JP6873936B2 - Analytical model construction support device and analytical model construction support method - Google Patents

Description

The present invention relates to a work procedure support system for constructing an analysis model for performing data analysis and simulation.

In recent years, IoT (Internet of Things) technology, that is, technology for collecting information from sensors installed in various facilities and equipment via a network, etc., monitoring the operating status of machines, and utilizing it for operation and maintenance has been developed. It is becoming popular. In this field, it is necessary to construct an analysis procedure for collecting sensor data and perform abnormality diagnosis as a program, and to construct a physical model for performing motion simulation of a target (hereinafter, they are collectively referred to). Is called an "analytical model").

These analysis models need to be customized according to the user's analysis purpose, operating status, and operating status, even if the equipment and devices are of the same type. Customization is specifically done by adjusting many parameters of the analytical model. Here, the parameter is a general term for various numerical values that determine the analysis model and type information related to the configuration, for example, options for adjustment such as which of the candidates is selected as the algorithm of the diagnostic model.

A typical method of adjusting parameters is machine learning using input data. For example, Patent Document 1 proposes a method of modifying a simulation model at any time based on data received from an actual plant in order to predict the operation of the actual plant using the simulation model.

However, the parameters of the analysis model must be determined in consideration of the business situation, in addition to those that can be determined by learning using sensing data, and must be determined according to the purpose of the customer. There are things.

Among them, the parameters determined in consideration of the business situation can be determined by referring to the business record data as long as it exists. For example, in a plant, if a test operation is performed during operation for a certain period of time, noise will be added to the sensor data, so the data during that period should not be used for machine learning. Therefore, the parameter for specifying the period of the usage data can be determined by configuring the process of excluding the period based on the information on whether or not there is maintenance during operation.

On the other hand, the process of determining parameters that cannot comply with any data and must be determined according to the customer's purpose basically depends on the records of the workers. For example, in the abnormality detection model, the target value of what percentage of false alarms and detection omissions are allowed is an example.

As an attempt to manually record and reuse the reasons for design decisions, for example, in Non-Patent Document 1, for and against nodes that distinguish the background of the discussion of design changes from problems, positions, opinions, etc. , Generalization, specialization, proposals, etc., have been proposed as a format for connecting and recording in a network.

Japanese Unexamined Patent Publication No. 2005-332360

Jeff Conklin, Michael L. Begeman, "gIBIS: a hypertext tool for exploratory policy computer," Proceedings of the 1988 Computer-supported Cooperative, 1988 ACM computer.

As described above, the parameters of the analytical model can be easily set by referring to the record when the model is reused, if the process of determination is intentionally recorded by the designer as in Non-Patent Document 1 above. It will be possible. However, since the amount of work for the designer to record the process of examination increases, the record of the process of examination is realized in the analysis model development, which is often created by a small number of people in a short period of time compared to the development of a computer program. There are not many cases.

Therefore, it is desirable to be able to record the process of determining the analysis model parameters as automatically as possible. As the first step, among the parameters of the analytical model, which ones were determined by learning with data and which ones were manually adjusted by the operator are automatically classified, and the information is used when reused. It is desirable to be able to utilize it.

One aspect of the present invention is an analysis model support device that supports the work of constructing an analysis model, and includes an input device, an output device, a storage device, and a processing unit. The device configuration may be a single device or a distributed system configured in the cloud. The storage device stores the operation history of the analysis model development program and the difference information of the analysis model version control program. The processing unit executes a history shaping program, a work classification program, and a classification result presentation program. The history shaping program associates the operation history with the difference information. The work classification program classifies the parameters of the analysis model appearing in the difference information based on the data reference relationship appearing in the operation history, and obtains the classification result. The classification result presentation program outputs information based on the classification result to the output device.

Another aspect of the present invention is to use an information processing device including an input device, an output device, a storage device, and a processing unit, an operation log of a model development program of an analysis model, and model difference data between versions of the analysis model. This is an analytical model construction support method that classifies the parameters of the analytical model based on. The operation log includes a plurality of lines, and each of the plurality of lines of the operation log includes an item of the operation content for the analysis model and an item of the reference data which is the data referred to in the operation as an item. The model difference data includes a plurality of rows, and each of the plurality of rows of the model difference data includes an item of the difference information of the analysis model as an item. In this method, a mapping step is performed in which a row of the operation log and a row of the model difference data are associated with each other based on a preset related rule that combines the contents of the operation log item and the contents of the model difference data item. Execute. In addition, a classification step is executed in which the parameters that are the contents of the difference information item in the predetermined line of the model difference data are classified based on the contents of the reference data item in the corresponding operation log line.

