Prediction device, prediction method, and program

Abstract

Conventionally, by using one or more types of data sets among four types of data sets of actual data, simulation result data, experiment result data, and calculation result data, one or more other types of data sets cannot be obtained.SOLUTION: Using a learning model created by performing a learning process using a training data group including two or more kinds of data sets of an actual data set, a simulation result set, an experimental result set, and a calculation result set of a subject; and the received explanatory variable group; A prediction device 1 includes a prediction unit 131 that performs prediction processing of machine learning and acquires a prediction result, and an output unit 14 that outputs the prediction result acquired by the prediction unit 131.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a prediction device or the like that performs machine learning prediction processing using a learning model obtained using two or more types of datasets of the same subject (e.g., simulation data and actual machine data) and a group of explanatory variables including one or more types of datasets, obtains prediction results, and outputs the prediction results. [Background technology]

[0002] Conventionally, there has been a technology called Graph Neural Network (GNN), which is a deep learning model specially designed for processing data with a graph structure (see Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] “Introduction to Graph Neural Networks (GNNs)”, [online], [Retrieved June 17, 2024], Internet [URL: https: / / qiita.com / ymgc3 / items / a809d98abde5251bea15] Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the prior art, it was not possible to use one or more types of data sets among the four types of data sets, namely, actual data, simulation result data, experimental result data, and calculation result data, to obtain one or more types of data sets other than the one or more types of data sets. [Means for solving the problem]

[0005] The prediction device of the first invention is a prediction device comprising: a reception unit that receives a group of explanatory variables, which is one or more of four types of datasets, i.e., an actual dataset, a simulation result set, an experimental result set, and a calculation result set, for a single subject; a prediction unit that performs a machine learning prediction process using a learning model created by performing a learning process using a teacher data group including two or more types of datasets, i.e., the actual dataset, the simulation result set, the experimental result set, or the calculation result set, for a single subject, and the group of explanatory variables received by the reception unit, to obtain a prediction result, which is a group of objective variables corresponding to the group of explanatory variables; and an output unit that outputs the prediction result obtained by the prediction unit.

[0006] With this configuration, one or more types of data sets other than the one or more types of data sets can be obtained by using one or more types of data sets out of the four types of data sets: actual data, simulation result data, experimental result data, and calculation result data.

[0007] Furthermore, in the prediction device of the second invention, compared to the first invention, the teacher data group includes the same type of datasets for one subject at two or more different points in time, the explanatory variable group received by the reception unit includes a dataset for one subject at one point in time, and the prediction unit performs a machine learning prediction process using a learning model and the explanatory variable group received by the reception unit, and obtains a prediction result that is the same type of dataset as the dataset received by the reception unit but includes a dataset at a point in time different from the one point in time.

[0008] With this configuration, it is possible to obtain a prediction result that includes a dataset of the same type as the received data, but at a different time from the time of the received data.

[0009] Furthermore, the prediction device of the third invention is a prediction device according to the first or second invention, in which the learning model is a graph neural network (GNN).

[0010] With this configuration, one or more types of data sets out of four types of data sets, namely, actual data, simulation result data, experimental result data, and calculation result data, are used to obtain one or more types of highly accurate data sets other than the one or more types of data sets.

[0011] In addition, compared to the third invention, the prediction device of the fourth invention further comprises: a teacher data management unit in which two or more teacher data including two or more types of datasets selected from an actual dataset, a simulation result set, an experimental result set, and a calculation result set for one subject are stored; a knowledge management unit in which knowledge graph data, which is data based on two or more types of datasets and is data that serves as the basis for constructing graph data to be given to a learning module that performs machine learning learning processing, and has two or more pieces of knowledge node information that are node information and one or more pieces of knowledge edge information that are edge information; a graph construction unit that constructs graph data to be given to the learning module using the knowledge graph data and the teacher data for each of the two or more pieces of teacher data; and a learning unit that provides the two or more graph data constructed by the graph construction unit to the learning module, executes the learning module, and constructs a learning model, and the learning model used by the prediction unit when performing machine learning prediction processing is the learning model constructed by the learning unit.

[0012] This configuration allows the construction of a learning model that is a graph neural network (GNN).

[0013] Furthermore, in contrast to the third invention, the prediction device of the fifth invention further includes a knowledge management unit in which knowledge graph data is stored, the knowledge graph data being data based on two or more types of datasets and being original data for constructing graph data to be given to a learning module that performs machine learning learning processing, the knowledge graph data having two or more pieces of knowledge node information that are node information and one or more pieces of knowledge edge information that are edge information; and the prediction unit includes graph construction means that constructs graph data using the group of explanatory variables received by the reception unit and two or more pieces of knowledge graph data, and prediction means that performs machine learning prediction processing and acquires a prediction result using the graph data constructed by the graph construction means and a learning model.

[0014] With this configuration, one or more types of data sets out of four types of data sets, namely, actual data, simulation result data, experimental result data, and calculation result data, are used to obtain one or more types of highly accurate data sets other than the one or more types of data sets.

[0015] Furthermore, the prediction device of the sixth invention is a prediction device according to any one of the first to fifth inventions, in which the two or more types of datasets include an actual dataset and a simulation result set, the group of explanatory variables includes the simulation result set, and the prediction result includes the actual dataset.

[0016] With this configuration, highly accurate actual data can be obtained using the received simulation result data.

[0017] Furthermore, the prediction device of the seventh invention is a prediction device according to any one of the first to sixth inventions, in which the two or more types of data sets include a simulation result set, and the simulation result set is one or more results of finite element analysis.

