Transportation route determination method, transportation route determination device, and computer program

Abstract

This invention provides a transportation route determination method, a transportation route determination device, and a computer program that utilize analog computing in part to accelerate the calculation of transportation routes. [Solution] The transportation route determination method involves a computer determining a transportation route from a plurality of routes that connect a plurality of points, defined for each predetermined region, including a shipping point for the goods, points that the transportation equipment transporting the goods can pass through, and a destination point for the goods within the region, in the order in which the transportation route will travel. In this transportation route determination method, a plasma generator having channels that mimic nodes corresponding to the plurality of points and edges between each node generates plasma in the channels by giving electrodes corresponding to each node a positive or negative potential corresponding to the amount of goods at each point, and selects a route that passes through the nodes and edges where plasma emission has occurred.

Description

Technical Field

[0001] The present invention relates to a transportation route determination method, a transportation route determination device, and a computer program for realizing high-speed derivation calculation of a transportation route.

Background Art

[0002] In the problem of optimizing transportation routes in the logistics industry, not only the optimization of the route of the goods to be transported itself, but also the optimization of transportation resources related to logistics, such as transportation equipment such as trucks for transporting goods, drivers of transportation equipment, and logistics materials on which goods are loaded, are included.

[0003] In order to improve the working environment of drivers, a method has been proposed for determining a route such that the goods to be transported change multiple transportation devices from the collection point (start within the area of the goods) to the delivery point (goal within the area). When it is assumed that the goods change multiple transportation devices, it is necessary to be able to determine in real time at the transfer point which transportation device the goods should be loaded onto next according to the change in the situation. Regarding this, a method has been proposed for distinguishing transportation routes between trunk transportation and regional transportation and determining the route of regional transportation for each region (Patent Document 1, etc.).

[0004] Patent Document 1 proposes a method in which goods change multiple transportation devices, that is, the transportation device can unload goods at a point where loading and unloading of goods is possible, load other goods, and travel around within the area. In Patent Document 1, the logistics route is divided for each area, and based on agent-based modeling in which agents (transportation devices) move between nodes with bases within the area as nodes, the optimal route of the transportation device within the area is derived by simulation.

Prior Art Documents

Patent Documents

[0005] [[ID=3B]]

Patent Document 1

Summary of the Invention

[0006] Many route search methods, including agent-based modeling disclosed in Patent Document 1, do not easily derive optimal solutions in real time, taking into account the speed of real-world logistics and the constantly changing circumstances. Reinforcement learning in agent-based modeling requires a lot of power for computation.

[0007] This disclosure is made in view of the above circumstances and aims to provide a transportation route determination method, a transportation route determination device, and a computer program that achieve high-speed calculation of transportation routes by partially utilizing analog computing. [Means for solving the problem]

[0008] A transport route determination method according to one embodiment of the present disclosure is a transport route determination method in which a computer determines a transport route from a plurality of routes that connect a plurality of points, defined for each predetermined region, including a shipping point for an item, points that a transport device that transports the item can pass through, and a destination point for the item within the region, in the order in which the transport device travels, wherein a plasma generator having channels that mimic nodes corresponding to the plurality of points and edges between each node generates plasma in the channels by giving electrodes corresponding to each node a positive or negative potential corresponding to the amount of items at each point, and selects a route that passes through the nodes and edges where plasma emission has occurred.

[0009] A transport route determination method according to one embodiment of the present disclosure is a transport route determination method in which a computer determines a transport route by performing a simulation process based on agent-based modeling, using a transport network defined for each predetermined region, which assigns a shipping point for goods, a point that transport equipment transporting the goods can pass through, and a destination point for the goods within the region to nodes, and assigns a route connecting the points to the edges between the nodes, with the transport equipment as agents moving between the nodes, wherein as one step of the simulation process, the computer assigns a virtual potential to each node, to the node or the edge connected to the node, and to the nodes or edges surrounding the node or the edge, according to the quantity of goods to be shipped from the point corresponding to the node, and assigns a virtual potential with the opposite polarity to the virtual potential to each node, to the node or the edge connected to the node, and to the nodes or edges surrounding the node or the edge, according to the quantity of goods to be transported to the point corresponding to the node, and has channels corresponding to the nodes and edges of the transport network The plasma generator generates plasma in the channel, identifies a portion corresponding to the emitted node or edge from an image of the generated plasma emission, corrects the virtual potential applied to the node or edge corresponding to the identified portion based on a correction amount for the virtual potential, applies a virtual charge to the agent according to the number of items being transported by the transport equipment corresponding to the agent, determines the probability of selecting an edge to one or more adjacent nodes based on the attractive or repulsive force from the corrected virtual potential of the edge to the adjacent node to the agent's virtual charge in order to move from the node where the agent is located to one or more adjacent nodes, probabilistically selects one edge from the edges to one or more adjacent nodes based on the probability, and performs a process to update the quantity of items at the node to which the agent moved via the selected edge and the quantity of items being transported by the transport equipment corresponding to the agent, and the computer repeats the process of the above one step until the transport of items from the shipping point to the arrival point is completed.The process involves storing the history of the agent's movement of the nodes, resulting from the repetition of the aforementioned step, as a single transport route.

[0010] In the transportation route determination method of this disclosure, virtual potentials are assigned to nodes or edges around the node corresponding to the departure point of the goods (the point where the goods are collected and dispatched) and to nodes or edges around the node corresponding to the destination point of the goods (the point of arrival). Agent-based modeling is then performed by assigning numerical values ​​(virtual charges) to agents (corresponding to transportation equipment) moving between nodes, such that if the cargo bed is empty, they are more likely to move towards the departure point due to attractive forces from the node corresponding to the departure point or the edges leading to that node, and conversely, they are more likely to avoid the destination due to repulsive forces from the node corresponding to the destination or the edges leading to that node.

[0011] Virtual charge corresponds to each moving load or its quantity, and virtual potential is a field induced at the node corresponding to each stationary point due to the load present at that point. Virtual potential is a field that generates attractive / repulsive forces with respect to virtual charge. The attractive / repulsive forces here are determined such that the larger the value of the virtual potential or virtual charge, the stronger the attractive force from surrounding nodes or edges of other nodes, the closer the selection probability of the edge leading to that node approaches 1, and the stronger the repulsive force from surrounding nodes or edges of other nodes, the closer the selection probability of the edge leading to that node approaches 0.

[0012] By setting virtual potentials and virtual charges in this way, it is possible to represent that, in order to determine the optimal route for the movement of transport equipment, empty transport equipment is more likely to head towards the origin (collection point) within the network of goods, and transport equipment loaded with goods is more likely to head towards the destination (arrival point) within the network of goods. By setting virtual potentials and virtual charges, an attractive force acts between each point and the transport equipment passing through each point, causing negative charges to gather in areas with high potential, and conversely, a repulsive force acts to avoid negative charges.

[0013] In the transportation route determination method of the present disclosure, furthermore, a plasma generator having channels corresponding to actual nodes and edges is used, a voltage corresponding to the number of articles is applied to the electrode corresponding to the node to identify the channels in which plasma is generated, and the strength of the virtual potential of the edges corresponding to the channels in which plasma is actually generated is greatly corrected.

[0014] This enables the agent-based modeling to find the optimal solution more quickly and makes it possible to determine in real time how far to transport the articles and where to pick up the articles according to the running state of the transportation equipment that changes every moment.

Effect of the Invention

[0015] According to the present disclosure, the acceleration of the derivation calculation of the transportation route can be realized.

Brief Description of the Drawings

[0016] [Figure 1] It is an explanatory diagram of the outline of the transportation route determination method of the present disclosure. <​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​It is a flowchart showing an example of the arithmetic processing of a transport route by a transport route determination device. [Figure 13] It is a flowchart showing an example of the arithmetic processing of a transport route by a transport route determination device. [Figure 14] It is a flowchart showing an example of the control procedure of a plasma generation device by a processing device. [Figure 15] It is a schematic diagram of a transport network set in a grid pattern. [Figure 16] It is a schematic diagram of a transport network. [Figure 17] It is a diagram showing the state of plasma emission in a plasma generation device. [Figure 18] An example of the virtual potential (reward) after correction based on the plasma emission image is shown. [Figure 19] It shows the state of the transport network. [Figure 20] It shows the state of the transport network. [Figure 21] It shows the state of the transport network.

Embodiments for Carrying Out the Invention

[0017] The present disclosure will be specifically described with reference to the drawings showing its embodiments.

