Electronic Device and Method for Clustered Power Generation Equipment Maintenance Scheduling Considering Multiple Performance Indices

Abstract

An electronic device and method for scheduling maintenance of a cluster of power generation facilities considering multiple performance indices are disclosed. The method for scheduling maintenance of a cluster of power generation facilities considering multiple performance indices includes the steps of receiving a maintenance period, a maintenance volume, and multiple performance indices for scheduling maintenance of a cluster of power generation facilities; for each of the multiple performance indices, obtaining a solution group for scheduling maintenance of a cluster of power generation facilities by obtaining a maintenance schedule of the cluster of power generation facilities that optimizes the performance index under a given maintenance period and maintenance volume based on deep reinforcement learning; and iteratively updating the solution group such that the maintenance schedules included in the solution group are in a limit state where they cannot increase the remaining performance indices without lowering one of the multiple performance indices.

Description

Technology Field

[0001] The present disclosure relates to maintenance scheduling of clustered power generation facilities. More specifically, the present disclosure relates to a method for performing maintenance scheduling of clustered power generation facilities based on deep reinforcement learning by considering multiple performance indices. Background Technology

[0002] Cluster power generation facilities are utilized to produce electricity from naturally occurring energy sources such as solar, wind, wave, and tidal power. In cluster power generation facilities, each unit converts naturally occurring energy into electrical energy to generate electricity. For example, in the case of a wind-powered cluster power generation facility, the kinetic energy of the wind is converted into electrical energy by each wind turbine.

[0003] Cluster power generation facilities possess complex interrelationships, where the performance of other facilities can be affected if some units are shut down for maintenance. Furthermore, the trend toward large-scale installation of cluster power generation facilities is further complicating these interrelationships. Consequently, a method is required to efficiently schedule maintenance for cluster power generation facilities. The problem to be solved

[0004] The problem that the present invention aims to solve is to provide an electronic device and method for performing maintenance scheduling of clustered power generation facilities by considering multiple performance indices. means of solving the problem

[0005] As a technical means for achieving the technical problems described above, a method for scheduling maintenance of a cluster power generation facility considering multiple performance indices according to one aspect of the present invention may include the steps of: receiving a maintenance period, a maintenance volume, and multiple performance indices for scheduling maintenance of a cluster power generation facility; for each of the multiple performance indices, obtaining a maintenance schedule of the cluster power generation facility that optimizes the performance index under a given maintenance period and maintenance volume based on deep reinforcement learning, thereby obtaining a solution group for scheduling maintenance of the cluster power generation facility; and iteratively updating the solution group such that the maintenance schedules included in the solution group are in a limit state where they cannot increase the remaining performance indices without lowering one of the multiple performance indices.

[0006] In the step of iteratively updating the solution group, the current update step may include: a step of obtaining a maintenance schedule of the clustered power generation facility that optimizes the performance index under the given maintenance period and maintenance volume based on deep reinforcement learning for each of the multiple performance indices; and a step of updating the solution group such that, among the maintenance schedules of the solution group updated in the previous update step and the maintenance schedules obtained in the current update step, the solution group includes maintenance schedules in a limit state where it is not possible to increase the remaining performance indices without lowering one of the multiple performance indices.

[0007] In the current update step, the step of updating the solution group may include updating the solution group to include maintenance schedules that satisfy the Pareto frontier for the multiple performance indices among the maintenance schedules of the solution group updated in the previous update step and the maintenance schedules obtained in the current update step.

[0008] In the above current update step, the step of obtaining the maintenance schedule of the cluster power generation facility may include the step of determining the number of power generation facilities for maintenance by timeframe of the maintenance period under the maintenance volume to optimize the performance index based on the first deep reinforcement learning model, and the step of determining the list of power generation facilities for maintenance by timeframe of the maintenance period under the determined number of power generation facilities to optimize the performance index based on the second deep reinforcement learning model.

[0009] The state of the first deep reinforcement learning model includes the number of power generation facilities for maintenance per timeframe of the maintenance period, and the action of the first deep reinforcement learning model may include a change in the number of power generation facilities for maintenance per timeframe of the maintenance period.

[0010] The state of the second deep reinforcement learning model includes a list of power generation facilities for maintenance by timeframe of the maintenance period, and the action of the second deep reinforcement learning model may include changing the list of power generation maintenance for maintenance by timeframe of the maintenance period.

[0011] The reward for moving from the current state to the next state of the first deep reinforcement learning model is calculated based on the difference between the value representing the performance index of the next state and the value representing the performance index of the current state, and the first deep reinforcement learning model can be trained such that the cumulative sum of the reward is maximized or minimized.

[0012] In the above current update step, the step of obtaining the maintenance schedule of the cluster power generation facility may include the step of determining the number and list of power generation facilities for maintenance by timeframe of the maintenance period under the maintenance volume, in order to optimize the performance index based on the third deep reinforcement learning model.

