KR102583178B1 - Method and apparatus for predicting power generation - Google Patents

Abstract

The embodiment relates to a method and device for predicting power generation. The power generation is predicted by classifying input variables into a plurality of characteristic data sets using the morphological characteristics of the graph of the input variable, and the relationship between the input variables and the predicted power generation is predicted using a graph. Provides a visually expressed method and device for predicting power generation.

Description

Method and device for predicting power generation {METHOD AND APPARATUS FOR PREDICTING POWER GENERATION}

Embodiments of the present invention relate to a method and device for predicting power generation, and more specifically, to a method and device for predicting power generation by deriving a graph of input data and predicting power generation using the shape of the graph.

Solar power generation is attracting attention as a powerful substitute for fossil fuels and as a key player in the smart grid. It is difficult to predict the amount of solar power generation because the amount varies greatly depending on weather conditions and the physical conditions of the power generation complex. This leads to an imbalance in energy production and distribution and reduces the stability of the power system. Therefore, highly accurate prediction of solar power generation is necessary for stable smart grid operation.

Multivariate time series data is collected from solar panels that produce solar energy. Multivariate time series data has nonlinearity and nonstationarity. Accurate prediction of multivariate time series data is difficult because statistical independent variables such as mean, variance, and covariance change over time and irregular fluctuations occur.

In particular, if there is a strong correlation between input variables, multicollinearity problems may occur during the learning process. Multicollinearity refers to a case where there is a strong correlation between input variables for learning a prediction model, rather than an independent relationship.

Therefore, if multicollinearity exists in the data, there is a problem that it is impossible to estimate the coefficients of the prediction model, or the prediction results of the prediction model may vary greatly depending on subtle changes in the data.

Additionally, in order to predict multivariate time series data, various studies have been conducted to predict future data changes by analyzing dependency relationships between the past and present in multivariate time series data. However, conventional prediction methods do not guarantee the order and continuity of learning data when training a prediction model, so there are limitations in time series prediction.

In addition, since it has a cyclical structure that preserves the past state, complexity increases as training of the prediction model is repeated, making it difficult to explain the internal structure of the model or the process of deriving prediction data.

The purpose of the power generation prediction method and device according to an embodiment of the present invention is to provide a power generation prediction method and device for predicting power generation using the form of an input data graph.

In addition, the method and device for predicting power generation according to an embodiment of the present invention are intended to provide a method and device for predicting power generation to prevent multicollinearity between input variables.

In addition, the power generation prediction method and device according to an embodiment of the present invention are intended to provide a power generation prediction method and device for visualizing and providing the process of deriving the power generation prediction result.

However, the technical challenge that this embodiment aims to achieve is not limited to the technical challenges described above, and other technical challenges may exist.

As a technical means for achieving the above-described technical problem, a method for predicting power generation includes generating a first graph for one or more input variable data among a plurality of input variable data, the input based on the form of the first graph. It includes classifying variable data into a plurality of feature data sets and predicting power generation by inputting the input variable data into a power generation prediction model learned using the plurality of feature data sets.

In addition, the step of generating the first graph according to the embodiment includes generating a second graph using one or more of the plurality of input variable data as an independent variable and/or a dependent variable, using the second graph deriving a correlation between the independent variable and the dependent variable, selecting a third graph whose correlation is greater than or equal to a preset reference value, and using one or more of the input variables corresponding to the third graph. 1 Includes the step of creating a graph.

In addition, the step of predicting the amount of power generation according to the embodiment includes the input variable data using one or more of the maximum value, minimum value, location of the inflection point, number of inflection points, slope, distribution, and average value of the first graph. It includes the step of classifying data sets.

In addition, the power generation prediction device according to the embodiment includes a memory for storing a power generation prediction program, and a processor for executing the power generation prediction program stored in the memory, wherein the processor executes the power generation prediction program to generate a plurality of input variables. Generating a first graph for one or more input variable data among the data, classifying the input variable data into a plurality of feature data sets based on the shape of the first graph, and learning using the plurality of feature data sets The power generation amount is predicted by inputting the input variable data into the generated power generation prediction model.

