Method for predicting amounts of power generated from distributed power plants by using reinforcement learning

Abstract

The present invention relates to a method for determining, on the basis of environmental variables collected from a new power plant, whether to subject a neural network model to additional reinforcement learning, the neural network model predicting the amounts of power generated from a plurality of distributed power plants. The method for predicting the amounts of power generated from distributed power plants by using reinforcement learning, according to an embodiment of the present invention, pertains to a method for predicting the amount of power generated from each of the dispersedly installed power plants by using a neural network model, wherein the method comprises the steps of: generating a plurality of reference clusters by clustering environmental variables cumulatively collected from each of the power plants; collecting new environmental variables from a new power plant and clustering the new environmental variables to generate a new cluster; and additionally using the new environmental variables in reinforcement learning of the neural network model on the basis of similarity between the plurality of reference clusters and the new cluster.

Description

A method for predicting power generation of distributed power plants using reinforcement learning

[0001] The present invention relates to a method for predicting power generation of multiple distributed power plants using a neural network model that undergoes continuous reinforcement learning.

[0002]

[0003] Distributed power plants refer to small-scale renewable energy power generation facilities, such as wind power, solar power, and small hydro power, that are distributed across different regions. They are generally installed near areas with high power demand and serve the role of supplying power.

[0004] In order to economically integrate and manage these distributed power plants, the power generation of each power plant must be accurately predicted, and recently, artificial intelligence models are being used to improve the accuracy of power generation prediction.

[0005] However, while an AI model once trained guarantees high performance for data with a certain tendency, distributed power plants are individually installed in different environments, so there is a limitation in that an AI model trained on existing data is difficult to use to predict the power generation of newly installed distributed power plants.

[0006]

[0007] The purpose of the present invention is to determine whether to perform additional reinforcement learning of a neural network model based on environmental variables collected from new power plants when predicting the power generation of multiple distributed power plants using a neural network model.

[0008] The objectives of the present invention are not limited to those mentioned above. Other objectives and advantages of the present invention not mentioned above can be understood through the following description and will be more clearly understood through the embodiments of the present invention. Furthermore, it will be readily apparent that the objectives and advantages of the present invention can be realized by the means and combinations thereof set forth in the claims.

[0009]

[0010] According to one embodiment of the present invention for achieving the above-described object, a method for predicting power generation of a distributed power plant using reinforcement learning is provided, which is a method for predicting power generation of each distributed power plant using a neural network model, the method comprising: a step of clustering environmental variables accumulated and collected from each power plant to create a plurality of reference clusters; a step of collecting new environmental variables from a new power plant and clustering the new environmental variables to create a new cluster; and a step of additionally utilizing the new environmental variables in reinforcement learning of the neural network model based on a similarity between the plurality of reference clusters and the new cluster.

[0011]

[0012] The present invention has the advantage of eliminating the need to spend a lot of time and resources training the neural network model each time a power plant to be managed is added when predicting the power generation of multiple distributed power plants through a neural network model, by determining whether to perform additional training of the neural network model based on environmental variables collected from new power plants.

[0013] In addition to the effects described above, specific effects of the present invention are described below while explaining specific details for carrying out the invention.

[0014]

[0015] FIG. 1 is a diagram illustrating a power generation prediction system of a distributed power plant according to one embodiment of the present invention.

[0016] FIG. 2 is a flowchart illustrating a method for predicting power generation of a distributed power plant using reinforcement learning according to one embodiment of the present invention.

[0017] Figure 3 is a diagram for explaining the reinforcement learning process of the neural network model of the present invention.

[0018] Figure 4 is a diagram showing how environmental variables accumulated and collected in an existing power plant are clustered.

[0019] Figures 5 and 6 are diagrams each showing the clustering of new environmental variables collected from a new power plant.

[0020]

[0021] The above-described objects, features, and advantages will be described in detail below with reference to the accompanying drawings, so that those skilled in the art can easily practice the technical idea of ​​the present invention. In describing the present invention, if it is determined that a detailed description of known technologies related to the present invention may unnecessarily obscure the gist of the present invention, a detailed description thereof will be omitted. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals are used to indicate the same or similar components.

[0022] Although terms like "first" and "second" are used herein to describe various components, these components are not limited by these terms. These terms are used merely to distinguish one component from another, and unless otherwise specified, a "first" component may also be a "second" component.