It is possible to understand how to adjust each parameter when reusing the model without increasing the burden on the analytical model developer. As a result, there are effects such as reduction of the amount of work and easy adjustment of the model even by an unskilled person.

It is a block diagram which shows the apparatus configuration example of Example 1. FIG. This table shows a conceptual example of the operation log, which is the operation log of the model development program. This is a table showing a conceptual example of model difference data output by the version control program. This is a table showing a conceptual example of the related rule list. It is a flowchart which shows the whole procedure of the recording operation in Example 1. FIG. It is a flowchart which shows the procedure of associating difference data with an operation log. It is a table which shows the conceptual diagram of the composite data which was associated with the difference data and the operation log association procedure. It is a flowchart which shows the procedure of the parameter classification for one parameter difference event using the difference data and the operation log correspondence result. It is a flowchart which shows the procedure which presents and supports the classification data recorded at the time of model reuse. It is an image diagram which shows the GUI example in Example 1. FIG. It is a block diagram which shows the apparatus configuration example of Example 2. FIG. This is a table showing a conceptual example of worker data. It is a flowchart which shows the whole procedure of the recording operation in Example 2. It is an image diagram which shows the GUI example in Example 2. FIG.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. However, the present invention is not construed as being limited to the description of the embodiments shown below. It is easily understood by those skilled in the art that a specific configuration thereof can be changed without departing from the idea or gist of the present invention.

In the configuration of the invention described below, the same reference numerals may be used in common among different drawings for the same parts or parts having similar functions, and duplicate description may be omitted.

The notations such as "first", "second", and "third" in the present specification and the like are attached to identify the components, and do not necessarily limit the number, order, or contents thereof. is not it. In addition, numbers for identifying components are used for each context, and numbers used in one context do not always indicate the same composition in other contexts. Further, it does not prevent the component identified by a certain number from having the function of the component identified by another number.

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

FIG. 1 shows an example of the device configuration of this embodiment. This configuration example is a configuration example in which the processing flow is designed by the server computer and at the same time, the classification support of the parameter correction work of the analysis model is performed. The analysis model construction server (101) is composed of a processing unit (102), a storage device (103), and a communication I / F (104), and is operated by a user at a work terminal (106) via a network (105). I do. The processing unit of the analysis model construction server contains programs such as a model development program (107), a version control program (108), a history shaping program (109), a work classification program (110), and a classification result presentation program (111). It is operational.

The model development program (107) is a program for the user to design the processing contents of the analysis model by using the model development program (107). For example, the model development program (107) is designed by the user changing the processing flow of the model while displaying the processing flow of the model by GUI. Is to do. As a concrete example, existing programs such as KNIME ™, which is an analysis flow description platform, can be used.

The version control program (108) is a program that records the difference in how the model design data (described later) in which the result designed on the model development program is described has changed. A known version control program for program development is available.

The history shaping program (109) performs a process of associating and extracting parameter correction and reference data using the model development log and version control difference information. The details of the procedure will be described later using the drawing of the processing flow.

The work classification program (110) labels which work pattern the parameter correction work is classified according to the type of reference data.

The classification result presentation program (111) supports model design by presenting the result of work classification to the user.

The storage device includes model design data (112), sensor data (113), business data (114), operation log (115), model difference data (116), related rule list (117), and work classification data (118). Each data of is stored.

The model design data (112) is data that describes the structure of the analytical model designed on the model development program, and corresponds to the program description. For example, the process of analysis describes where to read data, what kind of processing is performed, and what kind of output is performed.

The sensor data (113) is read by the analysis model as input data and used for learning some parameters and verifying the operation. That is, it is the data to be analyzed by the analysis model. It is assumed that this sensor data is collected from the device and temporarily stored in this server. As another configuration, it may be configured so that it can be referred to by connecting to a system that monitors the sensor value via a network.

The business data (114) is operational data other than sensor data, and describes operational information such as maintenance of the target equipment during a certain period of time, adjustment of output for some reason, and the like. It was done. Like the sensor data, the operations related to the operation are collected from the related system. In this embodiment, the business data (114) is also collected offline, but it may be configured to be connected to the system via a network and always referred to.

