Information processing device, information processing method, and program

Abstract

The present invention provides an information processing device, information processing method, and program that enable users to intuitively understand the differences between multiple process scenarios that are in a trade-off relationship, while suppressing the risk of hallucination, which generates physically impossible processes, and to make quick and evidence-based decisions. [Solution] The information processing device (10) includes a process scenario acquisition unit that acquires multiple process scenarios (602) in a trade-off relationship from an optimization processing unit (202) that calculates multiple optimal solutions that satisfy the constraints of the physical sequence between the work items and have different weighting conditions between the multiple evaluation axes, using an evaluation function (601) having multiple evaluation axes, and an explanation generation unit (203) that takes difference data (801) of evaluation values ​​between the multiple process scenarios (602) as input and outputs a comparative explanation text (702) between the multiple process scenarios (602) in natural language to a generation model (40).

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] In process control for construction work, it is required to plan an optimal schedule while comprehensively satisfying a plurality of management items such as quality (Q), cost (C), duration (D), safety (S), and environment (E). Conventionally, the creation and modification of a process schedule greatly depends on the experience and knowledge of skilled project managers, and every time a change occurs in some conditions, it is necessary to manually correct the entire schedule using spreadsheet software or the like, which requires a great deal of time and effort.

[0003] In contrast, in recent years, information processing techniques for automatically generating a work schedule based on work conditions registered in a database have been proposed (see, for example, Patent Document 1). In the technique of Patent Document 1, a method is adopted in which the optimality of a work schedule is evaluated, and the schedule is recreated if there are any unreflected work conditions. There is also known a technique of formulating a scheduling problem as a mathematical optimization problem and deriving a solution that strictly satisfies constraints using network flow or the like (see, for example, Patent Document 2).

Prior Art Literature

Patent Literature

[0004]

Patent Literature 1

Patent Literature 2

Summary of the Invention

Problem to be Solved by the Invention

[0005] However, while methods for automatically generating process schedules based on input conditions, such as those described in Patent Document 1, can quickly provide flexible process proposals, there may be working conditions that are not reflected in the generated results, potentially requiring the user to make manual corrections. Furthermore, when this type of method is combined with a generative approach using large-scale language models and applied to complex construction sites, the risk of generating processes that do not strictly satisfy the physical or logical relationships in real-world construction conditions, resulting in so-called hallucination, cannot be eliminated.

[0006] On the other hand, while methods that formulate the problem as a mathematical optimization problem, such as those described in Patent Document 2, can derive a highly feasible process that strictly satisfies the constraints, the resulting schedule data is limited to a collection of dry numerical data such as the operating status and output of the equipment. Therefore, when multiple evaluation axes such as construction period, cost, and resources are in a trade-off relationship, even if an optimization solver outputs a set of multiple numerical data (process scenarios), it is extremely difficult for process managers to immediately compare them and intuitively grasp the advantages and disadvantages of the differences between scenarios, such as whether to adopt "a plan with a shorter construction period but increased peak labor" or "a plan with a longer construction period but leveled labor," and make a decision. This presents a challenge in terms of high cognitive load on humans.

[0007] In view of the circumstances described above, at least some embodiments of the present invention aim to provide an information processing device, an information processing method, and a program that enable users to intuitively understand the differences between multiple process scenarios that are in a trade-off relationship, and to make quick and evidence-based decisions, while suppressing the risk of hallucination that generates physically impossible processes. [Means for solving the problem]

[0008] Information processing devices according to at least some embodiments of the present invention are An optimization processing unit calculates multiple optimal solutions for multiple work items constituting a construction process, using an evaluation function with multiple evaluation axes, satisfying constraints on the physical sequence of work items and with different weighting conditions between the multiple evaluation axes. From this optimization processing unit, a process scenario acquisition unit acquires multiple process scenarios that are in a trade-off relationship. At a minimum, the system includes an explanation generation unit that takes as input the difference data of evaluation values ​​for the multiple evaluation axes between the acquired multiple process scenarios, and outputs comparative explanatory text between the multiple process scenarios in natural language to the generation model, It is equipped with. [Effects of the Invention]

[0009] According to at least some embodiments of the present invention, multiple optimal process scenarios that are in a trade-off relationship while reliably satisfying physical constraints are obtained by mathematical algorithms, and only the "comparison of differences between scenarios," which is cognitively burdensome for humans, is explained in natural language by the generative model. This eliminates hallucination (generation of physically contradictory processes) which is inherent in generative AI, and allows managers to make decisions on the optimal process in a short amount of time. [Brief explanation of the drawing]

[0010] [Figure 1] This is an overall configuration diagram of a construction process management system according to one embodiment. [Figure 2] This is a hardware configuration diagram of an information processing device according to one embodiment. [Figure 3] This is a functional block diagram of an information processing device according to one embodiment. [Figure 4] This is a logical structure diagram (ER diagram) of a database according to one embodiment. [Figure 5] This is a flowchart of the process for calculating the required number of days according to one embodiment. [Figure 6] This is a diagram of a dependency resolution flow according to one embodiment. [Figure 7] This is a conceptual diagram of a multi-objective optimization process according to one embodiment. [Figure 8]It is a diagram of a multi-scenario comparative explanation generation flow according to an embodiment. [Figure 9] It is a conceptual diagram of prompt control according to an embodiment. [Figure 10] It is a user interface diagram of a scenario comparison screen according to an embodiment. [Figure 11] It is a user interface diagram of a Gantt chart screen according to an embodiment. [Figure 12] It is a conceptual diagram of 4D visualization simulation according to an embodiment. [Figure 13] It is a flowchart of a self-evolution cycle according to an embodiment. [Figure 14] It is an overall processing sequence diagram of an automatic process creation engine according to an embodiment. DETAILED DESCRIPTION OF EMBODIMENTS

[0011] Hereinafter, several embodiments of the present invention will be described in detail with reference to the drawings. The same constituent elements are denoted by the same reference numerals, and duplicate descriptions are omitted. In addition, the blocks and sequences shown in each drawing do not necessarily limit physical arrangements or strict processing orders, and can be appropriately integrated, distributed, or reordered without departing from the spirit of the present invention.

[0012] FIG. 1 is an overall configuration diagram of a construction process management system according to an embodiment. In some embodiments, as shown in FIG. 1, the information processing apparatus 10 is capable of cooperating with a generative model 40 (external API). The information processing apparatus 10 has a function of generating and acquiring a plurality of process scenarios in a trade-off relationship based on a plurality of evaluation axes for a plurality of work items constituting a construction process while satisfying constraints on physical context, and outputting comparative explanation text in natural language via the generative model 40 (external API).

[0013] In the embodiment shown in Figure 1, the construction process management system is built as a cloud-based platform and comprises a core information processing device 10 (remote command tower, MTU: Mission Technical Unit) and multiple site terminals 30A, 30B, 30C (hereinafter collectively referred to as site terminals 30) at multiple construction sites connected via a network 20. The information processing device 10 is, for example, a backend server implemented using an arbitrary web framework, equipped with a database such as a relational database, and centrally manages a large number of tables and a large number of key performance indicators (KPIs).