[0013] The above multiple performance indices may include at least two of maintenance costs, maintenance time, maintenance power loss, and maintenance priority.

[0014] As a technical means for achieving the aforementioned technical tasks, a computer program according to one aspect of the present invention is stored in a medium to execute a method for scheduling maintenance of clustered power generation facilities considering the aforementioned multiple performance indices using a computing device.

[0015] As a technical means for achieving the technical problems described above, an electronic device for scheduling maintenance of a cluster power generation facility considering multiple performance indices according to one aspect of the present invention comprises a memory configured to store one or more instructions and a processor, wherein the processor executes one or more instructions to: receive a maintenance period, a maintenance volume, and multiple performance indices for scheduling maintenance of a cluster power generation facility; for each of the multiple performance indices, obtain a maintenance schedule of the cluster power generation facility that optimizes the performance index under a given maintenance period and maintenance volume based on deep reinforcement learning, thereby obtaining a solution group for scheduling maintenance of the cluster power generation facility; and may be configured to iteratively update the solution group such that the maintenance schedules included in the solution group are in a limit state where they cannot increase the remaining performance indices without lowering one of the multiple performance indices.

[0016] The processor may be configured to obtain a maintenance schedule of the clustered power generation facility that optimizes the performance index under the given maintenance period and maintenance volume based on deep reinforcement learning for each of the multiple performance indices, and to update the solution group to include maintenance schedules in a limit state that cannot increase the remaining performance indices without lowering one of the multiple performance indices among the maintenance schedules of the solution group updated in the previous update step and the maintenance schedules obtained in the current update step.

[0017] The processor may be configured to update the solution group to include maintenance schedules that satisfy the Pareto frontier for the multiple performance indices among the maintenance schedules of the solution group updated in the previous update step and the maintenance schedules obtained in the current update step.

[0018] The processor may be configured to determine the number of power generation facilities for maintenance for each timeframe of the maintenance period under the maintenance volume in order to optimize the performance index based on a first deep reinforcement learning model, and to determine the list of power generation facilities for maintenance for each timeframe of the maintenance period under the determined number of power generation facilities in order to optimize the performance index based on a second deep reinforcement learning model.

[0019] The state of the first deep reinforcement learning model includes the number of power generation facilities for maintenance per timeframe of the maintenance period, and the action of the first deep reinforcement learning model may include a change in the number of power generation facilities for maintenance per timeframe of the maintenance period.

[0020] The state of the second deep reinforcement learning model includes a list of power generation facilities for maintenance by timeframe of the maintenance period, and the action of the second deep reinforcement learning model may include changing the list of power generation maintenance for maintenance by timeframe of the maintenance period.

[0021] The reward for moving from the current state to the next state of the first deep reinforcement learning model is calculated based on the difference between the value representing the performance index of the next state and the value representing the performance index of the current state, and the first deep reinforcement learning model can be trained such that the cumulative sum of the reward is maximized or minimized.

[0022] The above processor may be configured to determine the number and list of power generation facilities for maintenance for each timeframe of the maintenance period under the above maintenance volume, in order to optimize the above performance index based on the third deep reinforcement learning model.

[0023] The above multiple performance indices may include at least two of maintenance costs, maintenance time, maintenance power loss, and maintenance priority. Effects of the invention

[0024] According to the electronic device and method for maintenance scheduling of clustered power generation facilities considering multiple performance indices of the present invention, efficient maintenance scheduling is possible because the most superior maintenance schedules, when considering multiple performance indices comprehensively, can be selected from among maintenance schedules optimized for each performance index. Furthermore, by obtaining maintenance schedules that optimize each performance index based on deep reinforcement learning, it is possible to obtain maintenance schedules for large-scale clustered power generation facilities quickly and accurately.

[0025] The effects obtainable from the exemplary embodiments of the present disclosure are not limited to those mentioned above, and other unmentioned effects can be clearly derived and understood by those skilled in the art from the description of the exemplary embodiments of the present disclosure below. That is, unintended effects resulting from the implementation of the exemplary embodiments of the present disclosure can also be derived by those skilled in the art from the exemplary embodiments of the present disclosure. Brief explanation of the drawing

[0026] FIG. 1 is a drawing illustrating a cluster power generation facility according to one embodiment. FIG. 2 is a block diagram of an electronic device for scheduling maintenance of a cluster power generation facility considering multiple performance indices according to one embodiment. FIGS. 3 and 4 are flowcharts of a method for scheduling maintenance of clustered power generation facilities considering multiple performance indices according to embodiments. FIG. 5a is a diagram illustrating a deep reinforcement learning model according to one embodiment. FIG. 5b is a diagram illustrating a maintenance schedule obtained based on a deep reinforcement learning model according to one embodiment. FIGS. 6a to 6c are drawings for explaining the states and actions of the first to third deep reinforcement learning models according to embodiments. FIGS. 7 and 8 are drawings illustrating an update of a solution group based on the Pareto frontier according to embodiments. Specific details for implementing the invention

[0027] Below, various embodiments are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present disclosure.

