Data processing system, data processing method, and program

Abstract

To provide a data processing system, data processing method, and program capable of generating comprehensive scenarios. [Solution] The data processing system according to this disclosure includes a scenario generation unit that arranges agents in a multi-agent simulation, the simulation environment in which the agents are active, and the characteristics of the agents or the environment as nodes in a graph structure, and generates a graph structure as a simulation scenario by connecting a first node representing an agent or a second node representing the environment and a third node representing the characteristics.

Description

Technical Field

[0004] ,

[0005] , ,

[0001] The present disclosure relates to a data processing system, a data processing method, and a program.

Background Art

[0002] In recent years, technologies related to scenario generation for executing simulations have been disclosed. For example, Non-Patent Document 1 discloses a technique for generating a traffic scenario with a graph structure using an LLM (Large Language Model). In Non-Patent Document 1, vehicle data can be extracted from open data accumulated as real data, and a corner scenario, that is, a rare and dangerous scenario, can be generated. Open data is, for example, the video of a drive recorder.

Prior Art Documents

Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The scenarios generated in Non-Patent Document 1 are based on the positional relationships of vehicles and pedestrians on the road included in the open data. Therefore, the technique according to Non-Patent Document 1 cannot generate scenarios based on information that is not publicly available as open data. Therefore, the technique according to Non-Patent Document 1 has a problem that it cannot generate scenarios with high comprehensiveness.

[0005] This disclosure was made to solve these problems and aims to provide a data processing system, data processing method, and program that can generate comprehensive scenarios. [Means for solving the problem]

[0006] The data processing system relating to this disclosure includes a scenario generation unit that arranges agents in a multi-agent simulation, the simulation environment in which the agents are active, and the characteristics of the agents or the environment as nodes in a graph structure, and generates the graph structure as a scenario in the simulation by connecting a first node representing the agents or a second node representing the environment and a third node representing the characteristics.

[0007] The data processing method relating to this disclosure involves a computer arranging agents in a multi-agent simulation, the simulation environment in which the agents are active, and the characteristics of the agents or the environment as nodes in a graph structure, and generating a graph structure as a scenario in the simulation by connecting a first node representing the agents or a second node representing the environment and a third node representing the characteristics.

[0008] The program relating to this disclosure causes a computer to perform the following steps: arranging agents in a multi-agent simulation, the simulation environment in which the agents are active, and the characteristics of the agents or the environment as nodes in a graph structure; and generating a graph structure as a scenario in the simulation by connecting a first node representing the agents or a second node representing the environment and a third node representing the characteristics. [Effects of the Invention]

[0009] This disclosure provides a data processing system, data processing method, and program capable of generating comprehensive scenarios. [Brief explanation of the drawing]

[0010] [Figure 1] Figure 1 is a block diagram showing the configuration of the data processing system 1 related to this disclosure. [Figure 2] Figure 2 shows an example of a graph structure generated by the scenario generation unit 11. [Figure 3] Figure 3 is a flowchart showing an example of the processing operation of the data processing system 1. [Figure 4] Figure 4 is a block diagram showing the configuration of the data processing system 2 related to this disclosure. [Figure 5] Figure 5 is a table showing statistical information on the connection frequency between edges and nodes. [Figure 6] Figure 6 is a histogram showing statistical information on the connection frequency between edges and nodes. [Figure 7] Figure 7 is a flowchart showing an example of the processing operation of the data processing system 2. [Figure 8] Figure 8 is a flowchart showing an example of the flow when the scenario generation unit 21 generates a scenario based on similarity. [Figure 9] Figure 9 is a block diagram showing the configuration of the data processing system 3 related to this disclosure. [Figure 10] Figure 10 is a flowchart showing an example of the scenario generation process in the data processing system 3. [Figure 11] Figure 11 is a flowchart showing an example of scenario recording processing in data processing system 3. [Figure 12] Figure 12 is a flowchart showing an example of the scenario generation process in the data processing system 3. [Figure 13] Figure 13 is a schematic diagram showing an example of simulation results based on a scenario by the data processing system 3. [Figure 14]Figure 14 shows an example of the hardware configuration of the data processing system 4 related to this disclosure. [Modes for carrying out the invention]

[0011] (Embodiment 1) Embodiment 1 of the present disclosure will be described below with reference to the drawings. Figure 1 is a block diagram showing the configuration of the data processing system 1 according to the present disclosure. The data processing system 1 is a system for generating scenarios for executing multi-agent simulations. A multi-agent simulation is a simulation performed using a model in which multiple autonomous agents interact. A multi-agent simulation is used to simulate, for example, pedestrian flow or traffic flow. Here, the agents may be people, animals, cars driven by people, autonomous robots, or even microscopic things such as pollen or bacteria. In other words, the agents include any object whose individual characteristics and behaviors can be identified and designed. Hereafter, multi-agent simulation may be simply referred to as "simulation". The data processing system 1 includes a scenario generation unit 11.

[0012] The scenario generation unit 11 arranges agents, environments, and the characteristics of agents or environments in a multi-agent simulation as nodes in a graph structure, and generates a graph structure in which nodes representing agents or environments and nodes representing characteristics are connected, as a scenario in the simulation. In other words, the scenario generation unit 11 generates a scenario in the form of a graph structure. To put it another way, the scenario generation unit 11 generates a graph structure as a scenario in the simulation. A graph structure is a data structure composed of multiple nodes and edges that represent the relationships between nodes. In this embodiment 1, generating a graph structure and generating a scenario are synonymous.

[0013] The scenario generation unit 11 can generate a knowledge graph as a scenario. A knowledge graph is a graph structure that represents the connections between various types of knowledge. That is, the scenario generated by the scenario generation unit 11 may represent the connections between knowledge. Knowledge can also be expressed as external knowledge. For example, in the case of pedestrian flow simulation, the knowledge may be building information or road information.

[0014] The scenario generation unit 11 arranges agents and the environment as nodes. In other words, the scenario generation unit 11 defines agents and the environment as nodes. These nodes are respectively referred to as agent nodes and environment nodes. As described above, agent nodes may be people, animals, automobiles, or autonomous driving robots. Environment nodes are mainly environmental elements in a simulation based on geographical spatial information. That is, environment nodes are environmental elements within the simulation in which agent nodes act. Specifically, environment nodes are intersections and buildings. That is, while agent nodes are dynamic elements, environment nodes can be regarded as static elements.

[0015] The scenario generation unit 11 arranges the characteristics of agents or the environment as nodes. In other words, the scenario generation unit 11 defines the characteristics as nodes. This node is referred to as a characteristic node. The characteristics of an agent are the roles, attributes, decision-making types, personalities, etc. that an individual agent has. That is, the characteristics of an agent are elements that affect the behavior patterns of individual agents in a simulation. In other words, agents with different characteristics have different behavior patterns in a simulation.

[0016] Here, let's consider an example where the scenario generation unit 11 generates scenarios for disaster simulation. In this case, the roles of the agents may be classified by, for example, their occupation. That is, the roles of the agents may be, for example, "residents (ordinary citizens)" or "guides." Furthermore, the attributes of the agents may be classified by, for example, their state, age, driving speed, or field of vision. The attributes of the agents may be, for example, "elderly" or "disabled." In addition, the decision-making types of the agents may be classified by, for example, their familiarity with the area. The decision-making types of the agents may be, for example, "familiar with the area," "somewhat familiar with the area," or "unfamiliar with the area."