The power generation prediction method and device according to an embodiment of the present invention can provide a power generation prediction method and device for predicting power generation using the form of an input data graph.

In addition, the method and device for predicting power generation according to an embodiment of the present invention can provide a method and device for predicting power generation to prevent multicollinearity between input variables.

In addition, the power generation prediction method and device according to an embodiment of the present invention can provide a power generation prediction method and device for visualizing and providing the process of deriving the power generation prediction result.

1 is a configuration diagram of a power generation prediction device according to an embodiment.
Figure 2 is a flowchart of a method for predicting power generation according to an embodiment.
Figure 3 is a flowchart of a first graph generation method according to an embodiment.
Figure 4 is an example diagram of a second graph according to an embodiment.
Figure 5 is an example diagram of a power generation graph according to an embodiment.
Figure 6 is an example diagram of a first graph according to an embodiment.
Figure 7 is an example diagram of feature data set classification according to an embodiment.
Figure 8 is an example diagram of feature data set classification according to an embodiment.
Figure 9 is an example diagram of an ensemble model according to an embodiment.
Figure 10 is a conceptual diagram of data processing of an ensemble model according to an embodiment.
Figure 11 is a flowchart of a method for predicting power generation according to an embodiment.
Figure 12 is a flowchart of a fourth graph generation method according to an embodiment.
Figure 13 is an example diagram of a fifth graph according to an embodiment.
Figure 14 is an example diagram of a fifth graph according to an embodiment.
Figure 15 is an example diagram of a fifth graph according to an embodiment.
Figure 16 is an example diagram of a sixth graph according to an embodiment.
Figure 17 is an exemplary diagram of power generation prediction results according to an embodiment.
Figure 18 is an exemplary diagram of power generation prediction results according to an embodiment.
Figure 19 is an exemplary diagram of power generation prediction results according to an embodiment.

Below, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. In order to clearly explain the present invention in the drawings, parts that are not related to the description are omitted, and similar parts are given similar reference numerals throughout the specification.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only the case where it is "directly connected," but also the case where it is "electrically connected" with another element in between. . Additionally, when a part "includes" a certain component, this means that it may further include other components rather than excluding other components, unless specifically stated to the contrary.

In this specification, 'device' includes a unit realized by hardware, a unit realized by software, and a unit realized using both. Additionally, one unit may be realized using two or more pieces of hardware, and two or more units may be realized using one piece of hardware. Meanwhile, 'unit' is not limited to software or hardware, and the 'unit' may be configured to reside in an addressable storage medium or may be configured to reproduce one or more processors. Therefore, as an example, a 'unit' may include components such as software components, object-oriented software components, class components and task components, processes, functions, properties, procedures, Includes subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within the components and 'units' may be combined into a smaller number of components and 'units' or may be further separated into additional components and 'units'. Additionally, components and 'units' may be implemented to refresh one or more CPUs within the device.

In addition, the attached drawings are only for easy understanding of the embodiments disclosed in this specification, and the technical idea disclosed in this specification is not limited by the attached drawings, and all changes included in the spirit and technical scope of the present invention are not limited. , should be understood to include equivalents or substitutes.

Terms containing ordinal numbers, such as first, second, etc., may be used to describe various components, but the components are not limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

When a component is said to be "connected" or "connected" to another component, it is understood that it may be directly connected to or connected to the other component, but that other components may exist in between. It should be. On the other hand, when it is mentioned that a component is “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

Singular expressions include plural expressions unless the context clearly dictates otherwise.

In this application, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

Hereinafter, the configuration of the power generation prediction device 1 according to the embodiment will be described with reference to FIG. 1.

1 is a configuration diagram of a power generation prediction device according to an embodiment.

Referring to FIG. 1, the power generation prediction device 1 predicts power generation using input variable data and power generation data. Accordingly, power generation data can be received from the generator 100, or power generation data and input variable data accumulated and stored in the database 200 can be received.

That is, the generator 100 may transmit power generation data to the database 200 or memory 300. Generator 100 may include one or multiple generators. Additionally, the database 200 collects and stores power generation data received from the generator 100 or big data regarding power generation and input variables, and can transmit this to the memory 300.

