Information processing device, information processing method, and program

By transforming time-series data into a model with multidimensional vectors, the information processing device efficiently identifies nodes for quick command value generation, addressing the computational bottlenecks of existing methods and ensuring rapid response in systems with time-series data.

JP7577385B2Active Publication Date: 2024-11-05SOINN INC
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
JP2023542213
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2021-08-18
Filing Date
2022-03-24
Publication Date
2024-11-05
Estimated Expiration
2042-03-24

AI Technical Summary

Technical Problem

Existing methods for estimating command values in systems with time-series data, such as the k-nearest neighbor method, face challenges with high computational time due to the need to calculate distances between all data elements, which can hinder quick response times, especially in systems requiring immediate command value generation.

Method used

An information processing device that transforms the distribution structure of time-series data into a model using multidimensional vectors, allowing for the identification of nodes that approximate input data from a smaller number of nodes, thereby enabling quick command value generation by searching through a reduced set of nodes.

Benefits of technology

This approach significantly reduces the number of nodes requiring distance calculations, enhancing processing speed and enabling rapid command value output based on the system's operating status and state quantity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007577385000027
    Figure 0007577385000027
  • Figure 0007577385000028
    Figure 0007577385000028
  • Figure 0007577385000029
    Figure 0007577385000029
Patent Text Reader

Abstract

A model holding unit (1) holds a model trained by using training data based on time-series data including a plurality of data elements each expressed by a multidimensional vector including a first state quantity acquired in advance from a training subject and a first command value that corresponds to the first state quantity and is given to control the operation of the training subject. The model is created to represent training data distribution structure as a set of nodes each expressed by a multidimensional vector including a second state quantity and second command value based on the training result. A retrieval unit (2) receives input of a third state quantity acquired from a control subject as input data and retrieves a node that matches or is similar to the input data from among a smaller number of nodes than the number of data elements of the training data included in the model. An output unit (3) outputs a value based on the second command value of the retrieved node to the control subject as an output command value to be given to operate the control subject.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] Today, various data processing methods are used, and there are various types of data to be processed, such as data in which values ​​are systematically arranged. A typical example of such systematic data is time series data, which is a set of values ​​obtained by continuously or intermittently observing a certain phenomenon, that is, a set of values ​​that show the change of a phenomenon over time.

[0003] Time series clustering is known as a method for analyzing such time series data. There are three known time series clustering methods: Whole time-series clustering, Subsequence clustering, and Time point clustering (Non-Patent Documents 1 and 2). In Whole time-series clustering, clustering is performed by measuring the similarity for each of a plurality of time series data (time series set). In Subsequence clustering, one time series data (time series set) is divided into a plurality of segments, and clustering is performed for each segment. In Time point clustering, one time series data (time series set) is divided into points, and clustering is performed by measuring the similarity between each point. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2008-217246 A [Patent Document 2] JP 2014-164396 A [Non-patent literature]

[0005] [Non-Patent Document 1] @ZoneTsuyoshi, "Toeplitz Inverse Covariance-based Clustering ~Time Series Clustering~", December 6, 2018, Qiita, retrieved February 9, 2021,<URL: https: / / qiita.com / ZoneTsuyoshi / items / 53dad1d0f0d43512a913> [Non-Patent Document 2] Hiroshi, "I read a research survey paper on time series clustering," July 20, 2019, 10001 IDEAS, retrieved February 9, 2021,<URL: https: / / 10001ideas.com / 2019 / 07 / 20 / time-series-survey-paper / > Summary of the Invention [Problem to be solved by the invention]

[0006] For example, a method is widely used in which a state quantity of a certain system is acquired and a command value corresponding to the state quantity is given to control the behavior of the system. In such a control method, when the control process is performed automatically, a suitable command value is estimated based on the state quantity of the system acquired at an arbitrary time point, and the estimated command value is given to the system. In this case, learning data, which is time-series data consisting of, for example, a history of operations (command values) performed by a person and the state quantity of the system, which have been observed in advance, is learned by machine learning, and the command value is estimated based on the learning result.

[0007] In this case, for example, the k-nearest neighbor method is known for estimating a command value corresponding to the state quantity of the input data. The k-nearest neighbor method uses the learning data as it is, and outputs an operation value from k pieces of history data close to the input data. The output value can be, for example, the average value of the command value after a predetermined time has elapsed for each of the k pieces of history data close to the input data. The k-nearest neighbor method does not require learning compared to other supervised learning, so it does not require a huge amount of data and is used in various fields because the operation is within the range of the provided data. However, the k-nearest neighbor method has a problem that the amount of calculation is large and the time required for calculation is long because the distance is calculated between all data elements of the learning data and the input data.

[0008] This does not pose a problem in a system in which there is sufficient time between acquiring the state quantities of the system to be controlled and giving a command value; however, in a system such as a vehicle in which a command value must be given quickly in response to acquiring the state quantities, the length of calculation time becomes an obstacle to giving a command value quickly.

[0009] Therefore, there is a need to establish a method that can output command values ​​corresponding to the state quantities of the system at higher speed depending on the operating conditions of the system to be controlled.

[0010] The present invention has been made in view of the above circumstances, and has an object to quickly give a command value to a system according to the operating condition and state quantity of the system to be controlled. [Means for solving the problem]

[0011] An information processing device according to one aspect of the present invention includes: a model holding unit that holds a model created by learning learning data based on time-series data including a plurality of data elements represented by a multidimensional vector including a first state quantity previously acquired from a learning object and a first command value corresponding to the first state quantity given to control the operation of the learning object, the model holding unit holds a model created by learning a distribution structure of the learning data as a set of nodes represented as a multidimensional vector including a second state quantity and a second command value based on a learning result; a search unit that receives a third state quantity acquired from a control object as input data and searches for a node that matches or approximates the input data from the nodes that are fewer than the number of data elements of the learning data included in the model; and an output unit that outputs a value based on the second command value of the searched node to the control object as an output command value to be given to operate the control object. This makes it possible to quickly give a command value according to the operating situation and state quantity of the control object to the control object.

[0012] An information processing device according to one aspect of the present invention is the information processing device described above, in which the model is created by learning each of the learning data as a node, and the search unit desirably selects, from the nodes, nodes that are relatively close in time to the input data as the small number of the nodes, and searches for a node that is the shortest distance from the selected node to the input data. This makes it possible to create a model used for estimating a command value and identify the node that is the shortest distance when quickly giving a command value to a control object according to an operating situation and state quantity of the control object.

[0013] An information processing device according to one aspect of the present invention is the information processing device described above, in which the model is created by time-series clustering the learning data and classifying the nodes into a plurality of clusters, and the search unit preferably searches a cluster to which the input data belongs and a cluster immediately following the cluster to which the input data belongs, and searches for a node that is the shortest distance from the input data among the nodes that belong to the two searched clusters. This makes it possible to create a model used for estimating a command value when quickly giving a command value to a control object according to the operating situation and state quantity of the control object, and to specifically identify the node that is the shortest distance.

[0014] In the information processing device according to one aspect of the present invention, it is preferable that the search unit determines the output command value based on statistics calculated from command values ​​of some or all of the nodes in the cluster to which the searched node belongs. This makes it possible to determine an appropriate output command value as needed.

[0015] An information processing device according to one aspect of the present invention is the above information processing device, further comprising a model creation unit that creates the model by approximately learning a distribution structure of the data elements of the learning data using a smaller number of nodes than the data elements of the learning data, and outputs the created model to the model holding unit, and it is preferable that the search unit searches for a node that approximates the input data from the nodes included in the model. This makes it possible to create a model used for estimating a command value and identify a node with the shortest distance when quickly giving a command value to a control object according to the operating situation and state quantity of the control object.

[0016] In an information processing device according to an aspect of the present invention, the search unit preferably outputs a command value of the searched node as the output command value. This makes it possible to create a model used for estimating a command value and identify a node at the shortest distance when quickly providing a command value to a control object according to an operating situation and a state quantity of the control object.

[0017] In the information processing device according to one aspect of the present invention, it is preferable that the search unit determines the output command value based on a statistic calculated from a command value of the searched node and command values ​​of one or more nodes that are similar to the searched node. This makes it possible to appropriately determine the output command value.

[0018] An information processing device according to one aspect of the present invention is the above information processing device, and it is preferable that the one or more nodes close to the searched node are nodes within a predetermined distance from the searched node, or a predetermined number of nodes selected in order of proximity to the searched node. This makes it possible to determine an output command value as needed.

[0019] In the information processing device according to one aspect of the present invention, it is preferable that the statistical amount is any one of an average value, a median value, a maximum value, a minimum value, and a mode value, thereby making it possible to determine an output command value as required.

[0020] In an information processing device according to an aspect of the present invention, it is preferable that the second state quantity is the first state quantity previously acquired from the learning object, and the second command value is the first command value given to the learning object according to the first state quantity previously acquired. This makes it possible to create a model based on appropriate data.

[0021] In the information processing device according to one aspect of the present invention, it is preferable that the first command value is an actual value of a command value given to the learning object by an operator of the learning object based on the first state quantity previously acquired from the learning object, thereby making it possible to create a model based on appropriate data.

[0022] In an information processing device according to an aspect of the present invention, the first command value included in the data element of the learning data is preferably a command value given to the control target a predetermined time after the first state quantity previously acquired from the learning target, thereby making it possible to create a model that takes into account a time lag between acquisition of the state quantity and output of the command value.

[0023] An information processing method according to one aspect of the present invention is a method for learning learning data based on time-series data including a plurality of data elements expressed as a multidimensional vector, including a first state quantity previously acquired from a learning object and a first command value corresponding to the first state quantity given to control the operation of the learning object, and holds a model created as a set of nodes expressed as a multidimensional vector including a second state quantity and a second command value based on a learning result, a third state quantity acquired from a control object is input as input data, a node that matches or approximates the input data is searched for from the nodes that are fewer than the number of data elements of the learning data included in the model, and a value based on the second command value of the searched node is output to the control object as an output command value to be given to operate the control object. This allows a command value according to the operating situation and state quantity of the control object to be quickly given to the control object.