[0028] FIG. 1 is a drawing illustrating a cluster power generation facility (100) according to one embodiment.

[0029] Figure 1 illustrates an example of a cluster power generation facility (100) including 49 power generation facilities (1 to 49).

[0030] A power plant may include a cluster of power generation facilities (100) for converting naturally occurring energy sources into electrical energy. A solar power plant may include a cluster of power generation facilities (100) that generate power from solar energy, and in this case, each power generation facility may include a solar panel. A wind power plant may include a cluster of power generation facilities (100) that generate power from wind energy, and in this case, each power generation facility may include a wind turbine. A wave power plant may include a cluster of power generation facilities (100) that generate power from wave energy, and in this case, each power generation facility may include a floating facility that moves up and down according to the waves. A tidal power plant may include a cluster of power generation facilities (100) that generate power from tidal energy, and in this case, each power generation facility may include a tidal turbine.

[0031] For efficient maintenance scheduling of cluster power generation facilities (100), various performance indices need to be considered. Performance indices may include maintenance costs, maintenance time, maintenance power loss, and maintenance priority.

[0032] Maintenance costs may include labor costs, parts costs, and transportation costs. Maintenance time may include travel time to the power generation facility and time required to maintain the power generation facility. Maintenance loss power may represent the total power generated by a cluster power generation facility (100) in which some power generation facilities are not in operation due to maintenance, compared to the total power generated by a cluster power generation facility (100) in which all power generation facilities (1 to 49) are in operation. Maintenance priority may represent the remaining time until the maintenance parts need to be maintained. Maintenance priority may be set higher as the remaining time until the maintenance parts need to be maintained is shorter.

[0033] For example, the power loss from maintenance when power generation facilities 1 to 5 (1 to 5) are maintained may be the total power generated by the remaining power generation facilities (6 to 49) excluding power generation facilities 1 to 5 (1 to 5) from the total power generated by all power generation facilities (1 to 49).

[0034] For example, if the remaining time until maintenance of the maintenance parts of power generation facilities 1 to 5 (1 to 5) is 3 months, 13 months, 4 months, 16 months, and 4 months, respectively, the maintenance priority may be higher in the order of power generation facility 1 (1), power generation facilities 3 and 5 (3, 5), power generation facility 2 (2), and power generation facility 4 (4).

[0035] Since the power generation facilities (1 to 49) of the cluster power generation facility (100) are interconnected in mechanical, systemic, etc., the performance index may vary depending on which power generation facility is maintained.

[0036] For example, in wind power generation, when an object moves through a fluid, the effective wind speed of the downstream power generation facility decreases due to the influence of the wake, which is the fluid flow behind the object, or the effective wind speed of the downstream power generation facility recovers due to maintenance non-operation of the upstream power generation facility. Accordingly, depending on which power generation facility is maintained, there may be a difference in the total power generated by the cluster power generation facility (100).

[0037] For cluster power generation facilities comprising dozens or hundreds of units, performing maintenance scheduling while considering various performance indices is a complex issue. However, for efficient maintenance scheduling, it is necessary to perform maintenance scheduling by considering various performance indices.

[0038] In the following, embodiments of an electronic device and method capable of efficient maintenance scheduling for large-scale cluster power generation facilities are described, by considering multiple performance indices including at least two of maintenance costs, maintenance time, maintenance power loss, and maintenance priority.

[0039] FIG. 2 is a block diagram of an electronic device (200) for scheduling maintenance of a cluster power generation facility considering multiple performance indices according to one embodiment.

[0040] The electronic device (200) may be any electronic device including a processor (210) and a memory (220).

[0041] The processor (210) typically controls the overall operation of the electronic device (200). The processor (210) performs basic arithmetic, logic, and input / output operations and can, for example, execute program code stored in memory (220).

[0042] The memory (220) is a recording medium readable by the processor (210) and may include a permanent mass storage device such as RAM, ROM, and a disk drive. The memory (220) may store an operating system and at least one program or application code. The memory (220) may store program code for executing a method for scheduling maintenance of clustered power generation facilities considering multiple performance indices according to the present disclosure.

[0043] The electronic device (200) may further include, in addition to the processor (210) and memory (220), a communication module, an input / output device, or a storage device.

[0044] The processor (210) of the electronic device (200) can perform maintenance scheduling of a cluster power generation facility considering multiple performance indices by executing at least one instruction stored in memory (220). Below, embodiments of a method for maintenance scheduling of a cluster power generation facility considering multiple performance indices performed by the processor (210) are described.

[0045] FIG. 3 is a flowchart of a method for scheduling maintenance of clustered power generation facilities considering multiple performance indices according to one embodiment.

