JP2004151790A - Nurse wopk schedule preparation support software or the like - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide software capable of fulfilling a complicate evaluation rule, and preparing a nurse work schedule in a relatively short period of time without getting into any local solution. <P>SOLUTION: This nurse work schedule preparation support software comprises a step 10 for displaying a picture for promoting the input of an evaluation rule including a skill level/service series/daily allocated service/required number of persons associated with each nurse, a step 12 for carrying out automatic calculation by a computer to fulfill the evaluation rule as much as possible, and for displaying a temporary solution on the picture after predetermined end conditions are fulfilled, and a step 14 for promoting a user to manually correct the temporary solution and a step 18 for updating the evaluation rule. Then, the user ends a work when it is decided that the outputted a work schedule candidate is finally prepared one in a step 16. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to nurse work schedule creation software and the like.
[0002]
[Prior art]
The problem of creating a work schedule (nurse worktable) for a plurality of nurses (Nurse Scheduling Problem: NSP) is a complex combination optimization problem including a plurality of evaluation rules (constraint conditions). In general, nurse work schedules are created monthly by the nurses in each ward, and creating such a schedule requires a lot of time and effort, which is a great burden for busy nurses. To solve this problem, a work schedule creation system using a genetic algorithm (Genetic Algorithm: GA) (Yamamura, Kobayashi, Yamagishi, Ase, "Nurse Scheduling by Genetic Algorithm" by Hiroaki Kitano Genetic Algorithm pp 89-125, 1993).
[0003]
However, this conventional system employs a fully automatic algorithm for the purpose of reducing the human burden on the nurse. For this reason, there is a problem in that it is rather easy to fall into a local solution, and in order to obtain a solution that is satisfied by the nurse, it is necessary to start over from the beginning many times. And issues "Nursing Management Vol.3 No.1 pp38-42 1993).
[0004]
Thus, the present inventors have examined the above problems, and as a result, an interactive nurse in which a user and software cooperate to create a work table while the user is involved in the solution search process. Proposal of a work schedule creation support software and a system including this software (Inoue, Furuhashi, Maeda, Takaba "A Study on EA Engine for Nurse Scheduling Support System Using Interactive EA" The 16th Fuzzy System Symposium (Akita), September 6-8, 2000, Lecture Book pp619-622). This software is not a fully automatic algorithm, but a work schedule is presented to a user (eg, a nurse) during the solution search, and the user corrects and / or modifies the work schedule data. Alternatively, while making a fixed part of the work, a work schedule required by the user is created. For this reason, even if the evaluation rules are not sufficiently satisfied, the work schedule can be created in cooperation with the user and the software. When the work schedule is revised, the evaluation rules can be modified. The Bacterial Evolutionary Algorithm (BEA) used in this software is effective for improving work styles consisting of a short number of days (for example, three to five days). However, there was room for ingenuity in allocating the workload of each nurse equally over a month.
[0005]
[Non-Patent Document 1] Yamamura, Kobayashi, Yamagishi, Ase "Nurse Scheduling by Genetic Algorithm" by Hiroaki Kitano Genetic Algorithm pp. 89-125, 1993
[Non-Patent Document 2] Takemoto, "Current state and issues of creating a nurse work schedule using a computer", Nursing Management Vol. 3 No. 1 pp38-42 1993
[Non-Patent Document 3] Inoue, Furuhashi, Maeda, Takaba "A Study on EA Engine for Nurse Scheduling Support System Using Interactive EA" The 16th Fuzzy System Symposium (Akita), September 6-8, 2000, Lecture Book pp619-622
[0006]
[Problems to be solved by the invention]
The present invention has been made in view of the above circumstances, and an object thereof is to create a nurse work table in a relatively short time without falling into a local solution while satisfying a complicated evaluation rule. To provide possible software.
[0007]
[Means for Solving the Problems]
The inventors of the present invention have proposed a solution to the problem that it is difficult to evenly assign a nurse work table over a long period of time (for example, two weeks or more, about one month) using only bacterial evolutionary calculations. We worked diligently. As a result, (1) a cross-over operation is performed on a group of appropriately generated parent work tables (sometimes referred to as “parent individuals”), and (2) a child work table ( It has been found that the intended purpose can be achieved by applying mutation and / or heuristics to the vicinity of the crossover point of "child offspring." Was completed. In addition, in order to prevent further local solutions, fairness could be improved by generating mutations in the entire work schedule.
[0008]
That is, according to the study of the present inventors, the effect of equalizing the monthly work load of each nurse was expected by combining the crossover operation with the mutation for the entire work table. Nevertheless, there still remains a problem that the work form before and after the crossing point is largely destroyed by the crossover operation. Therefore, the present inventors further introduce a heuristic using the nature of the nurse work table creation problem to repair the crossover point, thereby making it possible to apply the crossover operation and avoiding a local solution to the entire work table. An efficient search method can be provided even for the evaluation item "fairness of work load" that requires a search.