[0024] A program according to one aspect of the present invention causes a computer to execute the following steps: learning learning data based on time-series data including a plurality of data elements represented by a multidimensional vector, including a first state quantity previously acquired from a learning object and a first command value corresponding to the first state quantity given to control the operation of the learning object, and holding a model created as a set of nodes represented as a multidimensional vector including a second state quantity and a second command value based on a learning result, a third state quantity acquired from a control object is input as input data, and a node that matches or approximates the input data is searched for from the nodes that are fewer than the number of data elements of the learning data included in the model, and a value based on the second command value of the searched node is output to the control object as an output command value to be given to operate the control object. This allows a command value according to the operating situation and state quantity of the control object to be quickly given to the control object. Effect of the Invention

[0025] According to the present invention, command values ​​according to the operating conditions and state quantities of a system to be controlled can be quickly given to the system. [Brief description of the drawings]

[0026] [Figure 1] FIG. 2 is a diagram illustrating an example of a system configuration for implementing an information processing device according to a first embodiment. [Diagram 2] 1 is a diagram showing an external configuration of an information processing device according to a first embodiment; [Diagram 3] FIG. 1 is a diagram illustrating a schematic configuration of an information processing device according to a first embodiment. [Figure 4] FIG. 1 is a diagram illustrating a situation in which an information processing device according to a first embodiment is used. [Diagram 5] FIG. 1 is a diagram showing an outline of a robot used in an experiment. [Figure 6] FIG. 13 is a diagram showing an example of an operation to be learned by a robot arm. [Figure 7]1A to 1C are diagrams illustrating forces acting in various states when the robot arm avoids an obstacle. [Figure 8] FIG. 2 is a diagram illustrating a format of time-series data. [Figure 9] FIG. 13 is a diagram showing an example of learning the movement of a robot arm. [Figure 10] FIG. 2 is a diagram showing nodes created by a model creation unit. [Figure 11] 4 is a flowchart of a process in the information processing device according to the first embodiment. [Figure 12] 11 is a flowchart of a modified example of the process in the information processing device according to the first embodiment. [Figure 13] FIG. 2 is a diagram showing nodes that are targets of distance calculation in the processing in the information processing device according to the first embodiment and in the processing in the k-nearest neighbor method. [Figure 14] FIG. 11 is a diagram illustrating a configuration of an information processing device according to a second embodiment. [Figure 15] FIG. 11 is a diagram illustrating a situation in which an information processing device according to a second embodiment is used. [Figure 16] 11 is a flowchart showing an operation of the information processing device according to the second embodiment. [Figure 17] FIG. 11 is a diagram illustrating a configuration of a model creating unit according to a second embodiment. [Figure 18] 1 is a flowchart of a learning process using the SOINN method. [Figure 19] 13 is a diagram showing an example of a distribution of nodes obtained by learning the learning data used in the first embodiment by the SOINN method in the information processing device according to the second embodiment; FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the drawings, the same elements are given the same reference numerals, and duplicated explanations will be omitted as necessary.

[0028] First embodiment FIG. 1 is a diagram showing an example of a system configuration for realizing an information processing device according to the first embodiment. The information processing device 100 can be realized by a computer 1000 such as a dedicated computer or a personal computer (PC). However, the computer does not need to be a single computer physically, and may be multiple computers when performing distributed processing. As shown in FIG. 1, the computer 1000 has a CPU (Central Processing Unit) 1001, a ROM (Read Only Memory) 1002, and a RAM (Random Access Memory) 1003, which are connected to each other via a bus 1004. Note that an OS software for operating the computer will not be described, but it is assumed that the computer that constructs this information processing device also has such software.

[0029] An input / output interface 1005 is also connected to the bus 1004. To the input / output interface 1005, for example, an input unit 1006 including a keyboard, a mouse, a sensor, etc., a display including a CRT, an LCD, etc., an output unit 1007 including a headphone, a speaker, etc., a storage unit 1008 including a hard disk, etc., and a communication unit 1009 including a modem, a terminal adapter, etc. are connected.

[0030] The CPU 1001 executes various processes, for example, processes of each unit of the information processing device 100 described later, in accordance with various programs stored in the ROM 1002 or various programs loaded from the storage unit 1008 to the RAM 1003 in this embodiment. A GPU (Graphics Processing Unit) may be provided separately from the CPU 1001 to perform the same processes as the CPU 1001. The GPU is suitable for performing routine processes in parallel, and can be applied to learning processes described later to improve the processing speed compared to the CPU 1001. The RAM 1003 also stores data necessary for the CPU 1001 and the GPU to execute various processes as appropriate.

[0031] The communication unit 1009 performs communication processing, for example, via the Internet (not shown), transmits data provided by the CPU 1001, and outputs data received from a communication partner to the CPU 1001, RAM 1003, and storage unit 1008. The storage unit 1008 exchanges information with the CPU 1001 and stores and erases information. The communication unit 1009 also performs communication processing of analog or digital signals with other devices.

[0032] The input / output interface 1005 is also connected to a drive 1010 as necessary, and, for example, a magnetic disk 1011, an optical disk 1012, a flexible disk 1013, or a semiconductor memory 1014 is appropriately attached, and computer programs read from them are installed in the memory unit 1008 as necessary.

[0033] The external configuration of the information processing device 100 will be described. Fig. 2 shows the external configuration of the information processing device 100 according to the first embodiment. The information processing device 100 has a processing unit 110, a display unit 120, and an input unit 130. The processing unit 110 is configured as hardware having the above-mentioned CPU 1001, ROM 1002, RAM 1003, bus 1004, input / output interface 1005, storage unit 1008, communication unit 1009, drive 1010, and the like. The display unit 120 corresponds to the above-mentioned output unit 1007, and is configured as a display device such as an LCD that displays images in a format that can be viewed by an operator. The input unit 130 corresponds to the above-mentioned input unit 1006, and is configured by various input means such as a mouse and a keyboard.

[0034] Next, the time series data TSD, which is input data to be learned by the information processing device 100, will be described. The time series data TSD is given as a set of multidimensional vectors representing data elements. If i is a parameter representing the order of the data elements, then the multidimensional vector a i is the time t i and a vector containing m components p, as follows: where i is an integer equal to 2 and m is an integer equal to or greater than 1.

number

number

[0035] The time series data TSD is stored in, for example, a storage unit (for example, the RAM 1003 or the storage unit 1008) provided in the information processing device 100.

[0036] Machine learning is applied to the information processing device 100 described below, and an overview of machine learning will be described as a prerequisite for this. The general machine learning described below is merely a prerequisite for understanding the control system described in the following embodiment, and the machine learning applied to the control system is not limited to this.

[0037] Machine learning can be broadly divided into supervised learning and unsupervised learning. Below, we will provide an overview of each method.

[0038] In general, supervised learning involves learning to predict a certain variable (objective variable) from given variables (explanatory variables). More specifically, supervised learning is a method of learning the relationship between input data (explanatory variables) and correct answer data (objective variable) by providing correct answer data for input data (explanatory variables).

[0039] For example, when the correct data is a continuous value, learning is performed by regression analysis. Note that the method of learning continuous data is not limited to regression analysis (e.g., linear regression). In regression analysis, it is possible to predict the output corresponding to the input data by fitting the input data with various functions.

[0040] In addition, when the correct answer data for the input data is label information, learning by classification is performed. In learning by classification, for example, methods such as regression (logistic regression, support vector machine), tree (decision tree, random forest), neural network, clustering (k nearest neighbor method, etc.) are used.

[0041] In unsupervised learning, the features of input data are learned without being given correct answer data. Unsupervised learning methods include clustering such as k-means and SOINN, dimensionality reduction such as PCA, and Hotelling's T 2 For example, clustering can extract similar features from input data and group them together.

[0042] Here, we will explain the SOINN method. The SOINN method is a learning technique that grows neurons as necessary during learning, and is known as a Self-Organizing Incremental Neural Network (SOINN). SOINN has many advantages, such as being able to learn non-stationary inputs by autonomously managing the number of nodes, and being able to extract an appropriate number of classes and topological structure even for classes with complex distribution shapes. As an example of an application of SOINN, in pattern recognition, after learning classes of hiragana characters, it is possible to additionally learn classes of katakana characters, etc.

[0043] As an example of such SOINN, a method called E-SOINN (Enhanced SOINN, Patent Document 1) has been proposed. E-SOINN allows online additional learning in which learning is added at any time, and has the advantage of being more efficient than batch learning. Therefore, E-SOINN allows additional learning even when the learning environment changes to a new environment. E-SOINN also has the advantage of being highly resistant to noise in input data. In addition, various SOINN methods have been proposed, such as LB-SOINN (Load Balance Self-Organizing Incremental Neural Network, Patent Document 2).

[0044] In the SOINN method, a neural network having a plurality of nodes is applied. Specifically, a non-hierarchical neural network in which nodes described by n-dimensional vectors (n is an integer equal to or greater than 1) are arranged is input to the information processing device 100. The neural network is stored in a storage unit such as the RAM 1003.

[0045] The neural network in the SOINN method is a self-replicating neural network that inputs an input vector into the neural network and automatically increases the number of nodes and edges placed in the neural network based on the input vector. By using a self-replicating neural network, it is possible to automatically increase the number of nodes.

[0046] The neural network in the SOINN method has a non-hierarchical structure. By adopting a non-hierarchical structure, additional learning can be performed without specifying the timing to start learning in other layers. In other words, additional learning can be performed online.

[0047] Clustering in the SOINN method refers to nodes and edges and performs class classification. Various classification methods can be applied to classify the nodes constituting the network. For example, processing similar to that of LB-SOINN in Patent Document 2 may be performed.

[0048] The following describes the configuration and operation of the information processing device 100 according to the embodiment 1. Fig. 3 shows a schematic configuration of the information processing device 100 according to the embodiment 1. The information processing device 100 includes a model holding unit 1, a search unit 2, and an output unit 3.

[0049] The model holding unit 1 reads and holds a model MD created from learning data DAT, which is time-series data. As the model holding unit 1, various storage means such as the RAM 1003 and the storage unit 1008 in FIG.

[0050] The search unit 2 is configured to be able to appropriately read the model MD from the model holding unit 1. The search unit 2 refers to the state quantity of data to be estimated (hereinafter referred to as input data) that is separately input, and searches for an approximate node from among the nodes included in the read model MD. Then, the search unit 2 outputs the search result to the output unit 3.

[0051] The output unit 3 outputs an output command value that is determined based on the search result and corresponds to the state quantity of the input data.

