JP2012068738A - Service time estimation device and service time estimation method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To estimate service time in consideration of chronological changes.SOLUTION: A service time estimation device is configured: to calculate an error between an estimated value and an actual value of the examination time for a patient who has been examined on the day (S301), specifically, to set a value obtained by [the actual value of the examination time/the estimated value of the examination time] as an error for a record of the patient who has been examined in examination time data; to perform timer-series analysis on an error of each patient to estimate an error on the examination time of the patient to be examined next (S302), specifically, to calculate the error of a patient to be estimated by use of a model expression obtained by the time-series analysis and set the calculated error as an error of a record of the patient to be estimated; and to calculate a new estimated value of the examination time for the patient to be examined next by the estimated value of the examination time obtained in advance in a neural network and the estimated error (S303), specifically, to calculate and set a new estimate value of the examination time by [the estimated value of the examination time×the error] of the record of the patient to be estimated.

Description

  The present invention relates to an apparatus and method for predicting service times using time series analysis.

  A technique for predicting the waiting time of a user that occurs in a hospital, a beauty salon, a restaurant, and the like has been developed. For example, Patent Literature 1 discloses a “waiting time prediction method, waiting time prediction system, and program” that predicts a hospital consultation time using a neural network and calculates a waiting time.

Japanese Patent No. 4,217,689

The patient's examination time on the day greatly depends on the number of patients and the disease name on the day. Therefore, in the method of Patent Document 1, in order to reflect such a situation on the current day in the predicted value of the examination time, an average value of the actual value of the examination time up to the current day on that day is used as one of the input data of the neural network. Added. However, in the neural network, even if the average value of the day's examination time is entered, it will not be immediately reflected in the predicted value. And the predicted value calculated by the neural network are combined using a coefficient α (0 ≦ α ≦ 1) (see paragraph [0093] of Patent Document 1).
Examination time = average value of examination time up to the current day x α + examination time output from neural network x (1-α)
However, since the actual examination time includes components that vary in time series (for example, due to changes in patient congestion, doctor fatigue, etc.) The time-series fluctuation component cannot be fully reflected in the predicted value of the examination time.

  The present invention has been made in view of the above problems, and its main object is to predict service time in consideration of time-series fluctuations.

  In order to solve the above problems, the present invention predicts a predicted service time, which is a time required for a user who receives a predetermined service, and a track record which is a time required for a user who actually receives the service. A service time prediction device for storing service time and predicting a service time for a user who next receives the service, the predicted service time and the results of a plurality of users who have already received the service Means for calculating an error from the service time; means for performing time-series analysis on each calculated error to predict an error relating to the next user; the predicted service time of the next user; Means for re-estimating a service time for the next user based on the error.

  The service time actually taken by the user includes time-series fluctuation components. According to this configuration, it is considered that the time series fluctuation component appears significantly in the error between the service time predicted without considering the time series fluctuation and the actual service time. By performing time series analysis on the error, it is possible to predict an error including a time-series fluctuation component regarding the next user. Then, the predicted service time value that does not include the time-series fluctuation component for the next user is corrected using the error prediction value that includes the time-series fluctuation component. According to this, it is possible to calculate a predicted value of service time including time-series fluctuation components, and to improve the accuracy of the predicted value.

  In the service time prediction apparatus of the present invention, the predicted service time may be a service time predicted by a neural network.

  Further, in the service time prediction apparatus of the present invention, an error between the predicted service time and the actual service time is a value obtained by dividing the actual service time by the predicted service time or the predicted service time, and the actual service. It may be a difference from time.

  The present invention includes a service time prediction method. In addition, the problems disclosed by the present application and the solutions thereof will be clarified by the description of the mode for carrying out the invention and the drawings.

  According to the present invention, the service time can be predicted in consideration of time-series fluctuations.

