JP2019212045A - Computer system and learning control method - Google Patents

Abstract

To automatically select an optimal plan history used in a re-learning process for updating information used to create a plan.SOLUTION: There is provided a computer system for managing a plan that regulates an order of plural processes for controlling an event, in which the processes include items related to controlling the event, and the plan is created based on a constraint condition defined from parameters and threshold values calculated from the items, and on a plan pattern composed of divergence reference values representing a tendency of adaptability of the constraint condition in a plan history. Further, in the computer system, a computer stores constraint condition information, plan pattern information, and plan history information for managing data in which the plan history is associated with an evaluation value, selects, when a first constraint condition is updated and based on the evaluation value, a re-using plan history from the plan history information, and executes a re-leaning process based on the re-using plan history information and the constraint condition to update the plan pattern information.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a computer system that generates a plan that defines the order of processing for controlling an event, and a learning control method thereof.

  In order to generate in advance plans such as product manufacturing plans, business system operation plans, and management plans, it is necessary to consider constraints on resources such as time and space, equipment, and personnel.

  Since time, knowledge, experience, etc. are necessary to generate a plan, it is difficult for a user who does not have knowledge to generate a plan manually. It was. Therefore, there is a need for a computer or system that automatically generates a plan of the same quality as a plan generated by an expert.

  A technique described in Patent Document 1 is known as a technique for generating a plan by adding tacit knowledge that is not known in advance as a constraint condition. Japanese Patent Laid-Open No. 2004-133826 describes a storage device that stores information on a plurality of created plans created for a predetermined event and each usage information used to create each of the created plans, and a constraint that has been found for the predetermined event. Applying conditions and each usage information to a predetermined algorithm to create multiple temporary plans, and the similarity between plans is less than the predetermined standard in each set of each temporary plan and the prepared plan corresponding to the temporary plan Apply the created plans and their usage information in dissimilar sets to a predetermined algorithm, generate new constraint conditions corresponding to dissimilar factors between plans, and set the new constraint conditions and already-known constraint conditions to a predetermined plan A plan creation support apparatus is described that includes an arithmetic unit that executes a constraint generation process provided to a generation unit.

Japanese Patent Laid-Open No. 2006-189079

  As described in Patent Document 1, tacit knowledge is specified by executing a learning process using previously known constraint conditions and plan history. When the constraint condition is changed, it is necessary to execute the learning process again. In this specification, the learning process executed after the second time is referred to as a relearning process.

  Since the plan history is a plan generated based on the constraint conditions before the update, the entire plan history cannot be used for the relearning process. Therefore, conventionally, it is necessary for the user to select a plan history that can be used or to collect a new plan history and then execute the relearning process.

  It takes time to select the plan history and collect the plan history. Therefore, it is difficult to automatically and quickly generate a new plan as the constraint condition is changed.

  The present invention realizes a computer system and method for automatically selecting a plan history to be used for the relearning process and executing the relearning process as the constraint conditions are updated.

  A typical example of the invention disclosed in the present application is as follows. That is, a computer system that manages a plan that defines an order of a plurality of processes for controlling an event, wherein the computer system is connected to an arithmetic device, a storage device connected to the arithmetic device, and the arithmetic device Wherein the process includes at least one item related to the control of the event, and the plan is defined from parameters and threshold values calculated from the at least one item. And at least one computer is generated based on a plan pattern composed of deviation reference values representing a tendency of conformity of the constraints in the plan history that is the plan generated in the past, Constraint condition information for managing the constraint condition, plan pattern information for managing the plan pattern, and The plan history information for managing the data associated with the evaluation value representing the quality of the plan, calculated based on the plan history and the constraint condition, is stored and managed based on the constraint condition information. When one constraint condition is updated, a reuse plan history is selected from the plan history information based on the evaluation value of the plan history, and a relearning process based on the reuse plan history and the constraint condition is executed. As a result, the plan pattern information is updated.

  According to one aspect of the present invention, it is possible to automatically select a plan history to be used for the relearning process and execute a quick relearning process. Problems, configurations, and effects other than those described above will become apparent from the description of the following examples.

1 is a diagram illustrating a configuration example of a system according to a first embodiment. 6 is a diagram illustrating an example of a data structure of constraint condition information according to Embodiment 1. FIG. It is a figure which shows an example of the data structure of the plan history information of Example 1. FIG. It is a figure which shows an example of the data structure of the plan of Example 1. FIG. It is a figure which shows an example of the data structure of the plan pattern information of Example 1. FIG. 3 is a flowchart illustrating processing executed by an update control unit according to the first embodiment. 3 is a flowchart illustrating processing executed by a learning unit according to the first embodiment. 3 is a flowchart illustrating processing executed by a learning unit according to the first embodiment. It is a flowchart explaining the process which the plan production | generation part of Example 1 performs.