[0046] In step S310, the processor receives a maintenance period, a maintenance volume, and multiple performance indices for scheduling maintenance of the cluster power generation facility.

[0047] The processor can receive maintenance period, maintenance volume, and multiple performance indices from user input, or receive maintenance period, maintenance volume, and multiple performance indices stored in memory.

[0048] The maintenance period may be the total period for which maintenance is to be performed. The maintenance volume may be the number of power generation facilities among the clustered power generation facilities for which maintenance is to be performed. Multiple performance indices may include at least two of maintenance cost, maintenance time, maintenance power loss, and maintenance priority.

[0049] For example, the maintenance period received by the processor may be 6 days, the maintenance volume may be 70 units, and the multiple performance indices may be maintenance costs and maintenance time. As another example, the maintenance period received by the processor may be 10 days, the maintenance volume may be 120 units, and the multiple performance indices may be maintenance costs, maintenance time, maintenance power loss, and maintenance priority.

[0050] In step S320, the processor obtains a solution group for the maintenance scheduling of clustered power generation facilities by obtaining a maintenance schedule of clustered power generation facilities that optimizes the performance index for each of the multiple performance indices under a given maintenance period and maintenance volume based on deep reinforcement learning.

[0051] For example, if multiple performance indices are maintenance cost and maintenance time, the processor can obtain a maintenance schedule that optimizes maintenance cost under a given maintenance period and volume based on deep reinforcement learning, obtain a maintenance schedule that optimizes maintenance time under a given maintenance period and volume based on deep reinforcement learning, and include the obtained maintenance schedules in a solution group.

[0052] The processor can obtain multiple maintenance schedules by performing deep reinforcement learning multiple times for each of the multiple performance indices.

[0053] For example, if multiple performance indices are maintenance power loss and maintenance priority, the processor can perform deep reinforcement learning multiple times to obtain multiple maintenance schedules that optimize maintenance power loss under a given maintenance period and maintenance volume, perform deep reinforcement learning multiple times to obtain multiple maintenance schedules that optimize maintenance priority under a given maintenance period and maintenance volume, and include the obtained multiple maintenance schedules in a solution group.

[0054] FIG. 5a illustrates an example of a deep reinforcement learning model (500) used in deep reinforcement learning. Referring to FIG. 5a, the deep reinforcement learning model (500) may include an input layer into which states are input, an output layer into which actions are output, and one or more hidden layers between the input layer and the output layer. The input layer, the output layer, and the hidden layer may be fully connected layers. The deep reinforcement learning model (500) may be trained such that the cumulative sum of rewards resulting from a transition from the current state to the next state is maximized or minimized. In cases where a larger reward value indicates a better performance index, the deep reinforcement learning model (500) may be trained such that the cumulative sum of rewards is maximized, and in cases where a larger reward value indicates a worse performance index, the deep reinforcement learning model (500) may be trained such that the cumulative sum of rewards is minimized.

[0055] FIG. 5b illustrates an example of a maintenance schedule obtained based on a deep reinforcement learning model (500). Referring to FIG. 5b, a maintenance schedule was derived using a deep reinforcement learning model (500) to optimize the performance index of maintenance loss power for a maintenance period of 7 days and a maintenance volume of 25 units. Deep reinforcement learning was performed to maximize the return, which is the cumulative sum of rewards.

[0056] As a result of deep reinforcement learning, a schedule was obtained to maintain 6 power generation facilities (Nos. 0, 2, 6, 8, 44, and 14) on day 1, 4 power generation facilities (Nos. 32, 4, 16, and 18) on day 2, 6 power generation facilities (Nos. 20, 12, 42, 34, 28, and 46) on day 3, 1 power generation facility (No. 10) on day 4, 6 power generation facilities (Nos. 26, 24, 38, 40, 22, and 36) on day 5, 1 power generation facility (No. 30) on day 6, and 1 power generation facility (No. 48) on day 7. It can be seen that when maintenance is performed according to the acquired maintenance schedule, the total power generated by the cluster power generation facility is 694.75 MW, and the power loss from maintenance can be minimized.

[0057] The first to third deep reinforcement learning models described below may have a structural format similar to the deep reinforcement learning model (500) of FIG. 5a.

[0058] The processor can obtain a maintenance schedule that optimizes the performance index under a given maintenance period and maintenance volume by using the first deep reinforcement learning model and the second deep reinforcement learning model.

[0059] The first deep reinforcement learning model can be trained to determine the number of power generation facilities for maintenance per timeframe of the maintenance period under maintenance volume in order to optimize the performance index. The state of the first deep reinforcement learning model includes the number of power generation facilities for maintenance per timeframe of the maintenance period, and the action of the first deep reinforcement learning model may include changing the number of power generation facilities for maintenance per timeframe of the maintenance period.