[0023] Additionally, in this specification, the phrase "upper (or lower)" or "upper (or lower)" of any component may mean that any component is positioned not only in contact with the upper surface (or lower surface) of the component, but also that other components may be interposed between the component and any component positioned on (or below) the component.

[0024] Additionally, when it is described herein that a component is “connected,” “coupled,” or “connected” to another component, it should be understood that the components may be directly connected or connected to each other, but that other components may be “interposed” between the components, or that each component may be “connected,” “coupled,” or “connected” through another component.

[0025] Additionally, the singular expressions used herein include plural expressions unless the context clearly dictates otherwise. In this application, terms such as "consist of" or "comprises" should not be construed to necessarily include all of the various components or various steps described in the specification, and should be construed to mean that some of the components or some of the steps may not be included, or that additional components or steps may be included.

[0026] Also, in this specification, when it says "A and / or B", it means A, B or A and B, unless otherwise specifically stated, and when it says "C to D", it means C or more and D or less, unless otherwise specifically stated.

[0027] The present invention relates to a method for determining whether to use reinforcement learning in a neural network model when predicting the power generation of multiple distributed power plants based on environmental variables collected from new power plants. Hereinafter, a method for predicting power generation of a distributed power plant using reinforcement learning (hereinafter, "power generation prediction method") according to one embodiment of the present invention will be described in detail with reference to FIGS. 1 through 6.

[0028] FIG. 1 is a diagram illustrating a power generation prediction system of a distributed power plant according to one embodiment of the present invention, and FIG. 2 is a flowchart illustrating a power generation prediction method of a distributed power plant using reinforcement learning according to one embodiment of the present invention.

[0029] Figure 3 is a diagram for explaining the reinforcement learning process of the neural network model of the present invention.

[0030] Figure 4 is a diagram illustrating a clustering of environmental variables accumulated and collected from an existing power plant, and Figures 5 and 6 are diagrams illustrating a clustering of new environmental variables collected from a new power plant, respectively.

[0031] Referring to FIG. 1, a power generation prediction system (1) of a distributed power plant according to one embodiment of the present invention may include a server (10), an external server (30), and a plurality of distributedly installed power plants (20). However, the power generation prediction system (1) illustrated in FIG. 1 is according to one embodiment, and the components constituting the system (1) are not limited to the embodiment illustrated in FIG. 1, and some components may be added, changed, or deleted as needed.

[0032] The power plant (20) may include any facility capable of generating electricity, such as a power plant using fossil fuels or a power plant using (new) renewable energy sources such as solar power, wind power, tidal power, and small hydro power. However, in order to ensure that the comparison between the environmental variables described below is meaningful, each power plant (20) may be of the same type, such as a solar power plant. Meanwhile, for the convenience of explanation, the multiple power plants (20) illustrated in FIG. 1 may be described separately as a plurality of existing power plants (21) and one new power plant (22).

[0033] The server (10) is a subject that performs the power generation prediction method of the present invention, and can be built within a management system that is connected to multiple distributed power plants and manages each power plant as illustrated in FIG. 1.

[0034] Such a server may be configured to include a computing device such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit) to perform the computational operations described below, and in addition, to perform the operations described below, the server (10) may include at least one physical element among ASICs (application specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field programmable gate arrays), processors, microprocessors, controllers, and micro-controllers.

[0035] Meanwhile, the external server (30) described in this specification is a subject that provides information that serves as the basis for the operation of the server (10), and may be a concept that includes different servers corresponding to the provided information. In other words, the external server (30) does not mean a single server, but may be a concept that collectively refers to multiple servers depending on the provided information.

[0036] Referring to FIG. 2, a method for predicting power generation according to one embodiment of the present invention may include a step (S10) of creating a reference cluster for environmental variables accumulated and collected from each power plant, a step (S20) of creating a new cluster for new environmental variables collected from a new power plant, and a step (S30) of additionally utilizing the new environmental variables in reinforcement learning of a neural network model that performs a power generation prediction task based on the similarity between the reference cluster and the new cluster.

[0037] However, the power generation prediction method illustrated in FIG. 2 is according to one embodiment, and each step of the invention is not limited to the embodiment illustrated in FIG. 2, and some steps may be added, changed, or deleted as needed.

