JP2021056918A - Data analysis device and data analysis method - Google Patents

Abstract

To achieve improvement in prediction accuracy of a work time.SOLUTION: A data analysis device is configured to implement: acquisition processing that acquires, for each work of a plurality of works, first learning data having an objective variable indicative of a work time, an explanation variable regarding a work environment, a worker variable indicative or presence or absence of a work implementation for each worker; prediction model generation processing that generates a first prediction model predicting the work time on the basis of the first learning data; calculation processing that inputs the first learning data to the first prediction model, and thereby calculates, for each work, a first prediction value of the work time and a first implication degree indicative of a degree in which the worker variable affects statistics of the first prediction value for each work; and learning data generation processing that generates an abstraction variable abstracting a work speed of a plurality of workers for each work on the basis of a distribution of respective first implication degrees of the plurality of works for each worker, and generates second learning data having the objective variable, explanation variable, and abstraction variable for each work.SELECTED DRAWING: Figure 5

Description

The present invention relates to a data analyzer and a data analysis method for analyzing data.

In recent years, the use of big data analysis has become widespread in various fields. In particular, in order to improve the efficiency of work performed by humans, such as product picking work in distribution warehouses and assembly work in factories, optimal product placement and work order are selected based on a large amount of past actual data. The service proposed as a business improvement measure is attracting attention. As one means of realizing such a service, it is possible to analyze the relationship between various conditions and working time in the working environment and work content from a large amount of past actual data, and predict the working time under arbitrary working conditions. A method of generating a work time prediction model and searching for the condition with the shortest work time from various work conditions using the prediction model can be mentioned.

For example, in Patent Document 1, a plurality of similar actual data are selected from the actual data stored in the actual data storage means for the specified assembly work time prediction product, and among the obtained actual data. , By performing multiple regression analysis with the type of parts of similar products as the explanatory variable and the assembly work time as the objective variable, an assembly work time prediction model is created, and the effectiveness of the prepared assembly work time prediction model is statistically statistic. Disclosed is an assembly work time prediction device that predicts the assembly work time from the assembly work time model when it is valid, and predicts the assembly work time by a conventional method when it is not valid.

Japanese Unexamined Patent Publication No. 07-164267

However, the work time for picking work in a distribution warehouse and assembly work in a factory largely depends on individuality, such as the skills of the worker himself and the existence of work conditions that the worker is not good at. .. Therefore, the model generated in Patent Document 1 does not reflect the individuality of the worker, and the accuracy of predicting the working time is limited. An object of the present invention is to improve the accuracy of predicting working time.

The data analyzer and the data analysis method which are one aspect of the invention disclosed in the present application are data analyzers including a processor for executing a program and a storage device for storing the program, and the plurality of processors are included. Acquisition process for acquiring the first learning data having an objective variable indicating the work time, an explanatory variable for the work environment, and a worker variable indicating whether or not the work is performed for each worker. And the prediction model generation process that generates the first prediction model that predicts the working time based on the first training data acquired by the acquisition process, and the first prediction model generated by the prediction model generation process. By inputting the first training data, the first predicted value of the working time and the first influence degree indicating the degree of influence of the worker variable on the statistic of the first predicted value for each work. Based on the calculation process calculated for each work and the distribution of the first influence degree of each of the plurality of works for each worker, an abstract variable that abstracts the work speed of the plurality of workers is obtained. It is characterized in that a training data generation process for generating a second training data having the objective variable, the explanatory variable, and the abstraction variable is executed for each work. To do.

According to a typical embodiment of the present invention, it is possible to improve the accuracy of predicting the working time. Issues, configurations and effects other than those described above will be clarified by the description of the following examples.

FIG. 1 is a block diagram showing a system configuration example of a data analysis system. FIG. 2 is a block diagram showing a hardware configuration example of the data analyzer. FIG. 3 is an explanatory diagram showing an example of business performance data. FIG. 4 is an explanatory diagram showing an example of product information. FIG. 5 is a flowchart showing a detailed processing procedure example of the prediction model generation processing shown in FIG. FIG. 6 is an explanatory diagram showing an example of worker ID use learning data. FIG. 7 is an explanatory diagram showing an example of worker attribute use learning data. FIG. 8 is an explanatory diagram showing an example of intermediate learning data. FIG. 9 is an explanatory diagram showing an example of the first influence degree table. FIG. 10 is an explanatory diagram showing influence degree distribution information in all operations. FIG. 11 is an explanatory diagram showing an example of learning data using abstract variables. FIG. 12 is an explanatory diagram showing an example of the second influence degree table. FIG. 13 is a graph showing an example of the influence distribution of the explanatory variables. FIG. 14 is an explanatory diagram showing an example of worker feature use learning data. FIG. 15 is a flowchart showing an example of a business improvement measure generation processing procedure for generating a business improvement measure using a worker feature usage prediction model generated by the prediction model generation processing. FIG. 16 is a flowchart showing an example of a processing procedure for properly using various prediction models in step S1504. FIG. 17 is an explanatory diagram showing an example of worker attribute / feature data.

Hereinafter, the data analysis system will be described with reference to the attached drawings. In the following description, the case where the target work is the product picking work in the distribution warehouse and the prediction target of the prediction model is the work time required for the picking work will be taken up. It should be noted that this embodiment can be applied to all services that generate a prediction model from work performance data and generate business improvement measures using the prediction model, and is not limited to the use cases to be described.

<System configuration example of data analysis system>
FIG. 1 is a block diagram showing a system configuration example of a data analysis system. The data analysis system 100 includes a business system 101 and a data analysis device 102. The business system 101 and the data analysis device 102 are communicably connected to each other via a network such as Local Area Network (; LAN), Wide Area Network (WAN), and the Internet.

The business system 101 has one or more calculators, and stores the execution and execution results of the business plan input from the outside as business performance data in the business performance DB 110. The data analysis device 102 executes the model generation process 121 and the business improvement measure generation process 122.

The model generation process 121 is a process of acquiring business performance data including business contents and business results from the business system 101 from the business performance DB 110 and generating a prediction model. The business improvement measure generation process 122 is a process of generating a business improvement measure for realizing improvement of the business evaluation index by giving new business performance data to the prediction model generated by the model generation process 121.

The business evaluation index is an index for evaluating business such as Key Performance Indicator (KPI). Business performance data includes a plurality of attributes that are parameters related to business. Attributes include characteristics such as timestamp, gender, amount, and working hours.

Based on the business improvement measures received from the data analysis device 102, the business system 101 changes the business plan to be implemented next, executes the business again, and accumulates the business performance data. The data analysis system 100 continuously executes the cycle of business execution and business improvement described above.

<Hardware configuration example of data analyzer 102>
FIG. 2 is a block diagram showing a hardware configuration example of the data analysis device 102. The data analyzer 102 includes a processor 201, a storage device 202, an input device 203, an output device 204, and a communication interface (communication IF) 205. The processor 201, the storage device 202, the input device 203, the output device 204, and the communication IF 205 are connected by the bus 206. The processor 201 controls the data analyzer 102. The storage device 202 serves as a work area for the processor 201. Further, the storage device 202 is a non-temporary or temporary recording medium that stores various programs and data that execute the model generation process 121 and the business improvement measure generation process 122. Examples of the storage device 202 include a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), and a flash memory. The input device 203 inputs data. The input device 203 includes, for example, a keyboard, a mouse, a touch panel, a numeric keypad, and a scanner. The output device 204 outputs data. The output device 204 includes, for example, a display and a printer. The communication IF205 connects to the network and transmits / receives data.

Specifically, the storage device 202 includes, for example, business performance data 221 and product information 222, worker ID usage learning data 223, worker attribute usage learning data 224, intermediate learning data 225, first influence degree table 226, and abstraction. Abstract variable usage learning data 227, second influence degree table 228, worker feature usage learning data 229, worker attribute / feature data 230, worker ID usage prediction model 231, worker attribute usage prediction model 232, intermediate prediction model 233 , Abstract variable usage prediction model 234, worker feature usage prediction model 235 are stored.

The business record data 221 is data extracted from the business record DB 110 and received by the data analyzer 102. The business performance data 221 is used to generate a prediction model by the model generation process 121, or is used to generate a business improvement measure by the business improvement measure generation process 122. Details of the business performance data 221 will be described later with reference to FIG.

The product information 222 stores information indicating the attributes of the product. Details of the product information will be described later in FIG.

The worker ID use learning data 223 is learning data indicating a work in which the objective variable is defined as the working time, the explanatory variable is the moving distance of the worker, the quantity, weight, volume, etc. of the picked article, and is used as the explanatory variable. Further, it is learning data to which a worker ID variable is added. The worker ID variable is a variable indicating whether or not the worker specified by the worker ID has performed the target work. The details of the worker ID use learning data will be described later with reference to FIG.

The worker attribute use learning data 224 is learning data indicating a work in which the objective variable is defined as the working time, the explanatory variable is the moving distance of the worker, the quantity, weight, volume, etc. of the picked article, and is used as the explanatory variable. It is learning data to which worker attribute variables are added. The worker attribute variable is a variable indicating the attribute of the worker such as the number of days of experience of the worker. The worker attribute use learning data does not include the worker ID variable. The details of the worker attribute use learning data will be described later in FIG.