The operation log (115) is an operation log output by the model development program (107), and records what kind of operation the designer has performed. The contents will be described later.

The model difference data (116) is data output by the version control program (108), and is a time-series record of the differences in the model design data. The contents will be described later.

The model difference data (116) may be simply referred to as the difference data (116) below. The related rule list (117) lists the relationship between which module in the analysis model was operated if there was a key phrase in the correspondence between the difference data (116) and the operation log (115). It is a thing.

The work classification data (118) is the work classification result extracted by the procedure of this embodiment described later. The contents will be described later.

In this embodiment, the device configuration is such that the program and data placed on the analysis model construction server (101) are operated from the work terminal (106) via the network, but even the above-mentioned various data can be acquired. If there is, a device configuration that uses the same program and data on a single computer may be configured instead of via the network. Further, the analysis model construction server (101) and the work terminal (106) may be configured by a single computer, or any part such as an input device, an output device, a processing device, and a storage device may be a network. It may be composed of other computers connected by. In this embodiment, the same function as the function configured by the program can be realized by hardware such as FPGA (Field Programmable Gate Array) and ASIC (Application Specific Integrated Circuit).

FIG. 2 shows a conceptual example of the operation log (115), which is the operation log of the model development program (107) described above. Each line is a work log, and the time (201), worker ID (202), operation target (203) indicating which analysis model was operated, operation content (204) indicating what kind of operation was performed, Reference data (205) indicating the referenced data name is recorded.

For example, in the second line of the log, at 10:36:21 on September 28, 2017, the worker of IDU1357 targeted the analysis model Pred_Test.xml and set the data period module (the module is a partial routine of the analysis model). (Pointing) was edited, and it is recorded that "Operation_data_1" was referenced at that time.

Also, in the third line of the log, at 11:12:32 on September 28, 2017, the worker of IDU4321 edited the learning module for the analysis model Pred_Test.xml, and at that time, "factor data". It is recorded that the reference was made.

Here, some of the reference data is passed as an argument in order from the module in the previous stage. An example is sensor data (113), which is the main target of processing. In that case, it is sufficient to determine the name by performing operation log shaping processing that goes back upstream of the processing and obtains the name from the data input module, or to know that the data is passed as an argument. In that case, only "argument data" is extracted. The example on the third line of FIG. 2 is the latter.

Further, the recording timing of this operation log depends on the specifications of the model development program, but in this embodiment, it is assumed that recording is performed for each change operation. This specification can be implemented depending on the settings of the model development program.

FIG. 3 shows a conceptual example of the model difference data (116) output by the version control program (108) described above. Each row is model difference data indicating a difference event in which a difference occurs in the model. The time (301), the worker ID (302), the target file (303), and the difference information (304) are recorded.

For example, in the second line, as a result of calculating the difference between the analysis model and the previous time at the date and time 2017/09/28 10:37:25, the data period parameter "begin_0" is applied to the analysis model called Pred_Test.xml by the worker IDU1357. It is recorded that the save time stamp has been changed. In this way, the difference event whose parameters are changed will be referred to as a "parameter difference event".

It should be noted that the recording timing of this model difference data is basically recorded at the timing when the operator himself explicitly performs the operation manually. However, in order to eliminate omission of records, for example, when a specific operation on the analysis model (for example, test execution of the analysis model) is performed, a subroutine that operates a version control program is called so that the analysis model is called. By describing it in itself, it is possible to record the difference of itself in conjunction with a specific operation. In this embodiment, it is assumed that the difference information is recorded each time the test operation is performed by such a setting.

FIG. 4 shows a conceptual example of the related rule list (117). As described above, this list is a list of whether the module (partial routine of the analysis model) was operated for each combination of keywords and key phrases on the operation log (115) side and the difference data (116) side. It is created by the user (administrator) depending on the processing program and version control program. The operation log and the difference data can be associated with a predetermined module by using the related rule list.

The related rule list (117) includes the module to be operated (401), the operation log key phrase (402), the difference data key phrase (403), the reference data description (404), the parameter description (405), and the search width (406). ) Data is recorded.