[0018] Figure 1 is an explanatory diagram illustrating the overview of the transportation route determination method of this disclosure. The transportation route determination method in this disclosure divides the routes of transportation equipment, such as transport trucks, trains, airplanes, and ships (in Figure 1, roads that transport trucks can travel on) into multiple regions shown by dashed lines in the figure, and determines the optimal route for the transportation equipment for each region according to the conditions. Each region is determined to include a base B that can be passed through on the main route shown by a thick line, and includes a collection and delivery point P that can be a collection point or arrival point for goods within the region. A collection point is a place where goods are gathered to be loaded onto the transportation equipment to be transported within the region, and corresponds to the starting point (S) of the transportation equipment within the region. An arrival point is a place where goods are unloaded from each transportation equipment and sorted in order to be delivered to the consignor's address within the region, and corresponds to the destination point (G) of the transportation equipment within the region. Base B may also be treated as a collection point for goods to be delivered to the region that have arrived from other regions via the main route. Location B can also serve as a receiving point for goods destined for other regions via trunk routes.

[0019] The transportation route determination method of this disclosure, apart from the method of determining transportation routes by transportation equipment on the trunk route shown by the thick line in Figure 1, focuses on a region including base B, collection and delivery point P, relay point R, and points passed through, as shown in the lower part of Figure 1, and derives the optimal solution for the transportation route of transportation equipment within that region for each region. The optimal solution for the transportation route in each region can be the optimal solution for the transportation route of the transportation equipment itself moving within the region, and by connecting the transportation route of the destination region, the trunk route, and the transportation route of the destination region, it is possible to derive the optimal travel route for each item (the route by which the item moves by changing transportation equipment). By determining the transportation route within a region, the route of the transportation equipment can be limited to staying within the region or only traveling on the trunk route between bases B, improving the working environment for the transportation equipment drivers. Optimizing the transportation route of the transportation equipment also makes it possible to save driving energy in the transportation equipment.

[0020] By determining the optimal route for transportation equipment within a region based on specific conditions, it becomes possible to calculate the optimal solution for the entire movement route from the dispatch of goods to their arrival at the consignor as quickly as possible, and also to calculate the optimal solution for the most efficient transportation routes for transportation equipment as quickly as possible. It enables real-time, step-by-step calculations of how far goods should be transported, where they should be picked up, and where they should be transported, based on the constantly changing driving conditions of the transportation equipment.

[0021] Figure 2 is an overview diagram of the transportation network set up within a region. Figure 2A shows an example of the source map data, Figure 2B shows the transportation network created from the map data, and Figure 2C shows a further subdivided transportation network. As shown in Figure A, the transportation network is created by assigning intersections and junctions, as well as points that transportation equipment can pass through, such as hub B, collection and delivery point P, and relay point R, to vertices (nodes) arranged in a grid pattern, rather than using a road network that is not a grid. The geographical length between each point does not correspond to the length in the transportation network, and should be stored as separate data simply as the distance between nodes. As shown in Figures 2B and 2C, edges are removed when there is no corresponding route (road, waterway, sea route, etc.) on the map.

[0022] For convenience, the transportation network in Figure 2B will be described as being divided into transportation networks of the size shown in Figure 2C. The actual size of the transportation network to be calculated should be set to a size that allows for plasma generation in the plasma generator described later; however, for the sake of simplicity, the following explanation will assume a 7x7 node size. By converting to such a grid-like transportation network, the road traffic network can be standardized as data.

[0023] In the transport route determination method of this disclosure, an agent corresponding to a transport device repeatedly performs the step of selecting an edge to an adjacent node on the nodes of a transport network divided as shown in Figure 2C to create a transport route, and agent-based modeling and / or reinforcement learning are performed to select the most efficient transport route. Even if reinforcement learning is used, which rewards the agent for creating an efficient transport route, agent-based modeling, which involves a large number of process patterns and multiple steps for each process, requires a considerable amount of computation time, making it difficult to respond to real-time changes in transport conditions. Therefore, in the transport route determination method of this disclosure, the results of generating a plasma channel in a plasma generator that has electrodes corresponding to a simplified transport network (Figure 2C) are reflected in the agent-based modeling and reinforcement learning to speed up the derivation of the optimal solution.

[0024] Figure 3 is a schematic diagram of the transport route determination device 100, and Figure 4 is a block diagram showing the configuration of the processing device 1. The transport route determination device 100 includes the processing device 1 and the plasma generator 2. The processing device 1 is a computer that controls the plasma generator 2 and performs reinforcement learning calculations using the results of processing by the plasma generator 2. The plasma generator 2 is a device in which electrodes are installed inside a vacuum chamber, making it possible to observe the generated plasma from the outside. In the plasma generator 2, the electrodes and the channels through which the generated plasma passes are formed to correspond to the 7x7 nodes and edges between them shown in Figure 2C (see Figure 6). The processing device 1 and the plasma generator 2 are connected to enable the exchange of signals. The imaging unit 13 of the processing device 1 is configured to image the plasma generated by the plasma generator 2. The processing device 1 can start a simulation using the operation unit 15 and output the processing results and the progress of the calculations to the display unit 14.

[0025] The processing unit 1 comprises a processing unit 10, a storage unit 11, an input / output unit 12, an imaging unit 13, a display unit 14, and an operation unit 15. The processing unit 1 may not only be configured to use a single computer (hardware), but may also be configured to distribute processing by connecting multiple computers via communication, or it may be one of multiple server computers (instances) virtually generated on a large computer. The processing unit 1 may perform calculations using a quantum computer.

[0026] The processing unit 10 includes one or more processors such as a CPU (Central Processing Unit), MPU (Micro-Processing Unit), and GPU (Graphics Processing Unit). The processing unit 10 also includes memory, which is a temporary storage medium such as SRAM (Static Random Access Memory) and DRAM (Dynamic Random Access Memory). The processing unit 10 may be configured as a single hardware (SoC: System On a Chip) integrating the processor, memory, and furthermore, the storage unit 11, the input / output unit 12, and communication means (not shown).

[0027] The memory unit 11 is a relatively large-capacity non-temporary storage medium such as a hard disk or SSD (Solid State Drive). The memory unit 11 stores the program (program product) necessary for the processing unit 10 to execute processing, and reference setting data. The program product includes an information processing program P1 and a reinforcement learning engine. The information processing program P1 is a program that, when read into memory and executed by the processing unit 10, causes a general-purpose computer to function as the transport route determination device 100 of this disclosure, which performs various processes described later using information obtained from the plasma generator 2.

[0028] The information processing program P1 stored in the memory unit 11 may be an information processing program P9 stored in a computer-readable non-temporary storage medium 9, which the processing unit 10 reads and stores in the memory unit 11. The information processing program P1 may also be an information processing program P1 that the processing unit 10 downloads from a download server via communication means and stores in the memory unit 11.

[0029] The input / output unit 12 inputs and outputs signals to and from the plasma generator 2 and performs digital-to-analog conversion. Based on the data provided by the processing unit 10, the input / output unit 12 outputs a drive signal to the drive unit 26 (see Figure 9) which adjusts the electrodes of the plasma generator 2, and outputs a signal to the switch 221 (see Figure 9) which applies voltage to the electrodes to generate plasma. The input / output unit 12 can adjust the position of the electrodes in the plasma generator 2 and refer to the state of the switch, and the processing unit 10 can recognize these states.

[0030] The imaging unit 13 uses an image sensor that corresponds to visible light and / or near-infrared light, and outputs image data of the captured image to the processing unit 10. The imaging unit 13 may also output video image data. The imaging unit 13 may be a camera that is detachable from the processing unit 1.

[0031] The display unit 14 is a user interface that enables input and output with the processing unit 10. The display unit 14 is a display such as a liquid crystal display or an organic EL (Electro-Luminescence) display. The display unit 14 may also be a touch panel-integrated display. The processing unit 10 may output the plasma image captured by the imaging unit 13 to the display unit 14 as a monitor.

[0032] The operation unit 15 is a user interface that can input and output to the processing unit 10. The operation unit 15 includes a pointing device such as a mouse and a keyboard. The operation unit 15 may be, for example, a touch panel built into the display unit 14, or it may include physical buttons, switches, and physical dials. The operation unit 15 may also be configured to accept voice commands using a microphone and a voice recognition processing unit.

[0033] Figure 5 is a schematic exploded perspective view of the plasma generator 2, and Figure 6 is a front view (top view) of the inside of the chamber 20. Figure 7 is a cross-sectional view of the inside of the chamber 20, and Figures 8A and 8B are enlarged views of a portion of the channel forming section 23 of the chamber 20.