The intermediate learning data 225 is intermediate learning data for generating the learning data 227 using abstract variables. Details of the intermediate learning data will be described later with reference to FIG.

The first influence degree table 226 is a table in which each influence degree of the explanatory variable, the worker attribute variable, and the worker ID variable is defined for each work. Details of the first impact table 226 will be described later in FIGS. 9 and 10.

The abstract variable use learning data 227 is learning data indicating a work in which the objective variable is defined as the working time, the explanatory variable is the moving distance of the worker, the quantity, weight, volume, etc. of the picked article, and is used as the explanatory variable. Furthermore, it is the training data to which the worker attribute variable and the worker ID abstraction variable are added. That is, it is the learning data in which the worker ID abstraction variable is added to the worker attribute use learning data.

The worker ID abstract variable is a variable that is abstracted by expressions such as "work is slow" or "work is fast" without specifying the worker specified by the worker ID by the worker ID. Details of the learning data using abstract variables will be described later in FIG.

The second influence degree table 228 is a data table output as a result of inputting the business performance data 221 to be improved into the prediction model generated by the model generation process 121 using the abstract variable use learning data 227. Details of the second impact table 228 will be described later in FIGS. 12 and 13.

The worker feature use learning data 229 is learning data in which a worker feature variable is added to the abstraction variable use learning data 227. The worker characteristic variable is a variable that indicates the characteristics of the worker, for example, "I am good at the upper shelf". Details of the worker feature use learning data 229 will be described later with reference to FIG.

The worker attribute / feature data 230 is data in which the worker attributes and worker characteristics are listed for the worker ID of each worker. Details of the worker attribute / feature data 230 will be described later with reference to FIG.

The worker ID use prediction model 231 is a prediction model for predicting the work time, which is generated by the model generation process 121 using the worker ID use learning data 223 (FIG. 6).

The worker attribute use prediction model 232 is a prediction model for predicting the work time, which is generated by the model generation process 121 using the worker attribute use learning data 224 (FIG. 7).

The intermediate prediction model 233 is an intermediate prediction model for predicting the working time, which is generated by the model generation process 121 using the intermediate learning data 225 (FIG. 8).

The abstract variable use prediction model 234 is a prediction model for predicting the working time, which is generated by the model generation process 121 using the abstract variable use learning data 227 (FIG. 11).

The worker feature use prediction model 235 is a prediction model for predicting the work time, which is generated by the model generation process 121 using the worker feature use learning data 229.

<Business performance data 221>
FIG. 3 is an explanatory diagram showing an example of business performance data 221. The work record data 221 has a work ID 301, a branch number 302, a worker ID 303, a product code 304, a place code 305, a number 306, a work start time 307, and a work end time 308. The worker ID 303, the product code 304, the location code 305, and the number 306 indicate a work environment such as the arrangement of products, the products to be handled, and the composition of workers.

The work ID 301 is identification information that uniquely identifies the work, and indicates one unit of the work work that the worker picks. For example, a series of picking operations for products of the same shipping destination are applicable.

The branch number 302 is a number obtained by subdividing the work specified by the work ID 301 by the worker ID or the product code. Hereinafter, the work ID 301 is represented by x (x is a number) or xy (y is a number of the branch number 302). When the work ID 301 is represented by x only, the entry including all the branch numbers 302 of the work ID 301 is targeted, and when the work ID 301 is represented by xy, the branch number 302: y of the work ID 301: x Entry is targeted.

The worker ID 303 is identification information that uniquely identifies the worker. The product code 304 is identification information that uniquely identifies the product picked by the worker specified by the worker ID 303. The place code 305 is identification information indicating a place where the product specified by the product code 304 is stored. The place code 305 is represented by, for example, a row-b series-c stage (a, b, c are character strings), and the c stage (top, middle, bottom, etc.) of the b series (depth) of the shelf in row a. It means the height of the shelf).

The number 306 is the number of picked products specified by the product code 304. The work start time 307 is a date and time when the work specified by the work ID 301: xy is started. The work end time 308 is a date and time when the work specified by the work ID 301: xy is completed.

<Product information>
FIG. 4 is an explanatory diagram showing an example of product information. The product information 222 has a product code 304, a weight 401 per piece, and a volume 402 per piece. The weight 401 per product is the weight per product specified by the product code 304. The volume 402 per product is the volume per product specified by the product code 304.

<Model generation process 121>
FIG. 5 is a flowchart showing a detailed processing procedure example of the model generation processing 121 shown in FIG. The model generation process 121 is a process of processing and generating learning data from the business performance data 221 to generate a prediction model for predicting the work time.

The data analysis device 102 activates the model generation process 121 (step S501), and acquires the business performance data 221 for the target period of the learning data when the prediction model is generated from the storage device 202 (step S502). Next, the data analysis device 102 processes the work performance data 221 to generate the worker ID use learning data 223 (step S503). As a result, the work performance data 221 is converted into the worker ID use learning data 223. An example of the worker ID use learning data 223 is shown in FIG.

[Worker ID usage learning data 223]
FIG. 6 is an explanatory diagram showing an example of the worker ID use learning data 223. The worker ID use learning data 223 has a work ID 301, an objective variable 601, an explanatory variable 602, and a worker ID variable 603. The objective variable 601 is a value to be predicted, and corresponds to the working time in this example. The work time is the time obtained by subtracting the work start time 307 from the work end time 308 of the entry of the same work ID 301. When a plurality of entries exist in the work ID 301 according to the branch number 302, the total work time of each entry is the work time 601.

The explanatory variable 602 is a set of variables generated by aggregating various values in the business performance data 221 and indicates the feature amount of the entry of the work ID 301. Specifically, for example, the explanatory variable 602 has a movement distance 621, a quantity 622, a weight 623, a volume 624, a shelf upper tier 625, a shelf middle tier 626, and a shelf lower tier 627. The movement distance 621 is the distance traveled by the worker who performed the work specified by the work ID 301 in the work. The data analysis device 102 has map information in the warehouse including the location information indicated by the location code 305, and can calculate the road distance between the picked locations.

The quantity 622 is the number or quantity of goods picked in the work specified by the work ID 301. The weight 623 is a weight of 622 minutes of the quantity of goods picked in the work specified by the work ID 301. The volume 624 is the size of the quantity of goods picked by the work specified by the work ID 301, which is 622 minutes. The upper shelf 625 is the quantity of products picked from the upper shelf. The middle shelf 626 is the quantity of products picked from the middle shelf. The lower shelf 627 is the quantity of products picked from the lower shelf. In the entry of the work ID 301, the total of the values of the upper shelf 625 to the lower shelf 627 is the quantity 622.

The data analysis device 102 can communicate with an inventory management system (not shown), and the inventory management system manages the weight and volume of each product and the storage location code 305. Therefore, the data analysis device 102 can acquire the quantity 622 to the lower shelf 627 for each product by accessing the inventory management system.

The worker ID variable 603 is a variable indicating which of the workers specified by the worker ID 303 (ID1, ID2, ... In FIG. 6) has performed the work specified by the work ID 301. “ID #” (where # is a number) indicates worker ID 303. For example, if the worker ID variable 603 is "1", it indicates that the work specified by the work ID 301 has been performed by the worker specified by the worker ID 303. If the worker ID variable 603 is "0", it indicates that the worker specified by the worker ID 303 is not performing the work specified by the work ID 301. Here, the description of FIG. 6 ends.

Returning to FIG. 5, the data analysis device 102 generates the worker ID use prediction model 231 using the worker ID use learning data 223 (step S504). This process is a process of generating the worker ID usage prediction model 231 as a regression model for predicting the working time, which is the objective variable 601 by machine learning, and the method itself of the machine learning algorithm does not matter.

In this machine learning algorithm method, for example, the explanatory variable 602 and the worker ID variable 603 are multiplied by an appropriate coefficient to obtain the sum of them as the predicted value, and the square error between the objective variable 601 of the training data and the predicted value is minimized. It may be a method of generating a linear regression equation for obtaining each coefficient such that. Further, the method of the machine learning algorithm may be a method of generating a prediction model by a non-linear expression such as DNN (Deep Neural Network). In this way, by generating the worker ID usage prediction model 231, for the work of the worker who has a sample of the work results sufficient for the learning data, the predicted value of the work time corresponding to the values of various explanatory variables 602. High accuracy can be achieved.

Next, the data analysis device 102 processes the work performance data 221 to generate the worker attribute use learning data 224 (step S505). FIG. 7 shows an example of the worker attribute use learning data 224.

[Worker attribute use learning data 224]
FIG. 7 is an explanatory diagram showing an example of worker attribute use learning data 224. The worker attribute use learning data 224 is learning data using the worker attribute variable 700. Specifically, for example, the worker attribute use learning data 224 has a work ID 301, an objective variable 601, an explanatory variable 602, and a worker attribute variable 700. The worker attribute variable 700 is a variable indicating the attributes of the worker.

The worker attribute variable 700 may use the information stored in the work performance data 221 as the worker attribute variable 700, or may be a newly created variable. In the former case, the age, gender, height, and worker level of the worker are examples. As for the latter, examples include 701 days of experience as shown in FIG. 7 and 702 days of experience of nearby workers.