[0060] Figure 6a illustrates an example of the state and action of the first deep reinforcement learning model. Referring to Figure 6a, for example, the first state of the first deep reinforcement learning model is the number of power generation facilities for maintenance by day during a total maintenance period of 4 days for a maintenance volume of 48 units, which may be 10 units on day 1, 13 units on day 2, 13 units on day 3, and 12 units on day 4. The second state of the first deep reinforcement learning model may be 9 units on day 1, 15 units on day 2, 12 units on day 3, and 12 units on day 4. The first action of the first deep reinforcement learning model is a change in the number of power generation facilities for maintenance on a daily basis for a total maintenance period of 4 days for a maintenance volume of 48 units, which may be an action of decreasing by 1 unit on the 1st day, increasing by 2 units on the 2nd day, decreasing by 1 unit on the 3rd day, and not changing the number on the 4th day. For example, if the value representing the maintenance cost of the first state is 100 and the value representing the maintenance cost of the second state is 120, the reward for moving from the first state to the second state according to the first action may be 120-100=20.

[0061] The second deep reinforcement learning model can be trained to determine a list of power generation facilities for maintenance for each timeframe of the maintenance period, under the number of power generation facilities per timeframe determined by the first deep reinforcement learning model, in order to optimize the performance index. The state of the second deep reinforcement learning model includes a list of power generation facilities for maintenance for each timeframe of the maintenance period, and the action of the second deep reinforcement learning model may include changing the list of power generation maintenance for each timeframe of the maintenance period.

[0062] Figure 6b illustrates an example of the state and action of the second deep reinforcement learning model. Referring to Figure 6b, for example, the first state of the second deep reinforcement learning model is a list of power generation facilities for daily maintenance during a total maintenance period of 4 days for a maintenance volume of 48 units, which may be power generation facilities 1 through 10 on day 1, power generation facilities 11 through 23 on day 2, power generation facilities 24 through 36 on day 3, and power generation facilities 37 through 48 on day 4. The second state of the second reinforcement learning model may be power generation facilities 2 through 11 on day 1, power generation facilities 1 and 12 through 23 on day 2, power generation facilities 24 through 36 on day 3, and power generation facilities 37 through 48 on day 4. The first action of the second deep reinforcement learning model is a change in the list of power generation facilities for daily maintenance during a total maintenance period of 4 days for a maintenance volume of 48 units, which may be an action of deleting power generation facility No. 1 and adding power generation facility No. 11 on Day 1, deleting power generation facility No. 11 and adding power generation facility No. 1 on Day 2, and not changing the list on Days 3 and 4. For example, if the value representing the maintenance time required for the first state is 50 and the value representing the maintenance time required for the second state is 40, the reward for moving from the first state to the second state according to the first action may be 40-50=-10.

[0063] Alternatively, the processor can obtain a maintenance schedule that optimizes the performance index under a given maintenance period and volume by using a third deep reinforcement learning model.

[0064] The third deep reinforcement learning model can be trained to determine the number and list of power generation facilities for maintenance for each timeframe of the maintenance period under maintenance volume, in order to optimize the performance index.

[0065] The third deep reinforcement learning model may include the states of the first and second deep reinforcement learning models and the actions of the first and second deep reinforcement learning models.

[0066] Alternatively, the state of the third deep reinforcement learning model includes the number and list of power generation facilities for maintenance by timeframe of the maintenance period, and the action of the third deep reinforcement learning model may include changing the number and list of power generation maintenance for maintenance by timeframe of the maintenance period.

[0067] Figure 6c illustrates an example of the state and action of the third deep reinforcement learning model. Referring to Figure 6c, for example, the first state of the third deep reinforcement learning model is the number and list of power generation facilities for maintenance by day for a total maintenance period of 4 days for a maintenance volume of 48 units, and may be 10 units on day 1, power generation facilities 1 through 10 on day 2, power generation facilities 11 through 23 on day 3, 13 units on day 3, power generation facilities 24 through 36 on day 4, and power generation facilities 37 through 48 on day 4. The second state of the third reinforcement learning model may be 9 units on day 1, power generation facilities 2 through 10, 14 units on day 2, power generation facilities 1 and 11 through 23, 13 units on day 3, power generation facilities 24 through 36, and 12 units on day 4, power generation facilities 37 through 48. The first action of the third deep reinforcement learning model may be a change in the number and list of power generation facilities for maintenance on a daily basis for a total maintenance period of 4 days for a maintenance volume of 48 units, which may be an action of deleting power generation facility 1 on day 1 to decrease by 1 unit, adding power generation facility 1 on day 2 to increase by 1 unit, and not changing the number and list on days 3 and 4. For example, if the value representing the maintenance priority of the first state is 600 and the value representing the maintenance priority of the second state is 540, the reward for moving from the first state to the second state according to the first action may be 540-600=-60.

[0068] In step S330, the processor iteratively updates the solution group such that the maintenance schedules included in the solution group are in a limit state where they cannot increase the remaining performance indices without lowering one of the multiple performance indices.