It is a figure which shows the hardware constitutions of the service time prediction apparatus. It is a figure which shows the structure of the examination time data 15A memorize | stored in the memory | storage part 15 of the service time prediction apparatus 1. FIG. It is a flowchart which shows the process of the service time prediction apparatus 1. FIG. It is a figure which shows the change of the difference | error of the predicted value of examination time, and a performance value. It is a figure which shows the example which calculates the error in the following time using the nearest neighbor method based on chaos theory.

  Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. The service time prediction apparatus according to the embodiment of the present invention calculates an error between the predicted value calculated by the neural network and the actual value on the day of the service time of each user who receives the service, and each of the users The next user's service is calculated by a neural network based on the predicted error by performing a time-series analysis on the error and predicting what the error related to the next user will be. It corrects the predicted value of time. According to this, since the predicted error includes a time-series fluctuation component, the time-series fluctuation component can be reflected in the predicted value of the service time of the next user.

  Below, the example which applied the service time prediction apparatus to the patient examination by the medical department doctor with a hospital is demonstrated.

≪Device configuration and overview≫
FIG. 1 is a diagram illustrating a hardware configuration of the service time prediction apparatus 1. The service time prediction device 1 includes a communication unit 11, a display unit 12, an input unit 13, a processing unit 14, and a storage unit 15, and each unit is connected so that data can be transmitted and received via a bus 16. The communication unit 11 is a part that performs IP (Internet Protocol) communication and the like with other devices via a network, and is realized by, for example, a NIC (Network Interface Card) or the like. The display unit 12 is a part that displays data according to an instruction from the processing unit 14, and is realized by, for example, a liquid crystal display (LCD). The input unit 13 is a part where an operator inputs data (for example, data of actual values of examination time), and is realized by, for example, a keyboard or a mouse. The processing unit 14 transfers data between each unit via a predetermined memory and controls the entire service time prediction apparatus 1. A CPU (Central Processing Unit) is stored in the predetermined memory. This is realized by executing the program. The storage unit 15 stores data from the processing unit 14 and reads the stored data, and is realized by a nonvolatile storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The Note that the service time prediction device 1 may be a stand-alone device (PC (Personal Computer) or the like), or may be a device (server or the like) that can communicate with a plurality of terminals via a network.

<< Data structure >>
FIG. 2 is a diagram illustrating a configuration of the examination time data 15 </ b> A stored in the storage unit 15 of the service time prediction apparatus 1. The examination time data 15A is data relating to the examination time of each patient on the day of the examination day. Patient number 15A1, examination start time 15A2, examination time predicted value 15A3, examination time actual value 15A4, error 15A5, and examination time new predicted value. Consists of records for each patient including 15A6. However, the record of the patient who has been examined (examined patient) includes patient numbers 15A1 to 15A5. Moreover, the record of the patient (prediction target patient) for which the examination time is predicted includes a consultation time predicted value 15A3, an error 15A5, and a new consultation time predicted value 15A6.

  The patient number 15A1 is a number unique to the patient, and is a sequential number starting from 1001 in FIG. The medical examination start time 15A2 is an actual value of the medical examination start time of the patient. The predicted examination time 15A3 is a predicted value of the examination time of the patient obtained by the neural network by the previous day (in advance). Note that the prediction of the examination time by the neural network is disclosed in, for example, [0072] of Japanese Patent No. 4194573 and paragraph [0080] of Japanese Patent No. 4217869.

  The examination time result value 15A4 is a result value of the examination time of the patient on the day of the examination day. The error 15A5 is an error between the predicted examination time value 15A3 and the actual examination time value 15A4 (actual error = actual examination time value 15A4 / predicted examination time 15A3) for the examined patient, and for the predicted patient. This is an error predicted from the actual error. The new consultation time predicted value 15A6 is a predicted value of a consultation time including a time-series fluctuation component related to a prediction target patient, and [an estimated consultation time 15A3 obtained by a neural network × an error 15A5 predicted from an actual error] Is calculated by

≪Device processing≫
FIG. 3 is a flowchart showing processing of the service time prediction apparatus 1. In this service time prediction apparatus 1, the processing unit 14 mainly predicts the consultation time of the next patient to be examined while referring to and updating the data in the storage unit 15.