The number of days of experience 701 is a worker attribute variable 700 indicating how many days have passed since the worker engaged in the work to be analyzed. For example, the target entry is obtained from the work performance data 221 over the entire past period. It is obtained by calculating the number of days from the base point to the work date and time of the corresponding entry, using the oldest work date and time in which the worker ID 303 exists as the base point date and time.

Further, the initial value of the number of experience days 701 may be given to each worker ID 303 from the outside, and the number of days of experience in the target entry may be calculated using the initial value as the base point date and time. As a result, it is possible to accurately reflect the number of days of experience 701 of a worker who has been engaged in the work before collecting the work performance data 221.

The number of days of experience 701 is not limited to the number of days, and may be expressed in minutes, hours, or months. Further, the type of the variable is not limited to the continuous value, and a classification value (for example, less than 1 month of experience, 6 months or more of experience, etc.) may be used. As a result, it is possible to make a variable that reflects the experience value that is more specialized in the characteristics of the business.

The number of experience days 702 of a nearby worker is a worker attribute variable that reflects the number of experience days 701 of a nearby worker (for example, within a predetermined range such as within a radius of 10 meters) when performing the work of the corresponding entry. It is 700. Specifically, for example, the data analyzer 102 refers to the location code 305, the work start time 307, and the work end time 308 of the business performance data 221, and the products to be picked exist in close locations and the work time zones overlap. Calculate the number of days of experience 701 of the worker in charge of another work. The result of this calculation is the number of days of experience of nearby workers 702.

When there are a plurality of the other workers, the data analyzer 102 calculates statistical values such as the average value, the median value, the maximum value, and the minimum value of the number of days of experience 701 of each different worker in the vicinity. The number of days of experience of the worker is set to 702. In particular, when the minimum value is selected, it is considered that a situation in which a new worker exists in the vicinity and supports the work can be reflected. Here, the description of FIG. 6 ends.

Returning to FIG. 5, the data analysis device 102 generates the worker attribute use prediction model 232 using the worker attribute use learning data 224 (step S506). Similar to the worker ID usage prediction model 231 in step S504, this process is a process of generating the worker attribute usage prediction model 232 as a regression model for predicting the working time, which is the objective variable 601 by machine learning. The algorithm method is the same as in step S504.

By generating the worker attribute usage prediction model 232 in this way, it is possible to obtain a universal prediction model in which the worker attribute possessed by the worker is used as an explanatory variable instead of the worker ID 303. As a result, when the work of a worker whose actual result does not exist in the learning data is predicted, the work time can be predicted accurately by giving the attribute information of the worker as an explanatory variable. The worker attribute usage prediction model 232 is used in FIG. 16, which will be described later.

Next, the data analyzer 102 generates the intermediate learning data 225 (step S507). FIG. 8 shows an example of the intermediate learning data 225.

[Intermediate learning data 225]
FIG. 8 is an explanatory diagram showing an example of the intermediate learning data 225. The intermediate learning data 225 is learning data in which the worker ID variable 603 of FIG. 6 is added to the worker attribute use learning data 224 of FIG. First, as shown in FIG. 8, the data analyzer 102 generates the intermediate learning data 225. Here, the description of FIG. 8 ends.

Returning to FIG. 5, the data analyzer 102 uses the intermediate learning data 225 to generate the intermediate prediction model 233 (step S508). Similar to the worker ID usage prediction model 231 in step S504, this process is a process of generating the intermediate prediction model 233 as a regression model for predicting the working time, which is the objective variable 601. Is the same method as in step S504.

Next, the data analyzer 102 uses the intermediate prediction model 233 to influence the influence of each variable on the prediction result when each work is an input (first influence degree) for all the works of the intermediate learning data 225. ) Is calculated (step S509). Specifically, for example, the data analyzer 102 creates the first impact table 226. FIG. 9 shows an example of the first influence degree table 226.

[First influence table 226]
FIG. 9 is an explanatory diagram showing an example of the first influence degree table 226. The first influence degree table 226 is information exemplifying the influence degree (first influence degree) of each variable on the predicted value calculated by using the prediction model generated by the intermediate learning data 225 in the form of a table in all the operations. Is. Specifically, for example, the first influence degree table 226 shows the work ID 301, the objective variable 601, the influence degree 901 of the explanatory variable 602, the influence degree 902 of the worker attribute variable 700, and the worker ID variable 603. It has an influence degree of 903 and a predicted value of 904.

The predicted value 904 is an output result (work) when the explanatory variable 602, the worker attribute variable 700, and the worker ID variable 603 of FIG. 8 are input to the prediction model generated by the intermediate learning data 225 for each work ID 301. Time).

The influence degree 901 of the explanatory variable 602, the influence degree 902 of the worker attribute variable 700, and the influence degree 903 of the worker ID variable 603 are such that the calculated predicted value 904 is the average value of the predicted values 904 of all works. It is information showing what kind of influence is received from each variable and determined. Specifically, for example, the degree of influence 901 of the explanatory variable 602 is set for each item of the explanatory variable 602 (movement distance 621, quantity 622, weight 623, volume 624, shelf upper tier 625, shelf middle tier 626, shelf lower tier 627, ...). Has an influence on.

The influence degree 902 of the worker attribute variable 700 has an influence degree for each item of the worker attribute variable 700 (experience days 701, experience days 702 of nearby workers, ...). The degree of influence 903 of the worker ID variable 603 has a degree of influence for each item (ID1, ID2, ...) Of the worker ID variable 603. The larger the impact value, the more influential the work is.

For example, when the work ID 301 is an entry of "1", the influence degree of the variable "ID1" of the worker ID variable 603 is "+55". This has an effect of "+55" on "300" which is the average value of all the predicted values 904, and similarly, the influence degree of other variables is added to finally "432" which is the predicted value 904. Means that has been decided. As a result, it becomes possible to quantitatively understand how the values of various variables affect the predicted value 904 in each work of the learning data.

As a method for calculating such an influence degree, for example, as a known technique, there are a method such as SHAP (Shapley Adaptive exPlanations) and LIMITE (Local Interpretable Model-agnostic Expressions).

As a result, the data analyzer 102 can obtain a quantitative value of how much each variable of the corresponding work affects the predicted value 904 with respect to the predicted value 904 output by an arbitrary prediction model. Here, the description of FIG. 9 ends.

Next, the data analyzer 102 generates the abstract variable use learning data 227 (step S510). Specifically, for example, the data analyzer 102 processes the worker ID variable 603 into the worker attribute variable 700 by using the first influence degree table 226. This will be specifically described with reference to FIG.

[Impact distribution information]
FIG. 10 is an explanatory diagram showing influence degree distribution information in all operations. The influence degree distribution information 1000 is information in which the worker ID variable name 1001 and the influence degree distribution 1002 are associated with each other. The data analysis device 102 obtains the distribution of the degree of influence in all the works (work IDs 301: 1, 2, ...) For each worker ID 303 by using the worker ID variable 603 of the worker and the degree of influence thereof.

That is, the influence degree distribution 1002 of a certain value (worker ID 303) of the worker ID variable name 1001 is the worker ID variable in the influence degree 903 of the worker ID variable 603 of the first influence degree table 226 shown in FIG. It is a column of influence degree 903 of 603. For example, if the worker ID 303 is "ID1", the influence distribution 1002 becomes the influence distribution 1002 of "+55, 0, 0, ..." In the ascending order of the work ID 301, and the worker ID 303 is "ID2". If there is, the influence degree distribution 1002 becomes the influence degree distribution 1002 of "0, -61, 0, ..." In the ascending order of the work ID 301.

In FIG. 10, in the influence degree distribution 1002, the influence degree 903 of the worker ID variable 603 in the work (non-execution work) in which the worker ID variable 603 of FIG. 8 is “0” is a black circle ●, and the worker ID variable 603 is “ The degree of influence 903 of the worker ID variable 603 in the work (implementation work) of 1 ”is indicated by a white circle.

For example, in the worker ID variable 603 of the intermediate learning data 225 in FIG. 8, the column of "ID1" is "1,0,0, ..." In the ascending order of the work ID 301. Therefore, in the influence degree distribution 1002 in which the worker ID 303 is "1", the value of the worker ID variable 603 corresponding to the influence degree "+58" in which the work ID 301 is "1" is "1", so that the work ID 301 is "1". The degree of influence "+58" of "1" is indicated as a white circle ○.

Further, since the value of the worker ID variable 603 corresponding to the influence degree “0” of the work ID 301 is “2” is “0”, the influence degree “0” of the work ID 301 “2” is indicated as a black circle ●. To. Similarly, since the value of the worker ID variable 603 corresponding to the influence degree “0” of the work ID 301 is “3” is “0”, the influence degree “0” of the work ID 301 “3” is indicated by a black circle ●. Will be done.

In this way, by obtaining the influence degree distribution 1002 for each worker ID variable name 1001, the worker ID 303 having a positively large influence degree 903 of the worker ID variable 603, the worker ID 303 having a negatively large influence degree, and the value being "0". It is possible to distinguish between the worker ID 303, which has almost no difference from the case of "."

For example, a worker whose worker ID 303 is "1" has a great positive effect when he / she participates in the work of an entry in which the worker ID variable 603 is "1" (for example, the work ID 301 is "1"). That is, it can be seen that the working time increases. That is, if variables other than the worker ID variable 603 are not used, the prediction accuracy will be lowered, and the influence peculiar to the worker will be taken into consideration. That is, if the worker ID variable 603 is used, the prediction accuracy will be higher by machine learning. It can be considered that the model has been trained. Therefore, it can be interpreted that the work of the worker whose worker ID 303 is "1" has an effect of prolonging the work time.