[0017] Environmental characteristics refer to the size and condition of individual environmental elements. Similarly, consider the case where the scenario generation unit 11 generates a scenario for disaster simulation. In this case, environmental nodes are, for example, intersections and buildings. In this case, the environmental characteristics may also be the size of the intersection or whether it is passable. Furthermore, the environmental characteristics may also be the structural type of the building, the number of floors in the building, the agent capacity, or the building's purpose.

[0018] Here, the scenario generation unit 11 does not need to provide environment characteristic nodes. In other words, the scenario generation unit 11 may define characteristics individually for each environment node.

[0019] The scenario generation unit 11, which has agent nodes, environment nodes, and characteristic nodes, connects each of the placed nodes by edges. Specifically, the scenario generation unit 11 connects any environment node to other environment nodes. The scenario generation unit 11 also connects any agent node to other agent nodes. Furthermore, the scenario generation unit 11 connects any environment node to any agent node. Furthermore, the scenario generation unit 11 connects any agent node or environment node to a characteristic node. The scenario generation unit 11 may or may not connect any characteristic node to other characteristic nodes.

[0020] Edges connecting any environment node to other environment nodes build the environment network in the scenario. For example, if two intersections are defined as environment nodes, connecting these intersections allows the connected edges to be considered a road. Additionally, relationships between agents can be defined by connecting any agent node to other agent nodes. For example, if two residents are defined as two agent nodes, connecting these residents defines that they know each other. This relationship may include friendships, family relationships, etc. Furthermore, the initial placement of agents can be defined by connecting any environment node to any agent node. For example, if an intersection and a resident are defined as an environment node and an agent node, respectively, connecting these nodes allows the initial placement of the resident to be defined as the intersection.

[0021] Furthermore, by connecting any agent node or environment node with characteristic nodes, the characteristics of each agent or environment can be defined. For example, consider a case where a resident is defined as an agent node, and multiple different characteristic nodes are defined as the agent's decision-making type. In this case, the decision-making type of that resident can be defined by connecting the agent node with any decision-making type.

[0022] Here, the scenario generation unit 11 may connect nodes and edges. For example, an agent node may be connected to an edge that connects an environment node to another environment node. In this case, the edge can be defined as the initial placement location of the agent node connected to that edge. In this case, the connection point of the agent node at the edge may be defined as a new environment node.

[0023] The scenario generation unit 11 connects nodes according to predetermined rules. These predetermined rules may also be referred to as predetermined preconditions. The predetermined rules prevent connections between specific nodes. For example, if there is an environment node that is not intended as the initial placement destination for a certain agent node, the predetermined rules shall include a provision that the agent node will not be placed in that environment node. The predetermined rules may also include a provision that there is only one connection between an agent node and an environment node, or a provision that prevents connections between environment nodes that are not intended as part of the simulation environment. The predetermined rules may be determined in advance or by a probability distribution. When the predetermined rules are determined by a probability distribution, the probability distribution may be set such that the connection probability between nodes whose connections are to be prevented is 0.

[0024] Furthermore, the scenario generation unit 11 may or may not randomly connect nodes to each other. In other words, the scenario generation unit 11 may randomly determine the connection relationships for a given node, or it may intentionally determine the connection relationships. That is, if there are multiple nodes of the same type that can be connected to a given node, the scenario generation unit 11 may determine which node to connect to using a uniform probability distribution, or it may determine which node to connect to using a weighted probability distribution. Here, the type of node may refer to classifications based on the type of node, such as agent nodes, environment nodes, and characteristic nodes, or it may refer to classifications based on the characteristics of an agent node or an environment node.

[0025] Here, a specific example of the graph structure generated by the scenario generation unit 11 will be explained using a diagram. Figure 2 is an example of a graph structure generated by the scenario generation unit 11. Figure 2 is the graph structure of a scenario in a human flow simulation during a disaster. In Figure 2, the agent node consists of three resident nodes and one guide node. The role of each agent is defined in each agent node. That is, the agent's attributes are not placed on the graph structure of Figure 2 as characteristic nodes. In addition, the environment node consists of five intersection nodes and one building (shelter) node.

[0026] In Figure 2, the characteristic node consists of three decision nodes and three attribute nodes. The decision nodes represent, for example, the agent's sense of direction. Specifically, the three decision nodes represent "having a sense of direction and moving to the nearest destination via the shortest route," "heading to the most famous destination," and "having no sense of direction and following the person in front of them." The attribute nodes represent, for example, the agent's state. Specifically, the three attribute nodes represent "able-bodied," "elderly," and "disabled."

[0027] The scenario generation unit 11 can define the environment in a scenario by connecting environment nodes. For example, a road network can be constructed by connecting intersection nodes to other intersection nodes. In addition, the location of buildings (shelters) on the road network can be defined by placing building (shelter) nodes.

[0028] Furthermore, the scenario generation unit 11 can determine the initial placement of agents by connecting agent nodes and environment nodes. For example, by connecting a resident node and an intersection node, residents can be placed at an intersection. The scenario generation unit 11 can also define the relationships between agents by connecting agent nodes. For example, by connecting two resident nodes, it can be defined that those residents are friends. This can be interpreted as the residents joining forces and acting together in the simulation.

[0029] Furthermore, the scenario generation unit 11 can define the characteristics of each agent by connecting agent nodes and characteristic nodes. For example, by connecting a resident node and a decision-making node, the resident's sense of the land can be defined. Also, by connecting a resident node and an attribute node, the resident's state can be defined. Note that in Figure 2, nodes indicating environmental characteristics are not placed.

[0030] Next, the processing flow of the data processing system 1 will be explained. Figure 3 is a flowchart showing an example of the processing operation of the data processing system 1. First, the scenario generation unit 11 places agent nodes, environment nodes, and characteristic nodes (S101). Here, the scenario generation unit 11 may place any of the agent nodes, environment nodes, or characteristic nodes in any order. In other words, the order in which the scenario generation unit 11 places the nodes is arbitrary. Next, the scenario generation unit 11 connects the nodes or edges according to predetermined rules (S102). That is, the scenario generation unit 11 connects any agent node, environment node, or characteristic node. Here, the scenario generation unit 11 may connect nodes and edges. Also, the scenario generation unit 11 may connect any of the agent nodes, environment nodes, or characteristic nodes in any order. In other words, the order in which the scenario generation unit 11 connects the nodes is arbitrary. Through these steps, the data processing system 1 generates a graph structure as a scenario.

[0031] Thus, the data processing system 1 according to this embodiment 1 can reflect the characteristics of agents or environments in a scenario by arranging the characteristics of agents or environments as nodes in a graph structure and connecting them to agent nodes or environment nodes. The graph structure based on the technology described in Non-Patent Document 1 is generated using data that has been utilized as open data, so it is possible to reflect physical situations such as the relationship between vehicles and roads and the positional relationship between vehicles in the scenario, but it is not possible to consider internal parameters such as the personality and function of agents and reflect them directly in the scenario. This is because such internal parameters cannot be found in open data. As a result, this technology may not be able to make the most of the functions that advanced simulations possess. This is because it is difficult to model agents etc. that take into account their personality and function from open data. In other words, this technology has difficulty directly generating scenarios that go beyond the information in open data.

[0032] The data processing system 1 according to this embodiment 1 can directly reflect internal parameters such as the role, attributes, personality, and decision-making of agents, which go beyond the information in open data, into the scenario. Therefore, the data processing system 1 can generate agents with various behaviors for each scenario, enabling the execution of more complex simulations. Furthermore, the data processing system 1 can define the structure of the graph in the scenario and the elements within the graph in accordance with the functions and computational processing of the simulation. Therefore, the data processing system 1 can make the most of the functions of advanced simulations. As a result, the data processing system 1 can generate comprehensive scenarios.