[0018] With this configuration, highly accurate actual data can be obtained using simulation result data that is the result of the received finite element analysis. [Effects of the Invention]

[0019] According to the prediction device of the present invention, one or more types of data sets among four types of data sets, namely, actual data, simulation result data, experimental result data, and calculation result data, are used to obtain one or more types of data sets other than the one or more types of data sets. [Brief explanation of the drawings]

[0020] [Figure 1] Conceptual diagram of prediction system A in embodiment 1 [Figure 2] Block diagram of the prediction system A [Figure 3] A flowchart illustrating an example of the operation of the prediction device 1. [Figure 4] A flowchart illustrating an example of the prediction process [Figure 5] A flowchart illustrating an example of the graph configuration process. [Figure 6] 10 is a flowchart illustrating an example of the knowledge graph search process. [Figure 7] Flowchart illustrating an example of a match determination process [Figure 8] Flowchart illustrating an example of similarity determination processing [Figure 9] A flowchart illustrating an example of the inference process [Figure 10] Flowchart illustrating an example of node joining processing [Figure 11] A flowchart illustrating an example of the learning process [Figure 12] A diagram showing an example of the knowledge data [Figure 13] A diagram showing an example of the training data [Figure 14] FIG. 10 shows an example of the graph data. [Figure 15] FIG. 10 shows an example of the graph data. [Figure 16] FIG. 10 shows an example of the graph data. [Figure 17] FIG. 10 shows an example of the graph data. [Figure 18] Block diagram of the computer system DETAILED DESCRIPTION OF THE INVENTION

[0021] Hereinafter, embodiments of a prediction device and the like will be described with reference to the drawings. Note that components with the same reference numerals in the embodiments perform similar operations, and therefore repeated description may be omitted.

[0022] (Embodiment 1) In this embodiment, we will describe a prediction device that performs machine learning prediction processing using a learning model created by a learning process using a group of teacher data, which is two or more types of data sets (e.g., actual data, simulation results, experimental results, calculation results) of the same subject, and a group of explanatory variables including one or more types of data sets, obtains prediction results, and outputs the prediction results.

[0023] The training data set here may include two or more homogeneous datasets of the same subject at different time points. The learning model here is preferably a graph neural network (GNN). The two or more datasets here are preferably actual data and simulation results, the explanatory variables are preferably simulation results, and the prediction results are preferably actual data.

[0024] In this specification, information X being associated with information Y means that information Y can be obtained from information X, or information X can be obtained from information Y, and the method of association is not important. Information X and information Y may be linked, may exist in the same buffer, information X may be included in information Y, or information Y may be included in information X, etc.

[0025] Furthermore, in this specification, selecting or determining information Z means obtaining information Z, obtaining a pointer to information Z, obtaining the ID of information Z, setting a flag on information Z, etc., and it is sufficient if information Z can be accessed.

[0026] 1 is a conceptual diagram of a prediction system A in this embodiment. The prediction system A includes a prediction device 1 and a learning device 2. Note that if the prediction device 1 has a learning function, the learning device 2 is not necessary.

[0027] The prediction device 1 is a device that acquires prediction results. The prediction device 1 is, for example, a server, but may also be a terminal. The prediction device 1 is, for example, a cloud server or an ASP server, but the type does not matter. The prediction device 1 is, for example, a smartphone, a tablet terminal, a personal computer, etc., but the type does not matter.

[0028] The learning device 2 is a device that configures a learning model, which will be described later. The learning device 2 is, for example, a server, but it may also be a terminal. The learning device 2 is, for example, a cloud server or an ASP server, but the type does not matter. The learning device 2 is, for example, a smartphone, a tablet terminal, a personal computer, etc., but the type does not matter.

[0029] 2 is a block diagram of a prediction system A in this embodiment. The prediction device 1 includes a storage unit 11, a reception unit 12, a processing unit 13, and an output unit 14. The storage unit 11 includes a model management unit 111 and a knowledge management unit 112. The processing unit 13 includes a prediction unit 131. The processing unit 13 may include a graph construction unit 23 and a learning unit 24. The prediction unit 131 includes graph construction means 1311 and prediction means 1312.

[0030] The learning device 2 includes a learning storage unit 21 , a learning reception unit 22 , a graph construction unit 23 , a learning unit 24 , and a learning output unit 25 .

[0031] The prediction device 1 and the learning device 2 may be an integrated device or may be separate devices. If they are separate devices, it is preferable that the two devices be able to communicate with each other via a network such as the Internet.

[0032] Various types of information are stored in a storage unit 11 included in the prediction device 1. The various types of information are, for example, learning models described below.

[0033] The model management unit 111 stores one or more learning models. A learning model is information configured by a machine learning learning process and is information used in machine learning prediction processes. The learning model is preferably a model acquired by the learning unit 24 or the learning device 2. The learning model may also be called a learner, a classifier, a classification model, or the like. The machine learning algorithm is preferably deep learning, but may also be a random forest, a decision tree, or the like. Furthermore, for machine learning, various machine learning functions such as the TensorFlow (registered trademark) library, the random forest module of the R language, and fastText, as well as various existing libraries, can be used.

[0034] The learning model here is a model created by performing a learning process using a group of training data including two or more types of datasets for one subject, such as an actual dataset, a simulation result set, an experiment result set, or a calculation result set. The learning model is preferably a graph neural network (GNN), but may also be other data structures such as a neural network.

[0035] The learning model here is a model for obtaining an actual data set using one or more types of data set selected from a simulation result set, an experiment result set, and a calculation result set, for example.

[0036] The learning model here is a model for obtaining an experimental result set using one or two types of data sets, for example, a simulation result set or a calculation result set.

[0037] The learning model here is, for example, a model for using a dataset including a dataset of a single subject at a single point in time to obtain a dataset of the same type as the dataset at that single point in time but at a different point in time.

[0038] The term "subject" refers to the object to be predicted. Examples of subjects include astronomy, meteorology, physical property simulations such as optimal material design, optimal structural design of buildings, and fluid design, chemical simulations such as optimization of contaminated water treatment and polymer design, biological simulations such as protein behavior, epidemics, economic models, social simulations such as traffic flow and human flow dynamics, high-dimensional image simulations, social media analysis and marketing analysis, demand forecasts based on modeling of product characteristics and functions, production line optimization through modeling of manufacturing processes, environmental impact assessments, and optimization of energy supply / demand for power generation / consumption. However, the subject is not limited.