For example, taking the first line as an example, if the key phrases "data period setting" on the operation log side and "Data Period" on the difference data side appear as the history of operating the module for "data period", that key. Make the phrase a table. In addition, the data name referred to at that time appears with the key phrase "Read Data From", and the parameters for the period of the data to be modified are the key "begin_0" for the start period and "end_0" for the end period. Indicates that it appears with the phrase. Further, the "search width" in the last column defines which range is associated before and after the parameter difference occurs. In the example of the first line, it is stipulated that the operation log up to 10 minutes before the difference occurs is associated, and the future direction is not associated.

The description in the reference data description (operation log) on the second line is "(argument data)", which indicates that the data is passed as an argument as described above.

The related rule list (117) is generally created by an operator or an administrator based on an empirical rule and stored in a storage device in advance. Alternatively, it may be automatically generated by machine learning. The data structure of FIG. 4 is an example, and the number of items may be added or reduced. In addition, the internal use of each item may be appropriately selected from the key phrase, data type, numerical range, and the like. Regarding the search width (406), the condition is set for each row in this example, but it may be set to a fixed value common to all rows, or as a rule to select the most recent one in terms of time. Good.

Next, the overall operation in this embodiment will be described. The operation of this embodiment is composed of two steps: a recording operation of automatically classifying and recording model parameters, and a procedure of presenting the classification data recorded when the model is reused. Each can be executed independently according to the usage timing of the user. Hereinafter, these two procedures will be described in order.

FIG. 5 shows the overall procedure of the recording operation in this embodiment. First, the operation log (115) is read by S501, and the model difference data (116) is read by S502. Next, using those data, the difference / operation log is associated (S503). The details of the process will be described later.

Next, parameter classification is performed for each parameter difference event extracted by this association (S504) (S505). When all the parameter difference events are completed (S506), the parameter classification is generally performed a plurality of times, so the latest classification in terms of time is adopted for each parameter (S507). In this embodiment, the latest classification is adopted, but the classification that occurs most frequently may be adopted depending on the convenience at the time of reuse.

FIG. 6 shows the procedure of the above-mentioned difference / operation log association S503. First, the related rule list (117) described above is read (S601). For each difference event in the difference data (116) (S602), it is determined whether there is a parameter difference (S603). If there is a parameter difference, it is determined by using the related rule list (117) whether or not there is a related log on the operation log (115) side within the range of the search width illustrated in FIG. 4 (S604). If it is related (S605), it is determined as a corresponding candidate (S606). Process all logs in the search width range (S607) and do this for all delta events (S608).

FIG. 7 shows a conceptual diagram of the composite data (700) associated in the association S503 between the operation log (115) and the difference data (116) described above performed by the history shaping program (109). The set to which each set # (701) is associated is shown. As shown in the column of origin (703), the row of "log" is the data derived from the operation log, and the row of "difference" is the data derived from the difference data. In the time (702), the time (201) (301) of the operation log and the difference data is stored. By this association, the operation content (706), the reference data (707), and the difference parameter (708) are associated with each other. In the example of FIG. 7, since the difference event of the set # 001 and 010 has no difference parameter, it is not included in the correspondence candidate of the result of the correspondence S503. The difference event of sets # 002 to 004 is a parameter difference event because there is a difference parameter.

The worker ID (704) is a worker ID (202) in the operation log (115) and a worker ID (302) in the difference data (116). The target (705) is the operation target (203) of the operation log (115) and the target file (303) of the difference data (116). These can be matched if the search width (406) is chosen appropriately.

In this example, it is described that the operation log side and the difference data side correspond to each other, but depending on the interval of the difference log, it is possible that a plurality of operations are combined and the difference is taken. In that case, it is difficult to know which parameter change is due to which operation, which is not desirable. Therefore, it is desirable to create the difference data as frequently as possible. One example is the procedure that is automatically created each time the test is executed.

FIG. 8 shows a procedure of parameter classification (S505) for one parameter difference event using the above-mentioned difference data and operation log association result performed by the work classification program (110). Here, the parameter classification is a classification from the viewpoint of what kind of work the parameter was determined or changed.

First, for each parameter difference event, it is determined whether or not there is associated reference data (S801). If there is no reference data, it is determined that the data was obtained from some external information, and a "other decision" flag is added (S802). If there is reference data, if it is argument data (S803), a data learning flag is added (S804). The flag may be given to the parameter itself, or may be indirectly given to the parameter by giving it to the operation content or module corresponding to the parameter.