  Embodiments of the present invention will be described below with reference to the drawings. However, the present invention is not construed as being limited to the description of the embodiments below. Those skilled in the art will readily understand that the specific configuration can be changed without departing from the spirit or the spirit of the present invention.

  In the configurations of the invention described below, the same or similar configurations or functions are denoted by the same reference numerals, and redundant descriptions are omitted.

  In the present specification and the like, the notations such as “first”, “second”, and “third” are given for identifying the constituent elements, and do not necessarily limit the number or order.

  The position, size, shape, range, and the like of each component illustrated in the drawings and the like may not represent the actual position, size, shape, range, or the like in order to facilitate understanding of the invention. Therefore, the present invention is not limited to the position, size, shape, range, and the like disclosed in the drawings and the like.

  FIG. 1 is a diagram illustrating a configuration example of a system according to the first embodiment.

  The system includes a computer 100 and a target system 105. The computer 100 and the target system 105 are connected to each other directly or via a network. The network is, for example, a LAN (Local Area Network) and a WAN (Wide Area Network). The network connection method may be either wired or wireless. The computer 100 may be connected to a terminal operated by the user via a network.

  The target system 105 is a system that controls events based on a plan. The target system 105 is, for example, a factory that manufactures a product, an air conditioning system, or the like.

  The computer 100 generates a plan that defines the order of processing for controlling events in the target system 105. The computer 100 according to the first embodiment learns a plan generated in the past, and generates a new plan based on the learning result and the constraint condition.

  The expert who controls the event in the target system 105 has knowledge that the event can be appropriately controlled even when some of the constraints are not satisfied, and the plan does not satisfy some of the constraints based on the knowledge. Is generated. In order to generate a plan that incorporates the above-described knowledge, the computer 100 learns the constraint conditions that allow violations and the degree of violation using the plans generated in the past. The computer 100 holds the learning result as a plan pattern. The computer 100 generates a new plan based on the plan pattern and the constraint conditions.

  The computer 100 includes an arithmetic device 111, a storage device 112, a network interface 113, and an IO interface 114 as hardware. Each hardware is connected to each other via an internal bus. The computer 100 may include a storage medium such as an HDD (Hard Disk Drive) and an SSD (Solid State Drive).

  The arithmetic device 111 is a processor or the like, and executes a program stored in the storage device 112. The arithmetic device 111 operates as a functional unit that realizes a specific function by executing processing according to a program. In the following description, when processing is described with a functional unit as the subject, it indicates that the arithmetic device 111 is executing a program that realizes the functional unit.

  The storage device 112 is a memory or the like, and stores a program executed by the arithmetic device 111 and information used by the program. The storage device 112 also includes a work area that is temporarily used by the program. The program and information stored in the storage device 112 will be described later.

  The network interface 113 is an interface for connecting to other devices via a network. The IO interface 114 is an interface for connecting to the input device 101 and the output device 102.

  The input device 101 is, for example, a keyboard, a pointing device such as a mouse, and a device such as a touch panel. The output device 102 is a device such as a display and a printer, for example.

  Here, the program and information stored in the storage device 112 will be described. The storage device 112 stores programs that implement the learning unit 120, the plan generation unit 121, the constraint condition management unit 122, and the update control unit 123. The storage device 112 also stores constraint condition information 130, plan history information 131, and plan pattern information 132.

  The constraint condition information 130 is information for managing a constraint condition related to an event. For example, there are constrained conditions regarding resources such as time, space, equipment, and personnel, and constrained conditions regarding quality such as product strength. The data structure of the constraint condition information 130 will be described with reference to FIG.

  The plan history information 131 is information for managing a plan generated in the past, that is, a plan history. The data structure of the plan history information 131 will be described with reference to FIG.

  The plan pattern information 132 is information for managing a plan pattern, which is information representing a tendency of conformity of constraint conditions in a plan generated in the past. The data structure of the plan pattern information 132 will be described with reference to FIG.

  The plan pattern is represented as an array of deviation reference values for each constraint condition. The deviation reference value of one constraint condition is an index representing the tendency of conformity of the constraint conditions in a plurality of plans generated in the past, and is also an index representing the importance of the constraint conditions. The reference deviation value of one constraint condition is calculated based on the deviation degree of one constraint condition in the plan history. The degree of deviation of one constraint condition is an index indicating the degree of deviation of one constraint condition in one plan.