Input variable data may include multivariate time series data related to weather. For example, when the generator 100 includes a plurality of solar power generators, the input variable data may include time series data regarding temperature, humidity, sunlight, river flow, wind speed, sunrise and sunset times, etc. Additionally, the input variable data may include not only measured data but also settings arbitrarily set by the user.

The above-mentioned solar power generator and weather data are merely examples described for convenience of explanation, and the embodiment is not limited thereto. The generator 100 may include a wind power generator, a thermal power generator, a hydroelectric power generator, a nuclear power generator, etc., and the input variable data includes not only weather data but also various time series data or data that can be expressed as a graph for predicting power generation as input variable data. Available.

The memory 300 stores a power generation prediction program. The power generation prediction program was established for convenience of explanation, and the name itself does not limit the function of the program. The memory 300 includes data generated or measured by the generator 100, data stored in the database 200, information and data required for functions performed by the processor 400, and data generated according to execution of the processor 400. At least one of them can be saved.

Memory 300 should be interpreted as a general term for non-volatile storage devices that continue to retain stored information even when power is not supplied and volatile storage devices that require power to maintain stored information. Additionally, the memory 300 may perform a function of temporarily or permanently storing data processed by the processor 400.

The memory 300 may include magnetic storage media or flash storage media in addition to volatile storage devices that require power to maintain stored information, but the scope of the present invention is not limited thereto. no.

In addition, the database 200 may receive and store accumulated data as a result of the power generation prediction program, data incidentally generated for predicting power generation, etc. from the memory 300. The database 200 may form part of the memory 300, but is not necessarily located inside the solar power generation prediction device 1 and may be located outside the device.

The processor 400 is configured to execute the power generation prediction program stored in the memory 300. The processor 400 may include various types of devices that control and process data. The processor 400 may refer to a data processing device built into hardware that has a physically structured circuit to perform functions expressed by codes or instructions included in a program. In one example, the processor 400 may include a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), or an FPGA ( It may be implemented in the form of a field programmable gate array, etc., but the scope of the present invention is not limited thereto.

The processor 400 is configured to execute the power generation prediction program and perform the following functions and procedures.

The processor 400 executes a power generation prediction program, visualizes input variable data that affects power generation, and classifies the input variable data into a plurality of feature data sets using morphological characteristics of the visualized data.

Creating a graph using input variable data can be used as a method of visualization. Therefore, the feature data set refers to data classified using visual features such as pattern separation and morphological features using a graph of input variable data. At this time, morphological features refer to one or more of the visually recognized features such as the maximum value, minimum value, location of inflection points, number of inflection points, slope, distribution, and average value of the graph.

The processor 400 executes a power generation prediction program to perform machine learning learning and power generation prediction using a feature data set.

Specifically, the processor 400 may execute a power generation prediction program to generate a power generation prediction model. The power generation prediction model may be an artificial intelligence model learned to derive power generation prediction results using a feature data set as learning data.

At this time, the power generation prediction model uses an ensemble model. Therefore, power generation prediction for each input variable data is performed using a feature data set corresponding to each of the plurality of input variable data.

In addition, the processor 400 executes a power generation prediction program to perform one or more of missing value processing, Yeo-Johnson Transformation, and min-max scaling for the feature data set. Preprocessing of the feature data set can be performed by performing .

Additionally, the processor 400 may execute a power generation prediction program and perform learning on input variable data using one or more of Recurrent Neural Networks (RNN), Long Short Term Memory (LSTM), and DENSE structures.

In addition, the processor 400 may execute a power generation prediction program to generate a graph regarding the causal relationship between input variable data or feature data sets and the power generation prediction result to explain the process of deriving the power generation prediction result.

That is, the processor 400 can execute the power generation prediction program and use intermediate result data derived in the process of deriving the power generation prediction result to express the process of deriving the power generation prediction result through visual means.

The method for predicting power generation using a power generation prediction program will be described in detail with reference to FIGS. 2 to 19 described later.