[0039] A real data set is one or more pieces of real data. Real data is real data. Examples of real data include data from a real machine or clinical trial data from humans. Real data is data collected directly from the real world. Real data is also called observational data, and includes field data and operational data. Specifically, real data includes, for example, weather observations, sensor logs, economic statistics, user behavior data, as well as image data, video data, and audio data.

[0040] A simulation result set is one or more simulation results for a target. A simulation result is information indicating the results of a simulation. The simulation result is preferably the result of a finite element analysis. The simulation result is data generated by using a computer model to mimic a real-world event. Examples of simulation results include physical simulations, weather simulations, economic models, etc.

[0041] An experimental result set is the results of one or more experiments on a subject. Experimental results are data obtained when an experiment is conducted in an experimental environment, not a real-world environment. Experimental results are data obtained from an experiment conducted under specific conditions in a controlled environment. Experimental results are data obtained, for example, from a scientific experiment or engineering test.

[0042] A calculation result set is one or more calculation results. A calculation result is data obtained by calculating an object.

[0043] Knowledge data is stored in the knowledge management unit 112. Knowledge data is data based on one or more types of data sets. Knowledge data is information for configuring graph data to be used in machine learning prediction processing or machine learning learning processing.

[0044] Knowledge data is, for example, graph data. Such graph data is called knowledge graph data. Knowledge graph data has two or more pieces of knowledge node information and one or more pieces of knowledge edge information. Knowledge node information is node information for constituting knowledge graph data. Knowledge edge information is edge information for constituting knowledge graph data.

[0045] Node information is information about the nodes that make up the graph data. Node information includes, for example, character strings and numerical values. In addition to character strings and numerical values, node information may also include multimodal data such as image data, video data, and audio data. Character strings and numerical values ​​are explanatory variables or objective variables. Character strings and numerical values ​​include, for example, the attribute name of a target or the attribute value of a target. Node information includes, for example, type information. Type information is information that indicates the type of data. Type information is, for example, an "actual data set," a "simulation result set," an "experiment result set," or a "calculation result set." Node information is associated with, for example, a node identifier. A node identifier is information that identifies a node. A node identifier is the node ID or name of a node.

[0046] Edge information is information that defines the relationship between two or more nodes. Edge information is information about edges that make up graph data. The edge information includes, for example, the node identifiers of each of the two nodes connected by the edge. Note that edges here are usually directed, but may be undirected.

[0047] The receiving unit 12 receives a group of explanatory variables. The receiving unit 12 may receive a prediction instruction having the group of explanatory variables. The receiving unit 12 may receive a learning instruction.

[0048] A prediction instruction is an instruction to perform a prediction process (to be described later) and obtain a prediction result. A learning instruction is an instruction to perform a learning process (to be described later) using two or more pieces of training data.

[0049] The explanatory variable group is a set of data used in the prediction process described below. The explanatory variable group is one or more of four types of data sets for one target: an actual data set, a simulation result set, an experiment result set, and a calculation result set. The receiving unit 12 receives the explanatory variable group, which is a digital signal.

[0050] Here, acceptance means, for example, receiving information transmitted via a wired or wireless communication line, but it may also be a concept that includes acceptance of information input from an input device such as a keyboard, mouse, or touch panel, or acceptance of information read from a recording medium such as an optical disk, magnetic disk, or semiconductor memory.

[0051] The processing unit 13 performs various types of processing. The various types of processing are, for example, processing performed by the prediction unit 131, the graph configuration unit 23 (described later), or the learning unit 24 (described later). The processing unit 13 processes digital signals.

[0052] The prediction unit 131 performs machine learning prediction processing using the learning model of the model management unit 111 and the group of explanatory variables received by the reception unit 12, and obtains a prediction result which is a group of objective variables corresponding to the group of explanatory variables.

[0053] The prediction unit 131 performs machine learning prediction processing, for example, using the learning model of the model management unit 111 and a group of explanatory variables accepted by the acceptance unit 12, and obtains a prediction result that includes a dataset of the same type as the dataset accepted by the acceptance unit 12 but at a point in time different from the one point in time.

[0054] It is preferable that the dataset for constructing the learning model includes actual data and simulation result data, the group of explanatory variables accepted by the accepting unit 12 includes a simulation result set, and the prediction result is an actual dataset.

[0055] The explanatory variable group received by the receiving unit 12 includes one or more types of data sets among a simulation result set, an experiment result set, and a calculation result set, and it is preferable that the prediction result is an actual data set.

[0056] For example, the explanatory variable group received by the receiving unit 12 includes one or two types of data sets, a simulation result set and a calculation result set, and the prediction result is an experiment result set.

[0057] For example, the explanatory variable group received by the receiving unit 12 includes one or two types of data sets from among an experiment result set and a calculation result set, and the prediction result is a simulation result set.

[0058] The machine learning algorithm is preferably deep learning using a GNN, but other algorithms such as deep learning using a neural network, random forest, or decision tree may also be used.

[0059] The graph construction means 1311 acquires graph data using the explanatory variable group received by the reception unit 12 and the knowledge data of the knowledge management unit 112. Details of the processing of the graph construction means 1311 will be described later.

[0060] It is preferable that the prediction means 1312 performs machine learning prediction processing and acquires a prediction result using the graph data constructed by the graph construction means 1311 and the GNN learning model. The means by which the prediction means 1312 performs prediction processing includes, for example, minimizing the error between the predicted value and the actual value using a loss function such as mean square error, mean absolute error, cross-entropy loss, or hinge loss, or evaluating the discrepancy between the predicted probability distribution of the model and the actual data distribution using negative log-likelihood, Kullback-Leibler divergence, Pearson's Chi-Squared, Jensen-Shannon divergence, etc.

[0061] The prediction means 1312 may, for example, construct a vector from the group of explanatory variables accepted by the accepting unit 12, and perform machine learning prediction processing using the vector and a learning model to obtain a prediction result. In such a case, the vector is, for example, a vector having each explanatory variable constituting the group of explanatory variables as an element. The learning model in such a case is a normal neural network or a folded neural network (CNN). When a learning model of a normal neural network or a folded neural network is used, the graph construction means 1311 is not necessary.