[0014] In this embodiment, the information processing device 10 functions as a "remote command center" that introduces the concepts of NASA's mission control and F1 pit walls to the construction industry. One information processing device 10 (or one MTU team operating it) simultaneously remotely manages the process, quality, safety, cost, etc., of 3 to 4 construction sites (site terminals 30) in parallel. This creates an environment where on-site staff can focus on making on-site, real-world judgments, and makes it possible to optimize the company's overall utilization rate (for example, reducing it from 100% to 65%).

[0015] Figure 2 is a hardware configuration diagram of an information processing device 10 according to one embodiment. In some embodiments, as shown in Figure 2, the information processing device 10 is configured as a computer comprising a control unit 101, a main memory 102, an auxiliary memory 103, and a communication interface 104, which are interconnected via a bus 105 so as to be able to communicate with each other. The information processing device 10 loads a predetermined program (such as a program for the automated process creation engine) stored in the auxiliary storage device 103 into the main storage device 102 and executes it in the control unit 101, thereby realizing the processing of each functional block described later.

[0016] In the embodiment shown in Figure 2, the control unit 101 is, for example, a CPU (Central Processing Unit), which controls the operation of the entire device and performs calculations such as calculating the number of days required for each work item, graph analysis, and triggering optimization calculations.

[0017] The main memory 102 consists of RAM (Random Access Memory) and functions as a work area for the control unit 101. The auxiliary storage device 103 is a non-volatile memory such as an HDD (Hard Disk Drive) or SSD (Solid State Drive), and stores the operating system, as well as database management systems such as relational databases, and numerous databases (quantity / work rate master, process baseline, event log, etc.).

[0018] The communication interface 104 is a communication controller that receives performance data (photos, KY records, etc.) from the field terminal 30 via the network 20, or sends prompts to the generation model 40 (external API) and receives the generated explanatory text.

[0019] The information processing device 10 may, if necessary, include an input device 106 (keyboard, mouse, touch panel, etc.) for receiving input from the user, and an output device 107 (display, etc.) for displaying processing results.

[0020] Furthermore, the information processing device 10 of this embodiment may also include a computing accelerator 108. The computing accelerator 108 is, for example, a GPU (Graphics Processing Unit), FPGA, ASIC (AI Accelerator), etc., and is used to speed up the calculation of a vast number of constraint satisfaction problems by the optimization processing unit (solver) described later, and the training process of machine learning models (such as work rate correction based on actual data) by the learning unit.

[0021] In other embodiments, the information processing device 10 does not need to be a single physical computer, but may be implemented as a distributed computing environment in which multiple server machines cooperate via a network, or as a virtual machine or container on the cloud.

[0022] Furthermore, the computing accelerator 108 may be an external resource available via the network (such as a GPU instance on the cloud).

[0023] In other embodiments not shown, the auxiliary storage device 103 may be a network storage (NAS) or cloud storage located in a physically separate location, and the information processing device 10 may be configured to access the database on the storage via the communication interface 104. This makes it possible to manage data from a vast number of construction projects safely and flexibly.

[0024] Returning to Figure 1, we will explain the components of the construction process management system other than the information processing device 10. The field terminal 30 is a wearable device such as a smartphone, tablet, or smart helmet carried by the site supervisor or worker. The field terminal 30 is equipped with an offline-compatible mobile app and, with intuitive operation within a few taps (for example, 3 taps), transmits performance data such as photos taken, hazard prediction (KY) activities, and self-inspection results to the information processing device 10 in real time, where it is stored as an event log.

[0025] The information processing device 10 is further configured to communicate with an external API server that provides a generative model 40 via a network 20. The generative model 40 is a generative AI such as a large-scale language model (LLM), and is responsible for generating and returning explanatory text in natural language using KPI difference data between scenarios transmitted from the information processing device 10 as input.

[0026] In other embodiments not shown, some or all of the functions of the information processing device 10 may be located on edge computing resources installed at each construction site or on an on-premises server. Furthermore, the generation model 40 is not limited to an external API, but may also be implemented as a model that operates locally within the information processing device 10. Furthermore, network 20 may be the internet, a dedicated line, a high-speed wireless communication network such as 5G / 6G, or a satellite communication network.

[0027] Figure 3 is a functional block diagram of an information processing device 10 according to one embodiment. In some embodiments, as shown in Figure 3, the information processing device 10 includes a process generation unit 201, an optimization processing unit 202, an explanation generation unit 203, a display control unit 204, and a learning unit 205.

[0028] Each functional block shown in Figure 3 (process generation unit 201, optimization processing unit 202, explanation generation unit 203, display control unit 204, and learning unit 205) is a functional module realized by the control unit 101 (CPU) executing a predetermined program stored in the auxiliary storage device 103, etc., and controlling hardware resources such as the main memory device 102, communication interface 104, or bus 105. In the embodiment shown in Figure 3, the information processing device 10 realizes a hybrid process creation architecture that achieves both reliable process calculations that comply with physical constraints and flexible natural language explanations using a generative model (generative AI) through the mutual cooperation of these functional modules.

[0029] The process generation unit 201 is responsible for calculating the baseline process (a directed graph including the critical path) which serves as the standard for the construction process. Specifically, the process generation unit 201 uses the quantity of each work item, the standard work rate, and correction coefficients based on site conditions (such as narrow space correction, nighttime correction, and skill level correction) to definitively calculate the number of days required for each work item using a predetermined arithmetic formula (for example, required days = quantity / (standard work rate × number of resources input × correction coefficient)). Furthermore, the process generation unit 201 automatically sets inter-task dependencies (FS / SS / FF / SF, etc.) from a vast number of rule-based templates, applies constraints such as limits on the number of simultaneous tasks within the same work area and curing periods, and then automatically detects and resolves cycles in the graph to construct a consistent process chart.

[0030] The optimization processing unit 202 has the function of calculating multiple optimal solutions that satisfy the constraints of the physical sequence between the work items and that have different weighting conditions between the multiple evaluation axes, using an evaluation function that has multiple evaluation axes. In this embodiment, the evaluation axis consists of at least six indicators, including total construction period, peak number of workers, temporary construction costs, specific downtime, quality risk, and notification risk. The optimization processing unit 202 is implemented as a constraint satisfaction solver, such as OR-Tools (registered trademark), and performs heavy optimization processing in parallel in the background, for example, using an asynchronous task queue.

[0031] Here, the "process scenario acquisition unit," which acquires multiple process scenarios, is realized by the control unit 101 executing a process to acquire multiple process scenarios. When the optimization processing unit 202 is located inside the information processing device 10 (for example, within the same program module), the control unit 101 realizes the function of acquiring multiple process scenarios by reading the calculation results calculated by the optimization processing unit 202 via the main memory 102 or the like. On the other hand, if the optimization processing unit 202 is located outside the information processing device 10 (for example, an external computing server or solver API service), the control unit 101 receives calculation results from the external device via the network through the communication interface 104, thereby realizing the function of acquiring multiple process scenarios.