[0052] Next, a situation in which the information processing device 100 is used will be described. Fig. 4 shows a schematic diagram of a situation in which the information processing device 100 according to the first embodiment is used. The information processing device 100 learns an operation performed by an operator 10 on an operation target device 20 (for example, a robot described later) in a learning phase, and outputs a command value estimated based on a learning result in an estimation phase to the operation target device 20, thereby causing the operation target device 20 to perform an appropriate operation. Note that, hereinafter, an object controlled by the information processing device 100, such as the operation target device 20, is also simply referred to as a control target.

[0053] In the following, to distinguish between data handled in the learning phase and data handled in the estimation phase, data handled in the learning phase will be written in uppercase letters, and data handled in the estimation phase will be written in lowercase letters. The state quantity handled in the learning phase will be written as P, and the command value will be written as Q. The state quantity handled in the learning phase, i.e., the input data, will be written as p, and the command value, i.e., the output value, will be written as q.

[0054] In the learning phase, the operator 10 operates the command device 11 to input a command INS in order to give a command value for commanding an appropriate operation to the operation target device 20. The command device 11 outputs a command value Q corresponding to the input command INS to the operation target device 20 and the model creation unit 12. The operation target device 20 performs an operation corresponding to the command value Q, and outputs a state quantity P to the model creation unit 12.

[0055] As a result, the model creation unit 12 acquires the command value Q given to the operation target device 20 and the state quantity P of the operation target device 20 at a certain time. Since the command value Q and the state quantity P are acquired sequentially, the model creation unit 12 can obtain time series data TSD including a set of the command value Q and the state quantity P and a plurality of data elements including the timing at which each set was acquired. Note that, hereinafter, the command value Q given to the operation target device 20 and the state quantity P of the operation target device 20 acquired in advance for model creation are also referred to as the first command value and the first state quantity, respectively.

[0056] The model creation unit 12 appropriately processes the time series data to create learning data, learns the learning data to generate a model MD, and outputs it to the model holding unit 1. As a result, the model MD is held by the model holding unit 1. That is, the learning data includes a plurality of data elements consisting of a first command value and a first state quantity acquired in advance. The model MD created by the model creation unit 12 is a model in which the distribution structure of the time series data is learned as a set of nodes described by a multidimensional vector including at least a state quantity and a command value as elements. In the following, the state quantity and the command value included in the multidimensional vector describing the node of the model MD are also referred to as a second state quantity and a second command value, respectively.

[0057] In the estimation phase, the search unit 2 appropriately reads the model MD from the model storage unit 1. The search unit 2 searches for one node N from among the nodes included in the model MD based on input data to a model based on the model MD, i.e., based on a state quantity p acquired from the operation target device 20. NEAR In the following, the state quantity acquired from the operation target device 20 in the estimation phase is also referred to as a third state quantity. NEAR The command value corresponding to the node N is read, and the read command value is output to the operation target device 20 as the command value Q. In this embodiment, the node N is selected from the nodes included in the model MD. NEAR However, multiple nodes may be determined from the nodes included in the model MD. From the command values ​​corresponding to the multiple determined nodes, the command value to be output may be calculated using, for example, a statistical method.

[0058] Next, as a premise for the processing performed by the information processing device 100, we will explain the time series data TSD consisting of the state quantity P and command value Q acquired by the model creation unit 12, and the model MD generated by the model creation unit 12 learning the learning data DAT created based on the time series data TSD.

[0059] The model creation unit 12 reads the learning data DAT, performs clustering, and learns the learning data DAT with the clustering information added thereto as a node. As will be described below, the learning data DAT is in the form of a collection of multidimensional vectors obtained by appropriately processing time-series data including the state quantity P, the command value Q, and information indicating timing, and each node is composed of the same multidimensional vector.

[0060] As a specific example, an example of learning about an experiment using a robot will be described. Figure 5 shows an outline of the robot used in the experiment. The robot 30 has a robot arm 31, from the tip of which a cylindrical pole 33 extends, and a force sensor 32 for detecting a force acting on the cylindrical pole 33 is provided at the connection between the robot arm 31 and the cylindrical pole 33. As a result, when an external force is applied to the cylindrical pole 33, the external force can be detected by the force sensor 32. The robot arm 31 can move in three dimensional directions within the horizontal plane (x and y directions) and in the height direction (z direction) by driving a beam 34 extending horizontally by a multi-joint drive mechanism (not shown) or the like.

[0061] In this example, the operator learned the operation of the robot arm 31 by moving the robot arm 31 using the command device 11. However, the learning of the operation of the robot arm 31 is not limited to this example, and the operator may learn the operation of the robot arm 31 while holding it in his / her hand, or may introduce a remote operation system or the like to operate the robot arm 31 remotely and learn the operation.

[0062] 6 shows an example of the motion to be learned by the robot arm 31. In this example, the robot arm 31 moves within a horizontal plane (xy plane) while coming into contact with an obstacle 40 and avoiding the obstacle 40.

[0063] 7 shows a schematic diagram of forces acting in each state when the robot arm 31 avoids the obstacle 40. The operation and forces acting in each state will be described below.

[0064] [Status 1] The robot arm 31 moves straight in the +y direction from the initial position and comes into contact with the obstacle 40. In this case, the external force acting on the robot arm 31 is substantially zero until it comes into contact with the obstacle 40 at the end of state 1.

[0065] [Status 2] When the robot arm 31 comes into contact with the obstacle 40, the robot arm 31 changes its moving direction to the -x direction and continues moving while being pressed against the obstacle 40. In this case, a reaction force fy2 in the -y direction due to being pressed against the obstacle 40 and a friction force fx2 in the +x direction act on the robot arm 31.

[0066] [Status 3] When the robot arm 31 reaches a corner (a corner on both the -x and -y sides) of the obstacle 40, it changes its moving direction to the +y direction and continues moving while being pressed in the +x direction against the obstacle 40. In this case, a reaction force fx3 in the -x direction and a friction force fy3 in the -y direction act on the robot arm 31 due to being pressed against the obstacle 40.

[0067] [State 4] When the robot arm 31 reaches a corner (a corner on the -x side and +y side) of the obstacle 40, it changes its moving direction to the +x direction and continues moving while being pressed in the -y direction against the obstacle 40. In this case, a reaction force fy4 in the +y direction due to being pressed against the obstacle 40 and a friction force fx4 in the -x direction act on the robot arm 31.

[0068] Next, the format of the time series data in this experiment will be explained. Figure 8 shows the format of the time series data TSD. Data elements of the time series data TSD are arranged vertically, and each row corresponds to one sampled data element. Each data element includes four regions DZ1 to DZ4.

[0069] The first field DZ1 indicates the sampling number i of the data element. Note that the number of samples is n (n is an integer equal to or greater than 1), so i is an integer equal to or greater than 1 and equal to or less than n. The second field DZ2 indicates the time (timing) t(i) at which each data element is sampled.

[0070] The third area DZ3 is the area where the external force (f x (i), f y (i)), velocity (v x (i), v y The values ​​included in this area DZ3 indicate values ​​related to information of the object to be controlled.

[0071] The fourth area DZ4 is the force (F x (i), F y (i)), and the amount included in this region DZ4 corresponds to the command value. Note that, although the command value is a force here, this is merely an example, and other values ​​such as a velocity and a position may be used as the command value.

[0072] Here, the sampling number i is used to indicate which data element the force, velocity, and position belong to, but this is merely an example, and the time t(i) may be used instead of the sampling number i.

[0073] In this example, as shown in the following equation, the force and velocity included in the third region DZ3 are used as the state quantity P(i) included in the learning data DAT.

number

[0074] Further, as shown in the following formula, a command value that is a predetermined time h in the future is used as the command value Q for the command value included in the above-mentioned fourth region DZ4.

number

number

[0075] The learning data DAT is expressed by the following formula.

number

[0076] In this embodiment, in equations [4] and [5], for command values ​​Q(n+1) to Q(n+h) that are future than the command value Q(n) sampled last in time, command value Q(n) is used. However, future command values ​​Q(n+1) to Q(n+h) are not limited to this, and other values ​​may be used as appropriate so that operation is favorable. In addition, data elements Di having command values ​​that do not originally exist in the time series data TSD, such as future command values ​​Q(n+1) to Q(n+h), may not be included in the learning data DAT.

[0077] In this embodiment, the number of data elements in the learning data DAT is the same as the number of samples n, but this is not limited to this. For example, only a part of the elements such as {D1, D3, D5...} may be used as the learning data DAT, or new data may be created to complement the area between Di and Di+1 and added to the learning data DAT.

[0078] Next, model creation in this example will be described. The model creation unit 12 performs clustering, which is unsupervised learning, on the learning data DAT, and generates the clustering result as a model. Here, the k-means method is used as the clustering method. Note that the clustering method is not limited to this, and various other clustering methods such as DBSCAN (Density-based spatial clustering of applications with noise) may be used. Also, in this example, Euclidean distance is used as the distance index, but other distance indexes may be used as appropriate.

[0079] FIG. 9 shows an example of learning the motion of the robot arm 31. In FIG. 9, a two-dimensional plane of fx and fy is extracted from the state quantities of the learning data and mapped for easy visualization. The regions separated by dashed lines are the respective clustering regions, and in this example, the region is separated into four regions. As shown in FIGS. 6 and 7, region CL1 corresponds to the state before collision with an obstacle (state 1), region CL2 corresponds to the state of moving leftward after collision (state 2), CL3 corresponds to the state of moving upward on the left side of the obstacle (state 3), and state CL4 corresponds to the state of moving rightward on the obstacle (state 4).

[0080] The model creation unit 12 creates nodes by adding the clustering results to the learning data DAT. Fig. 10 shows the nodes created by the model creation unit. As shown in Fig. 10, a node is created by adding one of CL1 to CL4 indicating the clustering result to each data element of the learning data DAT. That is, the i-th node is expressed by the following formula.

number

[0081] The model creation unit 12 outputs a model MD that is a set of nodes equal to the number of elements (n in this embodiment) of the learning data DAT expressed by the above formula. The model MD is expressed by the following formula.

number

[0082] Since the learning data is time-series data, the nodes obtained by learning naturally have a time-series order. Specifically, since the nodes can be linked to a quantity related to time, in this case the sampling order, it is possible to arrange the nodes in a time series. Note that the method of giving the nodes a time-series order is not limited to this, and they may be expressed by including a quantity related to time (i.e., time) as an element in a multidimensional vector representing the node.

[0083] In this example, the four clusters CL1 to CL4 are obtained in the chronological order of CL1, CL2, CL3, and CL4.