Here, even if the argument data is referenced, the parameters may be changed manually depending on the target module. For example, with respect to a module that reads sensor data (113) and selects data, a graph is drawn and the data manually selected by the user is changed based on the result. In this embodiment, if this can also be determined from the data (it is not necessary to hear external factors), the data learning flag is added. In another embodiment, if the policy is to distinguish it from data learning when human judgment intervenes, list the machine learning module and whether it was used or not (recorded in the operation log). ) May add a procedure to flag human judgment.

Next, it is checked whether or not the business data (114) is referred to (S805), and if it is referred to, the business data reference flag is added (S806). In addition, the content (data name) of the reference business information at that time is recorded (S807).

Here, the business data (114) is the sensor data (113) and all the data that is not the processed argument data. That is, when referring to data, it means that the argument data or business data is always referred to. In addition, as shown in this figure, the data learning flag and the business data determination flag are not exclusive, so regarding the parameter difference event, (1) other determination, (2) data learning, no business data reference, (3) data There are four cases: no learning, with business data reference, (4) with data learning, and with business data reference.

As described above, by automatically classifying the process of determining the parameters of the analysis model, the information can be utilized when the model is reused. Therefore, among the parameters of the analysis model, even if there are parameters that need to be changed according to the customer business situation and the customer purpose, they can be managed separately, not only determined by the sensor data reference.

FIG. 9 shows a procedure for presenting classification data recorded at the time of model reuse, which is another procedure in this embodiment, to support development.

First, the user specifies and inputs the model to be reused (S901). Next, the parameter classification data corresponding to the model, which was classified by the procedure described above during the previous development, is read (S902). For each parameter in the model (S903), if it is another determination parameter (S904), the corresponding confirmation item is presented (S905). Here, the confirmation items corresponding to the other determination parameters are for confirming that the parameter item needs to be determined from the outside and that the value of the parameter is prepared in advance. It is a hypothetical question sentence.

On the other hand, if the parameter is a data learning parameter (S906), learning-related information is presented (S907), and if it is a business data reference parameter (S908), the used business information is presented (S909). Here, the learning-related information includes the item name of the determined parameter, the name of the learning program (if any), the name of the usage data (if tracked on the processing model), and the like. Further, the used business information is information such as the name of the referenced business data (114).

FIG. 10 shows an example of GUI displayed to the user. This figure shows the processing contents of one analysis model "Abnormal diagnosis of power plant XX equipment". Each square (1001 to 1008) connected by an arrow is a processing module, argument data is passed in the order of the arrows, and processing is performed. Further, the information (1009 to 1011) displayed by the balloon in the figure is an example of the parameter classification and the additional information.

In the information (1009), it is shown that the parameter (set) name is the data reference period, the type is the business data reference parameter, and the reference information is the normal period data for the data selection module. It should be noted that the explanation is not for each parameter set as in this example, but for each individual parameter such as the start period and the end period, it may be displayed in detail. Similarly, in the information (1010), information on the determination cluster parameters (set) is presented with respect to the abnormality diagnosis learning module. Further, in the information (1011), information on the abnormality determination threshold value is presented with respect to the abnormality determination module. Since this parameter is the other decision parameter mentioned above, the comment "The value needs to be confirmed" is generated.

In the situation where the information (1011) is displayed, there is a changed parameter in the difference information item (304) in the predetermined line of the model difference data (116), and the operation log (115) associated with the parameter is displayed. This is the case when there is no reference data in the reference data item (205) of the row. In this case, in the parameter classification (S505), the "other determination" flag is added in association with the changed parameter "abnormality determination threshold value". Therefore, the information (1011) is displayed corresponding to the abnormality determination module (1007) associated with the parameter “abnormality determination threshold value”. As a result, the operator can recognize that he / she needs to perform the parameter setting work.

Note that FIG. 10 shows an example in which a plurality of parameters (sets) that have been modified in the past are presented all at once. As another embodiment, the order of parameter correction at the time of past recording is recorded, and the current parameter correction status is also extracted and sequential during the work as in the past recording, and their realization orders are compared. Then, when a certain parameter is determined, information on the next parameter, which has not been determined yet, may be presented in order. In this way, the presentation based on the order makes it possible to guide the order in which many parameters should be determined.