[0033] The agent characteristics that can be reflected in the scenarios generated by data processing system 1 include, for example, the agent's role, agent attributes, and agent decision-making type. By reflecting these as agent characteristics in the scenarios, the agent's behavior patterns during simulation can be changed for each scenario.

[0034] The technology described in Non-Patent Document 1 evaluates the feasibility of generated scenarios using an evaluation formula. This is to determine whether there are any inconsistencies in the procedure or content of events occurring in the generated scenarios, or whether they are appropriate scenarios to be considered. The technology described in Non-Patent Document 1 eliminates scenarios with low feasibility by evaluating the generated scenarios. However, the calculation using the evaluation formula becomes more complex depending on the scale of the simulation, and in the case of large-scale simulations, it becomes necessary to run the simulation in order to perform the evaluation, resulting in the problem of high computational costs for evaluation.

[0035] However, in the data processing system 1 according to this embodiment 1, since nodes are connected according to predetermined rules, the feasibility of the scenarios can be ensured by generating scenarios that conform to the rules. In other words, the data processing system 1 can ensure the quality of the generated scenarios. That is, by connecting nodes according to predetermined rules, the data processing system 1 can prevent connections between specific nodes. This eliminates the occurrence of impossible scenario conditions or their inclusion within a scenario. This allows simulations using the scenarios to operate appropriately. To put it another way, the data processing system 1 can pre-filter the scenarios to be generated by generating them according to the rules. As a result, the data processing system 1 can suppress the computational costs incurred by the technology described in Non-Patent Literature 1.

[0036] Furthermore, while the data processing system 1 according to this embodiment 1 generates scenarios using a graph structure, related technologies can generate scenarios in tabular format. In the case of tabular scenarios, it is usually necessary to prepare multiple tabular files. This is to organize the types of data, make them easier to understand, and to facilitate data modification. When preparing tabular data, multiple external knowledge sources are required to control the input of inappropriate values ​​(such as values ​​outside the definition). In this case, if the reference destination of one knowledge source in the scenario goes through another knowledge source, that is, if the data reference is a so-called multi-hop, the data processing becomes complicated. In contrast, in the case of graph-structured scenarios generated by the data processing system 1, the scenario can be represented by a single graph structure, eliminating the need to prepare multiple files and simplifying data processing.

[0037] Furthermore, even in the case of tabular scenarios, geospatial information may be required. Geospatial information data resources are often in the form of graphs, such as road networks. Therefore, when generating tabular scenarios using geospatial information, it becomes necessary to manually convert the data structure of the graph-structured geospatial information using a GIS (Geographic Information System) or similar method. On the other hand, by using data processing system 1, scenarios can be automatically generated using geospatial information.

[0038] Even when using tabular scenarios, it is possible to consider cases where graph-structured geospatial information is not used. In this case, it may not be possible to determine whether all the components of the scenario are included based solely on tabular data. For example, if data for a particular road is not present in the scenario for a road network, a scenario including that road will not be generated. On the other hand, by generating graph-structured scenarios using data processing system 1, it is possible to determine whether all the components of the scenario are included.

[0039] Furthermore, while related technologies include methods for generating scenarios in text (document) format, these scenarios need to be converted to tabular or other formats before being input into a simulation. Additionally, considering that not all the information necessary to run the simulation may be contained within the text, when using text-format scenarios, it is necessary to define a process for interpolating missing parts or to prepare a database in advance. On the other hand, generating graph-structured scenarios using data processing system 1 avoids these problems.

[0040] (Embodiment 2) Next, Embodiment 2 of the present disclosure will be described. Figure 4 is a block diagram showing the configuration of the data processing system 2 according to the present disclosure. The data processing system 2 comprises a scenario generation unit 21, a feature collection unit 22, a storage unit 23, and a generated scenario evaluation unit 24. The data processing system 2, like the data processing system 1, is a system for generating scenarios for executing multi-agent simulations. That is, the data processing system 2 has the addition of the feature collection unit 22, the storage unit 23, and the generated scenario evaluation unit 24 to the data processing system 1. Duplication between the scenario generation unit 21 in the data processing system 2 and the scenario generation unit 11 in the data processing system 1 will be omitted as appropriate.

[0041] The scenario generation unit 21 may generate scenarios based on data whose original data structure is a graph structure, or it may generate scenarios based on tabular data. In other words, the scenario generation unit 21 may generate a knowledge graph as a scenario based on tabular data. Hereafter, tabular and graph structures will be treated uniformly as scenarios.

[0042] The scenario generation unit 21 can generate scenarios based on the feature quantities of scenarios previously generated by the scenario generation unit 21. Feature quantities represent the characteristics of the scenarios generated by the scenario generation unit 21. In particular, feature quantities are used for comparison when comparing scenarios generated by the scenario generation unit 21. Feature quantities can also be called feature representations. There may be one or more feature quantities for a single scenario.

[0043] The reason why a single scenario may have multiple features is as follows: A single scenario is data that contains multiple parameters (conditions). Therefore, features may be defined for each of those parameters. Furthermore, the features of the scenario itself may be defined by a combination of multiple features.

[0044] Features may, for example, be connection distributions in a graph structure. A connection distribution is, for example, statistical information on the frequency of connections between nodes or edges in a graph structure. In other words, a connection distribution is, for example, statistical information on the frequency of connections between nodes and edges in a graph structure. The connection frequency statistics may be represented by a histogram or by a tabular format. The connection frequency statistics are typically discrete, but may also be continuous. In the following explanation, the connection frequency statistics will be treated as being represented by a discrete distribution.

[0045] This section describes statistical information regarding the connection frequency between nodes and edges in a graph structure. The scenario generation unit 21 may connect nodes to nodes, or nodes to edges. Here, when connecting nodes to nodes based on predetermined rules, it is assumed that there are multiple connectable nodes for a certain type of node. In this case, the scenario generation unit 21 may determine the nodes to connect using a uniform probability distribution, or it may determine the nodes to connect using a weighted probability distribution. The connection frequency between a node and other nodes is the degree to which a certain type of node repeatedly connects to the connectable nodes in question.

[0046] For example, consider a scenario generation in a human flow simulation where the scenario generation unit 21 determines the initial placement of residents by connecting agent nodes (residents) with environment nodes (intersections). Here, we assume there are multiple residents, and each resident can connect to multiple intersections. In this case, the scenario generation unit 21 connects each resident to an intersection according to some probability distribution. In this example, the connection frequency between nodes is, for example, the number of times a resident has connected to each intersection.

[0047] Furthermore, consider the case where the scenario generation unit 21 connects nodes and edges based on predetermined rules, and there are multiple connectable edges for a certain type of node. In this case, the connection frequency between nodes and edges is the degree to which a certain type of node repeatedly connects to that connectable edge.

[0048] For example, consider a scenario generation in a pedestrian flow simulation where the scenario generation unit 21 places residents on roads by connecting multiple resident nodes with roads, which are edges connecting multiple intersections. In this case, the connection frequency between nodes and edges may be the number of times a resident connects to each road.

[0049] Connection frequency statistics are aggregated data on the connection frequency between a node and other nodes, or between a node and an edge. The connection frequency statistics may be aggregated based only on the connection frequency in a single scenario, or they may be aggregated together with scenarios previously generated by the scenario generation unit 21.