  The learning unit 120 executes a learning process for generating or updating information used for generating a plan based on the plan history. More specifically, the learning unit 120 generates the plan pattern information 132 by executing a learning process before the target system 105 is operated. In addition, when the learning unit 120 receives an instruction from the update control unit 123, the learning unit 120 updates the plan pattern information 132 by executing the learning process again. In the following description, the learning process executed to update the plan pattern information 132 is described as a relearning process.

  When receiving the input data, the plan generation unit 121 generates a plan based on the constraint condition information 130 and the plan pattern information 132. Here, the input data is a set of data (processing data) relating to processing for controlling an event in the target system 105 in which the execution order is not defined. The plan generation unit 121 generates a plan by determining a processing order for realizing appropriate control. That is, the plan is a set of processing data in which the processing order is determined.

  The constraint condition management unit 122 manages the constraint condition information 130. The constraint condition management unit 122 adds, updates, and deletes constraint conditions.

  The update control unit 123 controls the learning process. The update control unit 123 receives the learning execution request and the constraint condition update request via the network interface 113 or the IO interface 114. The constraint condition update request is a request for updating the constraint condition, and includes constraint condition identification information and update contents. As the update content, for example, a value set as a new threshold value can be considered. Conditional expressions may be included as update contents.

  In addition, about each function part which the computer 100 has, a several function part may be put together into one function part, and one function part may be divided into a some function part for every function.

  FIG. 2 is a diagram illustrating an example of a data structure of the constraint condition information 130 according to the first embodiment.

  The constraint condition information 130 stores an entry including a condition ID 201, a conditional expression 202, and a threshold value 203. One entry corresponds to one constraint condition.

  The condition ID 201 is a field that stores identification information for identifying the constraint condition. For example, a number is stored in the condition ID 201.

  The conditional expression 202 is a field for storing a conditional expression set as a constraint condition. The conditional expression is defined based on a parameter calculated from items related to processing and a variable indicating a threshold value.

  The threshold value 203 is a field for storing a value set as a threshold value of the conditional expression.

  FIG. 3 is a diagram illustrating an example of a data structure of the plan history information 131 according to the first embodiment.

  The plan history information 131 stores an entry including a plan ID 301, a storage location 302, a time stamp 303, and an evaluation value 304. One entry corresponds to one plan history.

  The plan ID 301 is a field for storing identification information for identifying a plan. For example, a number is stored in the plan ID 301.

  The storage location 302 is a field for storing a value indicating the storage location of the plan. In the storage location 302, for example, an address of a storage area for storing the plan is stored.

  The time stamp 303 is a field for storing the time when the plan is generated. Note that the time stored in the time stamp 303 may be a time managed by the system or a time managed by the computer 100.

  The evaluation value 304 is a field for storing an evaluation value that is an index representing the quality of the plan. As the number of constraint conditions to be satisfied is larger, the plan is higher in quality, that is, is treated as an appropriate plan. A plan similar to the plan history may be treated as a high-quality plan. In this case, the evaluation value can be used as an index indicating similarity with the plan history.

  Here, the data structure of the plan will be described. FIG. 4 is a diagram illustrating an example of a data structure of a plan according to the first embodiment.

  The plan 400 stores an entry composed of an order 401 and items 402. One entry corresponds to one process (process data).

  The order 401 is a field for storing the execution order of processing.

  An item 402 is a field group that stores items related to processing. The item 402 includes one or more fields. The item 402 shown in FIG. 4 includes material number, weight, and length fields.

  FIG. 5 is a diagram illustrating an example of a data structure of the plan pattern information 132 according to the first embodiment.

  The plan pattern information 132 stores an entry composed of a condition ID 501 and a deviation reference value 502. One entry corresponds to a deviation reference value of one constraint condition. Condition ID 501 is the same field as condition ID 201.

  The deviation reference value 502 is a field for storing the deviation reference value of the constraint condition corresponding to the condition ID 501. The reference deviation value of one constraint condition is calculated based on the deviation degree of the constraint condition of the plan. The degree of divergence takes a large value in the case of a plan with a large divergence of constraints, and takes a small value in a plan with a small divergence of constraints. Note that the degree of deviation of the plan that satisfies the constraint condition is “0”.

  Here, the outline of the invention of Example 1 will be described. First, the relationship between the deviation reference value and the evaluation value will be described. In Example 1, the evaluation value of a plan is calculated using Formula (1).