[0062] The output unit 14 outputs the prediction result acquired by the prediction unit 131. The output unit 14 may output a portion of the information acquired by the prediction unit 131. Note that the portion of the information is also a prediction result. The processing unit 13 outputs the prediction result as a digital signal.

[0063] Here, output is a concept that includes displaying on a display, projection using a projector, printing on a printer, sound output, transmission to an external device, storage on a recording medium, and delivery of processing results to other processing devices or other programs.

[0064] Various types of information are stored in the learning storage unit 21 that constitutes the learning device 2. The various types of information include, for example, a group of teacher data and knowledge data.

[0065] The training data group includes two or more training data, which may be two or more types of datasets selected from an actual dataset, a simulation result set, an experiment result set, and a calculation result set for a single subject.

[0066] The knowledge data is the same information as the knowledge data in the knowledge management unit 112. The knowledge data is usually knowledge graph data.

[0067] The learning reception unit 22 receives various instructions and information. The various instructions and information are, for example, learning instructions. A learning instruction is an instruction to perform learning processing. The learning reception unit 22 receives various instructions and information that are digital signals.

[0068] The means for inputting various instructions and information may be any means, such as a touch panel, keyboard, mouse, or menu screen.

[0069] The graph construction unit 23 performs the same processing as the graph construction means 1311. The graph construction unit 23 constructs graph data for each of two or more pieces of teacher data.

[0070] The learning unit 24 performs machine learning learning processing using the teacher data group in the learning storage unit 21 to construct a learning model. An example of such learning processing is the construction of a GNN, which will be described later. However, the learning unit 24 may also construct a learning model such as a neural network or a decision tree.

[0071] It is preferable that the learning unit 24 performs machine learning learning processing and constructs a learning model using two or more graph data constructed for each of two or more teacher data by the graph construction unit 23. Algorithms used by the learning unit 24 to construct a learning model include an algorithm that generates a learning model by expressing features using spectral or spatial graph convolution, graph autoencoder, message passing, or graph embedding that combines these with an attention mechanism.

[0072] The learning unit 24 may construct a vector for each of two or more pieces of training data, provide the two or more vectors to a learning module that performs machine learning learning processing, execute the learning module, and construct a learning model. Such a learning model may be, for example, a normal neural network or CNN.

[0073] The learning output unit 25 outputs the learning model constructed by the learning unit 24. The output here is usually stored in a recording medium. The recording medium may be the learning storage unit 21 or a recording medium of an external device. The learning output unit 25 outputs the learning model as a digital signal.

[0074] The storage unit 11, model management unit 111, knowledge management unit 112, and learning storage unit 21 are preferably non-volatile recording media, but can also be realized as volatile recording media.

[0075] There is no restriction on the process by which information is stored in the storage unit 11 etc. For example, information may be stored in the storage unit 11 etc. via a recording medium, information transmitted via a communication line etc. may be stored in the storage unit 11 etc., or information input via an input device may be stored in the storage unit 11 etc.

[0076] The reception unit 12 and the learning reception unit 22 may be realized, for example, by a device driver for an input means such as a touch panel or a keyboard, control software for a menu screen, etc. The reception unit 12 may also be realized by a wireless or wired communication means.

[0077] The processing unit 13, prediction unit 131, graph construction means 1311, prediction means 1312, graph construction unit 23, learning unit 24, and learning output unit 25 can usually be realized by a processor, memory, etc. The processing procedures of the processing unit 13, etc. are usually realized by software, and the software is recorded on a recording medium such as a ROM. However, they may also be realized by hardware (dedicated circuit). The processor may be a CPU, MPU, GPU, etc., and the type does not matter.

[0078] The output unit 14 may be realized by, for example, driver software for an output device such as a display or speaker, or a combination of driver software for the output device and the output device, etc. The output unit 14 may also be realized by wireless or wired communication means.

[0079] Next, an example of the operation of the prediction device 1 will be described with reference to the flowchart of FIG.

[0080] (Step S301) The receiving unit 12 determines whether or not a prediction instruction has been received. If a prediction instruction has been received, the process proceeds to step S302, and if a prediction instruction has not been received, the process proceeds to step S304. Note that a prediction instruction usually has an explanatory variable group or a link to an explanatory variable group.

[0081] (Step S302) The prediction unit 131 acquires a group of explanatory variables from the prediction instruction, and performs machine learning prediction processing to acquire a prediction result using the group of explanatory variables and the learning model of the model management unit 111. An example of the prediction processing will be described with reference to the flowchart in FIG.

[0082] (Step S303) The output unit 14 outputs all or part of the prediction result acquired in step S302. Return to step S301. The prediction result acquired in step S302 is, for example, graph data. The output unit 14 may, for example, extract one or more pairs of attribute names and attribute values ​​from the prediction result, which is graph data acquired in step S302, and output the one or more pairs of information (attribute names and attribute values).

[0083] (Step S304) The reception unit 12 determines whether or not a learning instruction has been received. If a learning instruction has been received, the process proceeds to step S305, and if a learning instruction has not been received, the process returns to step S301.

[0084] (Step S305) The learning unit 24 performs a learning process. An example of the learning process will be described with reference to the flowchart of FIG.

[0085] (Step S306) The learning output unit 25 stores the learning model acquired in step S305. Return to step S301. The learning output unit 25 stores the learning model in the model management unit 111, for example.

[0086] In the flowchart of FIG. 3, the process ends when the power is turned off or an interrupt occurs to end the process.

[0087] Next, an example of the prediction process in step S302 will be described with reference to the flowchart in FIG.

[0088] (Step S401) The graph construction unit 1311 acquires a group of explanatory variables based on a prediction instruction. The group of explanatory variables based on a prediction instruction is, for example, a group of explanatory variables included in the prediction instruction or a group of explanatory variables that can be acquired from a link (for example, a URL) included in the prediction instruction.

[0089] (Step S402) The graph construction unit 1311 constructs graph data using the explanatory variable group. An example of such graph construction processing will be described with reference to the flowchart in FIG.