Similarly, a worker whose worker ID 303 is "2" is a worker who has a large negative influence, and a worker whose worker ID 303 is "45" has a special influence especially when this worker works. It turns out that it is a worker who has not. As a specific discrimination method, the data analyzer 102 has an average value (hereinafter, first average value) and variance of the degree of influence of the work group when the worker ID variable 603 is “0” (black circle ●). , The average value (hereinafter, the second average value) of the degree of influence of the work group in which the worker ID variable 603 is “1” (white circle ○) is obtained.

When the difference between the first average value and the second average value is a constant multiple of the variance or more, the data analyzer 102 determines that it has a large positive or negative effect. By performing the above processing, the worker is classified into three types: a worker who has a positive influence on the predicted value 904 of the objective variable 601 (working time), a worker who has a negative influence, and a worker who has no influence. It becomes possible.

Then, the data analysis device 102 generates the abstract variable use learning data 227 based on the present processing result. FIG. 11 shows an example of learning data 227 using abstract variables.

[Learning data using abstract variables 227]
FIG. 11 is an explanatory diagram showing an example of learning data 227 using abstract variables. The abstraction variable use learning data 227 has a work ID 301, an objective variable 601 and an explanatory variable 602, a worker attribute variable 700, and a worker ID abstraction variable 1101. That is, the abstraction variable use learning data 227 is the learning data in which the worker ID abstraction variable 1101 is added to the worker attribute use learning data 224 shown in FIG. 7.

The worker ID abstraction variable 1101 is a variable that abstracts the worker ID 303. Specifically, for example, the worker ID abstraction variable 1101 has variables "work is slow" and "work is fast". The data analysis device 102 determines the value of the worker ID abstraction variable 1101 based on the determination result of whether or not the above-mentioned influence is exerted.

For example, if the value of "work is slow" is "1", it means that the work of the work ID 301 was performed by a worker whose work is slow, and the value of "work is fast" is "1". If there is, it means that the work of the work ID 301 was performed by a worker who has a fast work. If the values of "slow work" and "fast work" are both "0", it means that the work of the work ID 301 was performed by an average speed worker. In this way, the combination of the values "slow work" and "fast work" allows the worker to be abstracted by his or her work ability. Here, the description of FIG. 11 ends.

Returning to FIG. 5, the data analyzer 102 generates the abstract variable usage prediction model 234 using the abstract variable usage learning data 227 (step S511). Similar to the worker ID usage prediction model 231 in step S504, this process is a process of generating the abstract variable usage prediction model 234 as a regression model for predicting the working time which is the objective variable 601 by machine learning. The method of the algorithm is the same as that of step S504.

In this way, by applying the worker ID abstraction variable 1101 as a new explanatory variable to the abstraction variable usage prediction model 234 instead of the worker ID variable 603, more accurate prediction can be realized. .. In particular, when the actual result of the worker to be predicted does not exist in the work actual data 221 and the worker is a newly joined worker as a new employee, the worker ID abstraction variable 1101 "work". By setting the value of "is slow" to "1", it is possible to make a prediction considering the influence of the work performed by an unfamiliar newcomer.

Next, the data analysis device 102 calculates the degree of influence (second degree of influence) of the work-specific variables on the abstraction variable usage prediction model 234 (step S512).

[Second influence table 228]
FIG. 12 is an explanatory diagram showing an example of the second influence degree table 228. The second influence degree table 228 is a table for storing the second influence degree calculated in step S512. Specifically, the second influence degree table 228, for example, the second influence degree table 228 has the work ID 301, the objective variable 601 and the influence degree 901 of the explanatory variable 602 as in the case of the first influence degree table 226. It has an influence degree 902 of the worker attribute variable 700 and a predicted value 904, and newly has an influence degree 1201 of the worker ID abstraction variable 1101. Since the calculation of each influence degree is the same as that of the first influence degree table 226, the description thereof will be omitted. This is the end of the description of FIG.

Returning to FIG. 5, the data analyzer 102 generates the worker characteristic variable using the second influence degree table 228 (step S513). Worker characteristic variables are variables that represent non-average and peculiar characteristics of each worker. When there are a plurality of worker work performance data 221 showing the same tendency, it can be said that the worker characteristic variable is a variable representing individuality that is characteristic but commonly exists. Here, a method of discovering specific worker characteristics will be described with reference to FIG.

[Influence distribution of explanatory variable 602]
FIG. 13 is a graph showing an example of the influence distribution of the explanatory variable 602. In the influence distribution 1300 of FIG. 13, among the explanatory variable 602, the worker attribute variable 700, and the worker ID abstraction variable 1101, the value of the upper shelf 625 of the explanatory variable 602 will be described as an example, but other The same is true for variables.

In FIG. 13, the horizontal axis of the influence degree distribution 1300 is the value of the variable (in this example, the value of the upper shelf 625), and the vertical axis is the influence degree (in this example, the second influence degree table 228 of FIG. 12). The value of the upper shelf 625 of the influence degree 901 of the explanatory variable 602).

The data analyzer 102 selects "upper shelf 625". The “shelf upper tier 625” is a variable indicating how many products are arranged on the shelf upper tier in the work of the corresponding work ID 301. As prior knowledge about the business, how the increase or decrease of the value of the corresponding variable affects the work time is defined in advance. For example, it is defined that the larger the number of products 306 on the upper shelf 625, the slower the work tends to be. The method of specifying such a tendency can be set by, for example, specifying a condition that the corresponding variable has a positive correlation with the objective variable 601.

The data analyzer 102 analyzes the influence distribution 1002 of FIG. 13 on the premise of the conditions specified in this way. For example, according to the influence degree distribution 1300, it can be seen that the influence degree when the value of the upper shelf 625 is “5” varies depending on the work. It can be seen that among these variations, there is work that does not have a large influence, contrary to the condition that the larger the value of the upper shelf 625, the larger the influence (the dotted line in FIG. 13). Circle).

This means that even if the product to be worked on is on the upper shelf 625, it can be said that the work has a work record that does not affect the work time. For example, it is possible to infer that the worker who carried out the work is tall. It becomes. In this way, the data analyzer 102 regards the work group having a tendency of influence contrary to the given prior knowledge as a group having similar characteristics (in this case, good at work on the upper shelf 625). , Worker feature variables are generated as new variables that represent the characteristics of individual workers. Here, the description of FIG. 13 ends.

Returning to FIG. 5, the data analyzer 102 generates the worker feature use learning data 229 using the worker feature variable (step S514). An example of the worker feature use learning data 229 is shown in FIG.

[Worker feature use learning data 229]
FIG. 14 is an explanatory diagram showing an example of worker feature use learning data 229. This is learning data in which a worker characteristic variable 1401 is added as an explanatory variable to the learning data 227 using abstract variables. If the value of "good at the upper shelf", which is an example of the worker characteristic variable 1401, is "1", it means that the work was performed by a worker who is "good at the upper shelf", and is set to "0". If there is, it indicates that it was not carried out by a worker who is "good at the upper shelf".

Further, the type of the worker characteristic variable 1401 may be a continuous value. In the case of continuous values, the data analyzer 102 calculates the distribution of the degree of influence for each constant or specific width, and similarly discovers the worker characteristics. For example, when the explanatory variable 602 "age" is selected, the data analyzer 102 may be 0-20 years old, 21-40 years old, 41-50 years old, 51-60 years old, 61 years old or older, and so on. The value of the degree of influence is summarized by the width, and the degree of influence distribution 1002 is generated. It is possible for the user or the like to define in advance which variable and what kind of analysis width is used for the above analysis. Here, the description of FIG. 14 ends.

Returning to FIG. 5, the data analysis device 102 generates the worker feature use prediction model 235 using the worker feature use learning data 229 (step S515). By generating the worker feature usage prediction model 235 in this way, the worker feature variable 1401 can be added as a new explanatory variable, and more accurate prediction can be performed. In particular, although the worker has no actual record in the work record data 221, his / her personal characteristics (for example, work on a tall and tall shelf is not bothered) are set as the value of the worker characteristic variable 1401 in advance. This enables more accurate prediction.

This completes the process of generating various prediction models for predicting the working time (step S516). When analyzing the worker characteristics, the data analysis device 102 excludes the work in which the difference between the actual value of the work time and the predicted value 904 in the work record data 221 is large from the work record data 221 to be analyzed. You may do so.

For example, work that is a constant multiple of the variance value or more is excluded based on the average value of the differences in all the work. As a result, it is possible to eliminate an adverse effect on the analysis when the cause of the work showing a tendency different from the prior knowledge is not due to personal characteristics but an abnormal value at the time of acquiring the actual business data.

Further, when two or more variables showing a tendency different from the prior knowledge are found for one work, they may be combined and generated as a worker characteristic variable 1401. For example, under the condition that the work on the upper shelf 625 is good and the picking of products whose weight 623 per piece is equal to or more than the threshold value is fast (as a prior knowledge, if the variable of weight 623 is large, the work becomes slow). When there are more than a predetermined number of worker's work (work showing the value of the degree of influence contrary to it), the data analyzer 102 sets the new worker characteristic variable 1401 as "good at the upper shelf & heavy luggage". Is good at. As a result, the data analyzer 102 can generate the worker characteristic variable 1401 that represents the personal characteristics of the worker in which a plurality of conditions are combined in more detail.