[0009]
In order to solve the above-mentioned problems, the first invention is to create a work schedule including a plurality of nurses' work processes over a long period of two weeks or more.・ Steps to display a screen prompting the user to enter evaluation rules including work series, assigned work on each day of the week, and the required number of people; (2) a step of automatically calculating by a computer so as to satisfy the evaluation rules as much as possible; and (3) an automatic calculation. Displaying a provisional solution on a screen after satisfying a predetermined termination condition, (4) a step of prompting a user to manually correct the provisional solution, and (5) a step of updating the evaluation rule. Wherein the (2) step includes: (a) a crossover step of crossing two randomly generated parent work tables; and (b) a predetermined neighborhood including the crossed points. The step of regenerating a slave roster the part by mutation and / or heuristics, characterized in that it comprises a step of replacing a child roster the inferior Pareto parent roster compared to (c) said child roster.
The step (2) further includes the step of (d) mutating the whole if the evaluation value is not improved after repeating the steps (a) to (c) a predetermined number of times. It is preferable to include
[0010]
<Nurse work table creation problem>
The nurse work schedule creation problem (NSP) is based on the constraints such as work constraints, quality of nursing, and requests from each nurse, etc., and the "day shift" (AM 8:00 to PM 4:00), "semi-night shift" ( This is a combination optimization problem in which three work modes, PM 4:00 to AM 0:00) and “midnight shift” (AM 0:00 to AM 8:00), are assigned to each nurse. FIG. 1 shows an example of a specific work schedule. This work schedule shows work schedules for 14 days for 10 nurses (Staff A to J). The symbols “day” in the table indicate “day shift”, “quasi” indicates “semi-night shift”, and “deep” indicates “midnight shift”. In addition, blanks in the work schedule indicate holidays. The leftmost column shows the level of skill of each nurse, and is ranked from A with high skill to C with little experience. The lower three rows show the number of nurses assigned to each work style on each day. The four columns on the right represent the number of days assigned to each nurse for each job.
[0011]
The work schedule evaluation rules include (1) work series, (2) assigned work on each day of the week, (3) required number of workers, and (4) assignment days. (1) When working on a single nurse, the work day spans multiple days, such as “day shift” to “late night shift” or “day shift” to “off” to “semi-night shift”. It means a series about work style. This work sequence should be reasonable for the health of the nurse. {Circle around (2)} The rules regarding the assignment work on each day of the week are restrictions that reflect the specialty of nurses, such as “Nurse A does not assign quasi-night work on Saturday”. {Circle around (3)} The required number is a constraint to maintain the quality of nursing, for example, one of the two late-night shifts must be assigned to a skilled nurse. (4) The number of days allocated is basically to be allocated evenly among nurses, but it is necessary to give a fair perspective from a broad perspective, taking into account the circumstances of each nurse and the number of days allocated last month. . Furthermore, in the field, there are cases where detailed restrictions such as compatibility between nurses and sudden schedule changes due to accidents are added, so that a flexible response to evaluation rules is required.
[0012]
<Nurse work schedule creation support software>
The inventors tried to solve the following three problems. {Circle around (1)} The evaluation rules for the nurse work schedule creation problem differ from hospital to hospital, and the user (eg, nurse) must define all the evaluation rules necessary for automatic creation of the work schedule. When the work schedule was manually created as in the past, the evaluation rules were often not organized, so that the evaluation rules were likely to be omitted. (Problem 1. Difficulty). (2) In the automatic calculation algorithm of the work schedule, the user tends to fall into a local solution, so that the user has to repeat the scheduling from the beginning many times in order to obtain a satisfactory solution (Problem 2. Solution search). Difficulties). {Circle around (3)} Since it takes a long time to search for a solution, a long calculation waiting time can be a psychological burden for busy users (especially nurses) (Problem 3. Length of calculation time).
In order to address these problems, an interactive work schedule creation support system shown in FIG. 2 is proposed. The work schedule creation by this system can be performed in the following procedure, for example, as shown in FIG.
[0013]
First, the user identifies evaluation rules as much as possible and inputs them to the system (Step 10. Input evaluation rules). Next, automatic scheduling is performed under the defined evaluation rules. However, before giving a psychological burden to a user waiting for the presentation of a solution candidate, the search is terminated and a provisional solution is presented (step 12. Mutation (“bacterial evolutionary”). Calculation (BEA) ”) is used to generate a work schedule candidate). That is, in executing the algorithm of the present invention, it is preferable in many cases that the termination determination at the time of execution of the automatic calculation is given by the number of repetition calculations (or it can be determined by the time during execution of the automatic calculation. In this case, the user can wait at most one to two minutes for displaying the result of the computer during the automatic calculation, so the number of repetitions of the automatic calculation or the automatic calculation time depends on the specifications of the computer used. It is practical to set an end condition, which can be understood by considering the size of the work calculation table.For example, to create a work table for 10 nurses for 30 days, There are four types of assignments to each cell: day shift, quasi-night shift, night shift, and holiday. ^10x30 A work schedule group of streets can be created. From these vast work schedule groups, those that satisfy the evaluation rules will be created.