[0069] In the first update, the processor can calculate multiple performance indices for each of the maintenance schedules of the solution group obtained in step S320. By comparing the multiple performance indices of the calculated maintenance schedules, the processor can select maintenance schedules that satisfy a limit state in which one performance index cannot be lowered while the other performance index cannot be raised. The processor can update the solution group so that the solution group includes only the selected maintenance schedules.

[0070] In any nth update satisfying n≥2, the processor can obtain a new maintenance schedule for clustered power generation facilities that optimizes the performance index for each of the multiple performance indices under a given maintenance period and volume based on deep reinforcement learning. For each of the new maintenance schedules, the processor can calculate the multiple performance indices. Among the new maintenance schedules and the maintenance schedules of the solution group, the processor can select maintenance schedules that are in a limit state where it is impossible to increase the remaining performance indices without lowering one performance index. The processor can update the solution group so that the solution group includes only the selected maintenance schedules.

[0071] In step S330, the processor may update the solution group based on the Pareto frontier. Maintenance schedules in a limit state where it is not possible to increase the other performance indices without decreasing one of the multiple performance indices may be identical to maintenance schedules that satisfy the Pareto frontier for the multiple performance indices. Accordingly, the processor may update the solution group to include maintenance schedules that satisfy the Pareto frontier for the multiple performance indices.

[0072] An example of a Pareto frontier is illustrated in FIG. 7. Referring to FIG. 7, for example, it is assumed that in step S320, maintenance schedules (711–713) are obtained to optimize a first performance index, and maintenance schedules (721–723) are obtained to optimize a second performance index. The processor can obtain a Pareto frontier (PF1) for the first and second performance indices by calculating the second performance index of the maintenance schedules (711–713) and the first performance index of the maintenance schedules (721–723). The processor can update the solution group to include maintenance schedules (711, 712, 721, 722) among the maintenance schedules (711–713, 721–723) that satisfy the Pareto frontier (PF1).

[0073] The processor may terminate the update of the solution group when a predetermined number of new maintenance schedules match the maintenance schedules of the solution group, or when the solution group has been updated a predetermined number of times.

[0074] Maintenance schedules derived to optimize individual performance indices are obtained in step S320, and in step S330, the most superior maintenance schedules can be selected from among the maintenance schedules optimized for individual performance indices by comprehensively considering multiple performance indices. Accordingly, efficient maintenance scheduling is possible. For example, when maintenance costs, maintenance time, maintenance power loss, and maintenance priority are considered as multiple performance indices, a maintenance schedule can be derived that satisfies the deadline for maintaining parts while reducing maintenance costs, maintenance time, and maintenance power loss.

[0075] In addition, by obtaining a maintenance schedule that optimizes each performance index based on deep reinforcement learning, it is possible to quickly and accurately obtain maintenance schedules for large-scale clustered power generation facilities.

[0076] FIG. 4 is a flowchart of a method for scheduling maintenance of clustered power generation facilities considering multiple performance indices according to one embodiment.

[0077] In step S410, the processor receives a maintenance period, a maintenance volume, and multiple performance indices for scheduling maintenance of the cluster power generation facility.

[0078] In step S420, the processor obtains a solution group for the maintenance scheduling of clustered power generation facilities by obtaining a maintenance schedule of clustered power generation facilities that optimizes the performance index for each of the multiple performance indices under a given maintenance period and maintenance volume based on deep reinforcement learning.

[0079] The description of step S310 of FIG. 3 may be applied to step S410, and the description of step S320 of FIG. 3 may be applied to step S420. For convenience of explanation, redundant descriptions are omitted.

[0080] In step S430, the processor determines whether the termination condition is satisfied.

[0081] The processor may determine that the termination condition is satisfied if the maintenance schedules obtained in step S420 match each other for more than a predetermined number, if the maintenance schedules obtained in step S440 match the maintenance schedules of the solution group for more than a predetermined number, or if steps S440 and S450 are performed a predetermined number of times.

[0082] If it is determined that the termination condition is satisfied, proceed to Termination. If it is determined that the termination condition is not satisfied, proceed to Step S440.

[0083] In step S440, the processor obtains a maintenance schedule for clustered power generation facilities that optimizes the performance index for each of the multiple performance indices under a given maintenance period and maintenance volume based on deep reinforcement learning.

[0084] In step S440, for each of the multiple performance indices of step S320 of FIG. 3, an explanation may be applied to obtain a maintenance schedule for clustered power generation facilities that optimizes the performance index under a given maintenance period and maintenance volume based on deep reinforcement learning. For the convenience of explanation, redundant explanations are omitted.

[0085] In step S450, the processor updates the solution group to include, among the maintenance schedules of the solution group and the maintenance schedules obtained in step S440, maintenance schedules that are in a limit state where they cannot increase the remaining performance indices without lowering one of the multiple performance indices.

