at this time, starting from node 5 and moving forward 20 meters, node 9 and node 10 can be reached. At this time, node 9 and node 10 are the demarcation points of the newly determined neighboring area. The passing routes determined by the neighboring area path planning model, the first estimated time consumption of each passing route, the second estimated time consumption determined by the far - neighboring area time estimation model, and the total estimated time are shown in Table 2, where the discount factor a is taken as 1.

[0075] Table 2

[0076] As shown in Table 2, starting from node 5, two passing routes are explored, namely passing route 5 (5→7→9) and passing route 6 (5→8→10). The first estimated time taken for passing route 5 is calculated by the near-neighbor area path planning model to be 7 seconds. The second estimated time taken from node 9 to the end point 99 is determined by the far-neighbor area time estimation model to be 35 seconds, and the total estimated time taken is 42 seconds. Correspondingly, the first estimated time taken for passing route 6 is 10 seconds, the corresponding second estimated time taken is 33 seconds, and the corresponding total estimated time taken is 43 seconds. Then, when planning the path this time, passing route 5 is selected as the optimal path for the overhead crane.

[0077] When the overhead crane moves close to or reaches the end point (demarcation point) of the optimal path each time, repeat the above process of path planning until the end point.

[0078] As attached Figure 10a - 10e shown, in an example of an orbital map of a 9x9 grid cell, each grid cell is numbered according to its corresponding combination of abscissa and ordinate. The overhead crane needs to move from the starting position (grid cell 11 in the lower left corner) to the end position (grid cell 99 in the upper right corner), and the overhead crane advances four grid cells each time. When planning the path for the first time, the end point of the optimal path determined when planning the path from grid cell 11 is grid cell 33. When planning the path for the second time, the end point of the optimal path determined when continuing to plan the path with grid cell 33 as the starting point is grid cell 64. Similarly, when planning the path for the third time, the end point of the optimal path determined when continuing to plan the path with grid cell 64 as the starting point is grid cell 77. When making the fourth decision, the end point of the optimal path determined when continuing to plan the path with grid cell 77 as the starting point is grid cell 99, and then the path planning stops. Embodiment 2

[0079] This embodiment discloses a path planning system, including: A starting point and end point acquisition unit for acquiring the starting point and end point of the path to be planned; A first estimation unit for determining, through a near-neighbor area path planning model, multiple passing routes from the starting point to multiple demarcation points in the near-neighbor area and the first estimated time taken to move on each passing route; A second estimation unit for determining, through a trained far-neighbor area time estimation model, the second estimated time taken from each demarcation point to the end point; An optimal path determination unit for determining which passing route is selected as the optimal path according to the first estimated time taken corresponding to each determined passing route and the second estimated time taken from the demarcation point of the passing route to the end point. Embodiment 3

[0080] This embodiment discloses a transportation controller, including a memory and a processor. The memory stores a program that can be executed by the processor. When the program is executed, the path planning method described in any of the above is implemented.

[0081] There are still various embodiments of the present invention. All technical solutions formed by using equivalent transformations or equivalent substitutions fall within the protection scope of the present invention.

Claims

1. A path planning method, characterized in that: The steps include: S1, obtain the starting point and end point of the path to be planned; S2, determining multiple routes from the starting point to multiple demarcation points in the neighboring area and a first estimated time for moving on each route through a neighboring area path planning model; S3, determining the second estimated time from each demarcation point to the end point by using the trained distant neighbor area time estimation model; S4, determining which pass route to select as the optimal path based on the first estimated time corresponding to each pass route and the second estimated time from the dividing point of the pass route to the end point.

2. The path planning method according to claim 1, characterized in that: In S2, when exploring a passable route in the neighboring area by using the neighboring area path planning model, if it is determined that the explored passable route includes an end point, then selecting a passable route with the shortest estimated time from the starting point to the end point as the optimal path; If it is determined that the explored travel route does not include the end point, multiple travel routes from the starting point to multiple boundary points in the neighboring area and a first estimated time consumed for moving on each travel route are determined.

3. The path planning method according to claim 1, characterized in that: The neighborhood area path planning model adopts an improved Dijkstra algorithm, and the weight of the edge of the improved Dijkstra algorithm is determined based on the number of overhead cranes on the edge, whether the edge is located on a branch controlled by a regional control unit, the number of overhead cranes in the contactless power supply area to which the edge belongs, and the number of available unloading platforms between two nodes of the edge.

4. The path planning method according to claim 1, characterized in that: A demarcation point in the neighboring area refers to the node that the overhead crane is closest to after the overhead crane moves forward by a predetermined distance from the starting point; Or a demarcation point in the said neighborhood area refers to the node where the overhead travelling vehicle finally reaches when it moves forward a predetermined number of nodes; Or a demarcation point in the adjacent area refers to the grid unit where the overhead travelling vehicle is located when it moves forward a predetermined number of grid units.

5. The path planning method according to claim 1, characterized in that: The distant neighbor area time estimation model is a deep neural network, which includes an input layer, an embedding layer, at least one hidden layer and an output layer.

6. The path planning method according to claim 1, characterized in that: The training data required for the distant neighbor area time estimation model training is obtained through simulation by a simulation system.

7. The path planning method according to claim 1, characterized in that: Determine the total estimated time corresponding to each route according to the following formula, and select the route with the shortest total estimated time as the optimal path; ； Among them, t0 is the total estimated time, and t1 is the first estimated time corresponding to a passable route; is the discount factor; t2 is the second estimated time from the dividing point of the pass route to the end point.

8. The path planning method according to any one of claims 1 to 7, characterized in that: Each time the overhead travelling vehicle approaches or moves to a demarcation point of the selected passage route, the demarcation point is taken as a starting point and steps S2 to S4 are executed.

9. A path planning system, characterized in that: include: A starting point and end point acquisition unit is used to obtain the starting point and end point of the path to be planned; A first estimation unit, used to determine a plurality of travel routes from the starting point to a plurality of demarcation points in the neighboring area and a first estimated time consumption for moving on each travel route through a neighboring area path planning model; A second estimation unit, used for determining a second estimated time from each of the demarcation points to the end point by using a trained distant region time estimation model; The optimal path determination unit is used to determine which pass route is selected as the optimal path according to the first estimated time corresponding to each pass route and the second estimated time from the demarcation point of the pass route to the end point.

10. A transport controller, comprising a memory and a processor, wherein the memory stores a program executable by the processor, characterized in that: When the program is executed, the path planning method as described in any one of claims 1 to 8 is implemented.

Images