[0032] The explanation generation unit 203 takes the difference data of evaluation values ​​for multiple evaluation axes between multiple process scenarios output from the optimization processing unit 202 as input and outputs comparative explanatory text in natural language via the generation model 40 (external API). The explanation generation unit 203 has a function to strictly control the output of prompts to the generation model 40, and is configured to exclude from prompts commands related to the generation process of the multiple process scenarios to be executed in the optimization processing unit 202, and the process for determining compliance with physical constraints, in order to prevent hallucination. In other words, the explanation generation unit 203 has a function to strictly control commands (prompts) to the generation model 40, and is configured to explicitly exclude from prompts tasks related to "deterministic logical calculations," such as the calculation of required days and the determination of compliance with physical constraints, in order to prevent hallucination.

[0033] The display control unit 204 generates display data (for example, data in DTO: Data Transfer Object format) to display the evaluation values ​​of multiple evaluation axes related to multiple acquired process scenarios and comparative explanatory text on the same screen, and distributes it to the field terminal 30. On the field terminal 30 side, the screen is rendered based on the display data distributed from the information processing device 10 by executing the web content (scripts such as HTML and React®) provided by the information processing device 10 in a browser. When a specific process scenario is selected by the user, the display control unit 204 generates update data to immediately reflect the result on the timeline of the Gantt chart screen and provides it to the field terminal 30, and also transmits control information to the user terminal to execute a 4D simulation using a three-dimensional building model (BIM), which will be described later.

[0034] The learning unit 205 stores on-site operation history and performance data on the system as an "immutable event log," and uses this as a data source for machine learning (deep learning and statistical analysis). The learning unit 205 automatically updates the work rate correction coefficient used in the process generation unit 201 based on actual data, or autonomously adjusts the weighting conditions in the optimization processing unit 202.

[0035] In other embodiments not shown, these functional blocks may be implemented as part of a "common extension mechanism" that is common across all 14 control dimensions, including quality, safety, and environment, as well as process control dimensions. Furthermore, the display processing by the display control unit 204 is not limited to rendering on a web browser, but may also be implemented as augmented reality (AR) display using a smart helmet or the like at the worksite.

[0036] Figure 4 is a logical structure diagram (ER diagram) of a database according to one embodiment. In some embodiments, as shown in Figure 4, the information processing device 10 includes a database that manages and correlates construction projects, process scenarios, work items, dependencies between tasks, process evaluation indicators, and actual data including operation history on the system. Here, the database shown in Figure 4 may be implemented as a relational database, for example, and may consist of a large number of tables (e.g., 100 or more) and a vast number of rows (e.g., 1000 or more) of data definitions (DDL).

[0037] In the embodiment shown in Figure 4, the database may include six entities: a project master 301, a scenario definition 302, a WBS item 303, a work dependency 304, performance indicator data 305, and an event log 306.

[0038] Project Master 301 holds basic information such as an ID and name that uniquely identifies a construction project, construction period, contract amount, and spatial hierarchy (L0~L8).

[0039] The scenario definition 302 manages multiple process scenarios generated by the optimization processing unit 202. Each scenario record stores configuration information such as the weighting conditions (weight array) between the evaluation axes (6 axes) used during optimization.

[0040] WBS item 303 holds data for each of the multiple work items that make up the construction process. WBS item 303 includes the ID and WBS code for each task, as well as attributes such as quantity (including units), standard work rate, number of teams, and correction coefficients based on site conditions, which are used to calculate the required number of days for each task item.

[0041] Task dependency 304 defines constraints on the physical dependencies between tasks and holds a large number of dependency rules (e.g., about 600) that are automatically generated from the template. Work dependency 304 includes connection types for FS, SS, FF, and SF, work section constraints, and settings for lag days such as curing periods.

[0042] The evaluation index data 305 stores the evaluation values ​​(evaluation values ​​for multiple evaluation axes) calculated by the optimization processing unit 202 for each process scenario. The evaluation values ​​are recorded as specific numerical data corresponding to, for example, six axes (total construction period, peak number of workers, temporary construction costs, specific downtime, quality risk, and notification risk), and serve as the data source for comparative display by the display control unit 204.

[0043] The event log 306 stores progress inputs and operation history from the field terminal 30 in an immutable format (event sourcing format) that only allows appending. The time-series data of the event log 306 is used as training data for machine learning by the learning unit 205.

[0044] In other embodiments not shown, the database may include extended schemas corresponding to all 14 management dimensions, such as the "material_instances" table for managing traceability at the individual material level and the "concrete_pour_records" table for managing concrete pouring and curing temperatures. Furthermore, to allow for flexible extension of attribute information, it may also have a structure that functions as a plug-in type extension schema (entity_extensions) that allows for the addition of extended attributes without modifying existing tables.

[0045] Figure 5 is a flowchart of the process for calculating the required number of days according to one embodiment. In some embodiments, as shown in Figure 5, the process generation unit 201 of the information processing device 10 calculates the required number of days for each work item using a predetermined arithmetic formula that uses at least the quantity of each work item, the standard labor cost, and a correction coefficient based on the site conditions, before processing by the optimization processing unit 202.

[0046] In the embodiment shown in Figure 5, the process generation unit 201 generates a process that will serve as the baseline for optimization. In step S401, it extracts the quantities of each task from a 3D building model such as BIM (Building Information Modeling) or estimation data (for example, 120t for foundation reinforcement, 500m for foundation formwork). 2 To obtain (etc.).

[0047] Next, in step S402, the work rate corresponding to the task is obtained from the standard work rate master for each job type (for example, 1.5t / day or 25m 2 The system obtains the number of days, etc., and in step S403, applies the input resources (number of teams, number of workers, number of heavy machines, etc.) allocated to the work.

[0048] Next, in step S404, the process generation unit 201 applies a field correction coefficient to realistically reflect the on-site environment and constraints. The on-site adjustment coefficient may include at least one of the following: a "narrow space adjustment (0.6~1.0)" that reflects the decrease in work efficiency in narrow spaces without scaffolding; a "nighttime adjustment (0.5~0.8)" that takes into account lighting conditions and noise restrictions; a "factory operation adjustment (0.7~0.9)" that reflects work restrictions in operating factories; and a "skill level adjustment (0.8~1.2)" that indicates the experience level of subcontractors and craftsmen.

[0049] Then, in step S405, the process generation unit 201 calculates the number of days required for each task using a deterministic arithmetic formula such as "Number of days required = Quantity ÷ (Standard work rate × Number of resources input × Correction coefficient)". Since the calculation of the number of days required for each task is performed without going through the generative model 40 (LLM), accurate numbers of days are derived, completely eliminating hallucination (calculation errors due to probabilistic output) in generative AI.