[0090] (Step S403) The prediction means 1312 acquires the learning model from the model management unit 111. Here, the learning model is a GNN.

[0091] (Step S404) The prediction means 1312 provides the graph data acquired in step S402 and the learning model acquired in step S403 to a prediction module that performs machine learning prediction processing, executes the prediction module, acquires the prediction result, and returns to the upper processing.

[0092] Next, an example of the graph construction process in step S402 will be described with reference to the flowchart in FIG.

[0093] (Step S501) The graph construction means 1311 creates one or more primitive graphs from the group of explanatory variables. A primitive graph is primitive graph data. A primitive graph has two nodes and an edge connecting the nodes. Here, the primitive graph has, for example, a first node, a second node, and an edge. The first node has node information having an attribute name and type information. The second node has node information having an attribute value and type information. The edge has edge information having a directed graph from the second node to the first node.

[0094] (Step S502) The graph construction unit 1311 performs knowledge graph search processing using one or more primitive graphs acquired in step S501. An example of the knowledge graph search processing will be described with reference to the flowchart in FIG.

[0095] (Step S503) The graph construction means 1311 performs processing to combine the nodes acquired in step S502 and acquire graph data. The processing returns to the upper level processing. An example of such node combining processing will be described with reference to the flowchart in FIG.

[0096] Next, an example of the knowledge graph search process in step S502 will be described with reference to the flowchart in FIG.

[0097] (Step S601) The graph construction means 1311 assigns 1 to a counter i.

[0098] (Step S602) The graph construction means 1311 determines whether the i-th node exists in one or more acquired primitive graphs. If the i-th node exists, the process proceeds to step S603; if not, the process returns to the upper level process.

[0099] (Step S603) Graph construction means 1311 acquires node information of the i-th node.

[0100] (Step S604) The graph construction unit 1311 uses the node information of the i-th node to search for a node that matches the i-th node from the knowledge graph data. An example of such a match determination process will be described with reference to the flowchart in FIG.

[0101] (Step S605) The graph construction means 1311 determines whether or not a matching knowledge node exists in step S604. If a matching knowledge node exists, the process proceeds to step S609; if not, the process proceeds to step S606.

[0102] (Step S606) The graph construction means 1311 uses the node information of the i-th node to search for nodes similar to the i-th node from the knowledge graph data. An example of such similarity determination processing will be described with reference to the flowchart in FIG.

[0103] (Step S607) The graph construction means 1311 determines whether or not there is a knowledge node whose similarity is equal to or greater than the threshold. If there is a knowledge node whose similarity is equal to or greater than the threshold, the process proceeds to step S608; if there is not, the process proceeds to step S611.

[0104] (Step S608) The graph construction unit 1311 acquires the knowledge node identifier of the knowledge node with the greatest similarity, associates the knowledge node information with the node information acquired in step S603, and temporarily stores them in a buffer (not shown).

[0105] (Step S609) The graph construction means 1311 acquires the knowledge node identifier of the matching knowledge node.

[0106] (Step S610) The graph construction means 1311 increments the counter i by 1. The process returns to step S602.

[0107] (Step S611) The graph construction unit 1311 performs inference processing. An example of the inference processing will be described with reference to the flowchart in FIG.

[0108] (Step S612) The graph construction means 1311 acquires the knowledge node identifier of one knowledge node.

[0109] Next, an example of the match determination process in step S604 will be described with reference to the flowchart in FIG.

[0110] (Step S701) The graph construction means 1311 assigns 1 to a counter i.

[0111] (Step S702) The graph construction means 1311 determines whether or not the i-th knowledge node exists in the knowledge management unit 112. If the i-th knowledge node exists, the process proceeds to step S703; if not, the process proceeds to step S708.

[0112] It is preferable that the graph construction unit 1311 does not adopt an already mapped knowledge node as the i-th knowledge node. An already mapped knowledge node is a node of knowledge node information temporarily stored in a buffer (not shown) in step S706.

[0113] (Step S703) The graph construction means 1311 acquires the knowledge node information of the i-th knowledge node from the knowledge management unit 112.

[0114] (Step S704) Graph construction means 1311 determines whether the two pieces of node information match. If they match, proceed to step S705, and if they do not match, proceed to step S707. The two pieces of node information here are the node information acquired in steps S603 and S703.

[0115] (Step S705) The graph construction means 1311 assigns "match" to the variable "determination result".

[0116] (Step S706) The graph construction unit 1311 acquires the knowledge node information of the i-th knowledge node, associates the knowledge node information with the node information acquired in step S603, and temporarily stores the information in a buffer (not shown). The process then returns to the upper level process.

[0117] (Step S707) The graph construction means 1311 increments the counter i by 1. The process returns to step S702.

[0118] (Step S708) The graph construction means 1311 assigns "mismatch" to the variable "determination result" and returns to the upper level process.

[0119] Next, an example of the similarity determination process in step S606 will be described with reference to the flowchart in FIG.

[0120] (Step S801) The graph construction unit 1311 acquires a vector of node information using a technique such as Word2Vec.

[0121] (Step S802) The graph construction means 1311 assigns 1 to the counter i.

[0122] (Step S803) The graph construction means 1311 determines whether or not the i-th knowledge node exists in the knowledge management unit 112. If the i-th knowledge node exists, the process proceeds to step S804; if not, the process returns to the upper level process.

[0123] It is preferable that the graph construction means 1311 does not adopt a knowledge node that has already been mapped as the i-th knowledge node.

[0124] (Step S804) The graph construction means 1311 acquires the i-th knowledge node information from the knowledge management unit 112. The graph construction means 1311 acquires the vector of the knowledge node information.

[0125] (Step S805) The graph construction means 1311 obtains the similarity between the two vectors and associates the similarity with the i-th knowledge node.

[0126] (Step S806) The graph construction means 1311 increments the counter i by 1. The process returns to step S803.

[0127] Next, an example of the inference process in step S611 will be described with reference to the flowchart in FIG.

[0128] (Step S901) The graph construction means 1311 assigns 1 to a counter i.