[0034] The plasma generator 2 houses a channel forming section 23 inside a chamber 20 for forming a channel 21 and positioning an electrode 22, and both ends are sealed with an upper cover 24 and a lower cover 25. The electrode 22 protrudes from the channel forming section 23 and the lower cover 25 via a drive section 26, and power can be applied from an external power source. A mesh 27 connected to the ground potential is placed on top of the upper cover 24.

[0035] The body of the chamber 20 is made of a metal such as stainless steel, and is short and cylindrical with flange-shaped diameters at both ends. The hollow portion of the chamber 20 is sealed with an upper cover 24 and a lower cover 25 to allow only the plasma gas to be filled inside.

[0036] A channel-forming portion (substrate) 23, made of a dielectric material such as fluororesin and shaped to be roughly cylindrical, is embedded in the hollow portion of the chamber 20. A trench 231 and a hole 232, which form the channel 21, are formed on one end face of the channel-forming portion 23. When the top cover 24 is fixed to the chamber 20 body, the space defined by the back surface of the top cover 24 and the inner surfaces of the trench 231 and hole 232 becomes the channel 21.

[0037] As described above, the holes 232 in the channel forming section 23 are arranged in a grid pattern corresponding to 7x7 nodes. The holes 232 are cavities formed in a flattened cylindrical shape. The dimensions of the holes 232 are, for example, 5 mm in diameter and 3 mm in depth, and they are formed at 10 mm intervals. The trenches 231 are grooves with a rectangular cross-section, formed to connect adjacent holes 232. The trenches 231 are formed to be shallower and wider than the diameter and depth of the holes 232. The dimensions of the trenches 231 are, for example, 1 mm in width, 1 mm in depth, and 5 mm in length. Two holes 232 that are as far apart as possible have holes 233 for gas introduction and discharge. On the other end face of the channel forming section 23, holes 234 are provided at each location corresponding to the back surface of the holes 232, which serve as insertion points for the electrodes 22. The distance between holes 232 and holes 234 is approximately 1 mm.

[0038] The disc-shaped upper cover 24 is made of quartz, at least in its central portion being transparent or translucent, allowing observation of the channel-forming portion 23 inside the chamber 20. Similarly, the disc-shaped lower cover 25 is made of metal. The upper cover 24 and lower cover 25 are structured to seal the inside of the chamber 20, trapping the gas inside the chamber 20.

[0039] An electrode 22 is provided extending from the lower cover 25 toward the inside of the chamber 20. The lower cover 25 is provided with a drive unit 26 that moves the electrode 22 up and down. The drive unit 26 comprises a nut 261 to which the electrode 22 is fixed, a screw shaft 262 fitted into the nut 261, and a motor 263 that rotates the screw shaft 262. By rotating the motor 263, the drive unit 26 can move the electrode 22 up and down together with the nut 261. The drive unit 26 includes a drive circuit 264 that receives a control signal from the processing unit 1 and turns the power supply to the motor 263 on / off based on the control signal. The drive unit 26 is provided with a substrate 265 on the lower cover 25 that has connectors etc. exposed to the outside while maintaining airtightness so that power can be supplied from the outside and control signals can be input and output to the drive circuit 264. A voltage circuit 220 (see Figure 9) that supplies power to the electrode 22, the drive circuit 264, etc. are mounted on the substrate 265.

[0040] The drive unit 26 includes a voltage circuit 220 that supplies voltage to each of the electrodes 22. Figure 9 shows a schematic diagram of the voltage circuit 220 that applies voltage to the electrodes 22. The voltage circuit 220 is connected to each of the electrodes 22 via signal lines and may be located outside the chamber 20. The voltage circuit 220 includes a switch 221 that switches whether or not to apply high voltage to the electrodes 22. Based on a control signal from the processing unit 1, the switch 221 switches between the high-voltage power source 222 and the resistor 223. The plasma generator 2 can generate plasma when connected to the high-voltage power source 222 by the switch 221.

[0041] When the electrode 22 of the plasma generator 2 is raised, the distance between the hole 232 and the electrode 22 becomes shorter. Conversely, when the electrode 22 of the plasma generator 2 is lowered, the distance between the hole 232 and the electrode 22 becomes longer. By connecting a predetermined resistor 223 to the electrode 22, the electrode 22 can be set to a reference potential (e.g., ground potential). The polarity of the potential of the plasma generator 2 can be reversed by interposing a circuit that inverts the output from the high-voltage power source 222. When the switch 221 is not connected to either the high-voltage power source 222 or the resistor 223 (floating state), plasma is less likely to be generated in the corresponding hole 232 than in other holes.

[0042] The plasma generator 2, configured in this way, generates plasma by applying positive and negative potentials corresponding to the quantity of goods at each location to electrodes 22 corresponding to each node in channels 21 that simulate nodes corresponding to multiple locations, including collection points and arrival points, and the edges between each node. The transport route determination device 100 generates a high electric field between the nodes corresponding to collection points and the nodes corresponding to arrival points by reversing the polarity of the high voltages applied to the nodes corresponding to collection points and arrival points in the plasma generator 2. The transport route determination device 100 determines the route by considering that the route of the part corresponding to the hole 232 and trench 231 where plasma emission occurred can be considered the shortest route when a high electric field is generated between the node corresponding to the collection point and the node corresponding to the arrival point in the channel 21 of the plasma generator 2. The processing device 1 adjusts the distance between the electrode 22 and the hole 232 by controlling the height of the electrode 22. The higher the height of the electrode 22, the shorter the distance between the electrode 22 and the hole 232. As the distance decreases, when a voltage is applied from the high-voltage power source 222 to the electrode 22, the strength of the electric field generated by the electrode 22 increases, making it easier to generate plasma. By using a plasma generator 2 in which the electrode 22 can be adjusted, it becomes possible to obtain the optimal solution early, as will be described later.

[0043] The processing unit 10 of the processing unit 1 outputs a control signal to adjust the height of the electrode 22 corresponding to the node of the collection / delivery point P (start or goal) that the agent corresponding to the transport equipment should head to, in units of a transport network including 7x7 nodes as shown in Figure 2C. The height of the electrode 22, i.e., the height of the applied potential, corresponds to the number of items to be shipped (remaining number) at the node corresponding to the collection point S, or the number of items to be delivered to the node corresponding to the arrival point G (the number that is insufficient to meet the required number). By controlling the switch 221, the processing unit 10 of the processing unit 1 connects the electrode 22 corresponding to the node that the agent can pass through to a high-voltage power source 222, generates plasma, and captures the process with the imaging unit 13, which is positioned with the top cover 24 facing forward. In addition, a high-voltage power source 222 with the opposite polarity to the node of collection / delivery point P may be connected to the node where the agent is located. This allows the plasma generator 2 to determine the path from the agent's starting point to the target collection / delivery point P that is most likely to generate plasma.

[0044] Figure 10 shows the plasma generation process in the plasma generator 2. Figure 10 shows the view from above the top cover 24. With the transparent top cover 24 in place, plasma gas is applied to the channels 21 (shown by dashed lines) formed by the grid-like arrangement of holes 232 and trenches 231 connecting the holes 232. When a high voltage is applied, the inside of the trenches 231 and holes 232 where plasma is generated emits light. In the example shown in Figure 10, the emitted light is represented by a thick line. In the example shown in Figure 10, the drive unit 26 connects a high-voltage power source 222 to the electrode 22 of the hole 232 corresponding to the node of collection point P in the transportation network shown in Figure 2C, and connects a reference potential to the electrode 22 of the other holes 232 corresponding to nodes in the transportation network shown in Figure 2C via a resistor 223. Electrodes 22 of holes 232 corresponding to nodes that do not exist in the transport network shown in Figure 2C (for example, the second column from the right, the first and second from the top) are not connected to anything (they are floating).

[0045] In the example shown in Figure 10, the portions of hole 232 to which high voltage is applied corresponding to base B and collection / delivery point P emit bright light, and the light is emitted along the trench 231 toward the hole 232 connected to the reference potential. Holes 232 in a floating state may also emit light.

[0046] In the transport route determination device 100, the processing unit 1 adjusts the height of the electrodes 22 corresponding to base B and collection / delivery point P in the transport network in the plasma generator 2, and uses the image captured by the imaging unit 13 (Figure 10) of the plasma generation process by switching the switch 221 to correct the edge selection probability and streamline the calculation by agent-based modeling. The transport route determination process procedure by the transport route determination device 100 will be described in detail below.