[0086] In the first update, the processor can calculate multiple performance indices for each of the maintenance schedules of the solution group obtained in step S420 and the maintenance schedules obtained in step S440. By comparing the multiple performance indices of the maintenance schedules, the processor can select maintenance schedules that satisfy a limit state in which one performance index cannot be lowered while the other performance index cannot be raised. The processor can update the solution group so that the solution group includes only the selected maintenance schedules.

[0087] In any nth update satisfying n≥2, the processor can calculate multiple performance indices for each of the maintenance schedules obtained in step S440. The processor can select maintenance schedules that are in a limit state among the maintenance schedules obtained in step S440 and the maintenance schedules of the solution group, in which it is not possible to increase the remaining performance indices without lowering one performance index. The processor can update the solution group so that the solution group includes only the selected maintenance schedules.

[0088] In step S450, the processor may update the solution group based on the Pareto frontier. Maintenance schedules in a limit state where it is not possible to increase the other performance indices without decreasing one of the multiple performance indices may be identical to maintenance schedules that satisfy the Pareto frontier for the multiple performance indices. Accordingly, the processor may update the solution group to include maintenance schedules that satisfy the Pareto frontier for the multiple performance indices.

[0089] An example of a Pareto frontier is illustrated in FIG. 8. Referring together with FIG. 8, for example, it is assumed that a solution group includes maintenance schedules (811–814), and in step S440, a maintenance schedule (821) for optimizing a first performance index and a maintenance schedule (822) for optimizing a second performance index are obtained. The processor can obtain a Pareto frontier (PF2) for the first and second performance indices by calculating the second performance index of the maintenance schedule (821) and the first performance index of the maintenance schedule (822). The processor can update the solution group to include maintenance schedules (811–813, 822) among the maintenance schedules (811–814, 821, 822) that satisfy the Pareto frontier (PF2).

[0090] In step S440, maintenance schedules derived to optimize individual performance indices are obtained, and in step S450, among the maintenance schedules with optimized individual performance indices, the best maintenance schedules can be selected when multiple performance indices are comprehensively considered. Accordingly, efficient maintenance scheduling is possible.

[0091] In addition, by obtaining a maintenance schedule that optimizes each performance index based on deep reinforcement learning, it is possible to quickly and accurately obtain maintenance schedules for large-scale clustered power generation facilities.

[0092] The various embodiments described above may be implemented in the form of a computer program that can be executed on a computer through various components, and such a computer program may be recorded on a computer-readable medium. In this case, the medium may continuously store the computer-executable program or temporarily store it for execution or download. Furthermore, the medium may be various recording or storage means in the form of a single or multiple hardware combinations, and is not limited to a medium directly connected to a computer system but may also exist distributed over a network. Examples of media may include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and media configured to store program instructions, including ROM, RAM, and flash memory. Additionally, other examples of media may include recording or storage media managed by app stores that distribute applications or sites and servers that supply or distribute various other software.

[0093] In this specification, "part," "module," etc. may be a hardware component, such as a processor or circuit, and / or a software component executed by a hardware component, such as a processor. For example, "part," "module," etc. may be implemented by components such as software components, object-oriented software components, class components, and task components, and by processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables.

[0094] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0095] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention. Explanation of the symbols

[0096] 100: Cluster Power Generation Facility 200: Electronic device 210: Processor 220: Memory 500: Deep Reinforcement Learning Model