<Business improvement measure generation process 122>
FIG. 15 is a flowchart showing an example of a business improvement measure generation processing procedure for generating a business improvement measure using the worker feature usage prediction model 235 generated by the model generation process 121. The data analysis device 102 first activates the business improvement measure generation process 122 (step S1501), and acquires the business performance data 221 to be improved from the storage device 202 (step S1502). The business performance data 221 to be improved is the business performance data 221 in the target period for which the business is to be improved. For example, if the business is to be improved on or after the next day when the business improvement measure generation process 122 is executed, the latest past One week's business performance data 221 and the like are targeted.

It is assumed that the working environment for the last week (for example, the arrangement of products, the products handled, and the composition of workers) is equal to the working environment after that. Measures to improve operations during the most recent past week are based on the idea that future operations are also expected to improve.

For example, the data analysis device 102 can continuously improve the business by repeating the business improvement measure generation process 122 every other week. The work environment to be improved is, for example, the arrangement of products, but at least one of the arrangement of products, the products to be handled, and the composition of workers may be the object of improvement.

The data analysis device 102 executes a process of changing the product placement location for the business performance data 221 to be improved (step S1503). Specifically, for example, the data analysis device 102 changes the location code 305 of the product of the business performance data 221 to be improved. The method of replacing the location code 305 is not particularly limited, and the data analyzer 102 may execute it randomly, or it may be executed by various algorithms suitable for the combinatorial optimization problem such as a genetic algorithm. ..

The data analysis device 102 recalculates the explanatory variable 602 of the worker ID use learning data 223 according to the business performance data 221 to be improved after the change. For example, since the location code 305 has been changed, the data analyzer 102 also changes the values of the upper shelf 625, the middle shelf 626, and the lower shelf 627. Further, the data analysis device 102 recalculates the travel distance 621 based on the changed location code 305. The recalculation target is not limited to the shelf upper tier 625, the shelf middle tier 626, the shelf lower tier 627, and the moving distance 621 as long as the explanatory variable 602 needs to be changed by changing the location code 305.

Also, because the location code 305 has been changed, the nearby workers will also be different. Therefore, the data analysis device 102 recalculates the number of days of experience 702 of the nearby worker, which is the worker attribute variable 700 of the worker attribute use learning data 224, according to the location code 305 of the degree of change. The recalculation target is not limited to the number of days of experience 702 of nearby workers as long as the worker attribute variable 700 needs to be changed by changing the location code 305.

The data analysis device 102 selects the prediction model to be applied from three types of prediction models, and inputs the recalculated worker ID usage learning data 223 obtained in step S1503 into the selected prediction model to perform the work. A time prediction value of 904 is obtained (step S1504). The proper use of the three types of prediction models will be described later in FIG.

The data analysis device 102 sums up the predicted work time 904 obtained in each work and calculates the total work time in one week (step S1505). The data analyzer 102 determines whether or not the obtained total working time matches a predetermined end condition (step S1506).

Here, the predetermined end conditions include conditions such as whether or not improvement has been made compared to the work time in the business performance data 221, the number of executions of step S1503, the degree of improvement due to the change in the product change arrangement, and the like. Applies. If the end condition is not satisfied (step S1506: No), the process returns to step S1503 again.

When the end condition is satisfied (step S1506: Yes), the data analyzer 102 determines the improvement measure (step S1507). The improvement measure is, for example, a work environment in which business performance data 221 to be improved can be obtained. When the business performance data 221 to be improved is obtained by changing the place code, the work environment to be the improvement measure is the arrangement state of the products after the change of the place code.

When the business performance data 221 to be improved is obtained by changing the product code of the product to be handled, the work environment to be the improvement measure is the arrangement state of the product after the change of the product code. When the work performance data 221 to be improved is obtained by changing the composition of the workers, the work environment to be the improvement measure is the arrangement of the workers after the composition of the workers is changed.

Then, the data analysis device 102 outputs the determined improvement measure (step S1508). Specifically, for example, the data analysis device 102 displays the determined improvement measures on a display or transmits the determined improvement measures to another computer such as the business system 101. Then, this process ends (step S1508).

In this explanation, the work of the last week is targeted for improvement, but it is not limited to this period. In addition, it can be applied to use cases where the work to be improved is the work content to be performed in the future and the work environment at the time of data analysis is improved. In this case, it can be realized by obtaining the conditions of the work environment (arrangement of items to be picked, etc.) in which the work time is the shortest when performing the work contents to be carried out in FIG. As a result, it becomes possible to improve the work environment each time according to the contents of the work implementation schedule, and a higher improvement effect can be expected.

<Processing for proper use of prediction model>
Here, the process of properly using the prediction model in step S1504 described above will be described.

[Proper use of prediction model]
FIG. 16 is a flowchart showing an example of a processing procedure for properly using various prediction models in step S1504. The data analysis device 102 is a worker who manages the worker attributes and worker characteristics corresponding to the worker ID 303 of all the workers existing in the work record data 221 based on the past work record data 221 held in advance. The attribute / feature data 230 is created or updated (step S1601). Step S1601 is periodically executed at a fixed timing such as every week. Here, the worker attribute / feature data 230 will be specifically described with reference to FIG.

[Worker attribute / feature data 230]
FIG. 17 is an explanatory diagram showing an example of worker attribute / feature data 230. The worker attribute / feature data 230 has a worker ID 303, an actual number of works 1701, a worker attribute variable 1702, a worker ID abstraction variable 1703, and a worker feature variable 1704.

The actual number of works 1701 includes 1711 in the learning data and 1712 within the latest one month. The value of 1711 in the learning data is the actual number of operations 1701 in the worker ID use learning data 223 when generating the worker ID usage prediction model 231 for the worker of the worker ID 303. The value of 1712 within the latest one month is 1701 of the actual number of works included in the work actual data 221 within the latest one month for the worker with the worker ID 303. In FIG. 17, the "latest one month" is set as an example, but the term is not limited to one month.

The values of 1711 in the training data and 1712 within the latest one month are, for example, the workers who are often included in the worker ID usage learning data 223 and the values of the latest work which are not included in the worker ID usage learning data 223. It is used to discriminate between a worker who has the work record data 221 and a worker who does not exist in the work record data 221 of the latest work.

The worker attribute variable 1702 includes, for example, the latest number of days of experience 1721 for the worker with work ID 301. The latest number of days of experience 1721 indicates the number of days of experience 701 at the latest time point for the worker of the work ID 301.

The worker ID use learning data 223 includes 1722, which is slow in work, and 1723, which is fast in work. The slow work 1722 indicates whether or not the work of the worker with the work ID 301 is slow. A value of "1" indicates that the work is slow, and a value of "0" indicates that the work is fast. Similarly, the fast work 1723 indicates whether or not the work of the worker with the work ID 301 is fast. A value of "1" indicates that the work is fast, and a value of "0" indicates that the work is slow.

As the values of the slow work 1722 and the fast work 1723, the values calculated at the time of generating the worker feature use learning data 229 are used. For example, in the influence degree distribution information 1000 of FIG. 10, if the average value av of the influence degree is a ≦ av ≦ b (a is less than 0, b is larger than 0), the work for the worker is slow. The values of 1722 and 1723, which works fast, are both set to "0".

If the average value av of the degree of influence is a> av, the value of 1722, which is slow for the worker, is set to "0", and the value of 1723, which is fast, is set to "1". If the average value av of the degree of influence is b <av, the value of 1722, which is slow for the worker, is set to "1", and the value of 1723, which is fast, is set to "0".

Further, as for the worker whose worker ID 303 is "40", the work performance data 221 does not exist in the learning data 1711 (the value of 1711 in the learning data is "0"), the worker learns as described above. It is not possible to set the values of 1722 for slow work and 1723 for fast work from the data. In this case, the data analysis device 102 generates the influence degree distribution information 1000 for the latest one-month work record data 221 and sets the values of 1722 for slow work and 1723 for fast work.

On the other hand, like the worker whose worker ID 303 is "55", the work performance data 221 is almost nonexistent, that is, for the workers having a predetermined number or less, the data analysis device 102 has a slow work of 1722 and a fast work. Set the value of 1723 to "0". However, this does not apply when the initial value of the worker attribute variable 1702 is defined for the worker who has no track record based on the prior knowledge about the work. Details will be described later.

Regarding the worker characteristic variable 1704, for example, the data analyzer 102 specifies the work ID 301 in which the value of the worker characteristic variable 1401 “good at the upper shelf” is “1” from the worker feature use learning data 229. Then, the data analysis device 102 sets "1" to the value of 1731, which the upper shelf is good at, in the entry of the worker who has performed the work of the specified work ID 301. Similarly, the data analysis device 102 identifies the work ID 301 in which the value of the worker feature variable 1401 “good at the upper shelf” is “0” from the worker feature use learning data 229. Then, the data analysis device 102 sets "0" to the value of 1731, which the upper shelf is good at, in the entry of the worker who performed the work of the specified work ID 301.