  First, the service time prediction apparatus 1 calculates an error between the predicted value of the examination time and the actual value for the patient who has undergone the examination on the day (S301). For example, as shown in FIG. 2, for the records with patient numbers 15A1 of 1001 to 1010 in the examination time data 15A, the examination time actual value 15A4 is fixed, so [diagnosis time actual value 15A4 / consultation time prediction] Value 15A3] is set as error 15A5. At this time, if the predicted value matches the actual value, the error is 1. If the predicted value is larger than the actual value, the error is smaller than 1. If the predicted value is smaller than the actual value, the error is larger than 1. The patient error 15A5 will vary in the vicinity of 1. FIG. 4 is a diagram showing changes in the error 15A5.

  Next, the service time prediction apparatus 1 performs a time series analysis on the error 15A5 of each patient calculated in S301, and predicts an error related to the examination time of the patient who will be examined next (S302). Time-series analysis is a method for analyzing time-series data, which is a data point sequence that is measured at predetermined time intervals as time passes. By finding the law of time-series data, future predictions can be made. Is to do. For time series analysis, an autoregressive (AR) model, a moving average (MA) model, an autoregressive moving average (ARMA) model, an autoregressive integrated moving average model (ARIMA: Autoregressive Integrated) Moving Average) model, nearest neighbor method based on chaos theory, etc. are used. Here, the error of the prediction target patient is calculated using the model formula obtained by the time series analysis. Thereby, as shown in FIG. 2, the error 15A5 (predicted value of error) of the record of the patient to be predicted is calculated and set.

  Then, the service time prediction apparatus 1 calculates a new consultation time predicted value 15A6 based on the consultation time predicted value 15A3 obtained in advance by the neural network and the error 15A5 predicted in S302 for the patient to be examined next. S303). As shown in FIG. 2, a new consultation time predicted value 15A6 (new predicted value) is calculated and set by [predicted consultation time value 15A3 × error 15A5] of the record of the patient to be predicted.

<Example>
FIG. 5 is a diagram illustrating an example of calculating an error at the next time point using a nearest neighbor method based on chaos theory. Hereinafter, an example of the next error calculation procedure performed by the service time prediction apparatus 1 will be described with reference to FIG.

(S1) Let y (t) be the error at time t, and set a set of error y (t-2) before 2 time points and error y (t-1) before 1 time point in a row after time t = 3. create.
(S2) From {y (t-2), y (t-1)} = {1, 0.8} at the time point (next time point) t = 11 of the prediction target, Find the distance to {y (t-2), y (t-1)}. The distance is calculated by squaring the difference between each element in the set at time t = 11 and the square root of the sum.
(S3) Two time points (here, t = 9 and t = 10) having a small distance are selected, and the reciprocal of the distance is obtained.
(S4) For two time points with a small distance, the weight is obtained by the ratio of the reciprocal of the distance.
(S5) The predicted value of the error at the next time is obtained by the sum of “past error × weight”. Here, the error at the time t = 11 and the error at the time t = 11 is calculated by adding the error × weight at the time t = 9 and t = 10.

  Then, a new predicted value (about 7.5) is calculated by multiplying the calculated error (about 0.94) and the initial predicted value (8).

  In the above embodiment, in order to cause each unit in the service time prediction apparatus 1 shown in FIG. 1 to function, a program executed by the processing unit 14 is recorded on a computer-readable recording medium, and the recorded program The service time predicting apparatus 1 according to the embodiment of the present invention is realized by causing a computer to read and execute. In this case, the program may be provided to the computer via a network such as the Internet, or a semiconductor chip or the like in which the program is written may be incorporated in the computer.