[0014]
As described below, the automatic calculation methods include (1) crossover operation between parent individuals, (2-1) heuristics near the intersection, (2-2) mutation near the intersection, and (3) ) Mutations in the entire individual can be used in any suitable combination.
In addition, when presenting a provisional solution, for a problem part that does not satisfy the evaluation rule, a background color different from other parts that satisfy the evaluation rule may be used so that the user can easily understand the problem part. preferable. At this time, if there are two or more stages in the evaluation rule, it is preferable to use different background colors (for example, as described later, a portion that does not satisfy the evaluation rule of absolute prohibition is “red”, A background color "yellow" can be used for a part that does not satisfy the evaluation rule of prohibition, and a blue color can be used for a part that satisfies the evaluation rule.) Since there is a trade-off between the quality of the solution and the operation time, it is necessary to employ an efficient search algorithm in order to shorten the operation time. Bacteria EA which is effective in searching for a local part (for example, 3 to 5 days) can be employed for searching for the nurse work schedule creation problem.
[0015]
Next, the user makes a manual correction directly to the solution candidate presented by the system (step 14. Correction and partial fixation of the work schedule candidate). In step 14, the user can perform operations such as changing or replacing part of the work schedule data. At this time, the solution candidates are not unique, and a plurality of solution candidates can be presented in parallel. Further, if there is a part that is partially satisfactory for a specific solution candidate, the part can be fixed and the recalculation part can be narrowed down.
[0016]
Here, the user makes a judgment on the output work table candidate (step 16), and terminates the work if it is determined to be final. If the user notices an evaluation rule that has not been noticed in step 10, the evaluation rule is updated (step 18: update of the evaluation rule), and a correction is made in step 14 under the updated evaluation rule. Using the completed work schedule as an initial individual, the process is executed again from step 12 (recalculation of an unfixed portion). At this time, in step S14, the part designated by the user to be fixed is excluded from the mutation.
Hereinafter, steps 12 to 18 described above are repeated until the user obtains a satisfactory work schedule. The finally completed work schedule is used in step 10 when the next work schedule is created.
[0017]
<Combination and Heuristics>
The problem 3 is the most important problem for the smooth cooperation between the system and the user. That is, in step 12, it is desired to quickly derive good-quality solution candidates in order to reduce the burden of the user's manual work. Here, since the quality of the solution and the operation time are in a trade-off relationship, an efficient search algorithm is required. The present inventors applied bacterial EA (BEA), which is effective for local search, to search for a nurse work schedule creation problem. BEA is an algorithm that refers to the evolutionary process of bacteria, and is effective in improving local portions of chromosomes. In the problem of creating a nurse work schedule, since there are many evaluation rules related to work series on two to four consecutive days, the dependency between separate days is low. For this reason, the bacterial operation of BEA, which aims to intensively improve the solution for each local portion of the chromosome, was effective. However, in the work schedule, "fairness in the number of assignment days" is required to equalize the work load of each nurse for one month. Thus, the present inventors have attempted to combine normal crossover with heuristics utilizing the nature of the nurse work schedule creation problem in order to quickly derive a good quality solution.
[0018]
The present inventors have focused on crossover in evolutionary computation in order to quickly present good-quality solution candidates. In this method, in order to perform the crossover efficiently, crossover operation and heuristics are combined to suppress the deterioration of the work sequence and to guarantee the diversity of the chromosome group by Pareto optimal solution search. Next, details of the method will be described.
[0019]
<Pareto optimal solution search>
Next, the Pareto optimal solution search will be described. The reason for performing the Pareto optimal solution search is that (1) diversity is given to the work table candidates presented to the user, and (2) individuals similar to each other at the time of crossover are formed by giving diversity to chromosome groups. There are two points to keep from becoming. In this method, a Pareto optimal solution search is performed using two evaluation axes, ie, the number of assignment days and the other evaluation items that are in a trade-off relationship. However, there are "absolute restrictions" that must be satisfied in the evaluation rules and "compromise restrictions" that may be compromised in view of the number of working days. Therefore, an absolute constraint is applied to both axes, and the evaluation function is defined as in the following equations (1) to (5).
[0020]
H ₁ = A + B (1)
H ₂ = A + C (2)
A = 1−1 / (1 + a) (3)
B = 1−1 / (1 + b) (4)
C = 1−1 / (1 + c) (5)
[0021]
Here, a is the number of absolute constraint violations, b is the number of compromise constraint violations, and c is the sum of absolute values of differences from the average number of working days. The constraint violation number means a number that does not satisfy the absolute constraint or the compromise constraint, and is added by +1 when the assignment in the work schedule violates some evaluation rule. H ₁ , H ₂ If no excellent individual exists in any of the above, the individual is determined to be a Pareto optimal individual. However, when the individual in the subchromosome group is Pareto-optimal, and there is no subpareto-optimal individual in the chromosome group, a preserved individual is selected according to the evaluation function of the following equation (6).