(Deviation degree) _i represents the deviation degree of the constraint condition whose identification information in one plan is “i”. (Difference reference value) _i represents a deviation reference value of a constraint condition whose identification information in the plan pattern is "i".

  The degree of deviation of the constraint condition is calculated based on the following process.

  (Calculation procedure 1) The computer 100 calculates the value of the parameter defined by the constraint condition using the processing data included in one plan history. The computer 100 determines whether or not the constraint condition is satisfied based on the comparison result of the parameter value and the threshold value.

  (Calculation procedure 2) When it is determined that the constraint condition is satisfied, the computer 100 calculates “0” as the degree of deviation. When it is determined that the constraint condition is not satisfied, the computer 100 calculates the absolute value of the difference between the parameter value and the threshold value, the ratio between the parameter value and the threshold value, and the like as the degree of divergence.

  In the first embodiment, the deviation reference value is calculated using Equation (2).

  The percentile is a percentile function whose value is determined from the array and the rate. d (i) represents an array of divergence degrees of constraint conditions whose identification information in each plan history is “i”, and p represents a percentile calculation parameter. c represents the number of elements of d (i). n represents the number of plan histories that do not satisfy the constraint condition in which the identification information is “i” in d (i), that is, the number of elements whose value is not “0”.

  In addition, the calculation method of the above-mentioned evaluation value, deviation degree, and deviation reference value is an example, and this invention is not limited to the calculation method of each value.

  The computer 100 according to the first embodiment selects a plan history to be used for the relearning process associated with the update of the constraint condition based on the evaluation value.

  Since the deviation degree and the deviation reference value are updated with the update of the constraint condition, the evaluation value is also updated. Therefore, it is necessary to select a plan history using the updated evaluation value. However, the divergence degree and the divergence reference value are calculated based on the updated constraint condition and the selected plan history. As described above, the plan history, the evaluation value, and the deviation reference value have a mutually dependent relationship, and thus a device is required to select the plan history based on the evaluation value.

  Hereinafter, processing for realizing the above-described function will be described. First, an outline of processing executed by the computer 100 will be described.

  The computer 100 executes a learning process in advance before generating a plan for the target system 105. Plan pattern information 132 is generated by the learning process. When the input data is received, the computer 100 generates a plan from the input data based on the constraint condition information 130 and the plan pattern information 133.

  When receiving a constraint update request, the computer 100 selects a plan history to be used for the relearning process from the plan history information 131, and executes the relearning process using the selected plan history. In the following description, the plan history used for the relearning process is also referred to as a reuse plan history. The computer 100 selects a reuse plan history in the following procedure.

  (Procedure 1) The computer 100 calculates the divergence degree of the updated constraint condition for each plan history, and selects a plan history having a small divergence degree as a plan history for updating. The computer 100 calculates a temporary deviation reference value for the updated constraint condition based on the deviation degree of the constraint condition to be updated in the plan history for update.

  (Procedure 2) The computer 100 calculates a temporary evaluation value based on the updated temporary deviation reference value of the constraint condition and the deviation reference value of another constraint condition. The computer 100 selects a reuse plan history based on the temporary evaluation value. Details of each process will be described below.

  FIG. 6 is a flowchart illustrating processing executed by the update control unit 123 according to the first embodiment.

  When receiving an instruction from the outside, the update control unit 123 starts processing described below.

  The update control unit 123 determines whether or not the received instruction is a constraint condition update request (step S101).

  When it is determined that the received instruction is not a constraint condition update request, that is, when it is determined that the received instruction is a learning execution request, the update control unit 123 outputs a learning execution request to the learning unit 120 ( Step S109) and the process is terminated.

  If it is determined that the received instruction is a constraint condition update request, the update control unit 123 outputs a constraint condition update request to the constraint condition management unit 122, and also outputs a temporary update request to the learning unit 120 (step). S102). Thereafter, the update control unit 123 shifts to a waiting state until an update completion notification is received from the learning unit 120. The provisional update request includes information on constraint conditions to be updated. The constraint condition information to be updated is, for example, a list of identification information of constraint conditions to be updated.

  Next, the update control unit 123 determines whether or not an update completion notification has been received from the learning unit 120 (step S103).

  When it is determined that the update completion notification has not been received from the learning unit 120, the update control unit 123 returns to step S103 after a predetermined time has elapsed.

  When it is determined that the update completion notification has been received from the learning unit 120, the update control unit 123 starts a loop process of the plan history (Step S104).