Further, in this embodiment, the form of presenting information on the flow diagram of the module is shown, but if the necessary information is presented, the drawing design is not limited to this drawing, and the drawing design is not limited to this drawing, and the table format and the user. You may use various GUIs such as a format that is displayed hierarchically by operation. In addition, the display contents can be displayed in detail for each parameter instead of being grouped together in a parameter set.

In this embodiment, in addition to the functions of the first embodiment, an embodiment in which the user's analysis model is classified and presented including the editing purpose is shown.

FIG. 11 shows the device configuration in this embodiment. The analysis model construction server (1100) prepares worker data (1101) in addition to the configuration of the first embodiment (FIG. 1). Further, the work classification program (1102) is added with a process of adding a work purpose or the like by referring to the worker data as described later. Along with this, the classification result presentation program (1103) also additionally displays the work purpose and the like.

FIG. 12 shows a conceptual example of worker data (1101). In the worker data, each line is information about the worker (operator), and the worker ID (1201), the role of the worker (1202), the operation purpose of the analysis model (1203), and the skill of the worker (1204). ) Etc. are recorded. Worker data is created and recorded by, for example, a user (administrator) who has the authority to supervise the worker.

The worker data (1101) and the operation log (115) or the difference data (116) can be associated with the worker ID. Therefore, it is possible to associate the target module with the worker.

FIG. 13 shows a procedure for recording the operation history in this embodiment. In addition to the procedure of the first embodiment (FIG. 5), the process of reading the worker data (1101) described above (1301) and adding the role of the worker of the event, the operation purpose, etc. to each parameter difference event (1302). ), And in connection with the process of finally selecting the latest classification flag for each parameter, a process of selecting the corresponding worker role, operation purpose, etc. has been added (1303).

In the process (1302), for example, with respect to the data of the parameter difference event shown in FIG. 7, the role (1201), the operation purpose (1203), and the skill (1204) of the worker data (1101) having the corresponding worker ID. To combine.

FIG. 14 shows an example of GUI in this embodiment. In addition to the display contents of the first embodiment, information about the purpose of correction and the role of the person in charge is additionally displayed (underlined portion 1401 to 1403). Note that this GUI is just an example, and it is possible to perform a display method such as listing by purpose.

In the above embodiment, the operation log of the model construction application and the difference information of the analysis model recorded by the version control program are used to provide a means for associating them. In addition, the reference data classification means is used to classify whether or not the data was referred to when modifying each parameter, and if so, the type of reference data. As a result of the above, a function is provided for automatically assigning three types of flags, data learning, business data reference, and other determination, to each parameter. In addition, we were able to provide a function to generate and present confirmation items according to the work stage when reusing the classification results as a model.

According to the above-described embodiment, the analytical model developer classifies the types of model parameters without keeping work records, and by what means adjusts each parameter when reusing the model. I know what to do. As a result, there are effects such as reduction of the amount of work and easy adjustment of the model even by an unskilled person.

The present invention is not limited to the above-described embodiment, and includes various modifications. For example, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is possible to add the configuration of another embodiment to the configuration of one embodiment. Further, it is possible to add / delete / replace the configurations of other examples with respect to a part of the configurations of each embodiment.

Model development program 107, version management program 108, history shaping program 109, work classification program 110, classification result presentation program 111, model design data 112, sensor data 113, business data 114, operation log 115, model difference data 116, related rules Listing 117, work classification data 118

Claims (15)