In this way, it is possible to manage the values of the latest worker attribute variable 1702, the worker ID abstraction variable 1703, and the worker feature variable 1704 of each worker at the time of executing the business improvement measure generation process 122. Here, the description of FIG. 17 ends.

Returning to FIG. 16, the data analyzer 102 works by using the worker attribute / feature data 230, the worker ID usage prediction model 231 and the worker attribute usage prediction model 232, and the worker feature usage prediction model 235. Execution of the time prediction is started in step S1504 (step S1601).

First, the data analysis device 102 determines whether or not the actual number of work 1701 in the learning data 1711 is equal to or more than a predetermined number (for example, 100) with respect to the worker ID 303 of the worker of the work to be predicted (step S1602). When there are a predetermined number or more (step S1602: Yes), the data analyzer 102 determines the prediction model to be applied to the worker ID usage prediction model 231 (step S1603).

In this case, in step S1505, the data analyzer 102 predicts the working time 904 by inputting the recalculated worker ID usage learning data 223 into the worker ID usage prediction model 231. ..

On the other hand, when there are no more than a predetermined number (step S1602: No), in the data analyzer 102, the number of workers to be predicted is 1712 within the latest one month, 1701 is present in a predetermined number (for example, 50) or more. It is determined whether or not (step S1604). When there are a predetermined number or more (step S1604: Yes), the data analyzer 102 determines the prediction model to be applied to the worker feature use prediction model 235 (step S1605).

In this case, in step S1505, the data analyzer 102 adds the worker ID abstraction variable 1703 and the worker feature variable 1704 to the worker ID usage learning data 223 after recalculation to the worker feature usage prediction model 235. By inputting the training data, the prediction of the working time 904 is executed.

On the other hand, when the number of cases does not exist more than a predetermined number (step S1604: No), the data analyzer 102 determines the prediction model to be applied to the worker attribute use prediction model 232 (step S1606).

In this case, in step S1505, the data analyzer 102 inputs the training data in which the worker attribute variable 1702 is added to the worker ID usage learning data 223 after recalculation into the worker attribute usage prediction model 232. , The prediction of working time 904 will be executed.

Through the above processing, the data analysis device 102 calculates the predicted work time of the work to be predicted, and ends this processing (step S1607). In this way, by appropriately using the prediction model to be used according to the worker to be predicted, it is possible to predict the work time with high accuracy according to the number of work performance data 221 of the worker.

That is, in this example, in step S1602, with respect to the worker whose number of past business record data 221 is equal to or more than a predetermined number (for example, 100) (step S1602: Yes), the data analysis device 102 is set to the individual worker. The working time is predicted by the worker ID usage prediction model 231 which is a specialized prediction model (step S1603).

Further, in step S1604, for an unknown worker whose latest one-month work record data 221 is less than a predetermined number (for example, 50), the data analysis device 102 is not a feature of the individual worker but a work. The working time is predicted using the member attribute usage prediction model 232 (step S1606).

In this case, the initial value may be defined by prior knowledge about the business regarding a specific attribute. For example, if it is possible to "consider that a worker with a small number of days of experience is a newcomer and work is slow" by prior knowledge, the data analysis device 102 previously performs work performance data 221 in the worker attribute / feature data 230. The value of "work is slow" for workers whose number of cases is less than a predetermined number (for example, 50 cases) is set to "1".

In the case of a worker who does not exist in 1711 in the training data (step S1602: No), but exists in 1712 within the latest one month in 1712 or more (for example, 100 cases) (step S1604: Yes), the data analyzer. In 102, the working time is predicted using the worker feature usage prediction model 235 using the values of the variables given in advance by the worker attribute / feature data 230 (step S1605).

As described above, in the case of any worker, the data analysis device 102 executes the prediction with the prediction model generated by the learning data having a sufficient amount of work. Therefore, it is possible to predict the working time with high accuracy. By using the data analysis device 102 described above, it is possible to predict the work time with high accuracy in consideration of the individuality of the worker who is the target of the work time prediction.

Further, the above-mentioned data analysis device 102 can also be configured as described in (1) to (15) below.

(1) In the data analysis device 102, the processor 201 has an objective variable 601 indicating the work time, an explanatory variable 602 regarding the work environment, and an operation indicating whether or not the work is performed for each worker for each of a plurality of work operations. The work of predicting the work time based on the acquisition process of acquiring the worker ID use learning data 223 having the member ID variable 603 and the worker ID use learning data 223 as the first learning data and the worker ID use learning data 223 acquired by the acquisition process. Working time by inputting worker ID usage learning data into the prediction model generation process that generates the worker ID usage prediction model 231 as the first prediction model and the worker ID usage prediction model 231 generated by the prediction model generation process. 904, and the worker ID variable 603, which indicates the degree of influence on the statistic of the first predicted value 904 for each work (for example, the average value or other statistic such as the median value). Work of a plurality of workers based on a calculation process for calculating one influence degree 903 for each work and a distribution of a first influence degree 903 (impact degree distribution information 1000) for each of a plurality of works for each worker. Worker ID abstraction variable 1101 that abstracts the speed is generated for each work, and for each work, there is an abstract variable use learning data having an objective variable 601, an explanatory variable 602, and a worker ID abstract variable 1101. The training data generation process of generating 227 as the second training data is executed.

As a result, it is possible to obtain an explanatory variable that abstracts a plurality of workers who have performed the work to the work speed for each work, and the abstract variable use prediction model 234 generated by using the abstract variable use learning data 227. It is possible to improve the accuracy of (third prediction model).

(2) In the data analysis device 102 of the above (1), the first learning data includes a worker attribute variable 700 indicating the number of experience days 701 of each of the plurality of workers who performed the work for each work. As a result, the first learning data becomes the intermediate learning data 225.

Further, in the prediction model generation process, the processor 201 generates the intermediate prediction model 233 as the first prediction model based on the intermediate learning data 225 which is the first training data including the worker attribute variable 700.

Then, in the calculation process, the processor 201 inputs the intermediate learning data 225 into the intermediate prediction model 233 for each work, so that the first predicted value 904 of the working time, the explanatory variable 602, the worker attribute variable 700, and the worker The first influence degree 903, which indicates the degree of influence that each of the ID variables 603 has on the statistic of the first predicted value 904 for each work, is calculated for each work.

Then, in the learning data generation process, the processor 201 generates the worker ID abstraction variable 1101 for each work based on the distribution of the first influence degree 903 for each worker (impact degree distribution information 1000), and abstracts it. A worker attribute variable 700 for each work is added to the variable use learning data 227. As a result, it is possible to generate an abstract variable usage prediction model 234 in consideration of the number of days of experience of the worker 701.

(3) In the data analysis device 102 of the above (2), the worker attribute variable 700 includes the number of experience days 702 of a worker within a predetermined distance from the worker and in the vicinity where the work time zones overlap. As a result, it is possible to reflect in the prediction model that the larger the number of days of experience 702 of a nearby worker is, the more likely the worker of that work is to receive work support from a nearby worker. The smaller the number of days of experience 702, the more the worker of the work can reflect the situation of providing work support to the nearby worker in the prediction model.

(4) In the data analysis device 102 of the above (2), in the prediction model generation process, the processor 201 has a working time based on the worker attribute use learning data 224, which is the first learning data including the worker attribute variable 700. The worker attribute usage prediction model 232 for predicting is generated.

As a result, the worker attribute usage prediction model 232 that considers the worker attribute such as the number of days of experience of the worker 701 allows the working time of the worker having such a worker attribute to be determined without specifying the individual worker. Can be predicted.

(5) In the prediction model generation process in the data analyzer 102 of the above (1), the processor 201 uses the abstract variable usage prediction model 234 that predicts the working time based on the abstract variable usage learning data 227. Generate as a prediction model.

Further, in the calculation process, the processor 201 inputs the abstract variable use learning data 227 into the abstract variable use prediction model 234 generated by the prediction model generation process, thereby explaining that the working time is the second predicted value 904. The second influence degree 901, which indicates the degree of influence that the variable 602 has on the statistic of the second predicted value 904 for each work, is calculated for each work.

Then, in the learning data generation process, the processor 201 generates a worker characteristic variable 1401 indicating the personal characteristics of the worker with respect to the explanatory variable 602 based on the distribution 1300 of the second influence degree 901 for each work, and abstracts it. The worker feature use learning data 229 in which the worker feature variable 1401 for each work is added to the abstract variable use learning data 227 is generated as the third learning data.

As a result, the worker feature use prediction model 235 (fourth prediction model) in consideration of the worker feature variable 1401 can be generated.

(6) In the prediction model generation process in the data analysis device 102 of the above (5), the processor 201 uses the worker feature use prediction model 235 that predicts the work time based on the worker feature use learning data 229. Generate as a prediction model.

As a result, the work time of a worker having such a characteristic is predicted without specifying the individual worker by the worker characteristic use prediction model 235 considering the individual characteristics of the worker such as being good at the upper shelf. be able to.

(7) In the data analyzer 102 of the above (1), the processor 201 aims at the first actual data (business actual data 221) having the worker who performed the work, the working time, and the working environment for each work. By generating the variable 601 and the explanatory variable 602 and generating the worker ID variable 603 from the worker for each work, the conversion process of converting the first actual data into the worker ID use learning data 223 is executed.