[0084] Next, a description will be given of the processing performed by the information processing device 100. Fig. 11 shows a flowchart of the processing in the information processing device 100 according to the first embodiment. The operation of the information processing device 100 is made up of the following steps SA1 to SA5.

[0085] Step SA1 The search unit 2 reads the model MD from the model storage unit 1 .

[0086] Step SA2 The search unit 2 acquires a state quantity p, which is input data.

[0087] Step SA3 The search unit 2 acquires which cluster of the model MD the state quantity p, which is the input data, belongs to. In the following, the cluster to which the state quantity p, which is the input data, belongs is referred to as the target cluster C TRG Here, the distance between the state quantity p, which is the input data, and the center of gravity of each cluster is calculated, and the cluster with the center of gravity that is the shortest distance is called the target cluster C. TRG In addition, the target cluster C TRG The method of determining is not limited to this, and other methods may be used as appropriate.

[0088] Step SA4 Search unit 2 searches target cluster C TRG and target cluster C TRG Cluster C immediately follows in time NEXT Among the nodes included in NEAR The search unit 10 searches for one node, and outputs the search result to the output unit 3. Here, the Euclidean distance is used as a distance index for searching for a node. In the following, the node N having the state quantity closest to the state quantity p, which is the input data, is called the node N. NEAR Simply, select the nearest node N NEAR It is called.

[0089] Here, only one closest node is searched, but multiple nodes close to the state quantity p, which is the input data, may be searched. In this case, a predetermined number of nodes may be searched in order of proximity to the state quantity p, or multiple nodes within a predetermined distance from the state quantity p may be searched.

[0090] The cluster to be searched is target cluster C. TRG and immediately following cluster C NEXT However, for example, the next cluster in time may be the target, or the immediately following cluster C NEXT Alternatively, only the first half of the nodes may be targeted.

[0091] Furthermore, although the Euclidean distance is used as a distance index for searching for nodes, other distance indexes may be used as appropriate.

[0092] Step SA5 The output section 3 is the closest node N NEAR The output command value q determined based on OUT For example, the output unit 3 outputs the nearest node N NEAR The command value q held by NEAR The output command value q OUT (first output command value determination method).

[0093] In addition, the output unit 3 outputs the output command value q OUT The method of determining the nearest node N NEAR The average command value of all nodes in the cluster to which OUT (a second output command value determination method).

[0094] Also, the output unit 3 is the closest node N NEAR The average command value of some nodes in the cluster to which belongs is the output command value q OUT (third output command value determination method). In this case, the closest node N NEAR Nodes within a given distance from the nearest node N NEAR The average of the command values ​​of the nodes selected by various selection methods, such as a certain number of nodes selected in order of distance from OUT It may be output as

[0095] Furthermore, if multiple nodes are searched for in step SA4, the average of the command values ​​of the multiple nodes is used as the output command value q OUT (fourth output command value determination method).

[0096] Although an average value is used in the second to fourth output command value determination methods, this is merely an example, and various values ​​that can be calculated by statistical methods, such as a median, a maximum value, and a minimum value, in other words, a value determined based on statistics, may also be calculated as the command value.

[0097] As described above, according to this configuration, when an output value corresponding to input data p is obtained, the nodes for calculating the distance to the input data p can be limited to nodes included in a limited cluster group, rather than all nodes. This makes it possible to significantly reduce the number of nodes to be the object of distance calculation, compared to general methods such as the k-nearest neighbor method, and to speed up the process of obtaining an output value corresponding to the input data p.

[0098] The basic operation of the information processing device 100 has been described above, but when a plurality of pieces of input data x are input consecutively, the operation described below may be performed. Fig. 12 shows a flowchart of a modified example of the processing in the information processing device 100 according to the first embodiment. In this modified example, for the second and subsequent pieces of input data p, the target cluster C determined based on the input data p input immediately before is used. TRG Using this, the closest node N NEAR In the operation shown in Fig. 12, steps SA11 to SA14 are added in addition to steps SA1 to SA5 in Fig. 11. Since SA1 to SA5 are the same as in Fig. 11, the added steps SA11 to SA14 will be explained.

[0099] Step SA11 Step SA11 is a step inserted before step SA. First, the search unit 2 sets the value of the initial flag FG to 0 to indicate that the operation is in the initial state, that is, that the input data x is being input for the first time. Note that the initial flag is not limited to this example, and may be any type of data as long as it can indicate that the operation is in the initial state.

[0100] Step SA12 Step SA12 is a step inserted between step SA2 and step SA3. The search unit 2 judges whether the operation is in the initial state, that is, whether the value of the initial flag FG is 0. If the value of the initial flag FG is 0 (initial state), the process proceeds to step SA3, and if the value of the initial flag FG is not 0 (not in the initial state), the process proceeds to step SA4.

[0101] Step SA13 Step SA13 is a step inserted between steps SA3 and SA4. The search unit 2 switches the value of the initial flag FG to 1 to indicate that the operation is not in the initial state.

[0102] Step SA14 The search unit 2 searches the nearest node N NEAR Based on the cluster to which it belongs, the target cluster C for the next process is TRG Here, the closest node N detected in this process is determined. NEAR Cluster C is the target cluster TRG In addition, the target cluster C in the next process is TRG The method of determining is not limited to this, and may be, for example, the closest node N detected in a predetermined number of past processes. NEAR Among the clusters to which belongs, the cluster with the highest occurrence frequency is selected as the target cluster C for the next process. TRG Furthermore, if multiple nodes are searched for in step SA4, the cluster with the highest frequency among the clusters to which the multiple nodes belong may be set as the target cluster C in the next process. trg For example, if five nodes are searched for in step SA4 and the cluster is {3, 3, 4, 4, 4}, the target cluster may be set to 4.

[0103] According to the above-described modified example, the closest node N corresponding to the previously input data p is NEAR The cluster to which p belongs is the target cluster C to be used for the next input data p. TRGIn general, time series data is expected to change relatively slowly and steeply between adjacent input data, and therefore the distance between adjacent input data is expected to be short.

[0104] Therefore, for the input data p to be input in the next process, the closest node N NEAR The distance from the cluster C to be used in the next process is considered to be short. TRG As a result, the target cluster C of the previous processing TRG Therefore, in this modified example, the closest node N NEAR The cluster to which C belongs is the target cluster for the next process. TRG By using it as the target cluster C for the second and subsequent processing TRG can be determined by a simple process. Specifically, the first target cluster C TRG Since the distance between the center of gravity of each cluster and the input data is calculated as in step SA3 (FIG. 11), the determination of the center of gravity of each cluster needs to calculate the center of gravity of each cluster. TRG Since the determination is performed by a simple process and there is no need to calculate the center of gravity of each cluster, the amount of calculation can be reduced and the processing time can be shortened. Note that, in this embodiment, the first target cluster Ctrg is determined by calculating the distance between the center of gravity of each cluster and the input data, but this is not limiting, and the first cluster in the chronological order (CL1 in this embodiment) may be the first target cluster Ctrg.

[0105] Next, a specific example of the processing in the information processing device 100 according to the first embodiment will be described. FIG. 13 shows nodes that are targets of distance calculation in the processing in the information processing device 100 according to the first embodiment and the processing in the k-nearest neighbor method. In the information processing device 100, as described above, only nodes that belong to two clusters CL2 and CL3 are targets of distance calculation for the input data p. In contrast, in the k-nearest neighbor method, all nodes included in the model MD are targets of distance calculation for the input data p. Note that, in this embodiment, for the sake of simplicity, a model including four clusters to which several tens of nodes belong is described as an example, but this is merely an example. Even in a model including a larger number of clusters and having a larger number of nodes belonging to one cluster, by applying this configuration, it is possible to limit the nodes that are targets of distance calculation for the input data to nodes that belong to limited clusters, so that even in a model with a larger number of clusters, the output processing of the command value can be made faster.

[0106] Therefore, according to this configuration, when an output value corresponding to input data p is obtained, the nodes for calculating the distance to the input data p are limited, thereby making it possible to improve the processing speed.

[0107] Embodiment 2 An information processing device 200 according to the second embodiment will be described. Fig. 14 typically shows the configuration of the information processing device 200 according to the second embodiment. Fig. 15 typically shows a situation in which the information processing device 200 according to the second embodiment is used. The information processing device 200 has a configuration in which a model creation unit 4 is added to the information processing device 100, and the search unit 2 is replaced with a search unit 6.

[0108] The model creation unit 4 reads the learning data DAT, which is time-series data, and performs learning by the SOINN method to create a model MD. The created model MD is output to the model holding unit 1. Learning in the model creation unit 4 will be described later.

[0109] The search unit 5 uses the model MD to refer to the state quantities of data to be estimated (hereinafter, referred to as input data) that is separately input, and searches for approximate nodes from among the nodes included in the model MD. Then, the search unit 5 outputs the search result to the output unit 3.

[0110] The data holding unit 1 and the output unit 3 are the same as those in the first embodiment, and therefore the description thereof will be omitted.

[0111] Next, a description will be given of the processing performed by the information processing device 200. Fig. 16 is a flowchart showing the operation of the information processing device 200 according to the second embodiment.

[0112] Step SB1 The model creation unit 4 creates a model MD by learning the learning data DAT obtained by receiving, storing, and processing the state quantities P and command values ​​Q using the SOINN method. In the SOINN method, input data described in a multidimensional vector is input, and nodes expressing the input data are generated, and a model MD is obtained as a network composed of these generated nodes. The learning process in the SOINN method will be described below.

[0113] Using the SOINN method, the model creation unit 4 creates a model consisting of a neural network with at least one layer structure in which nodes described by n-dimensional vectors are arranged. The neural network constituting the model created by the model creation unit 4 is a self-growing neural network with a one-layer structure in which an input vector is input to the neural network and the number of nodes arranged in the neural network is automatically increased based on the input vector.

[0114] This makes it possible to automatically increase the number of nodes using a self-organizing neural network, so that additional learning can be performed online by sequentially inputting input vectors.

[0115] The configuration of the model creation unit 4 will be described below. Fig. 17 shows a schematic configuration of the model creation unit 4 according to the second embodiment. The model creation unit 4 has input information acquisition means 41, winner node search means 42, similarity threshold calculation means 43, similarity threshold determination means 44, node insertion means 45, weight vector update means 46, node density calculation means 47, distribution overlap area detection means 48, edge connection determination means 49, edge connection means 50, edge deletion means 51, noise node deletion means 52 and output information display means 53.