An analytical model support device that supports the work of constructing an analytical model, including an input device, an output device, a storage device, and a processing unit.
The storage device is
The operation history of the development program of the analysis model and the difference information of the version control program of the analysis model are stored.
The processing unit
Execute the history shaping program, work classification program, and classification result presentation program,
The history shaping program
Corresponding the operation history with the difference information,
The work classification program
Based on the data reference relationship appearing in the operation history, the classification the parameters of the analytical model appearing in the difference information, the parameter is classified in terms of types of presence and the reference data of the reference data at the time of the decision or change Get results,
The classification result presentation program is
Outputs information based on the classification result to the output device.
An analysis model construction support device characterized by this. In the analysis model construction support device according to claim 1,
The storage device is
Store the related rule in which the combination pattern of the element in the operation history and the element in the difference information is set in advance.
The history shaping program
An analysis model construction support device characterized in that the operation history and the difference information are associated with each other by the pattern. In the analysis model construction support device according to claim 2.
The operation history is
Including the first time information
The difference information is
Includes second time information
The related rule is
A key phrase is used as the element, and the first time information and the search width information, which is a constraint condition to be satisfied by the second time information, are included.
The history shaping program
An analysis model construction support device characterized in that the operation history and the difference information are associated with each other based on the search width and the key phrase with reference to the related rule. In the analysis model construction support device according to claim 2.
The related rule is
The operation history and the difference information are further associated with the module of the analysis model.
The work classification program
An analysis model construction support device, characterized in that the parameters of an analysis model are classified according to whether or not the module whose parameters have been changed in the difference information refers to the data at the time of the change in the operation history. In the analysis model construction support device according to claim 4.
The work classification program
When the module whose parameter has been changed in the difference information refers to the data at the time of the change in the operation history, whether the data is the data input as the processing target of the analysis model or other external data . An analytical model construction support device characterized by classifying the parameters of an analytical model. In the analysis model construction support device according to claim 5.
The classification result presentation program is
An analysis model construction support device characterized in that information on the classification result of the parameter is presented for each module corresponding to the difference information in which the parameter appears. In the analysis model construction support device according to claim 6.
The classification result presentation program is
An analysis model construction support device characterized in that when presenting each module, it is clearly presented that it is necessary to determine by external information for parameters for which the corresponding module does not refer to data. In the analysis model construction support device according to claim 6.
The classification result presentation program is
An analysis model construction support device characterized by presenting the name of the data used when the data referred to by the relevant module is the external data when presenting each module. In the analysis model construction support device according to claim 6.
The classification result presentation program is
An analysis model construction support device characterized in that when presenting each module, corrections to be made next to the current work content are presented in order as compared with the current work content of the development program. In the analysis model construction support device according to claim 6.
At least one of the operation history and the difference information includes a related worker ID.
The storage device is
It has worker data corresponding to the worker ID and has worker data.
The processing unit
Run the work classification program,
The work classification program
By associating the module with the worker data, the purpose is estimated from the worker data for each change of the parameter appearing in the difference information.
The classification result presentation program is
An analysis model construction support device characterized by presenting the purpose associated with the module when presenting each module. Using an information processing device including an input device, an output device, a storage device, and a processing unit, the analysis model is based on the operation log of the model development program of the analysis model and the model difference data between the versions of the analysis model. It is an analysis model construction support method for classifying parameters.
The operation log contains multiple lines
Each of the plurality of lines of the operation log includes an item of the operation content for the analysis model and an item of reference data which is the data referred to in the operation as items.
The model difference data includes a plurality of rows and contains a plurality of rows.
Each of the plurality of rows of the model difference data includes an item of the difference information of the analysis model as an item.
A mapping step that associates the operation log line and the model difference data line with a combination pattern of the operation log item content and the model difference data item content based on a preset related rule. ,
The parameter is the content of the items of differential information of a predetermined row of the model difference data, based on the contents of the items of a data line of the corresponding operation log, to help in the parameter is determine or change Classification steps and classification steps that classify in terms of the presence or absence of data and the type of reference data,
An analytical model construction support method characterized by executing.
In the analysis model construction support method according to claim 11,
Each of the plurality of lines of the operation log further contains a first time information.
Each of the plurality of rows of the model difference data further contains a second time information.
The relevant rule further defines the relevance of the first time information and the second time information.
The association step further associates a row of operation logs with a row of model difference data, provided that the association is satisfied.
Analytical model construction support method. In the analysis model construction support method according to claim 11,
The classification step
The parameters are classified based on whether the content of the reference data item is the input data to be analyzed by the analysis model, the argument data processed by the input data, or the other business data.
Analytical model construction support method. In the analysis model construction support method according to claim 11,
With additional classification result presentation steps
The classification step
If there is a changed parameter in the difference information item of the specified line of the model difference data, and there is no reference data in the reference data item of the operation log line associated with it, the changed parameter is supported. Attach and flag
The classification result presentation step is
A predetermined output is output to the output device in association with the parameter.
Analytical model construction support method. In the analysis model construction support method according to claim 11,
With additional classification result presentation steps
The mapping step
The row of the operation log and the row of the model difference data are combined with a predetermined module of the analysis model based on a related rule in which a combination pattern of the contents of the operation log item and the contents of the model difference data item is set in advance. Correspond and
In the classification result presentation step,
Information based on the classification result of the parameters by the classification step is presented to the output device for each of the predetermined modules.
Analytical model construction support method.

Images