Further, in the acquisition process, the processor 201 converts the first actual data into the second actual data (business actual data 221 to be improved) by recalculating the working time based on the change in the working environment of the first actual data. Objective variable 601 indicating the work time recalculated for each work after conversion based on the second actual data, explanatory variable 602 for the changed work environment, and work for each worker based on the change in the work environment. The worker ID use learning data 223 having the worker ID variable 603 indicating whether or not the above is implemented is acquired as the fourth learning data.

Then, in the calculation process, the processor 201 inputs the fourth learning data regarding the worker to be predicted into the worker ID usage prediction model 231 to obtain the third predicted value 904 of the working time for the worker to be predicted. Calculate for each work.

As a result, it is possible to obtain a predicted value of the work time of each work when the worker to be predicted performs the improvement work specified from the work performance data 221 to be improved. Therefore, the user of the data analysis device 102 can specify whether or not the working time of the worker to be predicted has been improved.

(8) In the data analysis device 102 of the above (7), in the calculation process, the processor 201 has the first number of operations indicating that the worker to be predicted has performed in the worker ID variable 603 of the fourth training data. When there are more than a predetermined number (S1602: Yes), the third predicted value 904 of the working time for the worker to be predicted is calculated for each work by inputting the fourth learning data into the worker ID usage prediction model 231. ..

As a result, for the fourth learning data in which the work performed by the worker to be predicted is equal to or greater than the first predetermined number, the worker ID usage prediction model 231 is preferentially applied to predict the work time for the worker to be predicted. The accuracy can be improved.

(9) In the data analysis device 102 of the above (8), the processor 201 has a determination process (step S1507) for determining whether or not the third predicted value 904 of the working time satisfies a predetermined improvement condition, and a determination process. When it is determined that the improvement conditions are satisfied, a decision process (step S1508) for determining the work environment related to the second actual data as an improvement measure and an output process (step S1509) for outputting the decision result by the decision process are executed. To do.

As a result, the data analysis device 102 can present appropriate improvement measures for the work performed by the worker to be predicted.

(10) In the data analyzer 102 of the above (4), the processor 201 aims at the first actual data (business actual data 221) having the worker who performed the work, the working time, and the working environment for each work. By generating the variable 601 and the explanatory variable 602 and generating the worker ID variable 603 from the worker for each work, the conversion process of converting the first actual data into the worker ID use learning data 223 is executed.

Further, in the acquisition process, the processor 201 converts the first actual data into the second actual data (business actual data 221 to be improved) by recalculating the working time based on the change in the working environment of the first actual data. The objective variable 601 indicating the work time recalculated for each work after conversion based on the second actual data, the explanatory variable 602 for the changed work environment, and the work for each worker based on the change in the work environment. Acquires the fourth learning data having the worker ID variable 603 indicating whether or not the above is performed, and adding the worker attribute variable 1702.

Then, in the calculation process, the processor 201 inputs the fourth learning data regarding the worker to be predicted into the worker attribute use prediction model 232 to obtain the third predicted value 904 of the work time related to the worker to be predicted. Calculate for each work,

As a result, when the worker to be predicted performs the improvement work specified from the work performance data 221 to be improved, the predicted value of the work time of each work is calculated in consideration of the worker attributes of the worker to be predicted. Obtainable. Therefore, the user of the data analysis device 102 can accurately identify whether or not the working time of the worker to be predicted has been improved.

(11) In the data analysis device 102 of the above (10), in the calculation process, the processor 201 has the worker ID variable 603 of the fourth training data that the worker to be predicted is within a predetermined period (for example, the latest one month). When the number of works indicating that the work has been performed within 1712) does not exist in excess of the second predetermined number (step S1604: No), the fourth learning data regarding the worker to be predicted is input to the worker attribute usage prediction model 232. As a result, the third predicted value 904 of the working time for the worker to be predicted is calculated for each work.

As a result, for the fourth learning data in which the work performed by the worker to be predicted is less than the second predetermined number, the worker attribute usage prediction model 232 using the worker attribute is prioritized instead of specifying the individual worker. By applying it, it is possible to improve the prediction accuracy of the work time for the worker to be predicted.

(12) In the data analyzer 102 of the above (5), the processor 201 aims at the first actual data (business actual data 221) having the worker who performed the work, the working time, and the working environment for each work. By generating the variable 601 and the explanatory variable 602 and generating the worker ID variable 603 from the worker for each work, the conversion process of converting the first actual data into the worker ID use learning data 223 is executed.

Further, in the acquisition process, the processor 201 converts the first actual data into the second actual data (business actual data 221 to be improved) by recalculating the working time based on the change in the working environment of the first actual data. The objective variable 601 indicating the work time recalculated for each work after conversion based on the second actual data, the explanatory variable 602 for the changed work environment, and the work for each worker based on the change in the work environment. Acquires the fourth training data having the worker ID variable 603 indicating whether or not the above is performed, and adding the worker ID abstraction variable 1703.

Then, in the calculation process, the processor 201 inputs the fourth learning data regarding the worker to be predicted into the abstraction variable usage prediction model 234 to obtain the third predicted value 904 of the working time for the worker to be predicted. Calculate for each work.

As a result, when the worker to be predicted performs the improvement work specified from the work performance data 221 to be improved, the work time of each work is predicted in consideration of the work speed that abstracts the worker to be predicted. You can get the value. Therefore, the user of the data analysis device 102 can accurately identify whether or not the working time of the worker to be predicted has been improved.

(13) In the data analyzer 102 of the above (12), in the calculation process, in the calculation process, the processor 201 has the worker to be predicted in the worker ID variable 603 of the fourth training data within a predetermined period (for example, the latest one month). When the number of works indicating that the work has been performed within 1712) is equal to or greater than the second predetermined number (step S1604: Yes), the fourth learning data regarding the worker to be predicted is input to the abstract variable usage prediction model 234. As a result, the third predicted value 904 of the working time for the worker to be predicted is calculated for each work.

As a result, for the fourth training data in which the work performed by the worker to be predicted is equal to or greater than the second predetermined number, the abstract variable usage prediction model 234 that abstracts the work speed of the worker to be predicted is preferentially applied. As a result, it is possible to improve the prediction accuracy of the work time of the worker to be predicted.

(14) In the data analyzer 102 of the above (6), the processor 201 aims at the first actual data (business actual data 221) having the worker who performed the work, the working time, and the working environment for each work. By generating the variable 601 and the explanatory variable 602 and generating the worker ID variable 603 from the worker for each work, the conversion process of converting the first actual data into the worker ID use learning data 223 is executed.

Further, in the acquisition process, the processor 201 converts the first actual data into the second actual data (business actual data 221 to be improved) by recalculating the working time based on the change in the working environment of the first actual data. The objective variable 601 indicating the work time recalculated for each work after conversion based on the second actual data, the explanatory variable 602 for the changed work environment, and the work for each worker based on the change in the work environment. Acquires the fourth learning data having the worker ID variable 603 indicating whether or not the above is performed, and adding the worker characteristic variable 1704.

Then, in the calculation process, the processor 201 inputs the fourth learning data regarding the worker to be predicted into the worker feature use prediction model 235 to obtain a third predicted value of the work time related to the worker to be predicted. Calculated for each of the above operations.

As a result, when the worker to be predicted carries out the improvement work specified from the work performance data 221 to be improved, the predicted value of the work time of each work is taken into consideration in consideration of the personal characteristics of the worker to be predicted. Can be obtained. Therefore, the user of the data analysis device 102 can accurately identify whether or not the working time of the worker to be predicted has been improved.

(15) In the data analyzer 102 of the above (14), in the calculation process, in the calculation process, the processor 201 has the worker to be predicted in the worker ID variable 603 of the fourth training data within a predetermined period (for example, the latest past one month). When the number of works indicating that the work has been performed within 1712) is equal to or greater than the second predetermined number (step S1604: Yes), the fourth learning data regarding the worker to be predicted is input to the worker feature usage prediction model 235. As a result, the third predicted value 904 of the working time for the worker to be predicted is calculated for each work.

As a result, for the fourth learning data in which the work performed by the worker to be predicted is equal to or greater than the second predetermined number, the worker feature use prediction model 235 using the personal characteristics of the worker to be predicted is preferentially applied. As a result, it is possible to improve the prediction accuracy of the work time of the worker to be predicted.

The present invention is not limited to the above-described embodiment, and includes various modifications and equivalent configurations within the scope of the attached claims. For example, the above-described examples have been described in detail in order to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to those having all the described configurations. Further, a part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Further, the configuration of another embodiment may be added to the configuration of one embodiment. In addition, other configurations may be added, deleted, or replaced with respect to a part of the configurations of each embodiment.

Further, each of the above-described configurations, functions, processing units, processing means, etc. may be realized by hardware by designing a part or all of them by, for example, an integrated circuit, and the processor realizes each function. It may be realized by software by interpreting and executing the program to be executed.

Information such as programs, tables, and files that realize each function is recorded in a memory, hard disk, storage device such as SSD (Solid State Drive), or IC (Integrated Circuit) card, SD card, DVD (Digital Versaille Disc). It can be stored in a medium.

In addition, the control lines and information lines indicate those that are considered necessary for explanation, and do not necessarily indicate all the control lines and information lines necessary for implementation. In practice, it can be considered that almost all configurations are interconnected.