[0050] In other embodiments not shown, the formula for calculating the number of days required is not limited to the above. The formula for calculating the required number of days may incorporate, for example, a "weather correction coefficient" linked to meteorological data, a "seasonal correction coefficient" that takes into account the difference in strength development of concrete in cold and hot weather, or a "logistics correction coefficient" based on the distance of material transport on a vast site such as a large logistics facility.

[0051] Furthermore, the acquisition of work quantities (step S401) is not limited to automatic extraction from BIM; quantities and work rates estimated with high accuracy using a machine learning model (learning unit 205) from past similar project performance data accumulated in the event log 306 shown in Figure 4 may also be applied. This makes it possible to replace the estimation of the number of days, which previously relied on subjective experience and intuition, with data-driven arithmetic processing based on clear evidence.

[0052] Figure 6 is a diagram of a dependency resolution flow according to one embodiment. In some embodiments, as shown in Figure 6, the process generation unit 201 of the information processing device 10 forms a directed graph based on predetermined dependency rules and identifies the critical path. Furthermore, when forming the directed graph, the process generation unit 201 applies dependency rules to satisfy multiple constraints, including limitations on the number of simultaneous operations in the same work area and predetermined physical waiting periods, and automatically detects cycles in the directed graph to resolve inconsistencies.

[0053] In the embodiment shown in Figure 6, the process generation unit 201 obtains a list of work items and dependency rules (for example, a group of about 600 rules) extracted from a standard template in step S501.

[0054] Next, in step S502, four types of connection relationships are applied to define the physical order between tasks (FS: subsequent task starts after the preceding task is completed, SS: simultaneous start, FF: simultaneous end, SF: subsequent task ends after the preceding task has started). For example, in reinforced concrete (RC) construction, a typical dependency flow such as "rebar placement → (FS) → formwork → (FS) → concrete pouring → (FS) → curing → (FS) → formwork removal" is automatically set.

[0055] Next, in step S503, the process generation unit 201 adds spatial and temporal constraints. Specifically, measures such as "limiting the number of simultaneous tasks (zone constraints)" to prevent conflicts in work within the same work area (zone) and "setting lag days" to define physical waiting periods that do not involve work, such as concrete curing periods, are implemented.

[0056] Then, in step S504, the process generation unit 201 automatically detects whether there are any closed circuits (circular references, cycles) within the configured dependency network. In the construction process, a combination of rules can sometimes lead to a logical contradiction (deadlock) where "task B cannot start until task A is completed, and task A cannot start until task B is completed." If a closed loop is detected (Yes), the system either relaxes part of the rules or notifies the administrator of the error to resolve the contradiction, and then returns to step S502.

[0057] If no cycles exist or are resolved (No), in step S505, the process generation unit 201 forms a consistent and normal "directed graph" and calculates the ES (Early Start), EF (Early Finish), LS (Late Start), LF (Late Finish), and float (slack days) for each operation to identify the "critical path," which is the longest path that determines the overall construction period. The identified critical path and directed graph become the "baseline process (the basis for constraint conditions)" that is passed to the subsequent optimization processing unit 202.

[0058] In other embodiments not shown, the dependency rules may not only be based on static templates, but may also be dynamically inferred and applied by a machine learning model (learning unit 205) that has learned from event logs (actual data) of similar past projects, to propose new connection relationships that are optimal for a particular combination of work types or subcontractors (for example, a proposal for an SS relationship that performs two specific tasks in parallel).

[0059] Figure 7 is a conceptual diagram of a multi-objective optimization process according to one embodiment. In some embodiments, as shown in Figure 7, the optimization processing unit 202 of the information processing device 10 receives a baseline process 505, including the critical path and directed graph identified by the process generation unit 201, as the basis for constraint conditions. The optimization processing unit 202 then uses an evaluation function 601 having multiple evaluation axes to calculate multiple optimal solutions that satisfy the constraints of the physical sequence between work items while varying the weighting conditions between the multiple evaluation axes, thereby generating multiple process scenarios 602 that are in a trade-off relationship.

[0060] In the embodiment shown in Figure 7, the evaluation function 601 consists of six axes of indicators, including total construction period, labor peak, temporary cost, factory stop, quality risk, and permit risk. The optimization processing unit 202 may use an evaluation function 601, which is expressed as a linear sum (with coefficients a to f being weights), such as "Score = a × construction period + b × labor peak + c × temporary construction costs + d × factory shutdown + e × quality risk + f × notification risk," to search for a solution that optimizes (maximizes or minimizes) the score as a constraint satisfaction problem. By using the six-axis indicators described above, it becomes possible to quantitatively evaluate complex trade-offs unique to construction sites, such as not only minimizing the construction period, but also controlling the maximum number of personnel by trade (labor peak), controlling temporary construction costs based on overlapping days and daily compensation rates, reducing quality risks by considering inspection congestion and corrective buffers, and securing sufficient lead time (notification risk).

[0061] The optimization processing unit 202 automatically changes the weighting conditions (balance of coefficients a to f) in the evaluation function 601 and performs the calculation multiple times (or in parallel), thereby outputting multiple process scenarios 602 as shown in Figure 7.

[0062] The multiple process scenarios 602 may include a standard basic plan 602a and one or more alternative scenarios, each including at least one of the following: a shortened plan 602b that emphasizes minimizing the evaluation axis of the construction period, a leveling plan 602c that emphasizes minimizing the evaluation axis of the peak number of workers, or a minimal downtime plan 602d that emphasizes minimizing specific downtime periods. In the exemplary embodiment shown in Figure 7, the multiple process scenarios 602 include four scenarios: "Basic Plan 602a," which is a standard using standard work rates and basic number of teams; "Shortened Plan 602b," which prioritizes minimizing the evaluation axis of construction period and adds teams to critical work; "Leveling Plan 602c," which prioritizes minimizing the evaluation axis of peak labor numbers and suppresses job-specific peaks by consuming the float of non-critical work; and "Minimized Downtime Plan 602d," which prioritizes minimizing specific downtime periods and concentrates factory downtime work into specific periods.

[0063] In other embodiments not shown, the evaluation axes constituting the evaluation function 601 are not limited to the six axes described above. For example, a 7-axis or 8-axis evaluation function may be used that adds "CO2 emissions (Environment)" and "noise and vibration risk to neighboring areas (Community)" as evaluation axes for environmentally conscious projects.

[0064] Furthermore, the alternatives included in the generated process scenario 602 are not limited to the shortened option 602b, the leveling option 602c, or the minimum downtime option 602d. They may include one or more alternatives that are appropriate to the characteristics of the project, such as a "minimum cost option" for projects with strict budget limits, or a "minimum weather risk option" that best avoids weather risks such as rain.

[0065] Figure 8 is a diagram illustrating the flow of generating a comparative explanation of multiple scenarios according to one embodiment. Figure 9 is a conceptual diagram of prompt control according to one embodiment.