[0047] Figures 11-13 are flowcharts illustrating an example of the transportation route calculation process performed by the transportation route determination device 100. The transportation route determination device 100 pre-executes the following processes for each transportation network corresponding to each region and stores them in the storage unit 11.

[0048] The processing unit 10 sets delivery conditions for the target transport network, including the initial location and number of transport equipment acting as agents, the initial location (collection point S) and number of goods to be delivered within the region, and the destination point G for each item within the transport network (S101). In S101, the processing unit 40 initially associates the number of goods initially loaded onto the transport equipment (e.g., zero) with each agent. The number of agents may be one or more.

[0049] In S101, the delivery conditions are as follows: for example, there are 100 items to be delivered to a node corresponding to base B in the transportation network, i.e., collection point S of goods brought into the region via a main road (or the logistics materials if transported in units of pallets or folding containers). The delivery conditions also include the number of each item to be transported to collection point P (destination point G) within the transportation network. The delivery conditions further include the agent's initial position (not necessarily collection point S). The agent's initial position may be at base B or collection point P (collection point S or destination point G), or it may be at a relay point R. For the sake of simplicity, the following explanation will assume there is only one type of item, but there may be one or more types.

[0050] Based on the delivery conditions, the processing unit 10 assigns a virtual potential (a logical potential, a signed numerical value, a reward) to the edges of each node in the transport network that lead directly to a specific node corresponding to the starting point (collection point S) of goods within the region, according to the number of goods present at that specific node (S102). The magnitude of the numerical value assigned in S102 is determined according to the quantity of goods to be shipped from the starting point (collection point S) corresponding to the specific node.

[0051] The processing unit 10 assigns signed numbers not only to edges that point directly to a specific node, but also to edges that point to surrounding specific nodes, by decreasing the absolute value of the number according to the distance from the specific node (S103). The processing in S103 is performed by executing the calculation multiple times, so that it gradually spreads to each edge centered on the specific node.

[0052] Based on the delivery conditions, the processing unit 10 assigns a virtual potential (a logical potential, a signed numerical value, a reward) with the opposite polarity to the numerical value assigned in S102 to the edges of each node in the transport network that lead directly to a specific node corresponding to the destination point (destination point G) of the goods within the region (S104). In S104, the processing unit 10 assigns a virtual potential of a size corresponding to the quantity of goods to be transported to destination point G. In S104, the processing unit 10 updates the numerical value of each edge in stages so that the numerical value assigned in S103 propagates to the surrounding area.

[0053] The processing unit 10 assigns signed numbers not only to edges directly connected to the node corresponding to the arrival point, but also to surrounding edges, by decreasing the absolute value of the numerical value according to the distance from the arrival point node (S105). The processing in S105, like S103, is performed by executing multiple calculations so that it gradually spreads to each edge centered on a specific node.

[0054] The processing unit 10 may also assign a signed numerical value, inversely to the numerical value assigned in S102, to edges leading to nodes that can serve as transit points for transshipment of goods in transport equipment (nodes corresponding to locations such as warehouses). In this case, the magnitude of the numerical value should be a very small value.

[0055] The assignment of logical virtual potentials to each edge from S102 to S105 will be described in detail later. The processing unit 10 may assign a positive virtual potential with a valency corresponding to the quantity of goods to the edge heading towards collection point S where goods remain, and a negative virtual potential with a valency corresponding to the quantity of goods that are missing to the edge heading towards arrival point G where more goods should arrive. The positive and negative signs of the potentials may also be reversed.

[0056] Based on the delivery conditions, the processing unit 10 assigns a signed numerical value (virtual charge) to the agent corresponding to the transport equipment in its initial state, according to the empty state or loading rate of the cargo bed (S106). In S106, for example, if the numerical value associated with the agent is zero, i.e., the cargo bed of the transport equipment is empty, the processing unit 10 assigns a numerical value of a predetermined size with the opposite sign to that of S102, in order to facilitate the departure to collection point S where there are many goods. If the number of goods associated with the agent is the same as the number when fully loaded and the cargo bed is fully loaded, the processing unit 10 assigns a numerical value of a predetermined size with the same sign as in S102.

[0057] The numerical value assigned in S106 can also be interpreted as a virtual charge. An empty cargo bed is assigned a negative virtual charge of a predetermined valence to facilitate movement to the collection point S, while a fully loaded cargo bed is assigned a positive virtual charge of a predetermined valence to facilitate movement to the arrival point G.

[0058] The processing unit 10 may store the settings and assigned values ​​from S102-S106 as a multidimensional matrix (vector), representing the state at each node (number of items) or the values ​​assigned to the edges leading to each node.

[0059] The processing unit 10 sets the number of times the process is executed to an initial value (S107). The processing unit 10 executes the following process multiple times and, among the multiple executions, derives the path with the fewest number of executions (the number of steps the agent takes from being at one node to moving to the next node) or the shortest distance the agent travels, until the goods that were at specific nodes, namely collection point S and base B, are distributed to destination point G within the transport network.

[0060] The processing unit 10 sets the number of steps to an initial value (S108). The processing unit 10 performs processing based on agent-based modeling with the transport equipment as agents, as shown below, until the goods that were present at a specific node, which is the collection point S, have been distributed to each destination point G within the transport network.

[0061] The processing unit 10 initially assigns a reward to each node in the transport network according to the delivery conditions (S109). If there are no items loaded on the agent's cargo bed, the node corresponding to the originating point receives a reward that increases with the number of items present at the node. Conversely, if the agent's cargo bed is set to be nearly full, the node corresponding to the destination point receives a reward that increases with the number of missing items.

[0062] In that step, the processing unit 10 updates the virtual potential of each edge based on the number of items associated with the agent (the number of items loaded on the transport equipment, pallets, or folding containers) and the reward given to the node (S110).

[0063] In S110, the processing unit 40 assigns a signed numerical value (virtual potential) to the edges directly or indirectly connected to the node where the agent resides, with a magnitude corresponding to the number of items remaining at that node, and assigns the agent a signed numerical value (virtual charge) with a magnitude corresponding to the number of items it is associated with. The signed numerical values ​​on the edges leading to the surrounding nodes of the node whose signed numerical value has been updated should also be updated in accordance with the number of edges and distance connected to the node at the point of origin.

[0064] At each step, the processing unit 10 generates plasma in the plasma generator 2 with the electrodes 22 arranged according to the virtual potential updated in S110, and captures an image (S111). The processing unit 10 acquires an image showing the appearance of the captured plasma emission (S112), and identifies edges in the transport network corresponding to the strongly emitted parts from the acquired image (S113). The processing unit 10 corrects the virtual potential (reward) of the edge updated in S110 according to the intensity of the emission (S114). Note that the processing in S111-S114 may be skipped if the virtual potential is not updated in S110. The processing in steps S111-S113 may be executed multiple times, averaged, and then the edge corresponding to the most strongly emitted part may be identified.

[0065] The processing unit 10 determines the selection probability of an edge to an adjacent node based on the attractive or repulsive force derived from the signed numerical value (virtual potential) assigned to the edge toward the surrounding nodes of the agent and the signed numerical value (virtual charge) assigned to the target agent (S115).

[0066] In S115, the processing unit 10 evaluates the difference between the signed numbers as an attractive force if the numerical value (virtual charge) assigned to the agent and the numerical value (virtual potential) assigned to the edge toward the adjacent node have opposite signs. If the numerical value (virtual charge) assigned to the agent and the numerical value (virtual potential) assigned to the edge toward the adjacent node have the same sign, the processing unit 10 evaluates the sum of the absolute values ​​of these signed numbers as a repulsive force. The processing unit 10 may also represent the signed numbers as vectors having the direction assigned to that node, and calculate the attractive or repulsive force as the product of the dot product of the unit vector of the edge direction and the virtual charge assigned to the agent. The method of calculating the attractive or repulsive force is not limited to the method described above. The processing unit 10 may calculate the product of the virtual potential or virtual charge assigned to the edge, and if the calculated product is a positive value, it is a repulsive force; if it is a negative value, it is an attractive force, and the absolute value can be taken as the magnitude of the force. The processing unit 10 determines the selection probability of the edge with the greatest attractive force or the edge with the least repulsive force with the greatest selection probability. The processing unit 10 determines the selection probability based on the magnitude of the attractive or repulsive force, i.e., the product of the virtual potential and virtual charge, and whether the signs are opposite or not.

[0067] The processing unit 10 selects an edge to the next adjacent node to proceed to, based on the selection probability determined in S115 (S116). The processing unit 10 may probabilistically select an edge to the adjacent node based on the selection probability, or it may select the edge with the highest selection probability. In probabilistic selection, the processing unit 10 does not necessarily select the edge with the highest selection probability, but rather selects an edge by performing a weighted lottery using the selection probability as a weight.