[0129] (Step S902) The graph construction means 1311 determines whether or not the i-th knowledge node exists in the knowledge management unit 112. If the i-th knowledge node exists, the process proceeds to step S903; if not, the process returns to the upper level process.

[0130] It is preferable that the graph construction means 1311 does not adopt a knowledge node that has already been mapped as the i-th knowledge node.

[0131] (Step S 903 ) The graph construction means 1311 acquires the i-th knowledge node information from the knowledge management unit 112 .

[0132] (Step S904) Graph construction means 1311 determines whether the two pieces of node information correspond to each other. If the two pieces of node information correspond to each other, the process proceeds to step S905. If the two pieces of node information do not correspond to each other, the process proceeds to step S906.

[0133] The graph construction means 1311, for example, obtains the superordinate term of each of the two pieces of node information, and determines whether the two terms match or whether the similarity of the vectors obtained from each of the two terms is above a threshold.

[0134] The graph construction means 1311 constructs a prompt that includes, for example, two pieces of node information and asks whether the two pieces of node information have the same meaning, provides the prompt to the generation AI, and obtains an answer from the generation AI.

[0135] (Step S905) The graph construction means 1311 acquires the knowledge node information of the i-th knowledge node, associates the knowledge node information with the node information acquired in step S603, and temporarily stores the information in a buffer (not shown).

[0136] (Step S906) The graph construction means 1311 increments the counter i by 1. The process returns to step S902.

[0137] Next, an example of the node combining process in step S503 will be described with reference to the flowchart in FIG.

[0138] (Step S1001) The graph construction means 1311 assigns 1 to a counter i.

[0139] (Step S1002) The graph construction means 1311 determines whether the i-th node exists in the buffer (not shown) in which the knowledge node information is temporarily stored. If the i-th node exists, the process proceeds to step S1003; if not, the process returns to the upper process.

[0140] (Step S1003) The graph construction means 1311 assigns 1 to the counter j.

[0141] (Step S1004) Graph construction means 1311 determines whether or not another j-th node exists in a buffer (not shown). If another j-th node exists, the process proceeds to step S1005; if not, the process proceeds to step S1010.

[0142] (Step S1005) Graph construction means 1311 determines whether the two nodes were connected in the original graph. If they were connected, the process proceeds to step S1008, and if they were not connected, the process proceeds to step S1006.

[0143] (Step S1006) The graph construction unit 1311 acquires the distance between two nodes in the knowledge graph. Note that the distance between two nodes is calculated, for example, by calculating the shortest number of edges between the two nodes in the knowledge graph, the sum of the weights of the shortest edges between the two nodes in the knowledge graph, or the distance between the vectors of nodes obtained by convolving information about surrounding nodes, using a calculation such as cosine similarity.

[0144] (Step S1007) The graph construction means 1311 determines whether the distance acquired in step S1006 satisfies the join condition. If the join condition is satisfied, the process proceeds to step S1008; if not, the process proceeds to step S1009. The join condition is a condition for joining two nodes. For example, the join condition is that the distance is equal to or less than a threshold value.

[0145] (Step S1008) The graph construction unit 1311 constructs edge information of the edge that connects the nodes (two nodes) that constitute the primitive graph, which are nodes corresponding to the two nodes in the knowledge graph, and temporarily stores the edge information in a buffer (not shown).

[0146] (Step S1009) The graph construction means 1311 increments the counter j by 1. The process returns to step S1004.

[0147] (Step S1010) The graph construction means 1311 increments the counter i by 1. The process returns to step S1002.

[0148] Next, an example of the learning process in step S305 will be described with reference to the flowchart in FIG.

[0149] (Step S1101) The learning unit 24 assigns 1 to a counter i.

[0150] (Step S1102) The graph construction means 1311 determines whether or not the i-th teacher data exists. If the i-th teacher data exists, the process proceeds to step S1103, and if not, the process proceeds to step S1106.

[0151] (Step S1103) The graph construction means 1311 acquires the i-th teacher data from the storage unit 11.

[0152] (Step S1104) The graph construction unit 1311 converts the i-th teacher data into graph data. An example of such graph construction processing has been described with reference to the flowchart of FIG.

[0153] (Step S1105) The learning unit 24 increments the counter i by 1. The process returns to step S1102.

[0154] (Step S1106) The learning unit 24 performs learning processing using two or more graph data to obtain a learning model, and then returns to the upper-level processing.

[0155] Next, a description will be given of an example of the operation of the learning device 2. The example of the operation of the learning device 2 is the operations from S304 to S306 in FIG.

[0156] Specific examples of the operation of the prediction system A in this embodiment will be described below. The target here is, for example, contaminated water treatment. Two specific examples will be described. Specific example 1 is a learning process by the learning device 2. Specific example 2 is a prediction process by the prediction device 1.

[0157] (Example 1) It is assumed that the knowledge data shown in FIG. 12 is currently stored in the learning storage unit 21 of the learning device 2.

[0158] It is also assumed that the learning storage unit 21 stores a large amount of training data, including the training data shown in Fig. 13. The training data here is a set of simulation data, actual equipment data, and experimental data for contaminated water treatment.

[0159] Simulation data for contaminated water treatment is data used to virtually reproduce the treatment process of contaminated water. Such simulation data includes, for example, data on the types and concentration levels of specific contaminants contained in the contaminated water (e.g., radioactive materials, heavy metals, organic compounds, etc.), basic information on the quality of the contaminated water such as pH, temperature, amount of dissolved oxygen, and conductivity, parameters for evaluating the effectiveness of various treatment technologies (e.g., reverse osmosis, activated carbon adsorption, biological treatment, chemical precipitation), and data on the flow rate of water passing through the treatment system, the time required for treatment, or changes in the water flow rate over time. Actual equipment data in contaminated water treatment refers to actual measurement data obtained from actual treatment systems and equipment. Such actual equipment data includes, for example, treatment system operation data such as flow rate, treatment time, pressure, and temperature, inlet and outlet water quality data, equipment performance data such as treatment efficiency, energy consumption, and maintenance data, and environmental data such as ambient weather conditions and discharge data.