[0066] In some embodiments, as shown in Figures 8 and 9, the explanation generation unit 203 of the information processing device 10 takes difference data 801 of evaluation values ​​between multiple acquired process scenarios 602 as input and outputs it to a generation model 40 (external API) to generate comparative explanation text 702 in natural language. At that time, the prompt output via the explanation generation unit 203 is controlled to explicitly exclude commands (required days calculation formula, constraint determination command 802) related to the generation process of multiple process scenarios to be executed in the optimization processing unit 202, and the process of determining compliance with constraints on precedence.

[0067] In the embodiment shown in Figure 8, when the optimization processing unit 202 generates multiple process scenarios 602, the internal process of the information processing device 10 (for example, the difference data extraction unit 701) extracts "difference data" that shows how much the KPIs (evaluation values ​​on six axes such as construction period, labor peak, and temporary construction costs) of the alternative scenarios, including the shortened plan 602b, the leveling plan 602c, or the minimum stop plan 602d, have changed, based on the basic plan 602a. The explanation generation unit 203 constructs a prompt containing this difference data and sends it to an external generative model 40 (LLM) via the network 20. Based on the differences in the input KPIs, the generation model 40 generates comparative explanatory text 702 (such as a delay risk report or an explanation of the impact of the changes) in natural language that is easy for humans to intuitively understand, such as "The shortened plan can reduce the construction period by 16 days compared to the basic plan, but the labor peak will increase by 4 people," and returns it to the information processing device 10.

[0068] Here, we will explain, with reference to Figure 9, the mechanisms for realizing the "correct use of AI (limited application)" in several embodiments. The information processing device 10 (for example, the prompt control unit 803) strictly controls the content of the prompts sent to the generation model 40 at the system level. Specifically, the prompt control unit 803 is hardcoded to allow (OK) the inclusion of only "KPI / evaluation value difference data 801" that has already been determined and calculated by the solver in the prompt, while explicitly prohibiting (excluding) the inclusion of "required days calculation formula / constraint determination command 802" that would cause the generation model 40 to calculate the construction period itself or determine whether the dependency constraints are met in the prompt.

[0069] By incorporating this division of roles—"calculations to the solver, explanations to the LLM"—as an architectural boundary within the system (by sending only controlled prompts 804), it becomes possible to suppress hallucination (the risk of generating physically impossible processes) caused by probabilistic language generation, while using the superior natural language processing capabilities of the generative model 40 to support only the cognitively demanding task of comparing dry numerical data sets.

[0070] In other embodiments not shown, the explanation generation unit 203 may not only perform simple comparisons between scenarios, but also search for past trouble cases accumulated in the event log 306 shown in Figure 4 (for example, "In similar past projects, the risk of safety accidents increased by X% when labor peaks coincided at this time") using RAG (Retrieval-Augmented Generation) technology, etc., and incorporate the search results as additional elements into the prompt. As a result, the comparative explanatory text 702 output by the generative model 40 evolves beyond mere numerical explanations to include more sophisticated risk warnings (insights) based on the company's past failures and know-how.

[0071] Figure 10 is a user interface diagram of a scenario comparison screen according to one embodiment. Figure 11 is a user interface diagram of a Gantt chart screen according to one embodiment.

[0072] In some embodiments, as shown in Figures 10 and 11, the display control unit 204 of the information processing device 10 performs processing to display the evaluation values ​​(KPIs) for multiple evaluation axes related to multiple acquired process scenarios and the comparative explanatory text 702 on the same screen for comparison, and generates update data to immediately reflect the data of the process scenario selected by the user in the timeline of the Gantt chart screen executed on the field terminal 30.

[0073] In the embodiment shown in Figure 10, the display control unit 204 delivers web resources to the field terminal 30 to realize an SPA (Single Page Application) using a front-end technology such as React®. The browser on the field terminal 30 executes these resources and processes the DTO-format difference data provided sequentially from the display control unit 204, dynamically rendering the comparison table 901, highlighting the differences in evaluation values ​​of multiple evaluation axes between each scenario, and visually emphasizing (for example, changing the background color or adding a highlight frame) parts where differences in the process occur. This allows users to instantly understand which option can mitigate which risks from a vast amount of data.

[0074] At the top of screen 900 is a comparison table 901 that allows for the comparison of evaluation values ​​(six axes such as construction period and labor peaks) between multiple process scenarios, including the basic plan, shortened plan, leveling plan, and minimum stop plan. The display control unit 204 calculates the difference data between the standard basic plan and other plans, and performs visual emphasis 903 by applying dynamic styles such as CSS to parts where differences occur, such as "112 days (↓)" in the shortened plan or "6 people (↓)" in the leveled plan, by changing the background color or surrounding them with a border. This allows the user to instantly grasp "which plan can reduce which risks" from a vast amount of numerical data.

[0075] Furthermore, at the bottom of the screen, there is an AI report display area where comparative explanatory text 702 obtained from the generative model 40 is displayed.

[0076] Once the user has decided on the optimal plan on screen 900 and clicks the scenario selection button 904, an adoption request is sent from the field terminal 30 to the information processing device 10. In response, the display control unit 204 generates new process status data based on the selected scenario and provides it to the field terminal 30, and controls the immediate reflection (re-rendering) of the Gantt chart screen 1000 shown in Figure 11.

[0077] In the embodiment shown in Figure 11, the Gantt chart screen 1000 displays timelines (Gantt bars) 1002 for each WBS item 1001 based on the selected scenario. At this time, the critical path, which is the longest path that determines the overall construction period, is highlighted with a different color and thickness than other non-critical tasks. Additionally, a daily progress line (1004) is displayed, allowing for visual management of the discrepancy (delay) between the plan and actual results.

[0078] In other embodiments not shown, the Gantt chart screen 1000 may include an interactive editing function that, when a user directly changes the duration of a specific task using mouse operations (such as drag and drop), sends the change as a new constraint to the optimization processing unit 202 in the background, and immediately reflects the impact on related subsequent tasks and changes in KPIs by locally recalculating and re-optimizing them.

[0079] Figure 12 is a conceptual diagram of a 4D visualization simulation according to one embodiment. In some embodiments, as shown in Figure 12, the display control unit 204 links the timeline of the process scenario reflected on the Gantt chart screen 1000 with each component element of the 3D building model 1101 (BIM), causing the user terminal (site terminal 30) to perform a 4D visualization process that simulates the progress of construction in a virtual space over time.

[0080] In the embodiment shown in Figure 12, the display control unit 204 provides the field terminal 30 with model data for drawing the 3D building model 1101 and interface control information for 4D simulation, including a time axis slider 1102. The display control unit 204 (or a script running on the field terminal 30) executes the process data linkage process 1103 in response to the slider operation and performs drawing control on the screen of the field terminal 30, such as highlighting the component elements that are under construction at the specified date and time.