[0116] In the following explanation of the SOINN method, various alphabets such as i, j, k, and x are used. These are used for convenience in explaining the SOINN method, and even if they overlap with the alphabets used in embodiment 1, they will indicate different values.

[0117] The input information acquisition means 41 acquires an n-dimensional input vector as information to be given as an input to the model creation unit 4. Then, the acquired input vector is stored in a temporary storage unit (e.g., RAM 1003) and is sequentially input to the neural network stored in the temporary storage unit.

[0118] The winner node search means 42 searches for the input vector and nodes stored in the temporary storage unit, determining the node having the weight vector closest to the input vector as the first winner node and the node having the weight vector second closest to the input vector as the second winner node, and stores the results in the temporary storage unit. That is, for an n-dimensional input vector ξ, the means searches for nodes that satisfy the following equations stored in the temporary storage unit as the first winner node a1 and the second winner node a2, respectively, and stores the results in the temporary storage unit.

number

number

[0119] The similarity threshold calculation means 43 calculates the distance to the node that is the maximum distance from the node of interest among the adjacent nodes as the similarity threshold for the nodes and the similarity thresholds of the nodes stored in the temporary storage unit if there is a node (hereinafter referred to as an adjacent node) that is directly connected to the node of interest by an edge, and stores the result in the temporary storage unit, and if there is no adjacent node, calculates the distance to the node that is the minimum distance from the node of interest as the similarity threshold, and stores the result in the temporary storage unit. Specifically, for example, the similarity threshold for the node of interest is calculated as follows, and the result is stored in the temporary storage unit. [Step M_A1] The similarity threshold calculation means 43 calculates the similarity threshold T i is set to +∞ (a sufficiently large value) and the result is stored in the temporary memory. [Step M_A2] For the nodes stored in the temporary memory, if node i is the closest node or the second closest node to the input vector, it is determined whether node i has an adjacent node, and the result is stored in the temporary memory. [Step M_A3] If the result of the determination stored in the temporary storage unit indicates that the node has an adjacent node, the similarity threshold value T i is the maximum distance to the adjacent node, and the result is stored in the temporary memory. That is, for node i, a similarity threshold T i is calculated and the result is stored in the temporary storage unit.

number

number

[0120] The similarity threshold determination means 44 determines whether the distance between the input vector and the first winner node is greater than the similarity threshold of the first winner node, and whether the distance between the input vector and the second winner node is greater than the similarity threshold of the second winner node, for the input vector, node, and node similarity threshold stored in the temporary storage unit, and stores the results in the temporary storage unit. That is, as shown in the following equation stored in the temporary storage unit, when the distance between the input vector ξ and the first winner node a1 is greater than the similarity threshold T a1 The result is stored in the temporary storage unit, and the distance between the input vector ξ and the second winner node a2 is determined to be greater than or equal to the similarity threshold T a2 It is determined whether the value is greater than or equal to the first value, and the result is stored in the temporary storage unit.

number

[0121] Based on the judgment result by the similarity threshold judgment means 44 stored in the temporary storage unit, the node insertion means 45 inserts the input vector stored in the temporary storage unit as a new node in the same position as the input vector, and stores the result in the temporary storage unit.

[0122] The weight vector update means 46 updates the weight vectors of the nodes stored in the temporary memory unit so that the weight vector of the first winner node and the weight vector of the adjacent node of the first winner node are each closer to the input vector, and stores the results in the temporary memory unit. The update amount ΔW of the weight vector of the first winner node a1 a1 , and the update amount ΔW of the weight vector of the adjacent node i of the first winner node a1 ai is calculated based on the following formula, which is stored in the temporary storage unit, for example, and the result is stored in the temporary storage unit.

number

number

number

number

[0123] The node density calculation means 47 calculates the node density of a node of interest based on the average distance between its adjacent nodes for the node and node densities stored in the temporary storage unit, and stores the result in the temporary storage unit. Furthermore, the node density calculation means 47 has a unit node density calculation unit, which calculates the node density of the first winner node as a rate per unit number of inputs based on the average distance between the first winner node and its adjacent nodes for the first winner node and node densities stored in the temporary storage unit in order to accommodate additional learning, and stores the result in the temporary storage unit. Furthermore, the node density calculation means 47 includes a node density point calculation unit that calculates the node density point value of the first winner node based on the average distance between the first winner node and its adjacent nodes for the nodes and node density points stored in the temporary memory unit, and a unit node density point calculation unit that stores and accumulates the node density points in the temporary memory unit until the number of inputs of the input vector becomes a predetermined unit number of inputs, and when the number of inputs of the input vector becomes the predetermined unit number of inputs, calculates the node density of the node per unit number of inputs, and stores the result in the temporary memory unit.

[0124] Specifically, the node density point calculation unit calculates a node density point value pt i The point value pt given to node i is calculated and the result is stored in the temporary storage unit. i If node i becomes the first winner node, it is given a point value calculated based on the following formula, which is stored in the temporary storage unit. However, if node i is not the first winner node, no points are given to node i.

number

number

[0125] Here, if the average distance to adjacent nodes is large, it is considered that the area containing the node has few nodes, and conversely, if the average distance is small, it is considered that the area has many nodes. Therefore, the method for calculating the point value of node density is configured as described above so that if a node becomes the first winner node in an area with a large number of nodes, high points are awarded, and if a node becomes the first winner node in an area with few nodes, low points are awarded. This makes it possible to estimate the density of nodes in a certain range of area including the node, so that even if the node is located in an area with a high density distribution, a node density point can be calculated that has a density that more closely resembles the input distribution density of the input vector, compared to the conventional case in which the node density is calculated based on the number of times it has been the first winner.

[0126] The unit node density point calculation unit calculates the node density per unit input number of node i based on the following formula stored in the temporary storage unit, for example: i is calculated and the result is stored in the temporary storage unit.

number

number

[0127] In this embodiment, N is used instead of n to accommodate additional learning. This is to avoid the problem that in additional learning, points are often not given to nodes generated in previous learning, and when density is calculated using n, the density of previously learned nodes gradually decreases. In other words, by calculating node density using N instead of n, the density of a node can be maintained unchanged even when additional learning is performed for a long period of time, as long as the added data is not input near a previously learned node. This makes it possible to prevent the node density of the nodes from becoming relatively small even when additional learning is performed for a long period of time, and compared to conventional methods, it is possible to maintain and calculate a node density that is closer to the input distribution density of the input vector without changing it.

[0128] The distribution overlap area detection means 48 divides clusters, which are sets of nodes connected by edges, into sub-clusters, which are subsets of clusters, based on the node density calculated by the node density calculation means 47, for the nodes, edges connecting the nodes, and node density stored in the temporary storage unit, stores the results in the temporary storage unit, detects distribution overlap areas that are the boundaries of the sub-clusters, and stores the results in the temporary storage unit.

[0129] Furthermore, the distribution overlap area detection means 48 has a node search unit that searches for a node with a locally maximum node density based on the node density calculated by the node density calculation means 47 for the nodes, edges connecting the nodes, and node densities stored in the temporary storage unit; a first label assignment unit that assigns a label to the searched node that is different from labels already assigned to other nodes; a second label assignment unit that assigns, among nodes not assigned a label by the first label assignment unit, a label the same as the label of the node labeled by the first label assignment unit to nodes connected to the node by edges; a cluster division unit that, when there is a direct connection by an edge between the nodes assigned with different labels, divides a cluster, which is a set of nodes connected by the edges, into sub-clusters, which are subsets of the cluster; and a distribution overlap area detection unit that, when a node of interest and its adjacent nodes belong to different sub-clusters, detects an area including the node of interest and its adjacent nodes as a distribution overlap area that is the boundary of the sub-clusters.

[0130] Specifically, for the nodes, edges connecting the nodes, and node densities stored in the temporary storage unit, the overlapping areas of distribution that are the boundaries of sub-clusters are detected, for example, as follows, and the results are stored in the temporary storage unit. [Step M_B1] The node search unit searches for a node with a locally maximum node density based on the node density calculated by the node density calculation means 47 for the nodes and node densities stored in the temporary storage unit, and stores the result in the temporary storage unit. [Step M_B2] The first label assignment unit assigns a label to the node searched for in step M_B1 that is different from the labels already assigned to other nodes, for the nodes and the labels of the nodes stored in the temporary storage unit, and stores the result in the temporary storage unit. [Step M_B3] The second labeling unit assigns the same label as the label of the node labeled by the first labeling unit to the node not labeled by the first labeling unit in step M_B2, the edge connecting the nodes, and the node labels stored in the temporary storage unit, and stores the result in the temporary storage unit. In other words, the same label as the adjacent node with the locally highest density is assigned. [Step M_B4] The cluster division unit divides a cluster, which is a set of nodes connected by the edges stored in the temporary storage unit, into subclusters, which are subsets of clusters consisting of nodes with the same label, for the nodes, edges connecting the nodes, and node labels stored in the temporary storage unit, and stores the results in the temporary storage unit. [Step M_B5] The distribution overlap area detection unit detects an area including a node of interest and its adjacent node as a distribution overlap area that is the boundary of the subclusters, when the node of interest and its adjacent node belong to different subclusters, for the nodes, edges connecting the nodes, and node labels stored in the temporary storage unit, and stores the result in the temporary storage unit.

[0131] For the nodes, node densities, and distribution overlap areas stored in the temporary storage unit, the edge connection determination means 49 determines whether or not to connect an edge between the first winner node and the second winner node based on the node densities of the first winner node and the second winner node when the first winner node and the second winner node are nodes located in the distribution overlap area, and stores the result in the temporary storage unit.

[0132] Furthermore, the edge connection determination means 49 has a subcluster belonging determination unit that determines the subcluster to which a node belongs based on the nodes, node density, and node subclusters stored in the temporary storage unit, and an edge connection determination unit that determines whether or not to connect an edge between the first winner node and the second winner node based on the vertex density and node density of the subcluster to which the node belongs.

[0133] Based on the judgment result of the edge connection judgment means 49 stored in the temporary storage unit, the edge connection means 50 connects edges between the first winner node and the second winner node for the nodes and edges between the nodes stored in the temporary storage unit, and stores the results in the temporary storage unit.

[0134] The edge deletion means 51 deletes the edge between the first winner node and the second winner node for the nodes and edges between the nodes stored in the temporary storage unit based on the judgment result of the edge connection judgment means 49 stored in the temporary storage unit, and stores the result in the temporary storage unit.