  According to the embodiment of the present invention described above, the examination time predicted value 15A3 is based on a neural network, and time-series fluctuations are not taken into consideration. On the other hand, the examination time result value 15A4 has a time-series variation due to factors such as patient congestion and doctor fatigue. For example, when the patient is crowded, the examination time is shortened, and when there is a spare time for the next patient, the examination time becomes longer, and when the doctor gets tired, the examination time becomes longer. Accordingly, the error 15A5 calculated in S301 by the service time prediction apparatus 1 and the error 15A5 predicted in S302 include time-series fluctuation components. In S303, the predicted value of the examination time for the next patient that does not include the time-series fluctuation component is corrected by the prediction error that includes the time-series fluctuation component.

  Based on the above, it is possible to predict a patient's examination time in consideration of time-series fluctuations. According to this, when predicting a patient's waiting time, the precision of a predicted value can be improved.

<< Other embodiments >>
As mentioned above, although the form for implementing this invention was demonstrated, the said embodiment is for making an understanding of this invention easy, and is not for limiting and interpreting this invention. The present invention can be changed and improved without departing from the gist thereof, and equivalents thereof are also included in the present invention. For example, the following embodiments can be considered.

(1) In the above embodiment, an example in which the service time prediction device 1 is applied to a patient's examination by a doctor in a hospital has been described. However, the present invention is not limited to this, and other service providing facilities such as beauty salons and restaurants For example, the service time prediction apparatus 1 may be applied to services such as In the case of a hospital, the place where the service is provided is an examination room, but in the case of a beauty salon, it is a seat, and in the case of a restaurant, a table or a private room.

(2) In the above embodiment, a method using a neural network is applied as a method for calculating an initial predicted value. However, if a method that does not consider a time-series fluctuation component of the examination time is used, another method is used. You may use the predicted value by a method. For example, a statistical method of calculating an average value or a mode value of past performance data may be used.

(3) In the above-described embodiment, the difference between the predicted value of the consultation time for the 10 patients who have already been examined and the actual value of the consultation time for the prediction target patient among the patients on the day of the consultation date. However, for example, the examination time for the first patient on the current day is predicted using an error related to the patient on the previous day.

(4) In the above embodiment, the error is a value obtained by dividing the actual value of the examination time by the predicted value. However, another calculated value may be used as long as the value indicates an error. For example, the difference between the predicted value and the actual value may be used.

1 Service Time Prediction Device 14 Processing Unit 15 Storage Unit 15A Examination Time Data 15A3 Predicted Examination Time (Predicted Service Time)
15A4 Examination time results (actual service time)
15A5 Error 15A6 Examination time new predicted value (service time)

Claims (6)

Predicted service time that is predicted in advance for a user who receives a predetermined service and an actual service time that is time that is actually spent for the user who has received the service are stored, and then the service is stored. A service time prediction device for predicting service time for a receiving user,
Means for calculating an error between the predicted service time and the actual service time of a plurality of users who have already received the service;
A means to perform a time series analysis on each calculated error and predict an error for the next user;
Means for re-estimating the service time for the next user based on the predicted service time of the next user and the predicted error;
A service time prediction apparatus comprising: The service time prediction device according to claim 1,
The service time prediction device, wherein the predicted service time is a service time predicted by a neural network. The service time prediction apparatus according to claim 1 or 2,
The error between the predicted service time and the actual service time is
A service time prediction device, wherein the service time prediction is a value obtained by dividing the actual service time by the predicted service time or a difference between the predicted service time and the actual service time. Predicted service time that is predicted in advance for a user who receives a predetermined service and an actual service time that is time that is actually spent for the user who has received the service are stored, and then the service is stored. A service time prediction method using a computer for predicting service time for a receiving user,
The computer
Calculating an error between the predicted service time and the actual service time of a plurality of users who have already received the service;
Performing a time series analysis on each calculated error to predict the error for the next user;
Re-estimating the service time for the next user based on the predicted service time of the next user and the predicted error;
The service time prediction method characterized by performing. The service time prediction method according to claim 4,
The predicted service time is a service time predicted by a neural network. The service time prediction method according to claim 4 or 5,
The error between the predicted service time and the actual service time is
A service time prediction method, wherein a value obtained by dividing the actual service time by the predicted service time or a difference between the predicted service time and the actual service time.

Images