[0050] In other words, the scenario generation unit 21 can generate scenarios based on statistical information of connection frequencies between nodes and edges in a graph structure related to a scenario previously generated by the scenario generation unit 21. For example, the scenario generation unit 21 can consider this statistical information as a probability distribution and generate scenarios based on this probability distribution.

[0051] The above will be explained exemplified using diagrams. Figure 5 is a table showing statistical information on the connection frequency between edges and nodes. Figure 6 is a histogram showing statistical information on the connection frequency between edges and nodes. For example, consider the case where the scenario generation unit 21 generates a scenario in a human flow simulation. When generating a scenario, the scenario generation unit 21 places residents (agent nodes) on roads (edges) that connect intersections (environment nodes). As a result of the scenario generation unit 21 placing residents on the roads, residents are placed on each road.

[0052] In Figures 5 and 6, the scenario generation unit 21 is assumed to place 100 residents, and there are 5 roads to which they can be placed. As a result of the scenario generation unit 21 placing the residents, 20 residents are placed on road 1, 10 on road 2, 40 on road 3, 20 on road 4, and 10 on road 5. As shown in Figure 5, the relationship between the roads and the number of residents placed can be represented in a table.

[0053] Here, the scenario generation unit 21 can consider the statistical information of the connection frequency between roads and the number of residents in Figure 5 as a probability distribution. That is, the scenario generation unit 21 can consider that the probability of being placed on road 1 is 0.2, the probability of being on road 2 is 0.1, the probability of being on road 3 is 0.4, the probability of being on road 4 is 0.2, and the probability of being on road 5 is 0.1.

[0054] This probability distribution can be represented by a histogram, as shown in Figure 6. In the histogram in Figure 6, the horizontal axis represents roads and the vertical axis represents the probability of resident placement. That is, the histogram shows that the placement probability of road 1 is 0.2, the placement probability of road 2 is 0.1, the placement probability of road 3 is 0.4, the placement probability of road 4 is 0.2, and the placement probability of road 5 is 0.1. In this way, the scenario generation unit 21 can consider the statistical information of the connection frequency between nodes and edges in the graph structure related to a scenario previously generated by the scenario generation unit 21 as a probability distribution and generate a scenario based on that probability distribution.

[0055] When the scenario generation unit 21 generates scenarios based on a probability distribution, it can generate scenarios with a probability distribution whose similarity is greater than a predetermined value. In other words, the scenario generation unit 21 can generate scenarios with similar probability distributions to a given scenario. The similarity may be calculated as a distance based on optimal transport theory, or as KL (Kullback-Leibler) information (divergence). That is, the scenario generation unit 21 can generate scenarios with a small distance based on optimal transport theory to a given scenario.

[0056] Furthermore, the scenario generation unit 21 can generate scenarios with probability distributions whose similarity is smaller than a predetermined value. In other words, the scenario generation unit 21 can generate scenarios whose probability distributions are not similar to a given scenario. If the similarity is calculated as a distance based on optimal transport theory, the scenario generation unit 21 can generate scenarios with a large distance based on optimal transport theory for a given scenario.

[0057] Returning to Figure 4, we continue the explanation of the configuration of the data processing system 2. The feature collection unit 22 collects the feature quantities of the scenarios generated by the scenario generation unit 21. The feature collection unit 22 may also collect statistical information on the connection frequency between nodes or edges in the graph structure and other nodes or edges as feature quantities of the scenario. In other words, the feature collection unit 22 may also collect statistical information on the connection frequency between nodes and edges as feature quantities of the scenario. Here, the connection frequency between nodes and edges can be referred to as the co-occurrence between nodes.

[0058] The feature collection unit 22 may directly extract the features of the scenario generated by the scenario generation unit 21 from the graph structure itself. For example, the feature collection unit 22 may extract features using a graph neural network. In this case, the graph neural network uses, for example, the graph structure and predetermined features as training data. The feature collection unit 22 may then collect features by inputting the graph structure into the graph neural network and extracting the features.

[0059] The storage unit 23 is an example of a non-volatile storage device such as a hard disk or flash memory. The storage unit 23 stores the feature quantities of the scenarios collected by the feature quantity collection unit 22 for the scenarios generated by the scenario generation unit 21. That is, if the feature quantities are statistical information on the connection frequency between nodes and edges, the storage unit 23 stores that statistical information. The storage unit 23 may also store the scenarios. In this case, the storage unit 23 may store the scenarios in association with their feature quantities. If a single scenario contains multiple feature quantities, the storage unit 23 may organize and store them together.

[0060] Furthermore, the memory unit 23 may store information necessary for the scenario generation unit 21 to generate a scenario. For example, the memory unit 23 may store information about nodes, or it may store predetermined rules for connecting nodes. Specifically, the memory unit 23 may store information about environment nodes, agent nodes, and characteristic nodes.

[0061] The generated scenario evaluation unit 24 calculates the similarity between the feature quantities of the scenario generated by the scenario generation unit 21 and the feature quantities of the scenario stored in the memory unit 23. Here, the generated scenario evaluation unit 24 may target one feature quantity or multiple feature quantities in the scenario generated by the scenario generation unit 21. If the feature quantities of the scenario are statistical information of the connection frequency between nodes and edges, the generated scenario evaluation unit 24 may calculate the similarity by treating the statistical information as a probability distribution. For example, the generated scenario evaluation unit 24 may calculate the similarity as a distance based on optimal transport theory, or as KL information (divergence). The generated scenario evaluation unit 24 also compares the calculated similarity and determines whether the similarity is greater than or less than a predetermined value. This function performed by the generated scenario evaluation unit 24 is called similarity evaluation.

[0062] The generated scenario evaluation unit 24 can perform modification processing on the scenario generated by the scenario generation unit 21. For example, when the scenario generation unit 21 generates a scenario, if there are multiple nodes of the same type that can be connected to a certain node, the generated scenario evaluation unit 24 may perform processing to modify the probability distribution used to determine the connection node.

[0063] Furthermore, the generated scenario evaluation unit 24 can perform a search process for scenarios stored in the memory unit 23 during the similarity evaluation. For example, the generated scenario evaluation unit 24 can present scenarios stored in the memory unit 23 to the user. Specifically, the generated scenario evaluation unit 24 may present to the user some or all of the scenarios that have a similarity of a certain value or higher to the generated scenario. In addition, the generated scenario evaluation unit 24 can sort the scenarios that were compared, i.e., the scenarios stored in the memory unit 23, according to their similarity to the generated scenario.

[0064] Next, the processing flow of the data processing system 2 will be explained. Figure 7 is a flowchart showing an example of the processing operation of the data processing system 2. First, the scenario generation unit 21 arranges agent nodes, environment nodes, and characteristic nodes (S201). Here, the scenario generation unit 21 may arrange the agent nodes, environment nodes, and characteristic nodes in any order. In other words, the order in which the scenario generation unit 21 arranges the nodes is arbitrary. Next, the scenario generation unit 21 connects the nodes or edges according to predetermined rules (S202). That is, the scenario generation unit 21 connects any agent nodes, environment nodes, and characteristic nodes. Here, the scenario generation unit 21 may connect nodes and edges. Also, the scenario generation unit 21 may connect the agent nodes, environment nodes, and characteristic nodes in any order. In other words, the order in which the scenario generation unit 21 connects the nodes is arbitrary. After that, the feature collection unit 22 collects the features of the scenario (S203). Subsequently, the memory unit 23 stores the feature quantities of the scenario (S204).