[0135] Specifically, for the nodes, node densities, sub-clusters of nodes, and edges between nodes stored in the temporary storage unit, the edge connection determination means 49 determines whether or not to connect edges, for example, as follows, and the edge connection means 50 and edge deletion means 51 perform edge creation and deletion processing, and store the results in the temporary storage unit. [Step M_C1] The belonging sub-cluster determination unit determines the sub-clusters to which the first winner node and the second winner node belong, for the nodes and sub-clusters of the nodes stored in the temporary storage unit, and stores the results in the temporary storage unit. [Step M_C2] If the result of the judgment in procedure M_C1 stored in the temporary storage unit is that the first winner node and the second winner node do not belong to any sub-cluster, or if the first winner node and the second winner node belong to the same sub-cluster, edge connection means 50 connects the nodes by generating edges between the first winner node and the second winner node for the nodes and edges between the nodes stored in the temporary storage unit, and stores the result in the temporary storage unit. [Step M_C3] If the result of the judgment in procedure M_C1 stored in the temporary memory unit indicates that the first winner node and the second winner node belong to different sub-clusters, the edge connection judgment unit judges whether or not to connect an edge between the first winner node and the second winner node based on the vertex density and node density of the sub-cluster to which the node belongs, for the nodes, node densities, and edges between the nodes stored in the temporary memory unit, and stores the result in the temporary memory unit. [Step M_C4] If the result of the judgment by the edge connection judgment unit in procedure M_C3 stored in the temporary memory unit indicates that it is not necessary to connect an edge, then for the nodes and edges between the nodes stored in the temporary memory unit, the first winner node and the second winner node are not connected by an edge, and if the nodes are already connected by an edge, the edge deletion means 51 deletes the edge between the first winner node and the second winner node stored in the temporary memory unit for the nodes and edges between the nodes stored in the temporary memory unit, and stores the result in the temporary memory unit. [Step M_C5] If it is determined that an edge needs to be connected as a result of the judgment by the edge connection judgment unit in procedure M_C3 stored in the temporary memory unit, the edge connection means 50 generates an edge between the first winner node and the second winner node for the nodes and edges between the nodes stored in the temporary memory unit, and connects the nodes.

[0136] Here, the determination process performed by the edge connection determination unit will be described in detail. First, the edge connection determination unit 101 determines the node density of the first winner node from the nodes and node densities stored in the temporary storage unit. win and the second winner node density sec-win Among these, the minimum node density m is calculated based on the following formula, which is stored in the temporary storage unit, for example, and the result is stored in the temporary storage unit.

number

number

[0137] In this way, by comparing the minimum node density m of the first winner node and the second winner node with the average node density of the subclusters including the first winner node and the second winner node, respectively, it is possible to determine the degree of unevenness in the node density in the area including the first winner node and the second winner node. In other words, when the node density m of the valley of the distribution existing between subcluster A and subcluster B is greater than or equal to the threshold α A A max Or α B B max If it is greater, it can be determined that the shape of the node density is a small bump.

[0138] Here, α A and α B is calculated based on the following formula, which is stored in the temporary storage unit, and the result is stored in the temporary storage unit. B Regarding α A Since it can be calculated in the same manner, the explanation will be omitted here. i) A max / mean A If -1≦1, then α A =0.0. ii) 1 max / mean A If -1≦2, then α A =0.5. iii) 2 max / mean A In the case of -1, α A =1.0. A max / mean A In the case of i) where the value of is 1 or less, A max and mean A The values ​​of are similar, and it is determined that the unevenness in density is due to the influence of noise. Then, the value of α is set to 0.0 so that the sub-clusters are integrated. Also, A max / mean A In the case of iii) where the value of exceeds 2, A max is mean A ​​It is judged that the density unevenness exists because it is sufficiently large compared to . Then, the value of α is set to 1.0 so that the sub-clusters are separated. And A max / mean A In the case of ii) where the value of is other than the above-mentioned cases, the value of α is set to 0.5 so that sub-clusters are merged or separated according to the magnitude of density unevenness. In addition, mean A is the node density of node i in subcluster A. i The average value of N A is calculated as the number of nodes belonging to sub-cluster A based on the following formula, which is stored in the temporary storage unit, and the result is stored in the temporary storage unit.

number

[0139] In this way, when separating into sub-clusters, the degree of unevenness in the node density contained in the sub-clusters is determined, and two sub-clusters that satisfy certain criteria are merged into one, thereby preventing instability caused by excessive division of sub-clusters when detecting overlapping distribution areas.

[0140] The noise node deletion means 52 deletes a node of interest based on the node density calculated by the node density calculation means 47 and the number of adjacent nodes of the node of interest from the nodes, node density, edges between nodes, and the number of adjacent nodes stored in the temporary storage unit, and stores the result in the temporary storage unit.

[0141] Furthermore, the noise node removal means 52 has a node density comparison unit that compares the node density of a node of interest with a predetermined threshold value for the nodes, node density, edges between nodes, and number of adjacent nodes stored in the temporary memory unit, an adjacent node number calculation unit that calculates the number of adjacent nodes of the node of interest, and a noise node removal unit that considers the node of interest to be a noise node and removes it. Specifically, for example, the nodes, node density, edges between nodes, and number of adjacent nodes stored in the temporary memory unit are deleted based on the node density and the number of adjacent nodes of the node of interest, and the result is stored in the temporary memory unit as follows.

[0142] The noise node deletion means 52 calculates the number of adjacent nodes for a node i of interest using the adjacent node number calculation unit from the number of nodes, edges between nodes, and adjacent nodes stored in the temporary storage unit, and stores the result in the temporary storage unit. Then, the following processing is performed according to the number of adjacent nodes stored in the temporary storage unit. i) If the number of adjacent nodes stored in the temporary memory is 2, the node density comparison unit calculates the node density density of node i. i is compared with a threshold calculated based on the following formula, which is stored in the temporary storage unit, for example, and the result is stored in the temporary storage unit.

number

number

[0143] The output information display means 53 outputs information including the nodes stored in the temporary storage unit, that is, the model MD.

[0144] Next, the flow of the learning process using the SOINN method will be described. Fig. 18 shows a flowchart of the learning process using the SOINN method.

[0145] Step M1 The input information acquisition means 41 randomly acquires two input vectors, initializes the node set A as a set including only the two nodes corresponding to them, and stores the result in the temporary storage unit. Also, it initializes the edge set C ⊂ A × A as an empty set, and stores the result in the temporary storage unit.

[0146] Step M2 The input information acquisition means 41 inputs a new input vector ξ randomly selected from the learning data DAT and stores the result in the temporary storage unit. However, it goes without saying that an input vector once selected will not be selected again.

[0147] Step M3 The winner node search means 42 searches for the first winner node a1 having the weight vector closest to the input vector ξ and the second winner node a2 having the second closest weight vector for the input vector and nodes stored in the temporary storage unit, and stores the results in the temporary storage unit.

[0148] Step M4 The similarity threshold determination means 44 determines whether the distance between the input vector ξ and the first winner node a1 is greater than the similarity threshold T1 of the first winner node a1, and whether the distance between the input vector ξ and the second winner node a2 is greater than the similarity threshold T2 of the second winner node a2, for the input vectors, nodes, and node similarity thresholds stored in the temporary memory unit, and stores the results in the temporary memory unit. Here, the similarity threshold T1 of the first winner node a1 and the similarity threshold T2 of the second winner node a2 stored in the temporary memory unit are calculated by the similarity threshold calculation means 43 as shown in the above-mentioned steps M_A1 to M_A4, and the results are stored in the temporary memory unit.

[0149] Step M5 If the result of the judgment in step M4 stored in the temporary memory unit is that the distance between the input vector ξ and the first winner node a1 is greater than the similarity threshold T1 of the first winner node a1, or the distance between the input vector ξ and the second winner node a2 is greater than the similarity threshold T2 of the second winner node a2, the node insertion means 45 inserts the input vector ξ as a new node i in the same position as the input vector ξ for the input vectors and nodes stored in the temporary memory unit, and stores the result in the temporary memory unit.

[0150] Step M6 On the other hand, if the result of the judgment in step M4 stored in the temporary memory unit is that the distance between the input vector ξ and the first winner node a1 is less than or equal to the similarity threshold T1 of the first winner node a1, and the distance between the input vector ξ and the second winner node a2 is less than or equal to the similarity threshold T2 of the second winner node a2, the edge connection judgment means 49 judges whether or not to connect an edge between the first winner node a1 and the second winner node a2 based on the node densities of the first winner node a1 and the second winner node a2 for the nodes, node densities, and edges between the nodes stored in the temporary memory unit, and stores the result in the temporary memory unit.

[0151] Step M7 If the result of the judgment in step M6 stored in the temporary storage unit is to generate and connect an edge between the first winner node a1 and the second winner node a2, the edge connection means 50 connects edges between the first winner node and the second winner node for the nodes and edges between the nodes stored in the temporary storage unit, and stores the result in the temporary storage unit. Then, for the edges and edge ages stored in the temporary memory unit, the information processing device sets the age of the edge to 0 for newly generated edges and for edges that have already been generated between nodes, stores the results in the temporary memory unit, increments (increases by 1) the age of the edge that is directly connected to the first winner node a1, and stores the results in the temporary memory unit. On the other hand, if the result of the determination in step M6 stored in the temporary storage unit is that an edge is not to be connected between the first winner node a1 and the second winner node a2, the process proceeds to step M8, but if an edge has already been generated between the nodes, the edge deletion means 51 deletes the edge between the first winner node a1 and the second winner node a2 from the nodes and edges between the nodes stored in the temporary storage unit, and stores the result in the temporary storage unit. Note that the edge connection determination means 49, edge connection means 50, and edge deletion means 51 perform the processes as shown in the above-mentioned procedures M_C1 to M_C5. Next, for the node and node density point values ​​stored in the temporary memory unit, for the first winner node a1, the node density calculation means 47 calculates the node density point value of the first winner node a1 stored in the temporary memory unit and stores the result in the temporary memory unit, and adds the node density point value calculated and stored in the temporary memory unit to the point values ​​previously calculated and stored in the temporary memory unit, thereby accumulating it as node density points, and stores the result in the temporary memory unit. Next, the information processing device calculates the cumulative number M of times that the first winner node a1 has become the first winner node, which is stored in the temporary storage unit. a1 (increase by 1) and store the result in the temporary memory.