[0081] The process data for each WBS item is linked one-to-one or one-to-many in the database to the identifier of each component element that makes up the 3D building model 1101 (for example, ifc_guid in IFC format). The display control unit 204 responds to the user operating the time axis slider 1102 (moving the time axis) by executing the process data linkage process 1103, and sequentially highlights or switches between transparent and opaque rendering only the component elements that are under construction or completed at the specified date and time. This allows for visual confirmation of how the building will be assembled according to the selected process scenario, preventing rework due to spatial interference risks.

[0082] In other embodiments not shown, in addition to the 4D visualization process described above, a 5D simulation process may be performed that further links cost management data and displays the difference between planned costs and actual costs (EVM: Earned Value Management, etc.) at a specific date and time as a heatmap overlaid on the BIM model.

[0083] In another embodiment, the 4D visualization process described above may be linked to a just-in-time learning function. Specifically, when a user clicks on a specific component element on the 3D building model 1101, the system may include an educational interface that searches an educational content database (the extended table group in Figure 4) using the component's ifc_guid or product code as a key, and immediately displays a pop-up video (or a video with subtitles automatically summarized by the generation model 40) explaining the standard construction procedure for that component and the know-how of skilled workers. This allows users to intuitively confirm the correct procedure before starting work, contributing to the standardization of construction quality among junior workers and subcontractors.

[0084] Figure 13 is a flowchart of a self-evolution cycle according to one embodiment. In some embodiments, as shown in Figure 13, the information processing device 10 (learning unit 205) stores on-site operation history or performance data on the system as an immutable event log, and automatically updates the correction coefficient used to calculate the required number of days, or the weighting conditions between evaluation axes in the optimization processing unit, using machine learning based on the time-series data of the stored event log.

[0085] In the embodiment shown in Figure 13, first, each person in charge at the construction site performs daily operations and inputs such as registering the progress of work, taking photographs, inputting reasons for delays, or recording the occurrence of unexpected risks via the site terminal 30 (mobile app, etc.) (step S1200).

[0086] The data entered in S1200 is fully recorded in step S1201 in the event log storage base (event_log table) of the information processing device 10 as "immutable time-series data" indicating when, who, what, and how changes were made. This log has the characteristic of being append-only and not deletable, and functions as a highly accurate audit trail and data source for AI learning.

[0087] Next, in step 1202, the learning unit 205 processes the accumulated event log 306 as input for machine learning (e.g., regression analysis or reinforcement learning). The learning unit 205 analyzes the actual work speed under specific site conditions (narrow spaces, nighttime, etc.) and identifies the discrepancy between the "theoretical work rate" used in the initial arithmetic formula and the "actual work rate".

[0088] Then, in step 1203, the learning unit 205 automatically updates (optimizes) the correction coefficient used in the calculation of the number of days by the process generation unit 201 based on the analysis results, or autonomously adjusts the weighting conditions for the 6-axis evaluation in the optimization processing unit 202 (for example, setting a higher weight for quality risk in certain types of work). The updated parameters are immediately applied to subsequent process generation and optimization calculations. This allows the system to evolve as a business asset that becomes smarter the more it is used, unlike existing software which peaks at the time of implementation and becomes obsolete.

[0089] In other embodiments not shown, the learning unit 205 may comprehensively analyze not only work progress data but also on-site biological and environmental data acquired from IoT sensors (smart helmets, vital signs sensors, heavy equipment proximity sensors, etc.) as an event log. This makes it possible to automatically estimate sophisticated environmental correction factors, such as "work efficiency decreases by X% on extremely hot days."

[0090] Figure 14 is an overall processing sequence diagram of an automated process creation engine according to one embodiment. This diagram summarizes the flow of information between each component described so far (user terminal (field terminal 30), display control unit 204, process generation unit 201, optimization processing unit 202, and explanation generation unit 203).

[0091] First, when a user (such as a field project manager) performs a process creation operation via a dashboard screen running on the web browser of the field terminal 30, a "process creation request" (for example, an HTTP request to an API) is sent from the field terminal 30 to the information processing device 10 ("1. Process Creation Request").

[0092] In the information processing device 10, the display control unit 204 (e.g., a controller or API endpoint), which is responsible for communication with the field terminal 30 and receiving requests, receives a request and retrieves the necessary quantity, work rate, and master data from the database based on the content of the request, and passes them on to the process generation unit 201, which is a backend logical processing module ("2. Data Acquisition"). The process generation unit 201, based on the delivered data, performs a deterministic arithmetic calculation of the number of days and forms a directed graph including cycle detection to identify the baseline process (including critical path information) ("3. Deterministic Calculation and CP Identification").

[0093] Next, this baseline process is sent to the optimization processing unit 202, where four process scenarios (basic, shortened, leveled, and minimal stoppage) with trade-off relationships are generated in parallel using a 6-axis evaluation function based on multiple evaluation axes ("5. Multiple Scenario Generation"). The optimization processing unit 202 then sends only the "KPI difference data" extracted from these scenario data to the explanation generation unit 203 ("6. Sending Only KPI Differences").

[0094] The explanation generation unit 203 sends a request to the external generation model 40 using a controlled prompt that excludes computational instructions, and obtains an explanatory text in natural language ("7. Explanatory Text Generation").

[0095] Finally, the display control unit 204 distributes the numerical values ​​(with difference highlighting) and AI explanatory text for multiple scenarios to the field terminal 30, and displays them for comparison on the screen of the field terminal 30 ("8. Scenario + Explanatory Text" and "9. Comparison Screen Display (UI)").

[0096] When the user presses the "Adopt" button ("10. Scenario Adoption"), information for updating the Gantt chart and 4D / 5D DBIM model display is provided to the field terminal 30, and the changes are immediately reflected and the execution log is accumulated ("11. Immediate Gantt / 4D Reflection").

[0097] The characteristic configurations of the information processing device 10, information processing method, and program according to some of the embodiments described above can be summarized as follows.

[0098] [1] An information processing apparatus (10) according to at least some embodiments of the present invention is A process scenario acquisition unit acquires multiple process scenarios (602) that are in a trade-off relationship from an optimization processing unit (202) that calculates multiple optimal solutions for multiple work items constituting a construction process, using an evaluation function (601) having multiple evaluation axes, satisfying constraints on the physical sequence between the work items, and with different weighting conditions between the multiple evaluation axes. At a minimum, the explanation generation unit (203) takes the difference data (801) of the evaluation values ​​of the multiple evaluation axes between the acquired multiple process scenarios (602) as input and outputs a comparative explanation text (702) between the multiple process scenarios (602) in natural language to the generation model (40), It is equipped with.

[0099] According to the configuration described in [1] above, multiple optimal process scenarios with trade-off relationships are obtained by mathematical algorithms while reliably satisfying physical constraints, and only the "comparison of differences between scenarios," which is cognitively demanding for humans, is explained in natural language by the generative model. This eliminates hallucination (generation of physically contradictory processes) which is inherent in generative AI, and allows managers to make decisions on the optimal process in a short amount of time.