[0160] Experimental data in contaminated water treatment refers to data obtained from experiments conducted in laboratories or pilot plants, such as water quality characteristic data (e.g., initial contaminant concentration and water quality parameters), treatment condition data (e.g., treatment method and operating conditions), treatment efficiency data (e.g., removal rate, by-products, and water quality improvement), reaction kinetic data (e.g., reaction rate and reaction mechanism), or scale data.

[0161] Now, let us assume that the learning receiving unit 22 has received a learning instruction. Next, the graph construction unit 23 constructs graph data from the simulation data contained in the teacher data of FIG. 13, for example, by the above-described process using the knowledge data shown in FIG. 12. Note that any algorithm may be used to construct the graph data. Next, the graph construction unit 23 temporarily stores the graph data of the simulation data in a buffer (not shown). Such graph data is, for example, as shown in FIG. 14.

[0162] Furthermore, the graph construction unit 23 constructs graph data from the actual machine data included in the teacher data of Fig. 13 by the above-described process using the knowledge data shown in Fig. 12. Next, the graph construction unit 23 temporarily stores the graph data of the actual machine data in a buffer (not shown). Such graph data is, for example, as shown in Fig. 15.

[0163] Furthermore, the graph construction unit 23 constructs graph data from the experimental data included in the teacher data of Fig. 13 by the above-described process using the knowledge data shown in Fig. 12. Next, the graph construction unit 23 temporarily stores the graph data of the experimental data in a buffer (not shown). Such graph data is, for example, as shown in Fig. 16.

[0164] Next, the graph construction unit 23 combines, for example, the graph data of Figures 14, 15, and 16 using the knowledge data shown in Figure 12 to construct one piece of graph data, and temporarily stores it in a buffer (not shown). An example of such one piece of graph data is shown in Figure 17.

[0165] Then, the graph construction unit 23 performs the same process as above on other teacher data, constructs a large number of teacher data (graph data) as shown in FIG. 17, and stores them.

[0166] Next, the learning unit 24 provides the large amount of training data to a module that performs machine learning learning processing, executes the module, and acquires a learning model. Note that the learning model here is a GNN. Next, the learning output unit 25 accumulates the learning model. Note that the learning output unit 25 accumulates the learning model in, for example, the model management unit 111 of the prediction device 1.

[0167] (Example 2) For example, suppose that the receiving unit 12 of the prediction device 1 has received a prediction instruction having simulation data "Filter: 1μ filtration filter, Chemical Treatment: contamination_level: 100mg / L, Treatment efficiency: heavy metals 70%, organic compounds 50%, Input contamination level: 600mg / L, Output contamination_level: 500mg / L, ..." and which is an instruction to predict and output actual machine data.

[0168] Next, the graph construction means 1311 of the prediction device 1 constructs graph data having a structure as shown in FIG. 14 using the above-mentioned algorithm.

[0169] Next, the prediction means 1312 provides the graph data and the learning model accumulated in the model management unit 111 in specific example 1 to a prediction module that performs machine learning prediction processing, and executes the prediction module. Next, the prediction means 1312 acquires graph data of the actual machine having a structure as shown in FIG.

[0170] Next, the output unit 14 outputs the graph data, or / and part of the information in the graph data, or / and information obtained by flattening the graph data, "<Actual machine data> Filter: Cs colloid filter, Sr colloid filter, input Retention Time: 4 hours, output contamination_level 250 mg / L...".

[0171] As described above, according to this embodiment, one or more types of data sets among the four types of data sets, namely, actual data, simulation result data, experiment result data, and calculation result data, are used to obtain one or more types of data sets other than the one or more types of data sets.

[0172] Furthermore, according to this embodiment, it is possible to obtain a prediction result that includes a dataset of the same type as the received data, but which is a dataset at a time point different from the time point of the received data.

[0173] Furthermore, according to this embodiment, a learning model that is a graph neural network (GNN) can be configured.

[0174] Furthermore, according to this embodiment, highly accurate actual data can be obtained using the received simulation result data.

[0175] Furthermore, according to this embodiment, highly accurate actual data can be obtained using simulation result data, which is the result of the received finite element analysis.

[0176] The processing in this embodiment may be implemented by software. This software may be distributed by software download or the like. Furthermore, this software may be recorded on a recording medium such as a CD-ROM and distributed. This also applies to other embodiments in this specification. The software implementing the prediction device 1 in this embodiment is the following program. That is, this program causes a computer to function as: a reception unit that receives explanatory variable sets, which are one or more of four types of datasets for a single target: an actual dataset, a simulation result set, an experimental result set, and a calculation result set; a prediction unit that performs machine learning prediction processing using a learning model created by performing a learning process using a teacher data set including two or more types of datasets for the single target: the actual dataset, the simulation result set, the experimental result set, or the calculation result set; and the explanatory variable sets received by the reception unit, to obtain prediction results, which are target variable sets corresponding to the explanatory variable sets; and an output unit that outputs the prediction results obtained by the prediction unit.

[0177] FIG. 18 is a block diagram of a computer system 300 that executes the programs described in this specification to realize the prediction device 1 and the like according to the various embodiments described above.

[0178] In FIG. 18, a computer system 300 includes a computer 301 including a CD-ROM drive, a keyboard 302, a mouse 303, and a monitor 304.

[0179] 18, computer 301 includes, in addition to CD-ROM drive 3012, MPU 3013, bus 3014 connected to CD-ROM drive 3012 etc., ROM 3015 for storing programs such as a boot-up program, RAM 3016 connected to MPU 3013 for temporarily storing instructions of application programs and providing temporary storage space, and hard disk 3017 for storing application programs, system programs, and data. Although not shown here, computer 301 may further include a network card for providing connection to a LAN.

[0180] A program that causes the computer system 300 to execute the functions of the prediction device 1 and the like of the above-described embodiment may be stored on a CD-ROM 3101, inserted into the CD-ROM drive 3012, and then transferred to the hard disk 3017. Alternatively, the program may be transmitted to the computer 301 via a network (not shown) and stored on the hard disk 3017. The program is loaded into the RAM 3016 when executed. The program may also be loaded directly from the CD-ROM 3101 or the network.