  Specifically, the update control unit 123 selects a target plan history from the plan history information 131. As will be described later, when the update completion notification is received, the plan pattern information 132 is updated based on the updated constraint condition.

  Next, the update control unit 123 calculates a temporary evaluation value of the target plan history (Step S105).

  Specifically, the update control unit 123 calculates the divergence degree of each constraint condition for the target plan history. The update control unit 123 acquires a deviation reference value for each constraint condition from the plan pattern information 132. The plan generation unit 121 calculates a temporary evaluation value by substituting the deviation degree and the temporary deviation reference value of each constraint condition into Expression (1).

  Next, the update control unit 123 determines whether or not all the plan history processing has been completed (step S106).

  If it is determined that the processing of all the plan histories has not been completed, the update control unit 123 returns to step S104 and selects a new target plan history.

  When it is determined that all the plan history processing has been completed, the update control unit 123 selects a reuse plan history based on the temporary evaluation value (step S107).

  Specifically, the update control unit 123 calculates the average value and standard deviation of the temporary evaluation values of the plan history. The update control unit 123 selects, as a reuse plan history, a plan history whose temporary evaluation value is less than or equal to the average value of the temporary evaluation values and the total value of the standard deviations.

  Next, the update control unit 123 outputs an update request including information on the reuse plan history to the learning unit 120 (step S108), and ends the process. The reuse plan history information included in the update request is, for example, a list of reuse plan history identification information.

  The update control unit 123 may delete the plan history that has not been selected as the reuse plan history from the plan history information 131. At this time, the update control unit 123 may hold the deleted plan history as an archive.

  In addition, after the plan pattern information 132 is updated by the learning unit 120, the update control unit 123 calculates a new evaluation value based on the updated plan pattern information 132 and the degree of deviation of the constraint conditions in the reuse plan history. To do. The update control unit 123 sets the calculated evaluation value as the evaluation value 304 of each entry of the plan history information 131.

  7A and 7B are flowcharts for explaining processing executed by the learning unit 120 according to the first embodiment.

  When the learning unit 120 receives the request, the learning unit 120 executes processing described below.

  The learning unit 120 determines whether or not the received request is a learning execution request (step S201).

  When it is determined that the received request is a learning execution request, the learning unit 120 starts a loop process of the plan history (Step S202).

  Specifically, the learning unit 120 selects one target plan history from the plan history information 131.

  Next, the learning unit 120 starts a constraint loop process (step S203).

  Specifically, the learning unit 120 selects one target constraint condition from the constraint condition information 130.

  Next, the learning unit 120 calculates the divergence degree of the target constraint condition in the target plan history (step S204). Specifically, the following processing is executed.

  The learning unit 120 calculates the value of the parameter defined by the target constraint condition using the processing data included in the target plan history. The learning unit 120 determines whether or not the target constraint condition is satisfied based on the comparison result of the parameter value and the threshold value.

  When it is determined that the target constraint condition is satisfied, the learning unit 120 calculates “0” as the degree of divergence. When it is determined that the target constraint condition is not satisfied, the learning unit 120 calculates the absolute value of the difference between the parameter value and the threshold value, or the ratio between the parameter value and the threshold value as the divergence degree.

  In this way, an array of divergence degrees of each constraint condition is generated for one plan history. The above is the description of the process in step S204.

  Next, the learning unit 120 determines whether or not the processing of all the constraint conditions has been completed (step S205). That is, it is determined whether or not the divergence degrees for all the constraint conditions of the target plan history have been calculated.

  If it is determined that all the constraint conditions have not been processed, the learning unit 120 returns to step S203 and selects a new target constraint condition.

  When it is determined that all the constraint conditions have been processed, the learning unit 120 determines whether all the plan history processes have been completed (step S206).

  When it is determined that the processing of all the plan histories has not been completed, the learning unit 120 returns to step S202 and selects a new target plan history.

  When it is determined that all the plan history processes have been completed, the learning unit 120 generates the plan pattern information 132 (step S207), and then ends the process. Specifically, the following processing is executed.

  The learning unit 120 substitutes the divergence degree of each constraint condition of all the plan histories into Equation (2), and calculates a divergence reference value of each constraint condition.

  The learning unit 120 generates plan pattern information 132 and generates as many entries as the number of constraints in the plan pattern information 132. The learning unit 120 sets constraint condition identification information in the generated condition ID 501 of each entry, and sets a deviation reference value corresponding to the constraint condition in the deviation reference value 502 of each entry. The above is the description of the processing in step S207.