H = A + B + C (6)
[0022]
<A combined method of GA and heuristics>
Each evaluation item in the above-mentioned nurse work schedule creation problem influences each other in performing optimization by evolutionary calculation, which makes solution search difficult. Evaluation items other than the number of allocated days have low dependence between days separated from each other. Therefore, it is effective to use a technique in which mutations are concentrated on one part and a local part is searched, as in BEA, instead of performing mutation evenly on the entire chromosome. Searching for a local part by BEA is equivalent to the work of the work schedule creator, who limits the work schedule to a certain period (for example, several days of 3 to 5 days), changes the work assignment among nurses, and performs trial and error. I do. On the other hand, regarding the number of days assigned, it is necessary to consider the work load of each nurse in a month to be equal. Low effect. To solve this problem, both mutation and crossover effects on the entire chromosome are expected. However, when crossover is performed, the work sequence is interrupted before and after the crossover point, so that the evaluation value is significantly deteriorated. Therefore, the present inventors propose the following method as a method for performing a solution search while repairing the deterioration of the work sequence due to the crossover operation.
[0023]
A crossover operation is performed, and heuristics at the time of manual creation are used to repair the vicinity (for example, 3 to 5 days) of the crossing point. In addition, since this method has the risk of falling into a local solution, we decided to add more mutations. Next, an example of the algorithm of the method will be described with reference to FIGS.
[0024]
First, N initial populations are generated (step 20). Here, it is determined whether or not the entire mutation is required (step 21). In the following examples, when the methodology is examined, the method may be performed separately from the method including mutation to the whole (Yes direction in step 21). In the case of YES in step 21, the process proceeds to "A" to execute a mutation algorithm for the whole (described in detail later). On the other hand, in the case of NO (or in the case where the whole mutation algorithm is performed separately), two parent individuals are randomly selected from the chromosome group, and one new child individual is generated by one-point crossover. (Step 22). Also, in step 21, after performing the whole mutation algorithm, the algorithm may return from "B" to the heuristic algorithm again (described later). Thus, heuristics and global mutations can be used in combination with each other.
Next, a predetermined L _BM For the local part of the day (the length of re-creation centering on the intersection of the parent individual), the re-creation location of the child individual is recreated by heuristics (step 24: details of the heuristics will be described later). Next, for each day within the re-creation range of the offspring individual, the probability P _BM / L _BM (Step 26). Here, the evaluation value of the child individual is calculated and compared with the evaluation value of the chromosome group. If the child individual is a Pareto-optimal individual, one individual is randomly selected and replaced from among the poorly Pareto individuals of the chromosome group. (Step 28). It is determined whether or not the recalculated chromosome group satisfies the termination condition (step 30). If the result is YES, the operation is terminated. If the result is NO, steps 21 to 28 are repeated.
[0025]
<Heuristics modeling manual creation>
Next, heuristic procedures will be described. In this method, the task of sequentially determining the unassigned part of the work schedule was modeled. In the case where the assignment is manually performed one by one, the assignment is determined from a place where there are few selectable work modes due to restrictions such as the required number of employees and work series. In addition, when manually creating a work schedule, usually, priority is assigned from the earliest date. For this reason, the assignment priority of each nurse was determined for each date. First, the parameters used for the analysis will be described.
[0026]
The parameter L (s) means the nursing skill rank of the nurse s. The parameter W (d, s) indicates the contents of the assignment of the date d to the nurse s in the work schedule in the process of preparation, and the assignment of S (d, s) is based on the work form n∈ ｛“day shift”, “quasi-night shift”. ", Etc. and unassigned: Take the state of φ. The parameter R (d, n, l) indicates the number of insufficient personnel by rank with respect to the required number of nursing skills rank 1 in the work form n on the date d (however, if l = # (don't care), Insufficient number of personnel required). The parameter V (d, s, n) indicates a work sequence when the assignment to the nurse s on the date d is the work form n, the assignment work for each day of the week, and the assignment determination for the evaluation rule of the required staff. The parameter M (d, n) indicates the number of nurses who can satisfy the required staff restriction of the work mode n on the date d. The parameter U (d, s, n) indicates the priority for the assignment work with the work form n as the assignment to the nurse s on the date d.
Here, the assignment determination V (d, s, n) will be described. For V (d, s, n), the following judgment is made in accordance with each evaluation rule of work series, assigned work on each day of the week, and required staff.
[0027]
In the assignment determination, V (d, s, n) = 0 means absolute prohibition, and requires the violation of the absolute constraint of the work series, the violation of the assignment constraint on each day of the week, and the total number of employees or the number of employees by rank. In some cases, the number of employees has been exceeded. Also, V (d, s, n) = 1 means no prohibition of compromise, does not satisfy the condition of V (d, s, n) = “absolute prohibition”, violates the compromise constraint of the work series or assigns each day of the week. There have been cases where one of the work compromise restrictions has been violated. Also, V (d, s, n) = 2 means that allocation is possible, and a constraint violation may occur for any of the evaluation rules of the work series, the assigned work of each day of the week, and the evaluation rules of the required staff. If not.