Additionally, the processor 400 can transmit and receive information with the terminal 500 using the communication unit. Accordingly, input variable data can be received from the terminal 500, or the results of the power generation prediction program can be transmitted to the terminal 500 and displayed to the user.

The terminal 500 is, for example, a laptop equipped with a web browser, a desktop, a laptop, a wireless communication device that guarantees portability and mobility, or any type of device such as a smartphone, tablet PC, etc. It may refer to a handheld-based wireless communication device. In addition, the communication network may be a wired network such as a Local Area Network (LAN), Wide Area Network (WAN), or Value Added Network (VAN), a mobile radio communication network, or a satellite communication network. It can be implemented with all types of wireless networks.

Hereinafter, a method for predicting power generation according to an embodiment will be described with reference to FIG. 2.

Figure 2 is a flowchart of a method for predicting power generation according to an embodiment.

Referring to FIG. 2, in the first graph generating step (S100), a first graph is generated to visualize each of a plurality of input variable data. Then, in the feature data set classification step (S200), the input variable data is classified into a plurality of feature data sets using the morphological characteristics of the first graph. In the learning and prediction result derivation step (S300), the power generation prediction result is derived by learning the power generation prediction program using the feature data set.

Hereinafter, a method for generating a first graph according to an embodiment will be described with reference to FIGS. 3 and 8.

FIG. 3 is a flowchart of a method for generating a first graph according to an embodiment, FIG. 4 is an exemplary diagram of a second graph according to an embodiment, and FIG. 5 is an exemplary diagram of a power generation graph according to an embodiment.

Referring to FIG. 3, in the second graph creation step (S110), a second graph is generated using a plurality of input variable data with one or more of the input variables as an independent variable and/or a dependent variable.

Figure 4 shows that in order to predict solar power generation, the input variables are precipitation (Rain), temperature (Temprt), wind speed (Wnd_spd), humidity (Humdt), cloud amount (Cloud), and solar power generation (Pow). 2 This is an example of a graph.

The x-axis corresponds to the independent variable and the y-axis corresponds to the dependent variable, and a pair plot is created by selecting any one of the above-mentioned input variables as the independent variable and dependent variable. Therefore, 36 graphs are created with precipitation (Rain), temperature (Temprt), wind speed (Wnd_spd), humidity (Humdt), cloud amount (Cloud), and solar power generation (Pow) set as independent and dependent variables, respectively. .

The pair plot shown in FIG. 4 can visually express the correlation between independent variables and dependent variables. That is, the second graph is a graph representing the correlation between input variables. Specifically, the more similar the shape of the pair plot graph is to the shape of the y=x graph, the stronger the positive correlation between the independent variable and the dependent variable. However, the more similar the shape of the pair plot graph is to the shape of the y=-x graph, the stronger the negative correlation between the independent variable and the dependent variable.

Additionally, as shown in FIG. 4, the second graph according to the embodiment provides a brushing function to check the correlation of data in a specific section. The brushed data section is displayed in blue, and the correlation between data in a specific section can be immediately identified using color.

Therefore, using the second graph, one of the input variables can be selected as the target variable and the input variable with a high correlation with the target variable can be identified. In addition, learning of the power generation prediction device 1 can be performed by selecting input variables that have a strong correlation.

If there is a strong correlation between two input variables that are not target variables, multicollinearity problems may occur during the learning process. Multicollinearity refers to a case where there is a strong correlation between input variables for learning a prediction model, rather than an independent relationship.

Therefore, if multicollinearity exists in the data, it may not be possible to estimate the coefficients of the prediction model, or the prediction results of the prediction model may vary greatly depending on subtle changes in the data. Therefore, the occurrence of multicollinearity can be reduced by selecting one of the input variables as the target variable and performing learning on the other input variable that has a high correlation with the target variable.

To create the first graph, a graph whose shape is similar to that of the y=x graph or the y=-x graph among the second graphs may be selected as the third graph. Then, the first graph can be created using the input variable corresponding to the third graph. Additionally, the degree of correlation between the independent variable and the dependent variable can be derived using the form of the second graph, and a graph whose degree of correlation is above a preset reference value can be set as the third graph. Then, the first graph can be created using the input variable corresponding to the third graph.