  If it is determined in step S201 that the received request is not a learning execution request, the learning unit 120 determines whether the received request is a temporary update request (step S208).

  When it is determined that the received request is a temporary update request, the learning unit 120 starts a loop process of the plan history (Step S209).

  Specifically, the learning unit 120 selects one target plan history based on the constraint condition information 130 and information on the constraint condition to be updated included in the temporary update request.

  Next, the learning unit 120 starts a loop process for the constraint condition to be updated (step S210).

  Specifically, the learning unit 120 selects one target constraint condition from the constraint conditions to be updated.

  Next, the learning unit 120 calculates the deviation degree of the target constraint condition in the target plan history (step S211). The calculation method of the degree of deviation of the constraint condition is the same as that in step S204.

  Next, the learning unit 120 determines whether or not the processing of all the update target constraint conditions has been completed (step S212). That is, it is determined whether or not the divergence degrees for all the update target constraint conditions in the target plan history have been calculated.

  If it is determined that all the update target constraint conditions have not been processed, the learning unit 120 returns to step S210 and selects a new target constraint condition.

  When it is determined that all the update target constraint conditions have been processed, the learning unit 120 determines whether all the plan history processing has been completed (step S213).

  When it is determined that the processing of all the plan histories has not been completed, the learning unit 120 returns to step S209 and selects a new target plan history.

  When it is determined that the processing of all the plan histories has been completed, the learning unit 120 selects a plan history for update based on the divergence degree of the constraint conditions to be updated (step S214).

  Specifically, the learning unit 120 selects a predetermined number of plan histories as update plan histories in ascending order of the degree of deviation of the constraint conditions to be updated. If there are a plurality of update target constraint conditions, the learning unit 120 selects a predetermined number of plan histories as the update plan history in ascending order of the total value of the divergence degrees of the update target constraint conditions. Note that the number of plan histories to be selected is set in advance. However, the number of plan histories to be selected can be updated at an arbitrary timing.

  Next, the learning unit 120 calculates a temporary deviation reference value of the constraint condition to be updated based on the degree of deviation of the constraint condition to be updated in the plan history for update (step S215). The calculation method of the deviation reference value is the same as the calculation method used in step S207. However, the difference is that the temporary deviation reference value of the constraint condition to be updated is calculated using only the deviation degree of the plan history for update. The learning unit 120 refers to the plan pattern information 132 and overwrites the calculated temporary deviation reference value to the deviation reference value 502 of the entry of the constraint condition to be updated.

  Next, the learning unit 120 outputs an update completion notification to the update control unit 123 (step S216), and then ends the process.

  In step S208, when it is determined that the received request is not a temporary update request, that is, when it is determined that the received request is an update request, the learning unit 120 starts a loop process of a reuse plan history ( Step S217).

  Specifically, the learning unit 120 selects one target plan history from the reuse plan history.

  Next, the learning unit 120 starts a constraint loop process (step S218).

  Specifically, the learning unit 120 selects one target constraint condition from the constraint condition information 130.

  Next, the learning unit 120 calculates the divergence degree of the target constraint condition in the target plan history (step S219). The calculation method of the divergence degree is the same as the calculation method used in step S204.

  Next, the learning unit 120 determines whether or not the processing of all the constraint conditions has been completed (step S220).

  When it is determined that the processing of all the constraint conditions has not been completed, the learning unit 120 returns to step S218 and selects a new target constraint condition.

  When it is determined that all the constraint conditions have been processed, the learning unit 120 determines whether all the reuse plan history processes have been completed (step S221).

  If it is determined that the processing of all the reuse plan histories has not been completed, the learning unit 120 returns to step S217 and selects a new target plan history from the reuse plan histories.

  If it is determined that all the reuse plan history processes have been completed, the learning unit 120 updates the plan pattern information 132 (step S222), and then ends the process.

  Specifically, the learning unit 120 calculates a deviation reference value for each constraint condition based on the deviation degree of each constraint condition. The learning unit 120 refers to the plan pattern information 132 and overwrites the calculated deviation reference value to the deviation reference value 502 of the entries corresponding to all the constraint conditions.

  FIG. 8 is a flowchart illustrating the process executed by the plan generation unit 121 according to the first embodiment.

  When the computer 100 receives input data via the network interface 103 or the IO interface 104, the computer 100 starts processing described below.

  The plan generation unit 121 acquires input data (step S301), and generates a new plan by determining the processing order based on the constraint condition information 130, the plan pattern information 132, and the input data (step S302). . Two or more new plans may be generated for one input data.