[0028]
Next, a method of determining the priority U (d, s, n) will be described. U (d, s, n) = 0 means that allocation is not possible, which is a case where the condition {W (d, s) z {φ} is satisfied. U (d, s, n) = 1 means the lowest priority, and is a case where the condition {W (d, s) = φ and V (d, s, n) = 0} is satisfied. U (d, s, n) = 2 means that the priority is low, and is a case where the condition {W (d, s) = φ and V (d, s, n) = 1} is satisfied. U (d, s, n) = 3 means that the priority is medium, and the condition ｛W (d, s) = φ and V (d, s, n) = 2 and R (d, n, # ) = 0 and R (d, n, L (s)) = 0}. U (d, s, n) = 4 means high priority, and the condition ｛W (d, s) = φ and V (d, s, n) = 2 and (N (d, n) > 0 or R (d, n, L (s))> 0)}. U (d, s, n) = 5 means the highest priority, and the condition ｛W (d, s) = φ and V (d, s, n) = 2 and (N (d, n)> 0 or R (d, n, L (s))> 0) and (M (n) ≦ R (d, n, #) or M (n) ≦ R (d, n, L (s)))} Is satisfied.
Heuristics are assigned using the above-mentioned parameters. An example of the algorithm will be described with reference to FIG. This algorithm can be used, for example, during step 24 described above.
[0029]
First, the recreated portion is left blank as unallocated (step 40). Next, the earliest date on which unassigned data exists is selected and assigned as the assigned date D (step 42). Next, R (D, s, #) and R (D, s, l) on the date D are calculated (step 44). Next, the priority U (D, s, n) is calculated for all the nurses s and the work form n on the date D (step 46). Next, the maximum value U of the priority U (D, s, n) on the date D _max Is calculated (step 48). Further, on the date D, the priority is the condition {U (D, s, n) = U _max The minimum value of the number of nurses s that satisfies｝ is calculated, and the work form that takes the minimum value is assigned as work N (step 50). At this time, if there are a plurality of work forms that take the minimum value, one is randomly selected from these work forms and assigned as work assignment N.
[0030]
Next, on the date D, the priority is the condition {U (D, s, N) = U _max One is randomly selected from the nurses s satisfying｝, and is set as an assigned nurse S (step 52). Next, the assignment work N is assigned to the assignment W (D, S) in the work table (step 54). Here, it is determined whether or not unallocated data exists (step 56). If no data is allocated, the operation is terminated. If not, steps 42 to 54 are repeated.
[0031]
<Method using bacterial mutation for repair near crossing point>
For repairing the destruction of the work series due to crossover, a method of applying a bacterial mutation that concentrates mutations near the crossover point can be considered. An example of the algorithm will be described with reference to FIGS.
[0032]
First, after the chromosome group is initialized (step 59), it is determined whether or not a mutation is required for the whole (step 60). Here, in the case of Yes, a mutation algorithm for the entire chromosome described below is performed (in the following examples, when the methodology is examined, the method includes a method including mutation to the whole (step 60). Yes direction).
If No in step 60, two parent individuals are randomly selected from the chromosome group, and one new child individual is generated by one-point crossover (step 62). Next, the length L around the intersection of the parent individual _BM The local part of the day is selected as the local improvement range, and _C The individual chromosomes are multiplied to generate a subchromosome group (step 64). Next, one individual is randomly selected from the subchromosome group, one copy is generated, and the probability P for each day within the local improvement range of the selected individual is determined. _BM / L _BM (Step 66).
[0033]
Next, the mutant individuals are evaluated. That is, this individual is compared with an individual selected at random from the subchromosome pool, and individuals with good evaluation values are left, and inferior individuals are deleted (step 68). Here, steps 66 to 68 are set to N _ST This is repeated twice (step 70). Next, the evaluation value of the individual in the sub-chromosome group is compared with the evaluation value of the individual in the chromosome group, and the Pareto-optimal individual is replaced with a randomly selected one among the poorly Pareto individuals of the chromosome group (step 72). Steps 62 to 72 are repeated until the termination condition is satisfied (step 74).
[0034]
<Mutation to whole chromosome>
The search for fairness in working days may improve search efficiency by uniform mutations throughout the chromosome. In particular, when the diversity of the chromosome group is lost, the effect of the crossover operation is lost, and the solution search may be stagnated. Therefore, the selection operation (step 28 and step 72) of the parent individual in the above-described algorithm is performed a predetermined number of times (N _XT If the pareto-optimal offspring is not added to the chromosome group as a new parent individual during this time, and the evaluation value does not improve, a search is performed by mutation to the entire chromosome instead of crossover. . An example of an algorithm for that will be described with reference to FIG. The algorithm requires a global mutation in step 21 or step 60 (ie, a predetermined number of times (N _XT It can be carried out when it is determined that the evaluation value is not improved even by the selection operation of ())).