[0068] The processing unit 40 moves the agent via the selected edge (S117). The processing unit 10 determines whether the node to which the agent has moved via the selected edge corresponds to the arrival point G of the goods being transported (S118).

[0069] In S118, if it is determined that the node at the end of the selected edge is the destination G (S118: YES), the processing unit 10 consumes (unloads) the goods loaded on the cargo bed of the transport equipment, which is the agent, and consumes the reward attached to the node corresponding to the destination to which it has moved (S119). In S119, the processing unit 10 subtracts the number of goods associated with the agent and subtracts the number of goods that are missing at the destination.

[0070] If, in S118, it is determined that the node at the end of the selected edge is not the destination point G (S118: NO), the processing unit 10 skips S119 and proceeds to S120.

[0071] In S120, the processing unit 10 determines whether the node to which the agent has moved via the selected edge corresponds to the loading point and whether there is sufficient space on the agent's loading platform (S120). The processing unit 10 may also determine this by whether the absolute value of the charge corresponding to the number of items on the agent's loading platform is small or not.

[0072] If it is determined in S120 that the destination is a shipping point and that there is sufficient space on the agent's cargo bed (S120: YES), the processing unit 10 increases the number of items loaded on the cargo bed of the transport device, which is the agent, and consumes the reward assigned to the node corresponding to the destination shipping point (S121). In S121, the processing unit 10 adds the number of items associated with the agent and subtracts the number of items to be transported from the shipping point.

[0073] If it is determined in S120 that it is not a shipping point (S120: NO), the processing unit 10 adds the step number and stores it (S122). In S122, if distance data is associated with the edge that was selected and moved to, the processing unit 10 adds that distance to the cumulative travel distance.

[0074] Next, the processing unit 10 determines in S119 and S121 whether or not a delivery that meets the delivery conditions within the transport network has been completed (S123). In S123, the processing unit 10 may determine that the delivery is complete if all signed numerical values ​​(virtual potentials) assigned to each node become zero, or it may determine that the delivery is complete if the correct number of items that meet the delivery conditions have been unloaded at the node of arrival point G.

[0075] If it is determined that delivery has not been completed (S123:NO), the process returns to S110.

[0076] If it is determined that delivery is complete (S123:YES), the processing unit 10 stores the history of identification data of the nodes the agent passed through (transport route) and the number of steps (S124).

[0077] The processing unit 10 determines whether the processes in S110-S124 have been executed a predetermined number of times or more (S125). If it is determined that the number of executions is less than the predetermined number (S125: NO), the processing unit 10 adds to the execution count (S126) and returns the process to S108.

[0078] In S125, the processing unit 10 may determine whether the optimal solution (minimum number of steps or shortest cumulative distance) has been obtained through the executions so far, rather than determining whether the executions have been performed more than a predetermined number of times, or whether the optimal solution can be derived by performing further executions. In S125, the processing unit 10 may determine the optimal solution to be the route that has the highest evaluation amount converted into an indicator such as carbon dioxide emissions or cost, or driver working hours, and that is equal to or greater than a predetermined amount.

[0079] If it is determined in S125 that the process has been executed more than a predetermined number of times (S125: YES), the processing unit 10 extracts the transport route with the minimum number of steps or the shortest cumulative travel distance within the predetermined number of times (S127). The processing unit 40 stores the extracted transport route as the optimal solution under the delivery conditions (S128) and terminates the process.

[0080] The transport route determination device 100 may execute the processing procedure shown in the flowcharts of Figures 11-13 according to the actual delivery conditions, or it may execute it according to a plurality of predetermined delivery conditions.

[0081] Figure 14 is a flowchart showing an example of the control procedure for the plasma generator 2 by the processing device 1. The processing procedure shown in Figure 14 corresponds to the details of the S113 process in the processing procedure shown in the flowcharts of Figures 11-13.

[0082] First, the processing unit 10 adjusts the electrode 22 located in the hole 244 corresponding to the collection point S where the goods are gathered, according to the number of goods gathered (S301), based on the delivery conditions set in S101. In S301, the processing unit 10 controls the electrode to be higher when there are many goods and lower when there are relatively few.

[0083] The processing unit 10 adjusts the electrode 22 located in the hole 244 corresponding to the destination point G to which the goods should be delivered, according to the number of missing items (S302), based on the delivery conditions set in S101. In S302, the processing unit 10 controls the electrode to be higher when there are many missing items and lower when there are relatively few.

[0084] The processing unit 10 sets the height of the electrode 22 corresponding to other nodes in the target transport network (nodes corresponding to relay points R that transport equipment can pass through) to a predetermined height (S303).

[0085] The processing unit 10 outputs control signals to the drive unit 26 to control each switch 221 connected to each electrode 22 in order to generate plasma (S304). In S304, the processing unit 10 connects the electrode 22 corresponding to the node at collection point S to the high-voltage power source 222. The processing unit 10 connects the electrode 22 corresponding to the node at arrival point G to the high-voltage power source 222 whose polarity has been determined. The processing unit 10 grounds the electrodes 22 corresponding to the other nodes in the transport network via the resistor 223. The processing unit 10 leaves the electrodes 22 corresponding to nodes that do not exist in the target transport network (parts corresponding to locations that transport equipment cannot pass through) in a floating state. The processing unit 10 simultaneously activates the switches 221 corresponding to each electrode 22.

[0086] The processing unit 10 captures the channel 21 in the plasma generation state using the imaging unit 13, which has the top cover 24 in its field of view from above (S305), and terminates the plasma generation and image capture by the plasma generator 2.

[0087] Thus, the transport route determination device 100 of this disclosure combines a simulation process using agent-based modeling with analog computing of the plasma generator 2. The processing performed by the transport route determination device 100 will be explained below with specific examples, referring to a diagram illustrating an example of the method for deriving the optimal route.

[0088] Figure 15 is a schematic diagram of a transportation network set up in a grid. Each node, indicated by a circle, is assigned identification information in the form of a matrix number. Not all nodes are connected by thick lines, but as shown in Figure 2C, edges exist between nodes where roads actually exist and transportation equipment can move. These edges are directed edges. Edge data is represented by the matrix numbers of the starting and ending nodes.

[0089] In a transportation network, a transportation route can be represented as a column of node identification data (rows and columns). For example, the transportation route of an agent (transportation equipment) whose starting point for transportation within a region is the central point in Figure 15 can be represented as, for example, node (i,j) → node (i,j+1) → node (i-1,j+1) →… The transportation route of the agent from node (i,j) → node (i,j+1) → node (i-1,j+1) is shown by the white arrow in Figure 15.

[0090] In the initial state based on the delivery conditions, a specific node S, which is a collection point within the region, is node (i+1, j+1) and contains 100 items. As a result, in Figure 15, the edge leading to this node (i+1, j+1) is assigned a signed value (virtual potential) of "+100" in S102. The virtual potential corresponds to the number of items that should be piled up from that collection point S.

[0091] In the initial state based on the delivery conditions, the node that is the arrival point G within the region is node (i-2, j-2). Node (i-2, j-2) is assigned the condition that 40 items will be received. In Figure 15, the edge leading to this node (i-2, j-2) is assigned a virtual potential of "-40" in S104, for example, with a magnitude corresponding to the number of items to be transported.

[0092] Figure 16 is a schematic diagram of a transport network. The transport network shown in Figure 16 is an example of the transport network shown in Figure 15, but with virtual potentials assigned to nodes surrounding a specific node according to their distance from that node. In Figure 16, for node (i+1,j+1), which is assigned the number of items "+100" corresponding to collection point S, edges leading to adjacent nodes via edges are assigned a signed number with a gradient in magnitude, depending on the number of edges traversed or the sum of the distances of the corresponding edges. Nodes (i,j+1) and (i+1,j), which are connected to node (i+1,j) at collection point S via one edge, are each assigned a virtual potential of "+75". Additionally, edges leading to nodes (i,j+2), (i-1,j+1), and (i+2,j), which are connected to node (i+1,j+1) at collection point S via two edges, are each assigned a virtual potential of "+50". The virtual potentials assigned to the edges to these surrounding nodes are used solely for calculating attractive and repulsive forces between adjacent nodes. If each node is assigned features (number of items present at that node, virtual potential, ...), then the feature of node (i+1,j+1) assigned "+100" is (100,+100), and the feature of the adjacent nodes (i,j+1) and (i+1,j) is (0,+75). The feature of nodes (i,j+1) and (i+1,j) separated by two edges is (0,+50).