Claims

Claim 1 A method for scheduling maintenance of clustered power generation facilities considering multiple performance indices comprises: receiving a maintenance period, a maintenance volume, and multiple performance indices for scheduling maintenance of clustered power generation facilities; obtaining a solution group for scheduling maintenance of clustered power generation facilities by obtaining a maintenance schedule of clustered power generation facilities that optimizes the performance index under the given maintenance period and maintenance volume for each of the multiple performance indices based on deep reinforcement learning; and repeatedly updating the solution group such that the maintenance schedules included in the solution group are in a limit state where they cannot increase the remaining performance indices without lowering one of the multiple performance indices, wherein the current update step comprises: obtaining a maintenance schedule of clustered power generation facilities that optimizes the performance index under the given maintenance period and maintenance volume for each of the multiple performance indices based on deep reinforcement learning; A method comprising the step of updating the solution group such that, among the maintenance schedules of the solution group updated in the previous update step and the maintenance schedules obtained in the current update step, the solution group includes maintenance schedules in a limit state where one of the multiple performance indices cannot be lowered while the other performance indices cannot be raised; and in the current update step, the step of obtaining the maintenance schedule of the cluster power generation facility includes: the step of determining the number of power generation facilities for maintenance by timeframe of the maintenance period under the maintenance volume, so as to optimize the performance index based on a first deep reinforcement learning model; and the step of determining the list of power generation facilities for maintenance by timeframe of the maintenance period under the determined number of power generation facilities, so as to optimize the performance index based on a second deep reinforcement learning model. Claim 2 delete Claim 3 A method according to claim 1, wherein, in the current update step, the step of updating the solution group comprises the step of updating the solution group to include maintenance schedules that satisfy the Pareto frontier for the multiple performance index among the maintenance schedules of the solution group updated in the previous update step and the maintenance schedules obtained in the current update step. Claim 4 delete Claim 5 A method according to claim 1, wherein the state of the first deep reinforcement learning model includes the number of power generation facilities for maintenance per timeframe of the maintenance period, and the action of the first deep reinforcement learning model includes changing the number of power generation facilities for maintenance per timeframe of the maintenance period. Claim 6 A method according to claim 1, wherein the state of the second deep reinforcement learning model includes a list of power generation facilities for maintenance by timeframe of the maintenance period, and the action of the second deep reinforcement learning model includes changing the list of power generation maintenance for maintenance by timeframe of the maintenance period. Claim 7 A method according to claim 1, wherein the reward for moving from the current state to the next state of the first deep reinforcement learning model is calculated based on the difference between a value representing the performance index of the next state and a value representing the performance index of the current state, and the first deep reinforcement learning model is trained such that the cumulative sum of the reward is maximized or minimized. Claim 8 A method according to claim 1, wherein the step of obtaining the maintenance schedule of the cluster power generation facility in the current update step comprises the step of determining the number and list of power generation facilities for maintenance by timeframe of the maintenance period under the maintenance volume, in order to optimize the performance index based on the third deep reinforcement learning model. Claim 9 A method according to claim 1, wherein the multiple performance indices include at least two of maintenance costs, maintenance time, maintenance power loss, and maintenance priority. Claim 10 A computer program stored on a medium to execute the method of any one of claims 1, 3, 5 through 9 using a computing device. Claim 11 An electronic device for maintenance scheduling of clustered power generation facilities considering multiple performance indices, comprising: a memory configured to store one or more instructions; The system includes a processor, wherein the processor executes one or more instructions to: receive a maintenance period, a maintenance volume, and multiple performance indices for the maintenance scheduling of a cluster power generation facility; obtain a maintenance schedule of the cluster power generation facility that optimizes the performance index under the given maintenance period and maintenance volume based on deep reinforcement learning for each of the multiple performance indices, thereby obtaining a solution group for the maintenance scheduling of the cluster power generation facility; and be configured to repeatedly update the solution group such that the maintenance schedules included in the solution group are in a limit state where they cannot increase the remaining performance indices without lowering one of the multiple performance indices; and the processor obtains a maintenance schedule of the cluster power generation facility that optimizes the performance index under the given maintenance period and maintenance volume based on deep reinforcement learning for each of the multiple performance indices, and among the maintenance schedules of the solution group updated in the previous update step and the maintenance schedules obtained in the current update step, the maintenance schedules that are in a limit state where they cannot increase the remaining performance indices without lowering one of the multiple performance indices. An electronic device configured to update a solution group, wherein the processor determines the number of power generation facilities for maintenance by timeframe of the maintenance period under the maintenance volume to optimize the performance index based on a first deep reinforcement learning model, and determines the list of power generation facilities for maintenance by timeframe of the maintenance period under the determined number of power generation facilities to optimize the performance index based on a second deep reinforcement learning model. Claim 12 delete Claim 13 An electronic device according to claim 11, wherein the processor is configured to update the solution group to include maintenance schedules satisfying the Pareto frontier for the multiple performance indices among the maintenance schedules of the solution group updated in the previous update step and the maintenance schedules obtained in the current update step. Claim 14 delete Claim 15 An electronic device according to claim 11, wherein the state of the first deep reinforcement learning model includes the number of power generation facilities for maintenance per timeframe of the maintenance period, and the action of the first deep reinforcement learning model includes a change in the number of power generation facilities for maintenance per timeframe of the maintenance period. Claim 16 An electronic device according to claim 11, wherein the state of the second deep reinforcement learning model includes a list of power generation facilities for maintenance by timeframe of the maintenance period, and the action of the second deep reinforcement learning model includes changing the list of power generation maintenance for maintenance by timeframe of the maintenance period. Claim 17 An electronic device according to claim 11, wherein the reward for moving from the current state to the next state of the first deep reinforcement learning model is calculated based on the difference between a value representing the performance index of the next state and a value representing the performance index of the current state, and the first deep reinforcement learning model is trained such that the cumulative sum of the rewards is maximized or minimized. Claim 18 In claim 11, the electronic device configured such that the processor determines the number and list of power generation facilities for maintenance by timeframe of the maintenance period under the maintenance volume, in order to optimize the performance index based on the third deep reinforcement learning model. Claim 19 An electronic device according to claim 11, wherein the multiple performance indices include at least two of maintenance costs, maintenance time, maintenance power loss, and maintenance priority.

Images