  In principle, the plan generation unit 121 generates a new plan that satisfies all the constraint conditions. However, the plan generation unit 121 according to the first embodiment generates a new plan that does not satisfy some of the constraint conditions based on the plan pattern information 132. For example, the plan generation unit 121 corrects the constraint condition by adding or multiplying the deviation reference value to the threshold value of the constraint condition, and generates a new plan based on the corrected constraint condition. As a result, a plan that does not satisfy the constraint condition with a large degree of divergence is generated.

  As a plan generation method, for example, it is conceivable to use mathematical programming, linear programming, or the like. The present invention is not limited to the plan generation method. A plan generation method can be selected according to the constraint condition and the plan to be generated.

  Next, the plan generation unit 121 starts a loop process for a new plan (step S303).

  Specifically, the plan generation unit 121 selects a target new plan from the new plans.

  Next, the plan generation unit 121 calculates an evaluation value of the target new plan (step S304).

  Specifically, the plan generation unit 121 calculates the divergence degree of each constraint condition for the target new plan. The plan generation unit 121 acquires a deviation reference value for each constraint condition from the plan pattern information 132. The plan generation unit 121 calculates an evaluation value by substituting the deviation degree and the deviation reference value of each constraint condition into Expression (1).

  Next, the plan generation unit 121 determines whether or not processing has been completed for all new plans (step S305).

  When it is determined that the processing has not been completed for all the new plans, the plan generation unit 121 returns to step S303 and selects a new target new plan.

  When it is determined that processing has been completed for all new plans, the plan generation unit 121 selects a new plan to be output based on the evaluation value (step S306).

  For example, the plan generation unit 121 selects a new plan having the largest evaluation value. In addition, the plan generation unit 121 selects a predetermined number of new plans in descending order of evaluation values.

  Next, the plan generation unit 121 outputs the selected new plan (step S307). Thereafter, the plan generation unit 121 ends the process.

  At this time, the plan generation unit 121 stores the history of the new plan in the plan history information 131. Note that the history of all the generated new plans may be stored in the plan history information 131, or the history of the selected new plan may be stored in the plan history information 131.

  According to the first embodiment, when the constraint condition is updated, the computer 100 can automatically select a plan history to be used for relearning from the plan history. As a result, the relearning process associated with the update of the constraint condition can be quickly executed. Further, by executing the relearning process using the reuse plan history selected based on the temporary evaluation value, it is possible to prevent the quality of the plan from being deteriorated.

  In addition, this invention is not limited to an above-described Example, Various modifications are included. Further, for example, the above-described embodiments are described in detail for easy understanding of the present invention, and are not necessarily limited to those provided with all the described configurations. Further, a part of the configuration of each embodiment can be added to, deleted from, or replaced with another configuration.

  Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. The present invention can also be realized by software program codes that implement the functions of the embodiments. In this case, a storage medium in which the program code is recorded is provided to the computer, and a processor included in the computer reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiments, and the program code itself and the storage medium storing it constitute the present invention. As a storage medium for supplying such a program code, for example, a flexible disk, a CD-ROM, a DVD-ROM, a hard disk, an SSD (Solid State Drive), an optical disk, a magneto-optical disk, a CD-R, a magnetic tape, A non-volatile memory card, ROM, or the like is used.

  The program code for realizing the functions described in the present embodiment can be implemented by a wide range of programs or script languages such as assembler, C / C ++, perl, Shell, PHP, Java (registered trademark).

  Furthermore, by distributing the program code of the software that implements the functions of the embodiments via a network, the program code is stored in a storage means such as a hard disk or memory of a computer or a storage medium such as a CD-RW or CD-R. A processor included in the computer may read and execute the program code stored in the storage unit or the storage medium.

  In the above-described embodiments, the control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. All the components may be connected to each other.