[0093] In the transport network shown in Figure 16, virtual potentials are also assigned to the edges leading to the surrounding nodes of a specific node corresponding to destination G. In the example in Figure 16, the edges leading to the adjacent nodes (i-2,j-3), (i-2,j-1), and (i-1,j-2) of node (i-2,j-2), which are assigned "-40" and corresponding to destination G, are assigned a virtual potential of "-20". For destination G, the virtual potentials assigned to the edges leading to the surrounding nodes are used only to calculate the attractive and repulsive forces between adjacent nodes. When feature quantities (number of items present, virtual potential, ...) are assigned to each node, the feature quantity of node (i-2,j-2), which is assigned "-40", is initially (0,-40), and the feature quantities of the adjacent nodes (i-2,j-3), (i-2,j-1), and (i-1,j-2) are (0,-20).

[0094] Figure 17 shows the plasma emission in the plasma generator 2. Similar to Figure 10, Figure 17 shows the plasma emission as a diagram of an image taken from above the top cover 24. Figure 17 corresponds to the 7x7 node arrangement corresponding to the transport network shown in Figure 2C, Figure 15, or Figure 16. In Figure 17, the part corresponding to node (i+1,j+1) corresponding to collection point S emits the strongest light, followed by the part corresponding to node (i,j+1). Comparing the part corresponding to the edge from node (i+1,j) adjacent to node (i+1,j+1) corresponding to collection point S with the part corresponding to the edge from node (i,j+1), the plasma emission is stronger in the latter part corresponding to the edge from node (i,j+1).

[0095] In Figure 17, the portion corresponding to node (i-2, j-2) at the loading point G also emits strong light in a similar manner. Comparing the portion corresponding to the edge from node (i-2, j-3) adjacent to node (i-2, j-2) at loading point G, the portion corresponding to the edge from node (i-2, j-1), and the portion corresponding to the edge from node (i-1, j-2), the portion corresponding to the edge from node (i-2, j-1) emits the strongest plasma light.

[0096] The processing unit 10, upon recognizing the plasma emission shown in Figure 17, determines, for example, a correction amount of "+10" for the edge from node (i,j+1) to node (i+1,j+1), which is the collection point S, and a correction amount of "+5" for the edge from node (i+1,j) to node (i+1,j+1). The processing unit 10 also determines a correction amount of "+2" for the edge from node (i,j), which is the agent's starting point, to node (i,j+1), and a correction amount of "+5" for the edge from node (i,j) to node (i+1,j). The processing unit 10 determines a correction amount of "+5" for the edge from node (i-1,j+1), which is two edges away from collection point S, to node (i,j+1) heading towards collection point S, and determines a correction amount of "+2" for the two edges heading towards that node (i-1,j+1).

[0097] Similarly, the processing unit 10 determines a correction amount of "+2" for the edge from node (i-2, j-3) to the loading point G, node (i-2, j-2), "+10" for the edge from node (i-2, j-1) to node (i-2, j-2), and "+5" for the edge from node (i-1, j-2) to node (i-2, j-2).

[0098] Figure 18 shows an example of corrected virtual potential (reward) based on plasma emission images. It shows an example of virtual potential corrected by the correction amount shown in Figure 17. The value of the virtual potential has changed compared to Figure 16. For example, the virtual potential of the edge from node (i,j+1) to node (i+1,j+1) corresponding to collection point S is set to "+100" plus "+10" to get "+110". The virtual potential of the edge from node (i+1,j) to node (i+1,j+1) corresponding to collection point S is set to "+100" plus "+5" to get "+105". Similarly, the virtual potential of the edge from node (i,j) to node (i,j+1) is set to "+75" plus "+2" to get "+77", and the virtual potential of the edge from node (i,j) to node (i+1,j) is set to "+75" plus "+5" to get "+80".

[0099] There are two possible transport routes from node (i,j) to node (i+1,j+1) corresponding to collection point S: (i,j)→(i+1,j)→(i+1,j+1) and (i,j)→(i,j+1)→(i+1,j+1). In the initial edge selection, the latter has a higher virtual potential, and agents assigned a negative charge are more likely to choose this route. However, in a reinforcement learning method where the virtual potential is used as the reward (Q value) and the path is selected by anticipating the reward one step ahead, the sum of the two edge selections becomes "+187" for the former and "+185" for the latter, increasing the probability of choosing the former path. When selecting edges using reinforcement learning with rewards anticipated two or three steps ahead, the likelihood of choosing one path can be adjusted by setting how the Q value is calculated (the coefficient γ in equation 1, described later).

[0100] In Figure 18, similarly around the loading point G, the virtual potential of the edge from node (i-2, j-3) to node (i-2, j-2) corresponding to loading point G is given as "-42" by adding an absolute value of "+2" to "-40". The virtual potential of the edge from node (i-2, j-1) to node (i-2, j-2) corresponding to loading point G is given as "-50" by adding an absolute value of "+10" to "-40". The virtual potential of the edge from node (i-1, j-2) to node (i-2, j-2) corresponding to loading point G is given as "-45" by adding an absolute value of "+5" to "-40". The virtual potential of the edge from node (i-3, j-3) to node (i-2, j-3) is calculated as -22 by adding an absolute value of +2 to -20, and the virtual potential of the edge from node (i-3, j-1) to node (i-2, j-1) is calculated as -25 by adding an absolute value of +5 to -20.

[0101] Using the corrected virtual potential as shown in Figure 18, the processing unit 10 calculates the attractive or repulsive force to the virtual charge associated with the agent. As shown in Figure 15, if the virtual charge attached to the agent is "-50", the virtual potential attached to the edge to node (i,j+1) is "+77", and the virtual potential attached to the edge to node (i+1,j) is "+80". Therefore, the processing unit 10 calculates an "attractive force" of, for example, "127 (=+77-(-50))" from the edge to node (i,j+1) and an "attractive force" of "130 (=+80-(-50))" from the edge to node (i+1,j). The processing unit 10 uses these attractive forces as weights and probabilistically selects an edge to one of the nodes (S116). Since the processing unit 10 selects an edge by weighted lottery, it may still select the edge to node (i,j+1) even if the attractive force from the edge to node (i+1,j) is stronger than the attractive force from the edge to node (i,j+1).

[0102] Figure 19 shows the state of the transport network one step ahead from the state shown in Figure 18. In Figure 19, the agent moves from node (i,j) to node (i+1,j) by selecting the edge through S116. Since node (i+1,j) is neither the collection point S nor the arrival point G, there is no change in the goods at each node. In this case, plasma generation by the plasma generator 2 and imaging by the imaging unit 13 may also be skipped.

[0103] In the state shown in Figure 19, the agent at node (i+1,j) calculates the attractive or repulsive forces from the edges to the adjacent node (i+2,j), the edge to node (i+1,j+1), and the edge to node (i,j). The virtual charge attached to the agent is "-50", the corrected virtual potential attached to the edges to the adjacent node (i+2,j) and the edge to node (i,j) is "+52 (=+(50+2))", the virtual potential attached to the edge to node (i+1,j+1) is "+105 (=+(100+5))", and the corrected virtual potential attached to the edge returning to node (i,j) is "+55 (=+(50+5))". Therefore, in S115, the processing unit 10 calculates an attractive force of "105 (= +55 - (-50))" from the edge to the adjacent node (i+2, j), an attractive force of "155 (= +105 - (-50))" from the edge to the node (i+1, j+1), and an attractive force of "102 (= +52 - (-50))" from the edge to the node (i+2, j). The processing unit 10 uses these attractive forces as weights and selects an edge to one of the nodes probabilistically (S116). Because it is a weighted lottery, it does not necessarily select the edge with the strongest attractive force, but it will more often select the edge with the stronger attractive force.

[0104] Figure 20 shows the state of the transport network after one step has been taken from the state shown in Figure 19. Figure 20A shows the state immediately after the step has been taken, and Figure 20B shows the state after processing in S117. The agent has moved from node (i+1,j) to node (i+1,j+1) by selecting an edge leading to node (i+1,j+1). Node (i+1,j+1) is collection point S (S120:YES). At the time of Figure 20A, the number of items associated with the agent is "0 (zero)" because the cargo bed is empty. At the time of Figure 20A, the number of items at node (i+1,j+1) where the agent is located is "100". In S111, processing unit 10 loads "20" items to fill all of the empty cargo beds. As a result, the number of items associated with the agent after loading increases to "20" (S121). (S119).