[0100] [2] In some embodiments, in the configuration of [1] above, The information processing device (10) is Prior to processing by the optimization processing unit (202), the process generation unit (201) further includes a process generation unit (201) that calculates the required number of days for each work item using at least the quantity of each work item, the standard work rate, and a correction coefficient based on site conditions, and forms a directed graph based on predetermined dependency rules to identify the critical path. The optimization processing unit (202) is configured to calculate the plurality of process scenarios (602) using the critical path and the directed graph identified by the process generation unit (201) as criteria for constraints.

[0101] According to the configuration described in [2] above, the number of days required for each task is derived using a definitive arithmetic formula to identify the base critical path, and optimization processing is performed based on this. This allows for the stable generation of realistic process scenarios that do not depend on subjective experience and are free from logical inconsistencies.

[0102] [3] In some embodiments, in the configuration of [2] above, The process generation unit (201) is configured to apply dependency rules to form the directed graph in a manner that satisfies multiple constraints, including limitations on the number of simultaneous operations in the same work area and a predetermined physical waiting period, and to automatically detect cycles in the directed graph and resolve inconsistencies.

[0103] According to the configuration described in [3] above, complex constraints such as conflicts between work sections and curing periods are comprehensively applied, and circular references in the process (a situation where B cannot start until A is finished, and A cannot start until B is finished) are automatically detected and resolved. As a result, even for large-scale projects with numerous work items, a logically sound schedule can be constructed.

[0104] [4] In some embodiments, in any of the configurations described in [1] to [3] above, The aforementioned multiple evaluation axes consist of at least six indicators, including total construction period, peak number of workers, temporary construction costs, specific downtime periods, quality risk, and notification risk.

[0105] According to the configuration described in [4] above, evaluation is performed not only on simple construction period and cost, but also on six axes specific to construction sites, such as labor peaks, downtime, quality, and notifications. This allows for the quantitative evaluation of complex trade-offs at the site (e.g., a short construction period but a sharp increase in labor peaks) and the deriving of a truly optimal solution.

[0106] [5] In some embodiments, in any of the configurations [1] to [4] above, The process scenario acquisition unit is configured to acquire, as part of the plurality of process scenarios (602), at least one of the following alternative scenarios: a basic plan (602a) that serves as a standard criterion, and one or more alternative plans that include at least one of the plurality of evaluation axes: a shortening plan (602b) that emphasizes minimizing the evaluation axis of construction period, a leveling plan (602c) that emphasizes minimizing the evaluation axis of peak number of workers, or a minimum downtime plan (602d) that emphasizes minimizing a specific downtime period.

[0107] According to the above configuration [5], alternative plans such as shortening, leveling, and minimizing downtime, which are most frequently used in on-site management decisions, are obtained in comparison with the basic plan. This allows managers to select the most suitable option from the presented choices without having to create multiple plans themselves.

[0108] [6] In some embodiments, in any of the configurations [1] to [5] above, When outputting a prompt to the generated model (40) via the explanation generation unit (203), the system is configured to exclude from the prompt commands related to the generation process of the multiple process scenarios (602) to be executed in the optimization processing unit (202), and the process for determining compliance with the constraints of the physical sequence.

[0109] According to the configuration described in [6] above, tasks that should be handled by the optimization processing unit, such as calculating construction time and determining constraints, are clearly excluded from the prompts for the generative model. This ensures that calculation errors and logical inconsistencies, which are weaknesses of LLMs, can be reliably prevented at the architectural level.

[0110] [7] In some embodiments, in any of the configurations described in [1] to [6] above, The information processing device (10) is The system includes a display control unit (204) configured to display the evaluation values ​​of the multiple evaluation axes for the multiple process scenarios (602) acquired and the comparison and explanation text (702) on the same screen (900), and to generate display data to immediately reflect the data of any of the process scenarios (602) selected by the user on the timeline of the Gantt chart screen (1000).

[0111] According to the configuration described in [7] above, KPIs and natural language explanations for multiple scenarios can be intuitively compared on the same screen, and the adopted option can be immediately reflected in the Gantt chart. As a result, even inexperienced users can finalize the overall project plan in a short time, achieving extremely high usability.

[0112] [8] In some embodiments, in the configuration of [7] above, The display control unit (204) is configured to automatically visually highlight (903) the differences in evaluation values ​​of the multiple evaluation axes and differences in processes between each scenario when comparing and displaying the multiple process scenarios (602).

[0113] According to the configuration described in [8] above, differences between scenarios (such as which processes have been extended or how much labor has increased) are automatically highlighted, allowing administrators to instantly grasp changes and risks that require attention from a large amount of data.

[0114] [9] In some embodiments, in the configuration of [7] or [8] above, The display control unit (204) is configured to link the timeline of the process scenario (602) reflected on the Gantt chart screen (1000) with each component element of the 3D building model, and to cause the user terminal to perform a 4D visualization process that simulates the progress of construction in a virtual space over time.

[0115] According to the configuration described in [9] above, the display control unit provides control data to link the time axis and model elements, enabling the user terminal to perform advanced 4D simulations. This makes it easier to visually share the latest process plan, which is centrally managed by the server (information processing device 10), among stakeholders, while entrusting the specialized rendering capabilities to the terminal.

[0116]

[10] In some embodiments, in the configuration of [2] or [3] above, The information processing device (10) is The system includes a learning unit (205) configured to automatically update, by machine learning, the correction coefficient used to calculate the required number of days, or the weighting conditions between evaluation axes in the optimization processing unit (202), based on the time-series data of the accumulated event log (306).

[0117] According to the configuration described in

[10] above, the more the system is used, the more actual on-site performance data is automatically accumulated, and the AI ​​autonomously adjusts the correction coefficients for work rates and the weighting for optimization, thus realizing a process management platform that "improves in accuracy the more it is used."

[0118]

[11] Information processing methods according to at least some embodiments of the present invention are A method of information processing performed by a computer, A process scenario acquisition step involves obtaining multiple process scenarios (602) that are in a trade-off relationship from an optimization processing unit (202) that calculates multiple optimal solutions for multiple work items constituting a construction process, using an evaluation function (601) having multiple evaluation axes, satisfying constraints on the physical sequence between the work items, and with different weighting conditions between the multiple evaluation axes. At a minimum, the explanation generation step involves taking difference data (801) of evaluation values ​​for the multiple evaluation axes between the acquired multiple process scenarios (602) as input and outputting a comparative explanation text (702) between the multiple process scenarios (602) to a generation model (40) in natural language, and Includes.

[0119] According to the method described in

[11] above, multiple optimal process scenarios with trade-off relationships are obtained by mathematical algorithms while ensuring that physical constraints are met, and only the "comparison of differences between scenarios," which is cognitively demanding for humans, is explained in natural language by the generative model. This eliminates hallucination (generation of physically contradictory processes) which is inherent in generative AI, and allows managers to make decisions on the optimal process in a short amount of time.