[0035]
First, it is determined whether or not the entire mutation is necessary (step 78), and in the case of Yes, the algorithm of the following steps 80 to 90 is executed. On the other hand, if it is determined in step 78 that mutation to the whole is not necessary, the process proceeds to “B” and proceeds to step 22 (<combined method of GA and heuristics>) or step 62 (<crossing point Approach using bacterial mutation for repair of the neighborhood>). That is, the present method can be performed by combining (1) a combined method of GA and heuristics, or (2) a method using bacterial mutation for repair near the crossing point.
If Yes in step 78, a parent individual is randomly selected from the chromosome group (step 80), and the parent individual is copied to _C The sub-chromosomes are multiplied into individual chromosomes (step 82). Next, one individual is randomly selected from the subchromosome group, one copy is generated, and the probability P _BM A uniform mutation is performed at / (1 month) (step 84). The mutant individuals are then evaluated. That is, this individual is compared with an individual randomly selected from the subchromosome pool, and individuals with good evaluation values are left, and inferior individuals are deleted (step 86). Next, step 84 to step 86 _LT This is repeated twice (step 88). Then, the evaluation value of the individual in the sub chromosome group is compared with the evaluation value of the individual in the chromosome group, and if there is a Pareto-optimal individual, it is replaced with an inferior Pareto individual in the chromosome group (step 90). When the termination condition is satisfied, the algorithm is terminated (Yes in step 92), and when the termination condition is not satisfied, the process returns to step 78 (No in step 92).
[0036]
BEST MODE FOR CARRYING OUT THE INVENTION
Next, an embodiment of the present invention will be described in detail with reference to the drawings, but the technical scope of the present invention is not limited by the following embodiment, without changing the gist, Various modifications can be made. Further, the technical scope of the present invention extends to an equivalent range.
Next, a nurse work table was actually created by using the above-mentioned various algorithms alone or in combination.
[0037]
<Question setting>
The example of a ward with 20 nurses was used to create a nurse work schedule. The evaluation rules used at that time are shown below.
(1) Rules related to work schedule
What are the absolute and compromise restrictions on the rules regarding work series? Each is as follows.
First, the absolute constraints are: (1) Do not assign “midnight shift” after “quasi-night shift”, (2) Do not assign late night shift for more than 4 consecutive days, and (3) sandwich midnight shift and late night shift. No assignment other than late-night work, and (4) continuous work for more than 6 days is prohibited.
In addition, as a compromise constraint, (1) do not assign "midnight shift" after "rest", (2) do not assign "quasi-night shift", "day shift", "midnight shift" in this order, and (3) day shift. Leaders are assigned for three or four consecutive days, and (4) midnight shift is not assigned for three consecutive days.
[0038]
(2) Assignment rules for each day of the week
Absolute constraints and compromise constraints on the assignment rules for each day are as follows.
As absolute constraints, there are three points: (1) Stafff J is a day shift and a day shift leader only, (2) Stafff S is a weekend break, and (3) Stafff T is a weekend break.
Also, there are two compromise restrictions: (1) Staff A is not assigned except for day shifts, and (2) Staff A is holiday off.
[0039]
(3) Rules on the required number of people
The absolute restrictions on the rules regarding the required number of people are as follows.
Absolute restrictions are as follows: (1) Do not assign more than 2 rank C nurses to “Night shift”, (2) Do not assign more than 2 rank C nurses to “Mid night shift”, (3) “Near shift” One or more nurses of rank A are assigned to "night shift"; (4) One or more nurses of rank A are assigned to "midnight shift"; and (5) A leader of day shift is assigned rank A or B. The five points.
[0040]
(4) Rules regarding the number of days allocated
In addition to the above rules, (1) rules other than Stuff A and Stuff J are allocated as fairly as possible, with respect to the number of days to be allocated.
Next, parameters used in each experiment are shown. Local search / recreate width is L _BM = 4, the number of individuals in the chromosome group is N = 8, and the mutation parameter is P _BM = 0.05 and the number of individuals in the subchromosome group N _C = 2 and the number of local improvements N _ST = 100 and search change confirmation cycle N _XT = 50 and the number of global mutations N _LT = 100 and the respective numerical values were set.
[0041]
Where L _BM Is assumed to change when the solution search is stagnant. Each time the state where no new Pareto optimal individual is added to the chromosome group in the selection operation of the parent individual continues for 10 times, L _BM Is increased by 1, and when a new Pareto optimal individual is added to the chromosome group, the value is returned to the original value.
[0042]
<Example 1 A combined method of GA and heuristics>
The effect of the combined method of GA and heuristic described above was examined. The effect of the mutation in step 26 was also examined. (1) When only heuristics are used after crossover (when step 26 is skipped and executed), (2) When heuristics and mutation of step 26 are used after crossover, and (3) Step without crossover 24 L _BM When the local part of the day was set at random and the heuristics and the mutation in step 26 were performed, the obtained nurse work management tables were compared for each case. For comparison of each method, 50 operations were performed until the number of selection operations of the parent individual was 5,000, and the average of the evaluation values of the best individual according to equation (6) in the chromosome group was obtained. Changes in the evaluation value were examined. However, the determination in step 21 was not performed, and the uniform mutation to the entire chromosome shown in FIG. 9 was not performed.