Figure 5 is an example of a target variable graph, which is a graph of solar power generation according to time series changes. If the target variable is selected as solar power generation, a line chart of solar power generation can be created as shown in FIG. 5.

Specifically, (a) in FIG. 5 is a time unit (HOUR), (b) in FIG. 5 is a unit of year, (c) in FIG. 5 is a unit of season, and (d) in FIG. 5 is a unit of month. Shows solar power generation in (month) units.

However, the embodiment is not limited to this, and can be expressed using various graphs and methods that can effectively visualize data depending on input variables.

Figure 6 is an example diagram of the first graph.

Figure 6 shows a case where the first graph for wind speed, humidity, solar radiation, ground temperature, and cloud cover is generated when the target variable is solar power generation. That is, among the second graphs, solar power generation is the independent variable, and a graph with wind speed, humidity, solar radiation, ground temperature, and cloud cover as dependent variables is selected as the third graph, and wind speed, This is an example in which the first graph for humidity, solar radiation, ground temperature, and cloud cover was created.

The first graph is used to derive features of input variable data, and the input data can be classified into a plurality of feature data sets using the morphological features of the first graph.

Hereinafter, a feature data set classification method according to an embodiment will be described with reference to FIG. 7.

Figure 7 is an example diagram of feature data set classification according to an embodiment.

Figures 7(a) and 7(b) show a data division group in which input variable data related to ground temperature among the plurality of first graphs described above is classified into a plurality of feature data sets. In other words, it indicates dividing input variable data about ground temperature into four groups (Group0, Group1, Group2, Group3). To visually clearly express the division of each group, each group may be displayed in a different color.

Specifically, when selecting the first graph regarding the ground temperature shown in FIG. 6, one morphological feature such as the maximum value, minimum value, location of the inflection point, number of inflection points, slope, distribution, average value, etc. of the first graph is selected. Using the above, the ground temperature data is classified into four feature data sets.

Figure 7(c) shows a line chart of ground temperature, which is the selected input variable, and Figure 7(d) shows a line chart of solar power generation amount, which is a target variable. Therefore, it is possible to visually express that the input variable is classified into a plurality of feature data sets according to the morphological characteristics of the graph.

At this time, Figure 7 shows only an example of classifying the feature data set by selecting ground temperature as the input variable, but the embodiment of the present invention is not limited to this and all input variables can be selected. In addition, an example of classifying input variable data into four groups is shown, but the embodiment of the present invention is not limited to this, and the number of groups and feature data sets classified according to the morphological characteristics of the first graph may vary. there is.

Figure 8 is an example diagram of feature data set classification according to an embodiment.

Figure 8 shows an example of a feature data set when the input variables are weather patterns, seasonal climate, and power generation, respectively.

If the input variable is a weather pattern, the feature data set can be classified as SUNNY DATA, RAINY DATA, CLOUDY DATA, and WINDY DATA. Additionally, if the input variable is seasonal climate, the feature data set can be classified into SPRING DATA, SUMMER DATA, AUTUMN DATA, and WINTER DATA. When the input variable is power generation, the feature data set can be divided into low-power (LOW-POWER DATA) and high-power (HIGH-POWER DATA).

Hereinafter, the structure of the ensemble model according to the embodiment will be described with reference to FIGS. 9 and 10.

Figure 9 is an example diagram of an ensemble model according to an embodiment.

Referring to FIG. 9, the ensemble model refers to a learning model that performs learning in parallel on each of a plurality of input variable data and derives a power generation prediction result using each input variable data.

For example, sub-dataset 1 is a feature data set separated according to weather patterns, SUNNY DATA, RAINY DATA, and CLOUDY DATA, as shown in FIG. 8. , can respond to strong winds (WINDY DATA).

Sub-data set 1 is input to Sub-model 1, a sub-model of the ensemble model, and Sub-model 1 performs learning using feature data sets separated according to weather patterns. Then, the power generation prediction result (Prediction 1) is derived using a feature data set separated according to weather patterns.

In addition, if the input variable is seasonal climate, sub-dataset 2 will correspond to the characteristic data sets of SPRING DATA, SUMMER DATA, AUTUMN DATA, and WINTER DATA. You can.