[0038] Before explaining each step illustrated in Fig. 2, a method for predicting the power generation of each distributed power plant (20) using a neural network model by the server (10) will be explained.

[0039] The neural network model (100) of the present invention can perform a task of predicting power generation based on environmental variables, and for this purpose, can be subjected to reinforcement learning by the server (10). This neural network model (100) can be implemented as a variety of models used in value-based or policy-based reinforcement learning.

[0040] Using Figure 3 as an example, reinforcement learning can be implemented by defining an environment, an agent, a state, an action, and a reward. Here, the environment can refer to the space or background where learning takes place, and the agent can be a subject that interacts with the environment and takes action. The state can refer to the situation of a given agent within the environment, and the action can refer to the decision the agent makes within the given environment. The reward refers to the reward for the agent's actions in the current environment, and reinforcement learning can be an algorithm that induces the agent to act in a way that maximizes the reward.

[0041] In this structure, the neural network model (100) of the present invention can function as an agent, and the server (10) can provide a state value (S t ) and reward value (R t ) is provided to the neural network model (100), and the neural network model (100) performs its own action (A t ) according to the compensation value (R) t) can be set up to maximize the algorithm. Since the method of implementing reinforcement learning is already known in the art, further detailed explanation will be omitted.

[0042] Referring again to FIG. 3, in order to enable the neural network model (100) to perform the power generation prediction task, the server (10) converts the accumulated daily environmental variables collected from the existing power plant (21) into state values ​​(S) in the reinforcement learning step. t ) can be set. The neural network model (100) sets the expected power generation by date within a preset range as an action value (A t ) can be output, and at this time, the server (10) outputs a compensation value (R) that is inversely proportional to the difference between the expected power generation output from the neural network model (100) and the actual power generation by date. t ) can be set.

[0043] This reinforcement learning can be performed for each past date, and as the learning is repeated, the neural network model (100) generates a reward value (R t ) can be selected so that the expected power generation is maximized, that is, the difference between the expected power generation and the actual power generation is minimized.

[0044] The server (10) can apply the neural network model (100) learned in this way to all distributed power plants (20) to predict power generation. For example, in order to predict power generation of power plant A for a future target date, the server (10) can first collect predicted environmental variables for the target date from an external server (30) at the location where power plant A is installed. For example, if power plant A is a solar power plant, the server can collect environmental variables such as predicted sunshine hours, temperature, cloud cover, humidity, and precipitation at the location where power plant A is installed from an external server (30) managed by the Korea Meteorological Administration.

[0045] Next, the server (10) can input the collected environmental variables into the neural network model (100). Since the neural network model (100) has been trained to select the expected power generation corresponding to the environmental variables through the aforementioned reinforcement learning, it can input the predicted environmental variables of the target date that were not used in the learning and output the expected power generation of the target date.

[0046] In this way, the server (10) can apply a single neural network model (100) to predict the power generation of all distributed power plants.

[0047] Meanwhile, in the present invention, environmental variables may include any environmental factors that may affect the amount of power generated, and in addition to the aforementioned sunshine hours, temperature, cloud cover, humidity, and precipitation, may include various environmental factors depending on the power generation method of the power plant.

[0048] When using a neural network model (100) for power generation prediction, many environmental variables and resources for processing them are required for reinforcement learning. However, in a server (10) that controls multiple power plants (20), there is a limitation in that it is practically difficult to perform re-learning each time a new power plant (22) is introduced by additionally utilizing environmental variables collected from the new power plant (22).

[0049] The present invention was conceived with this problem in mind, and is characterized by determining whether to proceed with additional learning based on environmental variables collected from a new power plant (22). Each step illustrated in Figure 2 will be described in detail below.

[0050] The server (10) can cluster environmental variables accumulated and collected from each power plant (20) to create multiple reference clusters (S10). As previously described, environmental variables include environmental factors that can affect power generation, and these environmental variables can be acquired from multiple environmental sensors installed in each power plant (20).

[0051] The server (10) may collect environmental variables directly from environmental sensors, or may collect environmental variables through a system such as an Energy Management System (EMS) that manages the resources of each power plant (20). This collection operation may be performed continuously over time, and the server (10) may accumulate and store the collected environmental variables in a database. The accumulated and stored environmental variables in this manner may be utilized for training the neural network model (100) described above.