100 Data analysis system 101 Business system 102 Data analysis device 110 Business performance DB
121 Model generation processing 122 Business improvement measure generation processing 201 Processor 202 Storage device 221 Business performance data 222 Product information 223 Usage learning data 224 Worker attribute usage learning data 225 Intermediate learning data 226 First impact table 227 Abstract variable usage learning data 228 Second influence table 229 Worker feature usage learning data 230 Worker attribute / feature data 231 Usage prediction model 232 Worker attribute usage prediction model 233 Intermediate prediction model 234 Abstract variable usage prediction model 235 Worker feature usage prediction model

Claims (15)

A data analyzer having a processor that executes a program and a storage device that stores the program.
The processor
Acquisition of first learning data having an objective variable indicating the work time, an explanatory variable related to the work environment, and a worker variable indicating whether or not the work is performed for each worker for each work of a plurality of works. Processing and
A prediction model generation process that generates a first prediction model that predicts the working time based on the first learning data acquired by the acquisition process.
By inputting the first training data into the first prediction model generated by the prediction model generation process, the first prediction value of the work time and the worker variable are the first prediction value for each work. A calculation process for calculating the first degree of influence, which indicates the degree of influence on the statistic, for each work, and
Based on the distribution of the first influence degree of each of the plurality of works for each worker, an abstract variable that abstracts the work speeds of the plurality of workers is generated for each work, and for each work, A training data generation process for generating a second training data having the objective variable, the explanatory variable, and the abstraction variable.
A data analyzer characterized by performing. The data analyzer according to claim 1.
The first learning data includes a worker attribute variable indicating the number of days of experience of each of the plurality of workers who performed the work for each work.
In the prediction model generation process, the processor generates the first prediction model based on the first learning data including the worker attribute variable.
In the calculation process, the processor inputs the first learning data including the worker attribute variable into the first prediction model for each work, so that the first predicted value of the working time and the explanatory variable are obtained. , The first degree of influence indicating the degree of influence that each of the worker attribute variable and the worker variable has on the statistic of the first predicted value for each work is calculated for each work.
In the learning data generation process, the processor generates the abstract variable for each work based on the distribution of the first influence degree for each worker, and the second training data is used for each work. Add the worker attribute variable,
A data analyzer characterized by this. The data analyzer according to claim 2.
The worker attribute variable includes the number of days of experience of another worker who is within a predetermined distance from the worker and whose working hours overlap.
A data analyzer characterized by this. The data analyzer according to claim 2.
In the prediction model generation process, the processor generates a second prediction model that predicts the work time based on the first learning data including the worker attribute variable.
A data analyzer characterized by this. The data analyzer according to claim 1.
In the prediction model generation process, the processor generates a third prediction model that predicts the working time based on the second learning data.
In the calculation process, the processor inputs the second training data into the third prediction model generated by the prediction model generation process, so that the second prediction value of the work time and the explanatory variable are the work. The second degree of influence, which indicates the degree of influence on the statistic of the second predicted value for each, is calculated for each of the operations.
In the learning data generation process, the processor generates a worker characteristic variable indicating the personal characteristics of the worker with respect to the explanatory variable based on the distribution of the second influence degree for each work, and the first. 2 Generate third learning data by adding the worker characteristic variable for each work to the training data.
A data analyzer characterized by this. The data analyzer according to claim 5.
In the prediction model generation process, the processor generates a fourth prediction model that predicts the work time based on the third learning data including the worker characteristic variable.
A data analyzer characterized by this. The data analyzer according to claim 1.
The processor
The objective variable and the explanatory variable are generated based on the first actual data having the worker who performed the work, the work time, and the work environment for each work, and the work is performed from the worker for each work. By generating a member variable, a conversion process for converting the first actual data into the first learning data is executed.
In the acquisition process, the processor converts the first actual data into the second actual data by recalculating the working time based on the change in the working environment of the first actual data, and the second actual data. Based on the actual data, the objective variable indicating the recalculated work time for each work, the explanatory variable for the changed work environment, and the execution of the work for each worker based on the change in the work environment. Acquire the fourth training data having the worker variable indicating the presence or absence, and
In the calculation process, the processor inputs the fourth learning data regarding the worker to be predicted into the first prediction model, and thereby obtains the third predicted value of the work time for the worker to be predicted. Calculate for each,
A data analyzer characterized by this. The data analyzer according to claim 7.
In the calculation process, when the processor has more than the first predetermined number of operations indicating that the worker to be predicted has performed in the worker variable of the fourth learning data, the first prediction model By inputting the fourth learning data into, the third predicted value of the working time for the worker to be predicted is calculated for each of the tasks.
A data analyzer characterized by this. The data analyzer according to claim 8.
The processor
Judgment processing for determining whether or not the third predicted value of the working time satisfies a predetermined improvement condition, and
When it is determined by the determination process that the improvement condition is satisfied, the determination process for determining the work environment related to the second actual data as an improvement measure, and the determination process.
Output processing that outputs the determination result by the determination processing and
A data analyzer characterized by performing. The data analyzer according to claim 4.
The processor
The objective variable and the explanatory variable are generated based on the first actual data having the worker who performed the work, the work time, and the work environment for each work, and the work is performed from the worker for each work. By generating a member variable, a conversion process for converting the first actual data into the first learning data is executed.
In the acquisition process, the processor converts the first actual data into the second actual data by recalculating the working time based on the change in the working environment of the first actual data, and the second actual data. Based on the actual data, the objective variable indicating the recalculated work time for each work, the explanatory variable for the changed work environment, and the execution of the work for each worker based on the change in the work environment. Acquire the fourth training data having the worker variable indicating the presence or absence and adding the worker attribute variable.
In the calculation process, the processor inputs the fourth learning data regarding the worker to be predicted into the second prediction model, and thereby obtains the third predicted value of the work time for the worker to be predicted. Calculate for each,
A data analyzer characterized by this. The data analyzer according to claim 10.
In the calculation process, when the number of operations indicating that the worker to be predicted has performed within the predetermined period in the worker variable of the fourth training data does not exist in the second predetermined number or more, the processor said. By inputting the fourth learning data regarding the worker to be predicted into the second prediction model, the third predicted value of the working time for the worker to be predicted is calculated for each work.
A data analyzer characterized by this. The data analyzer according to claim 5.
The processor
The objective variable and the explanatory variable are generated based on the first actual data having the worker who performed the work, the work time, and the work environment for each work, and the work is performed from the worker for each work. By generating a member variable, a conversion process for converting the first actual data into the first learning data is executed.
In the acquisition process, the processor converts the first actual data into the second actual data by recalculating the working time based on the change in the working environment of the first actual data, and the second actual data. Based on the actual data, the objective variable indicating the recalculated work time for each work, the explanatory variable for the changed work environment, and the execution of the work for each worker based on the change in the work environment. Acquire the fourth training data having the worker variable indicating the presence / absence and adding the abstraction variable.
In the calculation process, the processor inputs the fourth learning data regarding the worker to be predicted into the third prediction model, and thereby obtains the third predicted value of the work time for the worker to be predicted. Calculate for each,
A data analyzer characterized by this. The data analyzer according to claim 12.
In the calculation process, when the processor has a second predetermined number or more of operations indicating that the worker to be predicted has performed within a predetermined period in the worker variable of the fourth training data, the said processor. By inputting the fourth learning data regarding the worker to be predicted into the third prediction model, the third predicted value of the working time for the worker to be predicted is calculated for each work.
A data analyzer characterized by this. The data analyzer according to claim 6.
The processor
The objective variable and the explanatory variable are generated based on the first actual data having the worker who performed the work, the work time, and the work environment for each work, and the work is performed from the worker for each work. By generating a member variable, a conversion process for converting the first actual data into the first learning data is executed.
In the acquisition process, the processor converts the first actual data into the second actual data by recalculating the working time based on the change in the working environment of the first actual data, and the second actual data. Based on the actual data, the objective variable indicating the recalculated work time for each work, the explanatory variable for the changed work environment, and the execution of the work for each worker based on the change in the work environment. Acquire the fourth training data having the worker variable indicating the presence or absence and adding the worker characteristic variable.
In the calculation process, the processor inputs the fourth learning data regarding the worker to be predicted into the fourth prediction model, and thereby obtains the third predicted value of the work time for the worker to be predicted. Calculate for each,
A data analyzer characterized by this. A data analysis method executed by a data analyzer having a processor that executes a program and a storage device that stores the program.
The data analysis method is
The processor
Acquisition of first learning data having an objective variable indicating the work time, an explanatory variable related to the work environment, and a worker variable indicating whether or not the work is performed for each worker for each work of a plurality of works. Processing and
A prediction model generation process that generates a first prediction model that predicts the working time based on the first learning data acquired by the acquisition process.
By inputting the first training data into the first prediction model generated by the prediction model generation process, the first prediction value of the work time and the worker variable are the first prediction value for each work. A calculation process for calculating the first degree of influence, which indicates the degree of influence on the statistic, for each work, and
Based on the distribution of the first influence degree of each of the plurality of works for each worker, an abstract variable that abstracts the work speeds of the plurality of workers is generated for each work, and for each work, A training data generation process for generating a second training data having the objective variable, the explanatory variable, and the abstraction variable.
A data analysis method characterized by performing.

Images