[0120]

[12] Programs according to at least some embodiments of the present invention are A process scenario acquisition function obtains multiple process scenarios (602) that are in a trade-off relationship from an optimization processing unit (202) that calculates multiple optimal solutions for multiple work items constituting a construction process, using an evaluation function (601) having multiple evaluation axes, satisfying constraints on the physical sequence between the work items, and with different weighting conditions between the multiple evaluation axes. At a minimum, the system includes an explanation generation function that takes difference data (801) of evaluation values ​​for multiple evaluation axes between the acquired multiple process scenarios (602) as input and outputs a comparative explanation text (702) between the multiple process scenarios (602) in natural language to the generation model (40), and This is a program that enables a computer to implement this.

[0121] According to the program described in

[12] above, multiple optimal process scenarios with trade-off relationships are obtained by mathematical algorithms while ensuring that physical constraints are met. Only the "comparison of differences between scenarios," which is cognitively demanding for humans, is explained in natural language by the generative model. This eliminates hallucination (generation of physically contradictory processes) which is inherent in generative AI, and allows managers to make decisions on the optimal process in a short amount of time. [Explanation of Symbols]

[0122] 10: Information Processing Device 20: Network 30: Field terminal 40: Generative Model 201: Process generation department 202: Optimization Processing Unit 203: Explanation generation unit 204: Display Control Unit 205: Learning Department 301: Project Master 302: Scenario Definition 303: WBS Item 304: Work Dependencies 305: Evaluation Metric Data 306: Event Log 505: Baseline Process 601: Evaluation function 602: Process Scenario 602a:Basic plan 602b: Shortened version 602c: Leveling plan 602d: Minimum stoppage plan 701: Difference Data Extraction Unit 702: Comparative Explanation Text 801: Difference Data 802: Constraint judgment instruction 803: Prompt Control Department 804: Controlled Prompt

Claims

1. An optimization processing unit calculates multiple optimal solutions for multiple work items constituting a construction process, using an evaluation function with multiple evaluation axes, satisfying constraints on the physical sequence of work items and with different weighting conditions between the multiple evaluation axes. From this optimization processing unit, a process scenario acquisition unit acquires multiple process scenarios that are in a trade-off relationship. At a minimum, the system includes an explanation generation unit that takes as input the difference data of evaluation values ​​of the multiple evaluation axes between the multiple process scenarios obtained, and outputs comparative explanatory text between the multiple process scenarios in natural language to the generation model, Equipped with, The explanation generation unit is configured to construct a prompt including the difference data, which causes the generation model to output the comparative explanation text based on the difference data, and to transmit it to the generation model. Information processing device.

2. Prior to processing by the optimization processing unit, the process generation unit further includes a process generation unit that calculates the required number of days for each work item using at least the quantity of each work item, the standard work rate, and a correction coefficient based on site conditions, and forms a directed graph based on predetermined dependency rules to identify the critical path. The optimization processing unit is configured to calculate the multiple process scenarios using the critical path and directed graph identified by the process generation unit as criteria for constraints. The information processing apparatus according to claim 1.

3. An information processing apparatus according to claim 2, The process generation unit is configured to apply dependency rules to form the directed graph in a manner that satisfies multiple constraints, including limitations on the number of simultaneous operations in the same work area and a predetermined physical waiting period, and to automatically detect cycles in the directed graph and resolve inconsistencies. Information processing device.

4. An information processing device according to any one of claims 1 to 3, The aforementioned multiple evaluation axes consist of at least six indicators, including total construction period, peak labor force, temporary construction costs, specific downtime, quality risk, and notification risk. Information processing device.

5. An information processing device according to any one of claims 1 to 3, The process scenario acquisition unit is configured to acquire, as part of the plurality of process scenarios, at least one of the following alternative scenarios: a basic plan that serves as a standard criterion, and at least one of the following alternative plans that emphasize minimizing the evaluation axis of construction period, a leveling plan that emphasizes minimizing the evaluation axis of peak labor numbers, or a plan that emphasizes minimizing specific downtime periods. Information processing device.

6. An information processing device according to any one of claims 1 to 3, When outputting a prompt for the generation model via the explanation generation unit, the system is configured to exclude from the prompt commands related to the generation process of the multiple process scenarios to be executed in the optimization processing unit, and the process for determining compliance with the constraints of the physical sequence. Information processing device.

7. An information processing device according to any one of claims 1 to 3, The system includes a display control unit configured to display the evaluation values ​​of the multiple evaluation axes for the multiple process scenarios acquired and the comparative explanatory text on the same screen, and to generate display data to immediately reflect the data of any process scenario selected by the user on the timeline of the Gantt chart screen. Information processing device.

8. An information processing apparatus according to claim 7, The display control unit is configured to automatically visually highlight and display the differences in evaluation values ​​of the multiple evaluation axes and the differences in processes between each scenario when comparing and displaying the multiple process scenarios. Information processing device.

9. An information processing apparatus according to claim 7, The display control unit is configured to link the timeline of the process scenario reflected on the Gantt chart screen with each component element of the three-dimensional building model, and to cause the user terminal to perform a 4D visualization process that simulates the progress of construction in a virtual space over time. Information processing device.

10. An information processing apparatus according to claim 2 or 3, The system includes a learning unit configured to automatically update, using machine learning, the correction coefficient used to calculate the required number of days, or the weighting conditions between evaluation axes in the optimization processing unit, based on the time-series data of the accumulated event logs, which are stored as immutable event logs. Information processing device.

11. A method of information processing performed by a computer, A process scenario acquisition step obtains multiple process scenarios that are in a trade-off relationship from an optimization processing unit that calculates multiple optimal solutions for multiple work items constituting a construction process, using an evaluation function with multiple evaluation axes, satisfying constraints on the physical sequence of work items, and with different weighting conditions for the multiple evaluation axes. At a minimum, the explanation generation step involves taking the difference data of evaluation values ​​for the multiple evaluation axes between the acquired multiple process scenarios as input and outputting a comparative explanatory text between the multiple process scenarios in natural language to the generation model, Includes, The explanation generation step includes constructing a prompt containing the difference data, which causes the generation model to output the comparative explanation text based on the difference data, and sending this prompt to the generation model. Information processing methods.

12. An optimization processing unit calculates multiple optimal solutions for multiple work items constituting a construction process, using an evaluation function with multiple evaluation axes, satisfying the constraints of the physical sequence between the work items, and with different weighting conditions between the multiple evaluation axes. From this optimization processing unit, a process scenario acquisition function acquires multiple process scenarios that are in a trade-off relationship. At a minimum, the system includes an explanation generation function that takes the difference data of evaluation values ​​for the multiple evaluation axes between the acquired multiple process scenarios as input and outputs a comparative explanation text between the multiple process scenarios in natural language to the generation model, A program to make a computer realize this, The explanation generation function includes constructing a prompt containing the difference data, which causes the generation model to output the comparative explanation text based on the difference data, and sending this prompt to the generation model. program.

Images