[0065] Next, we will explain the flow when the scenario generation unit 21 generates a scenario based on similarity. Figure 8 is a flowchart showing an example of the flow when the scenario generation unit 21 generates a scenario based on similarity. First, the scenario generation unit 21 generates a scenario based on predetermined rules and a predetermined probability distribution (S205). Next, the feature collection unit 22 collects the features of the generated scenario (S206). After that, the generated scenario evaluation unit 24 refers to the memory unit 23 and calculates the similarity between the features of the scenarios stored in the memory unit 23 and the features of the scenarios generated by the scenario generation unit 21 (S207). Here, the generated scenario evaluation unit 24 may target all scenarios stored in the memory unit 23, or it may target some of the scenarios. Also, the generated scenario evaluation unit 24 may calculate the similarity between all the features of the scenarios stored in the memory unit 23, or it may calculate the similarity between some of the features.

[0066] Next, the generated scenario evaluation unit 24 determines whether the purpose is to generate similar scenarios (S208). That is, the generated scenario evaluation unit 24 determines whether the scenario generation unit 21 aims to generate scenarios similar to those stored in the memory unit 23. Here, aiming to generate similar scenarios means aiming to generate scenarios with features similar to those stored in the memory unit 23. Conversely, not aiming to generate similar scenarios means aiming to generate scenarios with features that are not similar to those stored in the memory unit 23.

[0067] If the purpose of this scenario generation is to generate similar scenarios, the generated scenario evaluation unit 24 determines whether the similarity between scenarios is above a certain value (S209). Here, if the similarity is considered as the distance according to optimal transport theory, the generated scenario evaluation unit 24 determines whether the distance between scenarios is below a certain value. If the similarity is above a certain value, the storage unit 23 stores the feature quantities of the scenarios (S212). If the similarity is not above a certain value, the generated scenario evaluation unit 24 performs a correction process (S211). The correction process by the generated scenario evaluation unit 24 is, for example, a process to correct the probability distribution for determining the connection nodes. After the correction process, the scenario generation unit 21 generates a scenario again (S205). The flow is repeated until the similarity of the scenarios is above a certain value. If a scenario that satisfies the similarity requirement is generated, the storage unit 23 stores the feature quantities of the scenarios (S212). Here, the generated scenario evaluation unit 24 may present the scenarios stored in the memory unit 23 to the user, or it may sort the scenarios stored in the memory unit 23.

[0068] If the purpose of this scenario generation is not to generate similar scenarios, the generated scenario evaluation unit 24 determines whether the similarity between scenarios is below a certain value (S210). Here, if similarity is considered as distance according to optimal transport theory, the generated scenario evaluation unit 24 determines whether the distance between scenarios is above a certain value. If the similarity is below a certain value, the storage unit 23 stores the feature quantities of the scenarios (S212). If the similarity is above a certain value, the generated scenario evaluation unit 24 performs correction processing (S211). After the correction processing, the scenario generation unit 21 generates a scenario again (S205). The flow is repeated until the similarity of the scenarios is below a certain value. If a scenario that satisfies the similarity requirement is generated, the storage unit 23 stores the feature quantities of the scenarios (S212). Here, the generated scenario evaluation unit 24 may present the scenarios stored in the storage unit 23 to the user, or it may sort the scenarios stored in the storage unit 23.

[0069] Thus, the data processing system 2 according to this embodiment 2 collects feature quantities from scenarios generated by the scenario generation unit 21 and stores these feature quantities, thereby enabling comparison of scenarios based on their features. By comparing the feature quantities of scenarios generated by the scenario generation unit 21 with the feature quantities of scenarios stored in the storage unit 23, the scenario generation unit 21 can generate scenarios based on the feature quantities of previously generated scenarios.

[0070] Scenarios with similar features mean that the content and structure of the scenarios are similar. Conversely, scenarios with dissimilar features mean that the content and structure of the scenarios are not similar. In other words, by generating scenarios based on features, it is possible to generate scenarios based on comparisons with scenarios that have been generated so far.

[0071] In related technologies, the manual setting of scenario variables and parameters resulted in high labor costs for scenario generation. This was because the scenario generation process was meticulously carried out based on the expertise of specialists. Furthermore, the increasing sophistication of simulation capabilities led to a massive increase in the number of parameters included in scenarios, making it difficult to generate appropriate scenarios.

[0072] According to the data processing system 2 of this embodiment 2, the scenario generation process can be automated, thereby reducing the cost of simulation operation. Furthermore, by enabling scenario generation based on feature quantities, it becomes possible to generate scenarios that cannot be explored manually, resulting in the generation of more comprehensive scenarios.

[0073] The scenario generation evaluation unit 24 of the data processing system 2 can calculate the similarity between the feature quantities of the scenarios generated by the scenario generation unit 21 and the feature quantities of the scenarios stored in the storage unit 23, and evaluate the similarity between scenarios. As a result, the scenario generation unit 21 can generate scenarios that match the target similarity. In other words, the scenario generation unit 21 can generate scenarios with high similarity or scenarios with low similarity.

[0074] One of the purposes of the generated scenario evaluation unit 24 in evaluating scenarios with high similarity is to search for existing scenarios stored in the memory unit 23. For example, if the scenario generation unit 21 generates a scenario that defines a real-time situation, the generated scenario evaluation unit 24 can search the memory unit 23 for scenarios with high similarity and extract a scenario from the memory unit 23 that is close to the current situation.

[0075] On the other hand, one of the purposes of the generated scenario evaluation unit 24 in evaluating scenarios with low similarity is to search for scenarios stored in the memory unit 23 and generate scenarios with different conditions and content from those scenarios. In other words, by evaluating scenarios with low similarity, the generated scenario evaluation unit 24 can explore and generate unknown scenarios.

[0076] When the features related to the data processing system 2 are statistical information of the connection frequency between nodes and edges, the data processing system 2 can compare scenarios using this statistical information. In particular, by considering this statistical information as a probability distribution, similar scenarios can be generated using this probability distribution. For example, if it is desired to generate a scenario similar to one stored in the memory unit 23, the scenario generation unit 21 can generate a similar scenario by determining the connection relationships between nodes or edges using a probability distribution similar to the probability distribution related to that scenario. Conversely, if it is desired to generate a scenario that is significantly different from, or dissimilar to, one stored in the memory unit 23, the scenario generation unit 21 can generate a dissimilar scenario by determining the connection relationships using a probability distribution that is not similar to the probability distribution related to that scenario.

[0077] When the feature vectors are used as statistical information representing the connection frequency between nodes and edges, and this statistical information is considered as a probability distribution, and the generated scenario evaluation unit 24 calculates the similarity of the feature vectors as a distance based on optimal transport theory, the data processing system 2 can quantitatively calculate the similarity of the scenarios. As a result, the data processing system 2 can intentionally generate similar scenarios and untested scenarios.

[0078] Furthermore, data processing system 2 can compare scenarios from multiple perspectives by directly extracting features from the graph structure itself. Data processing system 2 can extract features using, for example, a graph neural network. For example, when comparing scenarios from a single perspective, comparison is possible using a histogram. However, when comparing scenarios from multiple perspectives, it may be more appropriate to compare them from a macro perspective than to compare individual histograms. In such cases, a graph neural network can extract the features of a scenario from the graph structure itself. Therefore, by using a graph neural network, it may be possible to extract even more latent elements as features that do not appear in histograms.