[0152] Step M8 The weight vector update means 46 updates the weight vector of the first winner node a1 and the weight vector of the adjacent node of the first winner node a1 for the nodes and node weight vectors stored in the temporary memory unit so as to bring them closer to the input vector ξ, and stores the results in the temporary memory unit.

[0153] Step M9 The information processing device calculates the edge size based on a threshold value age that is preset and stored in the temporary storage unit. t The edges with ages exceeding are deleted and the result is stored in the temporary memory. t is used to remove edges that are generated incorrectly due to noise or other factors. t By setting a small value for age, edges are more likely to be deleted, and the effects of noise can be prevented. However, if the value is too small, edges are deleted more frequently, making the learning results unstable. On the other hand, if the value is too large, t If you set it to , edges generated by noise cannot be properly removed. Taking these factors into consideration, the parameter age t is calculated in advance through experiments and stored in the temporary storage unit.

[0154] Step M10 The information processing device determines whether the total number of given input vectors ξ stored in the temporary storage unit is a multiple of λ that is preset and stored in the temporary storage unit, and stores the result in the temporary storage unit. If the result of the determination stored in the temporary storage unit shows that the total number of input vectors is not a multiple of λ, the process returns to step M2 and processes the next input vector ξ. On the other hand, when the total number of input vectors ξ is a multiple of λ, the following process is executed. Here, λ is the period for deleting nodes considered to be noise. Although noise processing can be performed frequently by setting a small value for λ, if the value is set too small, nodes that are not actually noise will be mistakenly deleted. On the other hand, if an extremely large value is set for λ, nodes generated due to the influence of noise cannot be properly removed. Taking these factors into consideration, the parameter λ is calculated in advance by experiment and stored in the temporary storage unit.

[0155] Step M11 The distribution overlap area detection means 48 detects the distribution overlap areas that are the boundaries of the sub-clusters for the sub-clusters and distribution overlap areas stored in the temporary memory unit, as shown in the above-mentioned steps M_B1 to M_B5, and stores the results in the temporary memory unit.

[0156] Step M12 The node density calculation means 47 calculates the accumulated node density points stored in the temporary memory unit as a ratio per unit number of inputs and stores the result in the temporary memory unit, calculates the node density of the nodes per unit number of inputs, and stores the result in the temporary memory unit.

[0157] Step M13 The noise node deletion means 52 deletes nodes that are deemed to be noise nodes from among the nodes stored in the temporary storage unit, and stores the result in the temporary storage unit. Note that the parameters c1 and c2 used by the noise node deletion means 52 in step M13 are used to determine whether or not to regard a node as noise. Normally, a node with two adjacent nodes is often not noise, so a value close to 0 is used for c1. Also, a node with one adjacent node is often noise, so a value close to 1 is used for c2, and these parameters are set in advance and stored in the temporary storage unit.

[0158] Step M14 The information processing device determines whether the total number of given input vectors ξ is LT, which is preset and stored in the temporary storage unit, and stores the result in the temporary storage unit. If the result of the determination stored in the temporary storage unit shows that the total number of input vectors is not a multiple of LT, the process returns to step M2 and processes the next input vector ξ. On the other hand, when the total number of input vectors ξ reaches LT, learning is stopped.

[0159] By executing the above steps M1 to M14, a model MD configured as a network of nodes created by the SOINN method is obtained. Fig. 19 shows an example of a distribution of nodes obtained by learning the learning data DAT used in the first embodiment by the SOINN method in the information processing device 200 according to the second embodiment. In Fig. 19, for easy visualization, a two-dimensional plane of fx, fy is extracted from the state quantities of the learning data and mapped.

[0160] In this example, by performing learning using the SOINN method, the learning results of the data elements contained in the training data are represented by representative nodes generated by the SOINN method. In other words, the number of nodes contained in the model MD is smaller than the number of data elements contained in the training data DAT.

[0161] As the SOINN method, for example, the SOINN methods described in Patent Documents 1 and 2 can be applied. In addition, various SOINN methods that learn the distribution structure of data elements included in the learning data as a distribution structure represented by a smaller number of nodes than the data elements may be applied.

[0162] Returning to FIG. 16, the process of step SB2 and subsequent steps in the information processing device 200 will be described.

[0163] Step SB2 The search unit 5 reads the model MD from the model storage unit 1 in the same manner as in step SA1 of FIG.

[0164] Step SB3 The searching unit 5 acquires the state quantity p, which is input data, in the same manner as in step SA2 of FIG.

[0165] Step SB4 The search unit 5 searches the model MD for the node N closest to the input data p in the same manner as in step SA4 of FIG. NEAR Explore.

[0166] Step SB5 The output unit 3 is the same as in the first embodiment as in step SA5 in FIG. NEAR The output command value q determined based on OUT As in the first embodiment, the output command value q OUT may be determined.

[0167] As described above, according to this configuration, when an output value corresponding to input data p is obtained, the number of nodes for calculating the distance to the input data p can be limited to a number of nodes fewer than the number of data elements of the learning data. This makes it possible to significantly reduce the number of nodes to be subjected to distance calculations compared to a general method such as a k-nearest neighbor method that uses learning data or the same number of nodes as the data elements of the learning data, and to speed up the process of obtaining a command value corresponding to the input data p.

[0168] Therefore, according to this configuration, similarly to the first embodiment, when an output value corresponding to input data p is obtained, the nodes for calculating the distance to the input data p are limited, thereby making it possible to improve the processing speed.

[0169] Other embodiments The present invention is not limited to the above embodiment, and can be modified as appropriate without departing from the spirit of the present invention. For example, in the above embodiment, the judgment of the magnitude of two values ​​has been described, but this is merely an example, and the case where the two values ​​are equal in the judgment of the magnitude of two values ​​may be handled as necessary. That is, for the judgment of whether the first value is equal to or greater than the second value or smaller than the second value, either may be adopted as necessary. For the judgment of whether the first value is equal to or less than the second value or larger than the second value, either may be adopted. In other words, when the magnitude judgment of two values ​​is performed to obtain two judgment results, the case where the two values ​​are equal may be included in either of the two judgment results as necessary.

[0170] In the above-mentioned embodiment, the present invention has been mainly described as a hardware configuration, but the present invention is not limited to this, and any process can be realized by making a CPU (Central Processing Unit) execute a computer program. The program includes instructions (or software codes) that, when loaded into a computer, cause the computer to execute one or more functions described in the above-mentioned embodiment. The program may be stored in a non-transitory computer-readable medium or a tangible storage medium. Examples of the non-transitory computer-readable medium or tangible storage medium include, but are not limited to, a random-access memory (RAM), a read-only memory (ROM), a flash memory, a solid-state drive (SSD), or other types of storage technology, such as a compact disc (CD)-ROM, a digital versatile disc (DVD), a Blu-ray disc (registered trademark) or other types of optical disk storage devices, a magnetic cassette, a magnetic tape, a magnetic disk storage device, or other magnetic storage devices. The program may be transmitted by a temporary computer-readable medium or a communication medium. Without being limited thereto, examples of transitory computer-readable media or tangible storage media may include electrical, optical, acoustic or other forms of propagated signals.

[0171] In the first embodiment, it has been described that the nodes obtained by learning the training data DAT are clustered. However, it goes without saying that clustering may also be performed on the nodes obtained by learning the training data DAT using the SOINN method described in the second embodiment.

[0172] A part or all of the above-described embodiments can be described as, but is not limited to, the following supplementary notes.

[0173] (Supplementary Note 1) An information processing device comprising: a model holding unit that holds a model created by learning learning data based on time-series data including a plurality of data elements expressed as a multidimensional vector, the data including a first state quantity previously acquired from a learning object and a first command value corresponding to the first state quantity given to control an operation of the learning object, and in which a distribution structure of the learning data is created as a set of nodes expressed as a multidimensional vector including a second state quantity and a second command value based on a learning result; a search unit that receives as input data a third state quantity acquired from a control object and searches for a node that matches or approximates the input data from among the nodes that are fewer than the number of data elements of the learning data included in the model; and an output unit that outputs a value based on the second command value of the searched node to the control object as an output command value to be given to operate the control object.

[0174] (Supplementary Note 2) The information processing device described in Supplementary Note 1, wherein the model is created by learning each of the learning data as a node, and the search unit selects, from the nodes, nodes that are relatively close in time to the input data as the small number of nodes, and searches for a node that is the shortest distance from the selected node to the input data.

[0175] (Appendix 3) The information processing device according to appendix 2, wherein the model is created by time-series clustering the learning data and classifying the nodes into a plurality of clusters, and the search unit searches for a cluster to which the input data belongs and a cluster that is temporally adjacent to the cluster to which the input data belongs, and searches for a node that is the shortest distance from the input data among the nodes belonging to the two searched clusters.

[0176] (Supplementary Note 4) The information processing device according to Supplementary Note 3, wherein the search unit searches for a cluster whose center of gravity is the shortest distance from the input data as a cluster to which the input data belongs.

[0177] (Supplementary Note 5) The information processing device described in Supplementary Note 3, wherein the search unit, for initially input input data, searches for a cluster whose center of gravity is the shortest from the initially input input data as the cluster to which the initially input input data belongs, and, for input data input after the initially input input data, searches for a cluster to which the node with the shortest distance searched for the input data input one step before belongs as the cluster to which the input data input after the initially input input data belongs.

[0178] (Supplementary Note 6) The information processing device according to Supplementary Note 4, wherein the search unit searches for, for initially input input data, a cluster whose center of gravity is the shortest relative to the initially input input data as a cluster to which the initially input input data belongs, and for input data input after the initially input input data, searches for a cluster with the highest appearance frequency among a plurality of clusters to which nodes with the shortest distances searched for a plurality of previously input input data belong as a cluster to which the input data input after the initially input input data belongs.

[0179] (Supplementary Note 7) The information processing device according to any one of Supplementary Notes 4 to 6, wherein the search unit determines the output command value based on statistics calculated from command values ​​of some or all of the nodes in a cluster to which the searched node belongs.

[0180] (Appendix 8) The information processing device according to appendix 7, wherein the portion of nodes is a node within a predetermined distance from the searched node in a cluster to which the searched node belongs, or a predetermined number of nodes selected in order of proximity to the searched node.