[0052] Next, the server (10) can cluster each of the accumulated collected and stored environment variables according to similarity to create multiple reference clusters.

[0053] Referring to FIG. 4, the server (10) can calculate the mutual similarity of individual environmental variables and cluster each environmental variable based on the similarity. Since each power plant (20) may be installed in a different location, the environmental variables collected from each power plant (20) may have a tendency based on geographical characteristics.

[0054] For example, as shown in Fig. 4, environmental variables collected from the first power plant (20) are grouped into the first cluster (C) with high mutual similarity. ref1 ) can be formed, and the environmental variables collected from the second power plant (20) also have high mutual similarity to the second cluster (C ref2 ) can be formed, and the environmental variables collected from the third power plant (20) also form a third cluster (C) with high mutual similarity. ref3 ) can be formed.

[0055] Referring back to FIG. 1, when a server (10) is managing a number of existing power plants (21), a situation may arise where the server (10) must additionally manage a new power plant (22) due to the installation of a new power plant (20) or the introduction of a power plant (20) that was not previously managed.

[0056] At this time, the server (10) can collect new environmental variables from the new power plant (22) and cluster the new environmental variables to create a new cluster (S20). When a new power plant (22) not previously registered in the database is added as a management target, the server (10) can collect new environmental variables from multiple environmental sensors installed in the new power plant (22). The collection period for the new environmental variables can be set by the user, and can be set to, for example, one month.

[0057] At this time, the new environmental variables collected may be of the same type as the environmental variables accumulated and collected from the existing power plant (21). For example, if the environmental variables accumulated and collected from the existing power plant (21) are sunshine hours, temperature, cloud cover, humidity, and precipitation, the new environmental variables collected from the new power plant (22) may also be sunshine hours, temperature, cloud cover, humidity, and precipitation.

[0058] Referring to Fig. 5, environmental variables collected from a new power plant (22) may have a tendency according to the geographical characteristics where the new power plant (22) is installed, and may be a new cluster (C) different from the environmental variables collected from an existing power plant (21). new ) can be formed.

[0059] The server (10) is a previously created reference cluster (C ref ) and new cluster (C new ) can be compared to determine whether to perform additional learning of the neural network model (100). Specifically, the server (10) compares the previously generated reference cluster (C ref ) and new cluster (C new ) can be additionally utilized in reinforcement learning of the neural network model (100) based on the similarity between them (S30).

[0060] As shown in the example in Fig. 5, a new cluster (C new ) is the reference cluster (C ref ) can be created in different locations. The server (10) can be created in each reference cluster (Cref ) and new cluster (C new ) can be used to calculate the similarity between them.

[0061] In one embodiment, the server (10) comprises a plurality of reference clusters (C ref ) and each representative value of the new cluster (C new ) can be used to calculate the distance between representative values, and determine the similarity inversely proportional to the calculated distance.

[0062] Specifically, the server (10) each reference cluster (C ref ) is the mean or median of the environmental variables included in each reference cluster (C ref ) can be set as a representative value of a new cluster (C new ) is the average or median of the environmental variables included in the new cluster (C new ) can be set as a representative value. Next, the server (10) sets each reference cluster (C ref ) and the representative values ​​of the new cluster (C new ) can be calculated for each representative value, and the similarity inversely proportional to this can be determined.

[0063] Additionally, in one embodiment, the server (10) comprises a plurality of reference clusters (C ref ) among the environment variables included in each reference cluster (C ref ) and the representative values ​​closest to the center of the new cluster (C new ) among the environment variables included in the new cluster (C new ) can be calculated to find the distance between the representative values ​​closest to the center, and the similarity inversely proportional to the calculated distance can be determined.

[0064] Specifically, as illustrated in FIG. 5, each cluster can be partitioned into a circle or an ellipse. The server (10) can identify the center of each cluster and set the environment variable closest to the center among the environment variables belonging to each cluster as a representative value. Meanwhile, if the cluster is not a circle or an ellipse, the server (10) can identify the center of gravity of the area formed by the cluster as the center of the cluster. Subsequently, the server (10) can identify each reference cluster (C ref ) and the representative values ​​of the new cluster (C new ) can be calculated for each representative value, and the similarity inversely proportional to this can be determined.