[0079] (Embodiment 3) Next, Embodiment 3 of the present disclosure will be described. Figure 9 is a block diagram showing the configuration of the data processing system 3 according to the present disclosure. The data processing system 3 comprises geospatial data 31, a scenario generation unit (static) 32, a scenario knowledge graph 33, a scenario generation unit (dynamic) 34, an urban dynamics calculation unit 35, a generated scenario recording unit 36, a scenario database 37, a scenario generation control unit 38, and a generated scenario evaluation unit 39. The data processing system 3 is a multi-agent simulation system, and in particular a system for performing human flow simulations. That is, the data processing system 3 is a specific configuration for realizing the data processing system 2. In the following description of each configuration in the data processing system 3, parts that overlap with the data processing systems 1 and 2 will be omitted as appropriate.

[0080] The geospatial data 31 stores information about environment nodes, i.e., static elements, and predetermined rules for connecting environment nodes. Specifically, the geospatial data 31 stores geospatial information-based data such as road networks and building locations. The geospatial data 31 may also store information about environmental characteristic nodes. Furthermore, the geospatial data 31 may store open data or data that is not publicly available. Specifically, the geospatial data 31 may store base map information. The geospatial data 31 may also have some of the functions of the storage unit 23 according to Embodiment 2.

[0081] The static scenario generation unit 32 refers to information about environment nodes stored in the geospatial data 31 and connects environment nodes based on predetermined rules. The static scenario generation unit 32 may also refer to information about environment characteristic nodes stored in the geospatial data 31 and connect environment nodes and characteristic nodes based on predetermined rules and predetermined probability distributions. In other words, the static scenario generation unit 32 generates static element scenarios for the scenario generation unit 11 according to Embodiment 1 and the scenario generation unit 21 according to Embodiment 2. The static scenario generation unit 32 may transmit the generated scenarios to the dynamic scenario generation unit 34.

[0082] The scenario knowledge graph 33 stores information about agent nodes, i.e., dynamic elements, agent characteristic nodes, and predetermined rules for connecting these nodes. Specifically, the scenario knowledge graph 33 stores agent roles, attributes, decision-making types, personalities, etc., as characteristic nodes. The information stored in the scenario knowledge graph 33 may vary depending on the simulation software. That is, the scenario knowledge graph 33 may determine the information to store in accordance with the functions of the simulation software. The scenario knowledge graph 33 may also have some of the functions of the storage unit 23 according to Embodiment 2.

[0083] Furthermore, the scenario knowledge graph 33 may store information about environment nodes and predetermined rules for connecting environment nodes. For example, if the connection state of environment nodes differs depending on the scenario, the scenario knowledge graph 33 can store information about rules corresponding to such states. For example, if the road closure state differs depending on the scenario, the scenario knowledge graph 33 can store information about rules corresponding to such states. In other words, environment nodes can also be treated as dynamic elements.

[0084] The dynamic scenario generation unit 34 refers to information about agent nodes and agent characteristic nodes stored in the scenario knowledge graph 33 and connects agent nodes and characteristic nodes based on predetermined rules and predetermined probability distributions. The dynamic scenario generation unit 34 may receive static element scenarios from the static scenario generation unit 32. That is, after the static scenario generation unit 32 generates static element scenarios, the dynamic scenario generation unit 34 may place agent nodes and characteristic nodes in the static element scenarios and connect these nodes. The dynamic scenario generation unit 34 generates dynamic element scenarios for the scenario generation unit 11 according to Embodiment 1 and the scenario generation unit 21 according to Embodiment 2.

[0085] If the scenario knowledge graph 33 stores information about environment nodes and predetermined rules for connecting environment nodes, the scenario generation unit (dynamic) 34 may connect the environment nodes.

[0086] The urban dynamics calculation unit 35 is a software program that executes a simulation using the graph structure, which is a scenario generated by the scenario generation unit (static) 32 and the scenario generation unit (dynamic) 34, as input data.

[0087] The scenario generation recording unit 36 ​​receives scenarios generated by the static scenario generation unit 32 and the dynamic scenario generation unit 34 from the urban dynamics calculation unit 35 and collects their features. In other words, the scenario generation recording unit 36 ​​has the same functions as the feature collection unit 22 according to Embodiment 2. Specifically, the scenario generation recording unit 36 ​​collects statistical information on the connection frequency between nodes or edges in the graph structure and other nodes or edges as features of the scenario. The scenario generation recording unit 36 ​​also represents this statistical information as a probability distribution. The scenario generation recording unit 36 ​​may also represent the probability distribution as a histogram. The probability distribution is typically a discrete distribution, but it may also be a continuous distribution. The scenario generation recording unit 36 ​​stores the scenarios, features, and histograms in the scenario database 37.

[0088] The scenario database 37 stores the scenarios generated by the static scenario generation unit 32 and the dynamic scenario generation unit 34, as well as the probability distributions collected by the generated scenario recording unit 36. The scenario database 37 stores the scenarios and probability distributions in association. The scenario database 37 has some of the functions of the storage unit 23 according to Embodiment 2.

[0089] The scenario generation control unit 38 controls scenario generation by the dynamic scenario generation unit 34 based on a probability distribution of connection frequencies between predetermined nodes. Specifically, when there are multiple nodes of the same type that can connect to a given node, the scenario generation control unit 38 determines which node to connect to based on a predetermined probability distribution. The scenario generation control unit 38 may also control scenario generation by the static scenario generation unit 32 based on the probability distribution. Furthermore, the scenario generation control unit 38 may use a uniform probability distribution or a weighted probability distribution. The scenario generation control unit 38 may also refer to a probability distribution stored in the scenario database 37. The scenario generation control unit 38 has some of the functions of the scenario generation unit 11 according to Embodiment 1 and the scenario generation unit 21 according to Embodiment 2.

[0090] Furthermore, the scenario generation control unit 38 determines whether to adopt the generated scenario. That is, the scenario generation control unit 38 determines whether to have the urban dynamics calculation unit 35 run a simulation using the scenarios generated by the scenario generation unit (static) 32 and the scenario generation unit (dynamic) 34, or to store the scenario in the scenario database 37.

[0091] The generated scenario evaluation unit 39 calculates the similarity between the scenarios generated by the static scenario generation unit 32 and the dynamic scenario generation unit 34 and the scenarios stored in the scenario database 37. The generated scenario evaluation unit 39 considers the statistical information of the connection frequency between nodes and edges in these scenarios as a probability distribution and calculates the similarity as a distance based on optimal transport theory. In other words, the generated scenario evaluation unit 39 has the same configuration as the generated scenario evaluation unit 24 according to Embodiment 2.

[0092] Next, the processing flow of the data processing system 3 will be explained. Figure 10 is a flowchart showing an example of the scenario generation process in the data processing system 3. First, the static scenario generation unit 32 extracts information about static elements from the geospatial data 31 (S301). Next, the static scenario generation unit 32 constructs a road network based on predetermined rules (S302). That is, the static scenario generation unit 32 constructs a road network by placing intersections and connecting predetermined intersections based on the information extracted from the geospatial data 31. After that, the static scenario generation unit 32 places objects such as buildings (S303). Here, the static scenario generation unit 32 may perform the construction of the road network and the placement of objects such as buildings simultaneously.

[0093] Subsequently, the dynamic scenario generation unit 34 extracts information about dynamic elements from the scenario knowledge graph 33 (S304). Then, the dynamic scenario generation unit 34 performs placement of residents and connection processing of characteristic nodes based on predetermined rules and probability distributions (S305). Here, the connection relationships between nodes may be determined by the scenario generation control unit 38.