[0105] As shown in Figure 20B, the S121 process reduces the number of items remaining at node (i+1,j+1) (reward) from "100" to "80", and the virtual potential applied to the edges leading to node (i+1,j+1) is basically reduced to "+80". As a result, the virtual potential of each edge is updated to a smaller value, especially around node (i+1,j+1). Based on the updated state of the virtual potential and virtual charge after stacking, the processing unit 10 generates plasma in the plasma generator 2 and takes an image. At this time, the processing unit 10 adjusts the height of the electrode 22 corresponding to node (i+1,j+1) from "100" to "80". As a result, the generated plasma channel may change slightly, and the correction amount for each edge may differ from the amount shown in Figure 17. Figure 20B shows the plasma channel generated after the update and the correction amount determined from the intensity of its emission on the right side.

[0106] In Figure 20B, the agent at node (i+1,j+1) calculates the attractive or repulsive forces from the edge toward the adjacent node (i+1,j) and the edge toward node (i,j+1). The virtual charge assigned to the agent is updated to "+50", and the virtual potential of the edge toward the adjacent node (i+1,j) is "+65 (=+(60+5))". The virtual potential assigned to the edge toward node (i,j+1) is "+70 (=+(60+10))". In S115, the processing unit 10 calculates a repulsive force of "115=+65+50" toward the edge toward the adjacent node (i+1,j) and a repulsive force of "120=+70+50" toward the edge toward node (i,j+1). The processing unit 10 uses these repulsive forces as weights and probabilistically selects an edge to one of the nodes (S116). Because it is a weighted lottery, it does not necessarily select the edge with the weakest repulsive force, but it will often select the edge with the weakest repulsive force.

[0107] Figure 21 shows the state of the transport network after six steps have passed from the state shown in Figure 20. Figure 21A shows the state immediately after the step has passed, and Figure 21B shows the state after processing S112. The agent moves through a total of eight steps, following the transport route: node (i,j) → node (i+1,j) → node (i+1,j+1) → node (i,j+1) → node (i-1,j) → node (i-1,j) → node (i-2,j) → node (i-2,j-1) → node (i-2,j-2). Node (i-2,j-2) is the destination point G. At the time shown in Figure 21A, the number of items associated with the agent is 20 (fully loaded). The number of items required at node (i-2,j-2) where the agent is located at the time shown in Figure 21A is "40". In S111, the processing unit 10 unloads all the items from the cargo bed. As a result, the number of items associated with the agent after loading and unloading becomes "0 (zero)", and the assigned virtual charge is updated to "-50", which corresponds to a full load (S119).

[0108] As shown in Figure 21B, the processing in S119 reduces the number of items required at node (i-2, j-2) from "40" to "20", and the virtual potential applied to the edge leading to node (i-2, j-2) is basically reduced to "+20". As a result, the virtual potential of each edge is updated to a smaller value, especially around node (i-2, j-2). Based on the updated state of the virtual potential and virtual charge after loading and unloading, the processing unit 10 generates plasma in the plasma generator 2 and takes an image. At this time, the processing unit 10 adjusts the height of the electrode 22 corresponding to node (i-2, j-2) by lowering it from "40" to "20". As a result, the generated plasma channel may change, and the correction amount for each edge may differ from the amount shown in Figure 17. Figure 21B shows on the right the characteristics of the plasma channel generated after the update and the correction amount determined from the intensity of its emission.

[0109] In Figure 21B, the agent at node (i-2, j-2) calculates the attractive or repulsive forces from the edges leading to the adjacent node (i-2, j-3), the edge leading to node (i-2, j-1), and the edge leading to node (i-1, j-2). The virtual charge assigned to the agent has been updated to "-50", and the virtual potential of the edge leading to the adjacent node (i-2, j-3) is "-15 (= -(10+5))". The virtual potential assigned to the edge leading to node (i-2, j-1) is "-15 (= -(10+5))". Similarly, the virtual potential assigned to the edge leading to node (i-1, j-2) is "-12 (= -(10+2))". In S115, the processing unit 10 calculates a repulsive force of "-65 (=-15-50)" for the edge leading to the adjacent node (i-2, j-3), "-65 (=-15-50)" for the edge leading to node (i-2, j-1), and "-62 (=-12-50)" for the edge leading to node (i-1, j-2). The processing unit 10 uses these repulsive forces as weights and probabilistically selects an edge to one of the nodes (S116). Because it is a weighted lottery, it does not necessarily select the edge with the weakest repulsive force, but it will often select the edge with the weakest repulsive force.

[0110] In Figures 20 and 21, the virtual charge for the agent corresponding to a transport device fully loaded with goods is set to "+50". However, the virtual charge may be set to "-50" to reverse the polarity of the virtual potential and make it easier for the agent to reach the destination point G. In this case, for example, in Figure 20B, the virtual potential of node (i+1, j+1) at the collection point S where the agent is located may be reversed from "+100" to "-80" instead of "+80".

[0111] As shown in Figures 16-21, the process based on step-by-step agent-based modeling calculates the optimal path that minimizes the number of items at collection point S and destination point G as quickly as possible, or maximizes the total of the rewards (virtual potentials) given to and collected by the selected edges as quickly as possible. The selection of each edge may also be reinforced by performing reinforcement learning on the revenue as the sum of the rewards.

[0112] The processing unit 10 of the processing unit 1 takes the virtual potential applied to each edge as a reward r, and the reward obtained in the next step t+1 is r t+1 An action value (value of choice) function Q, anticipating the outcome, can be calculated using Equation 1 below, and the agent may be guided forward using reinforcement learning based on this Q value. As shown in Equation 1, the action value function Q can be expressed as a combination of the reward r and the reward r+1 received for selecting an edge in the next step t+1, α. As shown in Equation 1, the action value function Q is a combination of the largest virtual potentials among the virtual potentials assigned to one or more edges, at a set ratio. The calculation of the action value function Q for each step is performed before S117 in the processing procedure shown in Figures 11-13. This allows the transport route of the transport device, which is the agent, to be derived early.

[0113]

number

[0114] Adjusting the learning rate α to around 0.1 was expected to shorten the time it took to derive the optimal solution. The discount rate γ mentioned above corresponds to how much importance is placed on the impact of rewards from the next step or the step after that, i.e., the expected future rewards. When deciding on subsequent actions in order to maximize potential future income, γ is increased. Conversely, when prioritizing immediate income, γ is decreased. Adjusting γ to around 0.9 was expected to shorten the time it took to derive the optimal solution.

[0115] In reinforcement learning, which calculates the action-value function Q described above, recovers virtual potential as reward r, and integrates the reward r as income to derive the optimal path, we found that incorporating analog computing using the plasma generator 2 can significantly reduce the computation time.

[0116] Thus, by incorporating analog computing into the plasma generator 2, even for paths that appear symmetrical and identical in the simplified transport network shown in Figure 2C, it is possible to introduce differences in selection probabilities and expedite the arrival of the optimal solution. Even though the plasma emission in the channels of the plasma generator 2 is symmetrical, there is a non-random bias in the generated channels, making it highly likely that these are the optimal paths. Agent-based modeling incorporating such analog computing makes it possible to find the optimal solution at high speed.

[0117] The processing device 1 stores information on the optimal transportation route calculated for each region, and the transportation route determination device 100 determines the overall transportation route connecting the regions, as well as the movement route from dispatch to arrival of the goods, based on the actual transportation request.

[0118] In the above embodiment, processing was performed on a single type of item. However, processing based on agent-based modeling can be performed on a transport route that carries multiple items, by moving multiple agents together one step at a time. In this case, the processing unit 1 can adjust the electrodes 22 according to the number of items of each type, emit plasma using the plasma generator 2 for each type of item, and perform calculations by superimposing the results.

[0119] Furthermore, when moving multiple agents simultaneously, it is advisable to move them one step at a time while calculating the attractive or repulsive forces with the virtual charges attached to other agents. This allows the process to proceed in a way that avoids agents attempting to move consecutively to the same collection point S or the same arrival point G.

[0120] In the above-described embodiment, a transport network consisting of nodes and edges arranged in a grid pattern was created, and a plasma generator 2 having channels 21 that mimicked this transport network was used. However, the channels 21 of the plasma generator 2 are not limited to mimicking a grid-like transport network as shown in Figure 2C. The region may be divided into small areas as shown in Figure 2B, and the positional relationships of the nodes may be scaled down relative to the actual geographical positions.