[0065] Additionally, in one embodiment, the server (10) comprises a plurality of reference clusters (C ref ) among the environment variables included in each reference cluster (C ref ) and the representative values ​​closest to the mean of the new cluster (C new ) among the environment variables included in the new cluster (C new ) can each calculate the distance between the representative values ​​closest to the average value.

[0066] Specifically, the server (10) can calculate the average value of the environmental variables included in each cluster, and set one environmental variable closest to the average value among the environmental variables belonging to each cluster as a representative value. Subsequently, the server (10) can calculate the average value of the environmental variables included in each cluster, and set the representative value of the environmental variables included in each cluster as a representative value. ref ) and the representative values ​​of the new cluster (C new ) can be calculated for each representative value, and the similarity inversely proportional to this can be determined.

[0067] Additionally, in one embodiment, the server (10) comprises a plurality of reference clusters (C ref ) for a new cluster (C new ) to calculate the silhouette value of multiple reference clusters (C ref ) and new cluster (C new) can be used to calculate the silhouette coefficient, and the similarity proportional to the calculated silhouette value or coefficient can be determined.

[0068] Specifically, the reference cluster (C ref ) to C J , and a new cluster (C new ) to C I When expressed as , the server (10) has multiple reference clusters (C) according to [Mathematical Formula 1] below. ref ) for a new cluster (C new ) can be used to calculate the silhouette value.

[0069] [Mathematical Formula 1]

[0070]

[0071]

[0072]

[0073] (s(i) is the silhouette value, i is the new cluster (C new ) is the environment variable included in the reference cluster (C ref ) is the environment variable included in d(i,j) is the distance between i and j)

[0074] Server (10) is a new cluster (C new ) I can calculate silhouette values ​​for each of my environment variables and determine a similarity proportional to their average value.

[0075] Furthermore, the server (10) creates a new cluster (C) in the same way. new ) as well as the reference cluster (C ref ) For each of my environmental variables, a silhouette value is calculated, and based on the average silhouette value of each cluster, the silhouette coefficient can be calculated as shown in [Mathematical Formula 2] below.

[0076] [Equation 2]

[0077]

[0078] (K is the total number of clusters, is the average silhouette value of the kth cluster)

[0079] The server (10) can determine a similarity proportional to the silhouette coefficient.

[0080] If the similarity calculated according to the method described above is less than the reference value, the server (10) can additionally utilize the new environmental variable in the reinforcement learning of the neural network model (100). That is, the more the environmental variable collected from the new power plant (22) is not similar to the environmental variable collected from the existing power plant (21), the less accurate the power generation prediction operation of the neural network model (100) for the new power plant (22) may be. On the other hand, if the environmental variable collected from the existing power plant (21) is similar to the environmental variable collected from the existing power plant (21), the power generation prediction accuracy of the neural network model (100) for the new power plant (22) may be high even without separate learning.

[0081] Considering this, if the similarity is lower than the reference value, the server (10) performs additional reinforcement learning on the neural network model (100) using new environmental variables, and conversely, if the similarity is higher than the reference value, the existing neural network model (100) can be used as is for the power generation prediction operation for the new power plant (22).

[0082] Meanwhile, among the examples of calculating similarity, the reference cluster (C ref ) and the representative values ​​of the new cluster (C new ) is calculated based on the distance of the representative value, the similarity can be calculated in multiple ways, and in this case, if all of the multiple similarities are lower than the reference value, the server (10) performs additional reinforcement learning on the neural network model (100), and if any one of the multiple similarities is higher than the reference value, the existing neural network model (100) can be used as is for the power generation prediction operation for the new power plant (22).

[0083] Referring again to Figure 5, a new cluster (C new) is the reference cluster (C ref ) is formed in a completely different location, the similarity may be calculated to be lower than the reference value. At this time, the server (10) creates a new cluster (C new ) can be used to perform additional reinforcement learning on the neural network model (100).

[0084] On the other hand, referring to Figure 6, a new cluster (C new ) is the reference cluster (C ref ) is formed in a very close location, the similarity can be calculated to be higher than the reference value. At this time, the server (10) can perform a power generation prediction operation by the existing neural network model (100) for each distributed power plant (20) without additionally training the neural network model (100).

[0085] Meanwhile, since the decision on whether to perform additional training of the neural network model (100) must be made quickly, as described above, the server (10) can only collect environmental variables from the new power plant (22) for a short period of time. However, as time passes, the environment of the new power plant (22) may change, and the environmental variables may also change.