[0094] Next, the processing of the generated scenarios will be described. Figure 11 is a flowchart showing an example of scenario recording processing in the data processing system 3. First, the generated scenario recording unit 36 ​​collects statistical information on the connection frequency between nodes and edges in the scenarios generated by the scenario generation unit (static) 32 and the scenario generation unit (dynamic) 34 (S306). Next, the generated scenario recording unit 36 ​​creates a histogram of the statistical information on the connection frequency between nodes and edges (S307). After that, the scenario database 37 stores the generated scenarios and histograms (S308).

[0095] Next, scenario generation based on the similarity of scenario features will be described. Figure 12 is a flowchart showing an example of the scenario generation process in the data processing system 3. First, the scenario generation control unit 38 performs input processing of the scenario histogram (S309). Next, the scenario generation unit (static) 32 and the scenario generation unit (dynamic) 34 generate scenarios based on the input histogram (S310). Here, the scenario generation unit (static) 32 does not have to generate the static element part of the scenario. That is, the scenario generation unit (dynamic) 34 may refer to the static element part of the scenario stored in the scenario database 37 and generate the dynamic element part of the scenario. Next, the generated scenario recording unit 36 ​​collects statistical information on the connection frequency between nodes and edges (S311). After that, the generated scenario evaluation unit 39 calculates the similarity with existing scenarios (S312). That is, the generated scenario evaluation unit 39 refers to the feature quantities of the scenarios stored in the scenario database 37 and calculates the similarity with the scenario generated this time. Here, the generated scenario evaluation unit 39 may calculate the similarity to all existing scenarios, or it may calculate the similarity to only some of the existing scenarios.

[0096] Subsequently, the generated scenario evaluation unit 39 determines whether the purpose is to generate similar scenarios (S313). If the purpose is to generate similar scenarios, the generated scenario evaluation unit 39 determines whether the similarity between scenarios is above a certain value (S314). Specifically, the generated scenario evaluation unit 39 determines whether the distance between scenarios based on optimal transport theory is below a certain value. If the similarity is above a certain value, the scenario database 37 stores the scenarios and histograms in association (S317). If the similarity is not above a certain value, the generated scenario evaluation unit 39 performs a correction process (S316). The correction process by the generated scenario evaluation unit 39 is a process of correcting the histogram. After the correction process, the scenario generation unit (static) 32 or the scenario generation unit (dynamic) 34 generates the scenarios again (S310). The flow is repeated until the similarity of the scenarios is above a certain value. If a scenario that satisfies the similarity requirement is generated, the scenario database 37 stores the scenario and histogram in association (S317).

[0097] If the purpose of this scenario generation is not to generate similar scenarios, the generated scenario evaluation unit 39 determines whether the similarity between scenarios is below a certain value (S315). Specifically, the generated scenario evaluation unit 39 determines whether the distance between scenarios based on optimal transport theory is above a certain value. If the similarity is below a certain value, the scenario database 37 stores the scenarios and histograms in association (S317). If the similarity is above a certain value, the generated scenario evaluation unit 39 performs correction processing (S316). After the correction processing, the scenario generation unit (static) 32 or the scenario generation unit (dynamic) 34 generates the scenarios again (S310). The flow is repeated until the similarity of the scenarios is below a certain value. If a scenario that satisfies the similarity requirement is generated, the scenario database 37 stores the scenarios and histograms in association (S317).

[0098] In the flow shown in Figure 12, if a scenario that satisfies the similarity requirement is generated, the scenario generation control unit 38 may decide whether to adopt the generated scenario. If the scenario generation control unit 38 decides to adopt the generated scenario, the urban dynamics calculation unit 35 may separately execute a simulation using that scenario. In addition, the generated scenario evaluation unit 39 may present the scenarios stored in the scenario database 37 to the user, or may sort the scenarios stored in the scenario database 37.

[0099] Thus, the data processing system 3 according to this embodiment 3 can generate highly comprehensive scenarios and perform human flow simulations. Specifically, the scenario generation control unit 38 controls scenario generation by the scenario generation unit (static) 32 or the scenario generation unit (dynamic) 34 based on the probability distribution of connection frequencies between nodes. As a result, the data processing system 3 can intentionally generate scenarios that are highly or less similar to existing scenarios stored in the scenario database 37.

[0100] Furthermore, the data processing system 3 calculates the similarity between scenarios as a distance based on optimal transport theory. This allows the data processing system 3 to quantitatively show and compare the similarity between scenarios, which are represented as a graph structure.

[0101] Specific examples of simulation results based on scenarios generated by the data processing system 3 will be explained using diagrams. Figure 13 is a schematic diagram showing an example of a simulation result based on a scenario by the data processing system 3. Figure 13 shows an aerial image of an island connected by multiple bridges and its surroundings. The arrows in the image represent the passage of time, and the point cloud in the image represents people, who are agents. The human flow simulation in Figure 13 is a simulation of a situation where, for example, during a disaster, it becomes necessary to move to a destination outside the island by crossing bridges. Figure 13 shows three patterns of simulation results.

[0102] In the scenario generation shown in Figure 13, three types of decision-making patterns are prepared as characteristics of people. Specifically, when generating scenarios in Figure 13, three decision-making nodes are defined as characteristic nodes for agents. Specifically, people's decision-making types are assigned to one of the following types: "familiar with the area and moves to the nearest destination via the shortest route," "heads to the most famous destination for now," or "unfamiliar with the area and follows the person in front of them." This assignment is made based on a histogram generated by the scenario generation control unit 38. These decision-making types are referred to as the "familiar with the area" type, the "somewhat familiar with the area" type, and the "unfamiliar with the area" type, respectively. Note that the initial placement of people does not change between scenarios.

[0103] In Figure 13, scenario (A) assumes that 60% of the population is "familiar with the area," 30% are "somewhat familiar with the area," and 10% are "unfamiliar with the area." Scenario (B) assumes that 40% of the population is "familiar with the area," 30% are "somewhat familiar with the area," and 30% are "unfamiliar with the area." Scenario (C) assumes that 20% of the population is "familiar with the area," 10% are "somewhat familiar with the area," and 70% are "unfamiliar with the area." In other words, (A) is a scenario with many people familiar with the area, (B) is a scenario with a diverse range of people, and (C) is a scenario with many people unfamiliar with the area.

[0104] As shown in Figure 13, in scenario (A), many people can travel to their destinations outside the island. In scenario (B), although it takes longer than in (A), many people can still travel to their destinations. In scenario (C), many people remain on the island and are unable to travel to their destinations outside the island.

[0105] The data processing system 3 can quantitatively compare scenarios by calculating the similarity between scenarios in Figure 13 as a distance based on optimal transport theory. For example, suppose the data processing system 3 calculates the distances between scenarios in Figure 13 and finds that the distance between scenario (A) and scenario (B) is 40.0, the distance between scenario (B) and scenario (C) is 70.0, and the distance between scenario (A) and scenario (C) is 100.0. Because the similarity between scenarios can be quantitatively shown, the data processing system 3 can determine, for example, that scenario (A) and scenario (B) are relatively similar, but scenario (A) and scenario (C) are not similar. In this way, the data processing system 3 can generate comprehensive scenarios and quantitatively compare them.

[0106] (Example hardware configuration) Figure 14 shows an example of the hardware configuration of a data processing system 4 according to this disclosure. In Figure 14, the data processing system 4 includes a processor 41 and a memory 42. The processor 41 may be, for example, a microprocessor, an MPU (Micro Processing Unit), or a CPU (Central Processing Unit). The processor 41 may include multiple processors. The memory 42 is composed of a combination of volatile memory and non-volatile memory. The memory 42 may include storage located away from the processor 41. In this case, the processor 41 may access the memory 42 via an I / O (Input / Output) interface, which is not shown.