[0121] In the above-described embodiment, the processing unit 1 was described as performing processing based on agent-based modeling using the plasma generation results from the plasma generator 2. However, without performing processing based on agent-based modeling, and further omitting the application of virtual potentials, the processing unit 1 may create a path that preferentially selects the edges where plasma was generated, using the results of plasma generation in the plasma generator 2 using channels that have positive and negative potentials on the electrodes according to the amount of goods to be sent from each point or the amount of goods to be delivered to each point.

[0122] The embodiments disclosed above are illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims, and all modifications within the meaning and scope equivalent to the claims are included. [Explanation of Symbols]

[0123] 100 Transportation route determination device 1 Processing Unit 10 Processing Unit 11 Storage section 13. Photography Department P1 Information Processing Program 2. Plasma generator 21 channels 22 electrodes 23 Channel formation section

Claims

1. In a method for determining a transportation route in which a computer determines a transportation route from a plurality of routes that connect a plurality of points, defined for each predetermined region, including a point of dispatch for an item, points that the transportation equipment transporting the item can pass through, and a point of arrival of the item within the region, in the order in which the transportation equipment travels, A plasma generator having channels that mimic nodes corresponding to the aforementioned multiple locations and edges between each node, wherein plasma is generated within the channels by giving electrodes corresponding to each node a positive or negative potential corresponding to the amount of material at each location, A transport route determination method that selects a route that passes through the node and edge where plasma emission occurs.

2. A transportation route determination method in which a computer determines a transportation route by performing a simulation process based on agent-based modeling, using a transportation network defined for each predetermined region, which assigns a shipping point for an item, a point that can be traversed by transportation equipment transporting the item, and a destination point for the item within the region to nodes, and assigns a route connecting the points to the edges between the nodes, wherein the transportation equipment is treated as an agent moving between the nodes, and the transportation route is determined by performing a simulation process based on agent-based modeling. The computer, as one step of the simulation process, For each node, a virtual potential is assigned to the node or the edge connected to the node, and to the surrounding nodes or edges, according to the quantity of goods to be shipped from the point corresponding to that node. For each node, a virtual potential with the opposite polarity to the virtual potential is applied to the node or the edge connected to the node, and to the surrounding nodes or edges, in accordance with the quantity of goods to be transported to the point corresponding to that node. A plasma generator having channels corresponding to the nodes and edges of the aforementioned transport network, wherein plasma is generated within the channels, From the captured image of the generated plasma emission, the portion corresponding to the emitted node or edge is identified. The virtual potential applied to the node or edge corresponding to the identified portion is corrected based on the correction amount of the virtual potential. The transport equipment corresponding to the agent applies a virtual charge to the agent in proportion to the number of items being transported. In order for the agent to move from the node where it is located to one or more adjacent nodes, the probability of selecting an edge to one or more adjacent nodes is determined based on the attractive or repulsive force to the agent's virtual charge from the corrected virtual potential of the edge to one or more adjacent nodes. Based on the aforementioned probability, one edge is probabilistically selected from the edges to the one or more other nodes. The quantity of goods at the node to which the agent has moved via the selected edge, and the quantity of goods being transported by the transport equipment corresponding to the agent are updated. Execute the process, The aforementioned computer, The process described in step 1 is repeated until the transportation of the goods from the shipping point to the destination point is completed. The history of the agent's movement of the node, obtained by repeating the above step, is stored as a single transport route. A method for determining the transport route to execute a process.

3. The computer, in each step, selects an edge using reinforcement learning based on an evaluation function obtained by combining the virtual potential applied to the selected edge with the largest virtual potential among the virtual potentials applied to one or more edges connected to the node beyond the selected edge, in a set ratio. A method for determining a transport route according to claim 2, which involves performing a process.

4. A transportation route determination device that determines a transportation route by executing a simulation process based on agent-based modeling, using a transportation network defined for each predetermined region, which assigns a shipping point for an item, a point that can be traversed by transportation equipment transporting the item, and a destination point for the item within the region to nodes, and assigns a route connecting the points to the edges between the nodes, wherein the transportation equipment is treated as an agent moving between the nodes, and the transportation route determination device determines the transportation route by executing a simulation process based on agent-based modeling, A processing unit that performs the aforementioned simulation process, The system includes a plasma generator having channels corresponding to the nodes and edges of the aforementioned transport network, The processing apparatus, as one step of the simulation process, For each node, a virtual potential is assigned to the node or the edge connected to the node, and to the surrounding nodes or edges, according to the quantity of goods to be shipped from the point corresponding to that node. For each node, a virtual potential with the opposite polarity to the virtual potential is applied to the node or the edge connected to the node, and to the surrounding nodes or edges, in accordance with the quantity of goods to be transported to the point corresponding to that node. The plasma generator generates plasma in the channel, From the captured image of the generated plasma emission, the portion corresponding to the emitted node or edge is identified. The virtual potential applied to the node or edge corresponding to the identified portion is corrected based on the correction amount of the virtual potential. The transport equipment corresponding to the agent applies a virtual charge to the agent in proportion to the number of items being transported. In order for the agent to move from the node where it is located to one or more adjacent nodes, the probability of selecting an edge to one or more adjacent nodes is determined based on the attractive or repulsive force to the agent's virtual charge from the corrected virtual potential of the edge to one or more adjacent nodes. Based on the aforementioned probability, one edge is probabilistically selected from the edges to the one or more other nodes. The quantity of goods at the node to which the agent has moved via the selected edge, and the quantity of goods being transported by the transport equipment corresponding to the agent are updated. Execute the process, The aforementioned processing apparatus is The process described in step 1 is repeated until the transportation of the goods from the shipping point to the destination point is completed. The history of the agent's movement of the node, obtained by repeating the above step, is stored as a single transport route. Transportation route determination device.

5. The plasma generator has a resin substrate into which channels are formed, including a plurality of holes which are bottomed cylindrical cavities corresponding to the nodes arranged in a grid pattern, and trenches corresponding to the edges formed to connect adjacent holes, and generates plasma in a chamber covered with a transparent lid. The transport route determination device according to claim 4.

6. The plasma generator includes, for each of the plurality of holes, an electrode whose height can be changed, positioned opposite to it across the bottom surface. The processing device adjusts the height of the electrode corresponding to each node according to the number of items to be shipped from the point corresponding to each node, or the quantity of items to be transported to the point corresponding to each node. A transport route determination device according to claim 5.

7. The plasma generator includes a switch to select whether to connect the electrode to a high-voltage power source, to ground it via a predetermined resistor, or to leave it in a floating state. The processing device connects the high-voltage power source to the electrodes corresponding to the nodes that are the shipping and receiving points, grounds the other electrodes, and leaves the electrodes corresponding to nodes that are not in the transport network floating. A transport route determination device according to claim 6.

8. A computer program that uses a transportation network defined for each predetermined region, assigning nodes to the shipping point of an item, points that can be traversed by the transportation equipment transporting the item, and the destination point of the item within the region, and assigning the routes connecting the points to the edges between the nodes, and causing the transportation equipment to execute a simulation process based on agent-based modeling as agents moving between the nodes, As one step of the simulation process, the computer will be configured to: For each node, a virtual potential is assigned to the node or the edge connected to the node, and to the surrounding nodes or edges, according to the quantity of goods to be shipped from the point corresponding to that node. For each node, a virtual potential with the opposite polarity to the virtual potential is applied to the node or the edge connected to the node, and to the surrounding nodes or edges, in accordance with the quantity of goods to be transported to the point corresponding to that node. A plasma generator having channels corresponding to the nodes and edges of the aforementioned transport network, wherein plasma is generated within the channels, From the captured image of the generated plasma emission, the portion corresponding to the emitted node or edge is identified. The virtual potential applied to the node or edge corresponding to the identified portion is corrected based on the correction amount of the virtual potential. The transport equipment corresponding to the agent applies a virtual charge to the agent in proportion to the number of items being transported. In order for the agent to move from the node where it is located to one or more adjacent nodes, the probability of selecting an edge to one or more adjacent nodes is determined based on the attractive or repulsive force to the agent's virtual charge from the corrected virtual potential of the edge to one or more adjacent nodes. Based on the aforementioned probability, one edge is probabilistically selected from the edges to the one or more other nodes. The quantity of goods at the node to which the agent has moved via the selected edge, and the quantity of goods being transported by the transport equipment corresponding to the agent are updated. The process is executed, and further, To the aforementioned computer, The process described in step 1 is repeated until the transportation of the goods from the shipping point to the destination point is completed. The history of the agent's movement of the node, obtained by repeating the above step, is stored as a single transport route. A computer program that executes a process.

Images