[0086] Taking this into consideration, even if the server (10) determines that additional learning of the neural network model (100) is not necessary because the previously calculated similarity exceeds the reference value, the server can collect new environmental variables again from the new power plant (22) after a preset period of time has elapsed.

[0087] After recollecting new environment variables, the server (10) can continuously determine whether to perform additional reinforcement learning of the neural network model (100) by repeating the operations of step (S20) and step (S30).

[0088] As described above, the present invention has the advantage of eliminating the need to spend a lot of time and resources to train the neural network model (100) each time a power plant (20) to be managed is added when predicting the power generation of a plurality of distributed power plants (20) through a neural network model (100), by determining whether to perform additional learning of the neural network model (100) based on environmental variables collected from a new power plant (22).

[0089] Although the present invention has been described with reference to the drawings exemplified above, it is to be understood that the present invention is not limited to the embodiments and drawings disclosed herein, and that various modifications may be made by those skilled in the art within the scope of the technical idea of ​​the present invention. Furthermore, even if the operational effects according to the configuration of the present invention have not been explicitly described while describing the embodiments of the present invention, it is natural that the effects predictable by the corresponding configuration should also be acknowledged.

Claims

1. A method for predicting the power generation of each power plant by using a neural network model, wherein a server within a control system that is connected to multiple distributed power plants and manages each power plant, A step in which the server clusters environmental variables accumulated and collected from each power plant to create multiple reference clusters; The step of the above server collecting new environment variables from a new power plant and clustering the new environment variables to create a new cluster; and The server further includes a step of additionally using the new environmental variable in reinforcement learning of the neural network model if the similarity between the plurality of reference clusters and the new cluster is less than the reference value, and re-collecting the new environmental variable from the new power plant after a preset period of time if the similarity exceeds the reference value. A method for predicting power generation of distributed power plants using reinforcement learning.

2. In paragraph 1, The above neural network model Reinforcement learning is performed by using the daily environmental variables as state values, the daily expected power generation as action values, and the reward value inversely proportional to the difference between the daily actual power generation and the above expected power generation. A method for predicting power generation of distributed power plants using reinforcement learning.

3. In paragraph 1, The steps for creating the above reference cluster are A step of accumulating and collecting the environmental variables from multiple environmental sensors installed in each of the above power plants, A step of generating multiple reference clusters by clustering each of the above accumulated collected environmental variables according to similarity. A method for predicting power generation of distributed power plants using reinforcement learning.

4. In paragraph 1, The steps to collect the above new environment variables are: A step of collecting new environmental variables of the same type as the accumulated collected environmental variables from multiple environmental sensors installed in a new power plant that is not registered in the database. A method for predicting power generation of distributed power plants using reinforcement learning.

5. In paragraph 1, The steps to additionally utilize the above new environment variables in reinforcement learning are: A step of calculating the distance between each representative value of the plurality of reference clusters and the representative value of the new cluster, respectively. A method for predicting power generation of distributed power plants using reinforcement learning.

6. In paragraph 1, The steps to additionally utilize the above new environment variables in reinforcement learning are: A step of calculating the distance between the representative values ​​closest to the center of each reference cluster among the environmental variables included in the above multiple reference clusters and the representative value closest to the center of the new cluster among the environmental variables included in the new cluster. A method for predicting power generation of distributed power plants using reinforcement learning.

7. In paragraph 1, The steps to additionally utilize the above new environment variables in reinforcement learning are: A step of calculating the distance between the representative values ​​closest to the average value of each reference cluster among the environmental variables included in the above multiple reference clusters and the representative value closest to the average value of the new cluster among the environmental variables included in the new cluster. A method for predicting power generation of distributed power plants using reinforcement learning.

8. In paragraph 1, The steps to additionally utilize the above new environment variables in reinforcement learning are: A step of calculating a silhouette coefficient for the above multiple reference clusters and the new cluster. A method for predicting power generation of distributed power plants using reinforcement learning.

9. In paragraph 1, The steps to additionally utilize the above new environment variables in reinforcement learning are: A step of calculating a silhouette value of the new cluster for the plurality of reference clusters is included. A method for predicting power generation of distributed power plants using reinforcement learning.

Images