[0181] (Appendix 9) The information processing device according to any one of Appendices 4 to 6, wherein the search unit determines the output command value based on statistics calculated from a command value of the searched node and command values ​​of one or more nodes that are approximate to the searched node in the cluster to which the input data belongs and in a cluster immediately following in time the cluster to which the input data belongs.

[0182] (Supplementary Note 10) The information processing device according to Supplementary Note 9, wherein the search unit, for initially input input data, searches for a cluster whose center of gravity is the shortest with respect to the initially input input data as a cluster to which the initially input input data belongs, and, for input data input after the initially input input data, searches for a cluster with the highest appearance frequency among a cluster to which a node searched for in the input data input immediately before belongs and a cluster to which one or more nodes approximate to the searched node belong, as a cluster to which the input data input after the initially input input data belongs.

[0183] (Appendix 11) The information processing device according to appendix 1, further comprising a model creation unit that creates the model by approximately learning a distribution structure of the data elements of the training data using a smaller number of nodes than the data elements of the training data, and outputs the created model to the model holding unit, wherein the search unit searches for a node that approximates the input data from the nodes included in the model.

[0184] (Supplementary Note 12) The information processing device according to any one of Supplementary Notes 2 to 4 and 11, wherein the search unit outputs a command value of the searched node as the output command value.

[0185] (Appendix 13) An information processing device described in any one of Appendices 2 to 4 and 11, wherein the search unit determines the output command value based on statistics calculated from a command value of the searched node and command values ​​of one or more nodes that are close to the searched node.

[0186] (Appendix 14) An information processing device according to any one of Appendices 9, 10 and 13, wherein the one or more nodes similar to the searched node are nodes within a predetermined distance from the searched node, or a predetermined number of nodes selected in order of proximity to the searched node.

[0187] (Supplementary Note 15) The information processing device according to any one of Supplementary Notes 7 to 10, 13 and 14, wherein the statistical amount is any one of an average value, a median value, a maximum value, a minimum value and a mode value.

[0188] (Appendix 16) An information processing device according to any one of Appendices 1 to 15, wherein the second state quantity is the first state quantity previously acquired from the learning object, and the second command value is the first command value given to the learning object in accordance with the first state quantity previously acquired.

[0189] (Appendix 17) The information processing device described in Appendix 16, wherein the first command value is an actual value of a command value given to the learning object by an operator of the learning object based on the first state quantity previously acquired from the learning object.

[0190] (Appendix 18) An information processing device as described in Appendix 16, wherein the first command value included in a data element of the learning data is a command value given to the control object a predetermined time after the first state quantity previously acquired from the learning object.

[0191] (Supplementary Note 19) The information processing device according to any one of Supplementary Notes 1 to 17, wherein the information processing device is provided in the learning object or the control object.

[0192] (Supplementary Note 20) The information processing device according to any one of Supplementary Notes 1 to 17, wherein the learning object and the control object are the same.

[0193] (Supplementary Note 21) The information processing device according to Supplementary Note 20, wherein the information processing device is provided in the controlled object.

[0194] (Supplementary Note 22) An information processing method comprising: learning training data based on time-series data including a plurality of data elements expressed as a multidimensional vector, the data including a first state quantity previously acquired from a learning object and a first command value corresponding to the first state quantity given to control an operation of the learning object; retaining a model created as a set of nodes, the distribution structure of which is a set of nodes expressed as a multidimensional vector including a second state quantity and a second command value based on a learning result; receiving a third state quantity acquired from a control object as input data; searching for a node that matches or approximates the input data from the nodes that are fewer than the number of data elements of the learning data included in the model; and outputting a value based on the second command value of the searched node to the control object as an output command value to be given to operate the control object.

[0195] (Appendix 23) A program that causes a computer to execute the following steps: learning learning data based on time-series data including a plurality of data elements expressed as a multidimensional vector, the data including a first state quantity previously acquired from a learning object and a first command value corresponding to the first state quantity given to control an operation of the learning object; retaining a model created as a set of nodes represented as a multidimensional vector including a second state quantity and a second command value based on a learning result, the distribution structure of the learning data being based on the learning result; inputting a third state quantity acquired from a control object as input data, searching for a node that matches or approximates the input data from among the nodes that are fewer than the number of data elements of the learning data included in the model; and outputting a value based on the second command value of the searched node to the control object as an output command value to be given to operate the control object.

[0196] (Supplementary Note 24) An information processing system comprising: a model creation unit that learns learning data based on time-series data including a plurality of data elements expressed as a multidimensional vector, the data including a first state quantity previously acquired from a learning object and a first command value corresponding to the first state quantity given to control an operation of the learning object, and creates a model of a distribution structure of the learning data as a set of nodes expressed as a multidimensional vector including a second state quantity and a second command value based on a learning result; a model holding unit that holds the created model; a search unit that receives input of a third state quantity acquired from a control object as input data, and searches for a node that matches or approximates the input data from among the nodes that are smaller than the number of data elements of the learning data included in the model; and an output unit that outputs a value based on the second command value of the searched node to the control object as an output command value to be given to operate the control object.

[0197] Although the present invention has been described above with reference to the embodiment, the present invention is not limited to the above. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the invention.

[0198] This application claims priority based on Japanese Patent Application No. 2021-133415, filed on August 18, 2021, the disclosure of which is incorporated herein in its entirety. [Explanation of symbols]

[0199] 1 Model holding section 2 Exploration Department 3 Output section 4. Model Creation Section 5 Exploration Department 10 Operator 11 Command device 20 Operation target device 30. Robot 31 Robot Arm 32 Force Sensor 33 Cylinder Pole 41 Input information acquisition means 42 Winning node search method 43 Similarity threshold calculation means 44 Similarity threshold determination means 45 Node Insertion Method 46 Weight Vector Update Method 47 Node Density Calculation Method 48 Distribution overlap area detection means 49 Edge connectivity determination means 50 Edge connection means 51 Edge deletion method 52 Noise node removal method 53 Output information display means 100 Information processing device 110 Processing section 120 Display section 130 Input section 200 Information processing device 1000 Computers 1001 CPU 1002 ROM 1003 RAM 1004 Bus 1005 Input / Output Interface 1006 Input section 1007 Output section 1008 Storage section 1009 Communications Department 1010 Drive 1011 Magnetic disk 1012 Optical disc 1013 Flexible Disk 1014 Semiconductor memory

Claims

1. a model holding unit that holds a model learned as a set of nodes classified into a plurality of clusters by time-series clustering each of the learning data based on time-series data including a plurality of data elements expressed as a multidimensional vector, the data elements including a first state quantity previously acquired from a learning object and a first command value given to control an operation of the learning object, as a node expressed as a multidimensional vector including a second state quantity and a second command value; and a search unit that receives input of a third state quantity acquired from a control target as input data, selects nodes belonging to a cluster to which the input data belongs and a cluster immediately following the cluster to which the input data belongs, as the nodes that are smaller in number than the number of data elements of the learning data included in the model, and searches for a node that is closest to the input data from the nodes belonging to the two selected clusters as a node that matches or approximates the input data; an output unit that outputs a value based on the second command value of the searched node to the control object as an output command value to be given to operate the control object, Information processing device.

2. The search unit determines the output command value based on statistics calculated from command values ​​of some or all of the nodes in a cluster to which the searched node belongs. The information processing device according to claim 1 .

3. The search unit outputs a command value of the searched node as the output command value. The information processing device according to claim 1 .

4. the searching unit determines the output command value based on statistics calculated from a command value of the searched node and command values ​​of one or more nodes similar to the searched node; The information processing device according to claim 1 .

5. The one or more nodes similar to the searched node are nodes within a predetermined distance from the searched node, or a predetermined number of nodes selected in order of proximity to the searched node. The information processing device according to claim 4.

6. The statistic is any one of an average value, a median value, a maximum value, a minimum value, and a mode value.

6. The information processing device according to claim 4 or 5.

7. the second state quantity is the first state quantity previously acquired from the learning object, and the second command value is the first command value given to the learning object in response to the first state quantity previously acquired. The information processing device according to claim 1 .

8. the first command value is an actual value of a command value given to the learning object by an operator of the learning object based on the first state quantity previously acquired from the learning object; The information processing device according to claim 7.

9. The first command value included in the data element of the learning data is a command value given to the control target after a predetermined time with respect to the first state quantity previously acquired from the learning target. The information processing device according to claim 7.

10. a first state quantity previously acquired from a learning object and a first command value corresponding to the first state quantity given to control an operation of the learning object, the learning data being based on time-series data including a plurality of data elements expressed as a multidimensional vector, the data including the first state quantity previously acquired from the learning object and a first command value given to control an operation of the learning object, the learning data being time-series clustered as a node expressed as a multidimensional vector including a second state quantity and a second command value, and a distribution structure of the learning data being held as a learned model as a set of the nodes classified into a plurality of clusters; a third state quantity acquired from a control target is input as input data, and nodes belonging to a cluster to which the input data belongs and a cluster immediately following the cluster to which the input data belongs are selected as the nodes that are smaller in number than the number of data elements of the learning data included in the model, and a node that is closest to the input data from the nodes belonging to the two selected clusters is searched for as a node that matches or approximates the input data; outputting a value based on the second command value of the searched node to the control object as an output command value to be given for operating the control object; Information processing methods.

11. a process of time-series clustering each of the learning data based on time-series data including a plurality of data elements expressed as a multidimensional vector, the data including a first state quantity previously acquired from a learning object and a first command value given to control an operation of the learning object, as a node expressed as a multidimensional vector including a second state quantity and a second command value, thereby retaining a model learned as a set of the nodes classified into a plurality of clusters, the distribution structure of the learning data; a process of inputting a third state quantity acquired from a control target as input data, selecting nodes belonging to a cluster to which the input data belongs and a cluster immediately following the cluster to which the input data belongs as the nodes that are fewer in number than the number of data elements of the learning data included in the model, and searching for a node that is closest to the input data from the nodes belonging to the two selected clusters as a node that matches or approximates the input data; outputting, to the control object, a value based on the second command value of the searched node as an output command value to be given for operating the control object; program.

Citation Information

Patent Citations

  • Device and method for image retrieval and image display device

    JP2000276484A

  • Information processor, information processing method, and program

    JP2008217246A

  • Pattern recognition device, pattern recognition method, and program

    JP2008299640A

  • Information processing apparatus, information processing method and program

    JP2014164396A

  • Drilling machine, method for controlling drilling machine, and drilling system

    WO2016114240A1