[0181] The program does not necessarily include an operating system (OS) or a third-party program that causes the computer 301 to execute the functions of the prediction device 1 of the above-described embodiment. The program only needs to include instructions that call appropriate functions (modules) in a controlled manner to achieve the desired results. How the computer system 300 operates is well known, and a detailed description thereof will be omitted.

[0182] In addition, in the above program, the steps of transmitting information and receiving information do not include processing performed by hardware, such as processing performed by a modem or interface card in the transmission step (processing that can only be performed by hardware).

[0183] The computer that executes the program may be a single computer or a plurality of computers, that is, it may perform centralized processing or distributed processing.

[0184] Furthermore, in each of the above embodiments, each process may be realized by centralized processing in a single device, or may be realized by distributed processing in a plurality of devices.

[0185] The present invention is not limited to the above-described embodiment, and various modifications are possible, and it goes without saying that these modifications are also included within the scope of the present invention. [Industrial Applicability]

[0186] As described above, the prediction device 1 according to the present invention has the effect of being able to use one or more types of data sets out of four types of data sets, namely, actual data, simulation result data, experiment result data, and calculation result data, to obtain one or more types of data sets other than the one or more types of data sets, and is useful as a server or the like that receives prediction instructions from a terminal device and returns prediction results. [Explanation of symbols]

[0187] 1 Prediction device 2 Learning device 11 Storage area 12 Reception 13 Processing section 14 Output section 21 Learning storage unit 22 Learning Reception Department 23 Graph Construction 24 Learning Department 25 Learning output section 70 Treatment efficiency: Heavy metals 111 Model Management Department 112 Knowledge Management Department 131 Prediction Department 1311 Graph Construction Methods 1312 Prediction Methods

Claims

1. a receiving unit that receives a group of explanatory variables, which is one or more of four types of datasets, namely, an actual dataset, a simulation result set, an experiment result set, and a calculation result set, for one subject; a prediction unit that performs a machine learning prediction process using a learning model created by performing a learning process using a teacher data group including two or more types of datasets selected from an actual dataset, a simulation result set, an experiment result set, and a calculation result set for the one target, and the explanatory variable group received by the reception unit, and acquires a prediction result that is a target variable group corresponding to the explanatory variable group; an output unit that outputs the prediction result obtained by the prediction unit.

2. The training data group includes the same type of data set of the one subject at two or more different time points, the group of explanatory variables received by the receiving unit includes a dataset of the one subject at one time point, The prediction unit The prediction device according to claim 1, wherein a machine learning prediction process is performed using the learning model and the group of explanatory variables received by the reception unit, and a prediction result is obtained that includes a dataset of the same type as the dataset received by the reception unit but at a point in time different from the one point in time.

3. 3. The prediction device according to claim 1, wherein the learning model is a graph neural network (GNN).

4. a teacher data management unit in which two or more teacher data sets including two or more types of data sets selected from an actual data set, a simulation result set, an experiment result set, and a calculation result set for one subject are stored; a knowledge management unit in which knowledge graph data is stored, the knowledge graph data being based on the two or more types of data sets and being source data for configuring graph data to be provided to a learning module that performs machine learning learning processing, the knowledge graph data having two or more pieces of knowledge node information that are node information and one or more pieces of knowledge edge information that are edge information; a graph construction unit that constructs graph data to be provided to a learning module using the knowledge graph data and the teacher data for each of the two or more teacher data; a learning unit that provides the two or more graph data constructed by the graph construction unit to the learning module, executes the learning module, and constructs a learning model; The prediction device according to claim 3 , wherein the learning model used by the prediction unit when performing machine learning prediction processing is the learning model constructed by the learning unit.

5. a knowledge management unit in which knowledge graph data is stored, the knowledge graph data being based on the two or more types of data sets and being source data for configuring graph data to be provided to a learning module that performs machine learning learning processing, the knowledge graph data having two or more pieces of knowledge node information that are node information and one or more pieces of knowledge edge information that are edge information; The prediction unit graph construction means for constructing graph data using the explanatory variable group and the two or more pieces of knowledge graph data received by the reception unit; The prediction device according to claim 3 , further comprising: a prediction means for performing machine learning prediction processing using the graph data constructed by the graph construction means and the learning model, and acquiring the prediction result.

6. 6. The prediction device according to claim 1, wherein the two or more types of data sets include an actual data set and a simulation result set, the group of explanatory variables includes the simulation result set, and the prediction result includes the actual data set.

7. the two or more types of data sets include a simulation result set; The prediction device according to claim 1 , wherein the set of simulation results is one or more results of a finite element analysis.

8. A prediction method implemented by a reception unit, a prediction unit, and an output unit, a receiving step in which the receiving unit receives a group of explanatory variables, which is one or more types of datasets selected from four types of datasets, namely, an actual dataset, a simulation result set, an experiment result set, and a calculation result set, for one subject; a prediction step in which the prediction unit performs a machine learning prediction process using a learning model created by performing a learning process using a teacher data group including two or more types of datasets selected from an actual dataset, a simulation result set, an experiment result set, and a calculation result set for the one target, and the explanatory variable group received by the reception unit, to obtain a prediction result which is a target variable group corresponding to the explanatory variable group; an output step in which the output unit outputs the prediction result obtained by the prediction unit.

9. Computer, a receiving unit that receives a group of explanatory variables, which is one or more of four types of datasets, namely, an actual dataset, a simulation result set, an experiment result set, and a calculation result set, for one subject; a prediction unit that performs a machine learning prediction process using a learning model created by performing a learning process using a teacher data group including two or more types of datasets selected from an actual dataset, a simulation result set, an experiment result set, and a calculation result set for the one target, and the explanatory variable group received by the reception unit, and acquires a prediction result that is a target variable group corresponding to the explanatory variable group; A program for causing the prediction unit to function as an output unit that outputs the prediction result acquired by the prediction unit.

Images