[0107] In the above example, the program can be stored and provided to the computer using various types of non-transitory computer-readable medium. Non-transitory computer-readable medium includes various types of tangible storage medium. Examples of non-transitory computer-readable medium include magnetic storage media (e.g., magneto-optical disks), CD-ROMs, CD-Rs, CD-R / Ws, and semiconductor memory (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, RAMs). Alternatively, the program may be provided to the computer using various types of transient computer-readable medium. Examples of transient computer-readable medium include electrical signals, optical signals, and electromagnetic waves. Transitory computer-readable medium can supply the program to the computer via wired communication channels such as electric wires and optical fibers, or via wireless communication channels. Computers include various information processing devices such as PCs, servers, CPUs, MPUs, FPGAs (Field Programmable Gate Arrays), and ASICs (Application Specific Integrated Circuits).

[0108] Although the present disclosure has been described above with reference to embodiments, the present disclosure is not limited to the embodiments described above. Various modifications to the structure and details of the present disclosure can be made as can be understood by those skilled in the art within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0109] Each drawing is merely illustrative to illustrate one or more embodiments. Each drawing may be associated with one or more other embodiments rather than with only one specific embodiment. As those skilled in the art will understand, various features or steps described with reference to any one drawing can be combined with features or steps shown in one or more other drawings, for example, to create embodiments not explicitly shown or described. Not all features or steps shown in any one drawing to illustrate an exemplary embodiment are necessarily required, and some features or steps may be omitted. The order of steps shown in any of the drawings may be changed as appropriate.

[0110] Some or all of the above embodiments may also be described as follows, but are not limited to the following: (Note 1) The system includes a scenario generation unit that arranges agents in a multi-agent simulation, the simulation environment in which the agents are active, and the characteristics of the agents or the environment as nodes in a graph structure, and generates a scenario in the simulation by connecting a first node representing the agents or a second node representing the environment and a third node representing the characteristics, in the graph structure. Data processing system. (Note 2) The scenario generation unit generates the scenario by connecting the first node or the second node and the third node according to predetermined preconditions. The data processing system described in Appendix 1. (Note 3) A feature collection unit collects the feature quantities of the scenario generated by the scenario generation unit, The system further comprises a storage unit for storing the aforementioned feature quantities, The data processing system described in Appendix 1. (Note 4) The feature collection unit collects statistical information on the connection frequency between the nodes and edges as features of the scenario. The data processing system described in Appendix 3. (Note 5) The system further includes a generated scenario evaluation unit that calculates the similarity between the feature quantities of the scenario generated by the scenario generation unit and the feature quantities stored in the memory unit. The data processing system described in Appendix 3 or 4. (Note 6) The scenario generation unit considers the statistical information as a probability distribution and generates the scenario based on the statistical information stored in the memory unit. The data processing system described in Appendix 4 or 5. (Note 7) The scenario generation unit generates the scenario having the probability distribution in which the similarity is greater than a predetermined value. The data processing system described in Appendix 6. (Note 8) The scenario generation unit generates the scenario having the probability distribution in which the similarity is smaller than a predetermined value. The data processing system described in Appendix 6. (Note 9) The aforementioned scenario generation evaluation unit considers the statistical information as a probability distribution and calculates the similarity as a distance based on optimal transport theory. A data processing system as described in any one of the items from Appendix 5 to Appendix 8. (Note 10) The characteristics of the agent include the role of the agent, A data processing system as described in any one of the items from Appendix 1 to Appendix 9. (Note 11) The characteristics of the agent include the attributes of the agent, A data processing system as described in any one of the items from Appendix 1 to Appendix 9. (Note 12) The characteristics of the agent include the decision-making type of the agent. A data processing system as described in any one of the items from Appendix 1 to Appendix 9. (Note 13) The feature collection unit extracts the features using a graph neural network. The data processing system described in Appendix 3. (Note 14) Computers In a multi-agent simulation, the agents, the simulation environment in which the agents operate, and the characteristics of the agents or the environment are arranged as nodes in a graph structure. A graph structure is generated as a scenario in the simulation by connecting a first node representing the agent or a second node representing the environment, and a third node representing the characteristics. Data processing method. (Note 15) The steps include: arranging agents in a multi-agent simulation, the simulation environment in which the agents operate, and the characteristics of the agents or the environment as nodes in a graph structure; The steps include generating a graph structure as a scenario in the simulation, which connects a first node representing the agent or a second node representing the environment and a third node representing the characteristics, A program that causes a computer to execute something.

[0111] Some or all of the elements (e.g., configuration and function) described in Appendices 2 to 13 that are dependent on Appendice 1 may also be dependent on Appendices 14 and 15 in the same manner as those described in Appendices 2 to 13. Some or all of the elements described in any appendice may be applicable to various hardware, software, recording means, systems, and methods for recording software. [Explanation of symbols]

[0112] 1. Data Processing System 2. Data processing system 3. Data Processing System 4. Data Processing System 11 Scenario Generation Unit 21 Scenario Generation Unit 22 Feature Collection Unit 23 Memory section 24. Generation Scenario Evaluation Department 31. Geospatial Data 32 Scenario Generation Unit (Static) 33 Scenario Knowledge Graph 34. Scenario Generation Unit (Dynamic) 35. Urban Dynamics Calculation Department 36. Scenario Generation Recording Unit 37 Scenario Database 38 Scenario Generation Control Unit 39. Generation Scenario Evaluation Department 41 processors 42 memory

Claims

1. The system includes a scenario generation unit that arranges agents in a multi-agent simulation, the simulation environment in which the agents operate, and the characteristics of the agents or the environment as nodes in a graph structure, and generates a scenario in the simulation by connecting a first node representing the agents or a second node representing the environment and a third node representing the characteristics, in the graph structure. Data processing system.

2. The scenario generation unit generates the scenario by connecting the first node or the second node and the third node according to predetermined preconditions. The data processing system according to claim 1.

3. A feature collection unit collects the feature quantities of the scenario generated by the scenario generation unit, The system further comprises a storage unit for storing the aforementioned feature quantities, The data processing system according to claim 1.

4. The feature collection unit collects statistical information on the connection frequency between the nodes and edges as features of the scenario. The data processing system according to claim 3.

5. The system further includes a generated scenario evaluation unit that calculates the similarity between the feature quantities of the scenario generated by the scenario generation unit and the feature quantities stored in the memory unit. The data processing system according to claim 3.

6. The scenario generation unit considers the statistical information as a probability distribution and generates the scenario based on the statistical information stored in the memory unit. The data processing system according to claim 4 or 5.

7. The scenario generation unit generates the scenario having the probability distribution in which the similarity is greater than a predetermined value. The data processing system according to claim 6.

8. The scenario generation unit generates the scenario having the probability distribution in which the similarity is smaller than a predetermined value. The data processing system according to claim 6.

9. Computers In a multi-agent simulation, the agents, the simulation environment in which the agents operate, and the characteristics of the agents or the environment are arranged as nodes in a graph structure. A graph structure is generated as a scenario in the simulation by connecting a first node representing the agent or a second node representing the environment and a third node representing the characteristics. Data processing method.

10. The steps include: arranging agents in a multi-agent simulation, the simulation environment in which the agents operate, and the characteristics of the agents or the environment as nodes in a graph structure; The steps include generating a graph structure as a scenario in the simulation by connecting a first node representing the agent or a second node representing the environment and a third node representing the characteristics, A program that causes a computer to execute something.

Images