[0043]
FIG. 10 and FIG. 11 show display screens of software used in each embodiment. This software is provided with three types of menu tabs 15, each of which is a menu screen (FIG. 10) for executing a work table creation, a menu screen for setting a work table (FIG. 11), and an extraction rule setting screen. ing.
[0044]
Of these, 20 nurses (Staff_A to Staff_T) are shown in the left column 10 of the work creation execution menu screen (FIG. 10), and to the right there is approximately one month's worth (on the screen illustrated). Shows the work process matrix data 11 for (37 days). The lower column 12 of the data 11 shows the number of day shift staff, the number of quasi-night shift staff, the number of late night shift staff, and the number of leaders for each day. Further, the right column 13 of the data 11 shows the number of holidays, the number of day shifts, the number of quasi-night shifts, the number of late night shifts, and the number of leaders per month for each nurse.
[0045]
The work schedule setting menu screen (FIG. 11) is roughly divided into five frames 16 to 20. In the leftmost frame 16 among them, settings relating to each staff (for example, skills, adaptation of leaders, fairness of work assignment, maximum number of interlocks) are made. In the upper center frame 17, individual prohibited work restrictions for specific staff (for example, (1) cannot work on a specific day of the week (or only work on a daily basis)), leaders can be assigned for four days. The data is a combination of the content such as "impossible for the last night shift for 4 consecutive days" and (2) the priority of the constraint.). In the upper right frame 18, the upper limit number and the lower limit number of persons having predetermined skills (specified for each staff in the left frame 16) are set for day shift, semi-night shift, late night shift, and leader. In the lower frames 19 and 20, a minimum number of assignments and a maximum number of assignments are set for day shift, semi-night shift, late night shift, leader, and rest on each day of the week. In this manner, predetermined calculations and manual processing are performed on the menu screen for executing work creation based on the evaluation rules set on the menu screen for setting work schedule.
[0046]
FIG. 12 shows the transition of the improvement value of the total evaluation value of the equation (6), FIG. 13 shows the transition of the improvement value of the evaluation value only for fairness of the working days, and FIG. The change of the improvement value of the evaluation value except fairness was shown. From these results, it was found that crossover was effective in improving the evaluation value of fairness, and mutation was effective in improving the evaluation value excluding fairness.
[0047]
<Example 2 Effect of heuristics>
A comparison was made between the case where heuristics in Steps 20 to 30 were used for repairing the disruption of the work sequence after crossover and the case where bacterial mutations in Steps 60 to 74 were used. In the algorithm of Steps 20 to 30, the mutation of Step 26 was not used in order to confirm the effect of purely heuristics. In addition, the judgment in step 21 was not performed, and the uniform mutation to the entire chromosome shown in FIG. 9 was not performed.
[0048]
In each case, 50 operations are performed until the number of selection operations of the parent individual is up to 5,000, and the average of the evaluation values of the best individual according to Expression (6) in the chromosome group is obtained. The transition of the improvement value was examined. However, uniform mutation (steps 78 to 90) to the entire chromosome was not performed. Note that the local improvement frequency N, which is an experimental parameter, _ST Is calculated so that the operation time required to perform 5,000 selection operations of the parent individual is substantially equal in each case. _ST = 100. FIG. 15 shows the transition of the improvement value of the evaluation value from the initial individual.
As a result, it was shown that heuristics had better search efficiency than bacterial mutations.
[0049]
<Example 3 Effect of Mutation on Whole>
Next, the effect of the mutation on the entire chromosome was examined. After the crossover, the effect of the use of the heuristics and the mutations in Steps 20 to 30 together with the mutation was examined for the effect of the presence or absence of the uniform mutation on the entire chromosome in Steps 78 to 92. The chromosome evaluation calculation is performed 50 times up to 5,000 times, the average of the evaluation values of the best individual according to the formula (6) is obtained in the chromosome group, and the transition of the improvement value of the average evaluation value from the initial individual is calculated. Examined. However, when performing mutation to the whole, the mutation to the whole is performed in consideration of the fact that the evaluation operation time is about 20 times faster than when performing crossover using heuristics. In this case, the number of evaluation calculations was changed from 20 to 1, once.
[0050]
FIG. 16 shows the transition of the improvement value of the overall evaluation value of the equation (6), FIG. 17 shows the transition of the improvement value of the evaluation value of only the fairness of the working days, and FIG. The transition of the improvement of the evaluation value excluding the difference is shown.
As a result, it was found that by adding uniform mutations to the entire chromosome, the search efficiency for fairness in working days was improved. However, the method of adding only mutations to the entire chromosome significantly reduced the search efficiency.
[0051]
<Discussion>
According to the study of the present inventors, it has been shown that an efficient search for fairness in the number of working days can be realized by combining heuristics with the repair of the working sequence near the crossing point. Further, by using heuristics and mutation together, even better results could be obtained. This was thought to be due to the fact that algorithms using only heuristics that were inspired by manual creation tended to fall into local solutions, but by using mutations together, such local solutions could be avoided.