100 Computer 101 Input Device 102 Output Device 103 Network Interface 104 IO Interface 105 Target System 111 Computing Device 112 Storage Device 113 Network Interface 114 IO Interface 120 Learning Unit 121 Plan Generation Unit 122 Constraint Condition Management Unit 123 Update Control Unit 130 Constraint Condition Information 131 Plan history information 132 Plan pattern information

Claims (8)

A computer system that manages a plan that defines the order of a plurality of processes for controlling an event,
The computer system includes at least one computer having an arithmetic device, a storage device connected to the arithmetic device, and an interface connected to the arithmetic device,
The processing includes at least one item related to the control of the event,
The plan includes a constraint condition defined from a parameter calculated from the at least one item and a threshold value, and a deviation reference value representing a tendency of conformity of the constraint condition in a plan history which is the plan generated in the past. Generated based on a planning pattern consisting of
The at least one calculator is:
Corresponds to constraint condition information for managing the constraint condition, plan pattern information for managing the plan pattern, and an evaluation value representing the quality of the plan calculated based on the plan history and the constraint condition Holds plan history information for managing the attached data,
When the first constraint condition managed based on the constraint condition information is updated, based on the evaluation value of the plan history, select a reuse plan history from the plan history information,
A computer system, wherein the plan pattern information is updated by executing a re-learning process based on the reuse plan history and the constraint conditions. The computer system according to claim 1,
The at least one calculator is:
For each plan history stored in the plan history information, calculate a divergence degree representing the degree of divergence of the first constraint condition in the plan history,
Based on the degree of divergence of the first constraint in the plan history, a temporary divergence reference value of the first constraint is calculated,
By storing the temporary deviation reference value of the first constraint condition in the plan pattern information, the plan pattern information is updated,
Based on the updated plan pattern information and the plan history, a temporary evaluation value of the plan history is calculated,
A computer system, wherein the reuse plan history is selected from the plan history information based on the temporary evaluation value. The computer system according to claim 2,
The at least one calculator is:
Based on the degree of deviation of the first constraint condition of each of the plurality of plan histories, select a plurality of plan histories for update from the plan history information,
A computer system that calculates a temporary deviation reference value of the first constraint condition based on a deviation degree of the first constraint condition in the plurality of update plan histories. The computer system according to claim 3,
The at least one calculator is:
Based on the reuse plan history and the updated constraint information, the degree of deviation of the constraint conditions in the reuse plan history is calculated,
Based on the degree of deviation of the constraint condition in the reuse plan history, calculate a deviation reference value of the constraint condition,
The computer system, wherein the plan pattern information is updated based on a deviation reference value of the constraint condition. A learning control method in a computer system that manages a plan that defines an order of a plurality of processes for controlling an event,
The computer system includes at least one computer having an arithmetic device, a storage device connected to the arithmetic device, and an interface connected to the arithmetic device,
The processing includes at least one item related to the control of the event,
The plan includes a constraint condition defined from a parameter calculated from the at least one item and a threshold value, and a deviation reference value representing a tendency of conformity of the constraint condition in a plan history that is the plan generated in the past. Generated based on a planning pattern consisting of
The at least one computer is calculated based on constraint condition information for managing the constraint condition, plan pattern information for managing the plan pattern, and the plan history and the constraint condition. Holds plan history information for managing data associated with quality evaluation values,
The learning control method includes:
When the first constraint managed based on the constraint condition information is updated, the at least one computer selects a reuse plan history from the plan history information based on the evaluation value of the plan history. The first step;
A learning control method comprising: a second step in which the at least one computer updates the plan pattern information by executing a re-learning process based on the reuse plan history and the constraint condition. . The learning control method according to claim 5,
The first step includes
A third step in which the at least one computer calculates, for each plan history stored in the plan history information, a divergence degree representing a degree of divergence of the first constraint condition in the plan history;
A fourth step in which the at least one computer calculates a temporary deviation reference value of the first constraint condition based on a deviation degree of the first constraint condition in the plan history;
A fifth step in which the at least one computer updates the plan pattern information by storing a temporary deviation reference value of the first constraint condition in the plan pattern information;
A step in which the at least one computer calculates a temporary evaluation value of the plan history based on the updated plan pattern information and the plan history;
A learning control method comprising: a seventh step in which the at least one computer selects the reuse plan history from the plan history information based on the temporary evaluation value. The learning control method according to claim 6, wherein:
The fourth step includes
The at least one computer selecting a plurality of plan histories for update from the plan history information based on a divergence degree of the first constraint condition of each of the plurality of plan histories;
The at least one computer includes a step of calculating a temporary deviation reference value of the first constraint condition based on a deviation degree of the first constraint condition in the plurality of plan histories for update. Learning control method. The learning control method according to claim 7,
The second step includes
The at least one computer calculates a deviation degree of the constraint condition in the reuse plan history based on the reuse plan history and the updated constraint information;
The at least one computer calculates a deviation reference value of the constraint condition based on a deviation degree of the constraint condition in the reuse plan history;
The at least one computer includes a step of updating the plan pattern information based on a deviation reference value of the constraint condition.

Images