It is considered that the cause of the local solution when heuristics alone is used is that the algorithm for heuristics is incomplete and the solution search direction is incorrect. However, according to the study results of the present inventors, it was possible to compensate for incomplete heuristics by using a mutation that is a stochastic search in combination with a heuristic that is a search using prior knowledge. It is shown that.
[0052]
In addition to the above improvements, the addition of uniform mutations throughout showed that the search for fairness in working days was more efficient. This suggests that uniform mutations throughout are effective for global searches such as fairness.
In addition, it was shown that the search performance was not very good when only bacterial mutation was performed without using heuristics for repair near the crossing point.
As described above, according to the present embodiment, it is possible to create a nurse work table with a fair working time in a relatively short time while satisfying a complicated evaluation rule and without falling into a local solution. .
[0053]
【The invention's effect】
According to the inventions of the present inventors, it is possible to create a nurse work schedule in a relatively short time without updating a complicated evaluation rule and satisfying it, without falling into a local solution.
[Brief description of the drawings]
FIG. 1 is a diagram showing an example of a specific nurse work schedule. In the left column (skill level) in the figure, "A" indicates skill, "B" indicates unskilled, and "C" indicates a newcomer. Further, “day” indicates day shift (AM 8:00 to PM 4:00), “quasi” indicates semi-night shift (PM 4:00 to AM 0:00), and “deep” indicates midnight shift (AM 0:00 to AM8: 00).
FIG. 2 is a diagram showing an outline of an interactive nurse work table creation support system.
FIG. 3 is a diagram showing a procedure for creating a work table by the nurse work table creation support system described in the embodiment.
FIG. 4 is a diagram showing an algorithm of a nurse work table creation support software by a combined method of BEA and heuristics.
FIG. 5 is a flowchart of nurse work schedule creation support software based on a combined method of BEA and heuristics.
FIG. 6 is a flowchart of a heuristic modeling manual creation.
FIG. 7
FIG. 4 is a diagram showing an algorithm of a technique using a bacterial mutation for repairing near an intersection.
FIG. 8 is a flowchart of a method using a bacterial mutation for repair near an intersection.
FIG. 9 is a flowchart for adding a mutation to the whole.
FIG. 10 is a software execution menu screen 1 used when each embodiment is executed.
FIG. 11 is a software execution menu screen 2 used when executing each embodiment.
FIG. 12
It is a graph which shows transition of the improvement value of a comprehensive evaluation value. In the figure, ■ indicates an algorithm using crossover, heuristics and mutation, ▲ indicates an algorithm using crossover and heuristics, and × indicates an algorithm using heuristics and mutation. The results of each calculation are shown.
FIG. 13 is a graph showing a change in the improvement value of the fairness evaluation value in the first embodiment. In addition, the meaning of the symbol in a figure is the same as that of FIG.
FIG. 14 is a graph showing a transition of an improvement value of an evaluation value excluding fairness in the first embodiment. In addition, the meaning of the symbol in a figure is the same as that of FIG.
FIG. 15 is a graph showing a transition of an evaluation value improvement value in Example 2. In the figure, ◆ shows the results of calculation using an algorithm that uses crossover and heuristics in combination, and ■ shows the results of calculation using an algorithm that uses crossover and bacterial mutation in combination.
FIG. 16 shows a third embodiment.
It is a graph which shows transition of the improvement value of total evaluation value. In the figure, ◆ shows the results obtained by using an algorithm that uses crossover, heuristics, mutation, and total shift, and ■ shows the results obtained by using an algorithm that uses crossover, heuristics, and mutation. .
FIG. 17 is a graph showing a change in the improvement value of the fairness evaluation value in the third embodiment. The meanings of the symbols in the figure are the same as those in FIG.
FIG. 18 is a graph showing a transition of an improvement value of an evaluation value excluding fairness in the third embodiment. The meanings of the symbols in the figure are the same as those in FIG.

Claims (2)

In order to create a work schedule with multiple nurses working over a long period of two weeks or more, (1) include the level of skill, work sequence, assigned work on each day, and the required number of workers for each nurse A step of displaying a screen for prompting the user to input an evaluation rule; (2) a step of automatically calculating by a computer so as to satisfy the evaluation rule as much as possible; and (3) a screen of a provisional solution after a predetermined termination condition is satisfied by the automatic calculation. (4) a step of prompting the user to manually correct the provisional solution, and (5) a step of updating the evaluation rule.
The step (2) includes (a) a crossover step of crossing two randomly generated parent work tables, and (b) a mutation or / and / or heuristics of a predetermined neighborhood including the crossed points. Regenerating a child work table by using the following method: (c) replacing a parent work table, which is inferior to the child work table, with a child work table. The step (2) further includes the step of (d) mutating the whole if the evaluation value is not improved after repeating the steps (a) to (c) a predetermined number of times. The software according to claim 1, wherein:

Images