Sub-data set 2 is input to Sub-model 2, a sub-model of the ensemble model, and sub-model 2 performs learning using the classified feature data set when the input variable is seasonal climate. In addition, the power generation prediction results (Prediction 2) are derived using feature data sets separated according to seasonal climate.

When the input variable is power generation, sub-dataset 3 may correspond to the characteristic data sets LOW-POWER DATA and HIGH-POWER DATA.

Sub-data set 3 is input to Sub-model 3, a sub-model of the ensemble model, and sub-model 3 performs learning using the classified feature data set when the input variable is power generation. Then, the power generation prediction result (Prediction 3) is derived using a feature data set separated according to the power generation amount.

In addition, the ensemble model may have a structure in which N sub-models are provided in parallel to perform learning on N input variables. Therefore, each sub-model performs learning on the input data using the classified feature data set and derives power generation prediction results for each feature data set.

Therefore, the method and device for predicting power generation according to the embodiment perform learning by separating input variables, and thus perform learning and power generation prediction to prevent multicollinearity between input variables.

The power generation prediction results (Prediction 1, 2, 3, ..., N) derived from each sub-model (Sub-model 1, 2, 3, ..., N) are stored in memory 300 and/or database ( 200). Then, the processor 400 can execute the power generation prediction program to derive the loss rate or prediction accuracy for each prediction result (Prediction 1, 2, 3, ..., N).

The loss rate or error rate can be derived using a bias-variance tradeoff, and the prediction result with the lowest loss rate or error or the prediction result with the highest prediction accuracy can be derived as the final power generation prediction result. You can.

Figure 10 is a conceptual diagram of data processing of an ensemble model according to an embodiment.

Referring to FIG. 10, (a) of FIG. 10 shows a pre-processing step. Therefore, the feature data set input as a substructure of the ensemble model goes through a preprocessing step. The preprocessing step may perform one or more of missing value processing, Yeo-Johnson Transformation, and min-max scaling on the feature data set.

Figure 10(b) shows the layer structure for learning and result prediction. The feature data set that has gone through the preprocessing stage is used for learning and power generation prediction using a layer structure. The layer may include one or more of Recurrent Neural Networks (RNN), Long Short Term Memory (LSTM), and DENSE structures.

Additionally, Figure 10(c) illustrates the hyperparameter settings. In the hyperparameter setting stage, shape, units, activation functions, bias, etc. can be set. At this time, tanh or ReLu can generally be used for activation functions.

Figure 10(d) visualizes the layer structure. Therefore, in the preprocessing step, preprocessing is performed on the feature data set input as a substructure of the ensemble model. The preprocessed feature data set is used for learning and power generation prediction through layer layers (LSTM, DENSE).

And, the power generation prediction value derived from each sub-model may be stored in the memory 300 and/or the database 200. In addition, the processor 400 uses a bias-variance tradeoff to determine the loss or error rate for each prediction result (Prediction 1, 2, 3, ..., N). can be derived. The prediction result with the lowest loss rate or error or the prediction result with the highest prediction accuracy can be derived as the final power generation prediction result.

Hereinafter, a fourth graph generation method according to an embodiment will be described with reference to FIGS. 11 to 19.

Figure 11 is a flowchart of a method for predicting power generation according to an embodiment.

Referring to FIG. 11, the method for predicting power generation according to an embodiment may further include a fourth graph generating step (S400). The fourth graph refers to a graph to graphically express the process of deriving power generation prediction results from input variables. In other words, the fourth graph refers to a graph representing the causal relationship between input variable data and power generation prediction results.

Figure 12 is a flowchart of a fourth graph generation method according to an embodiment.

Referring to FIG. 12, the contribution is derived in the contribution derivation step (S410) between the input variable and the prediction result. Then, in the fifth graph creation step (S420), the fifth graph is created using the contribution between the input variable and the prediction result.

Specifically, the contribution between input variables and prediction results can be derived using SHAP (shapley additive explanations) value. A fifth graph using SHAP value can be derived as shown in FIGS. 13, 14, and 15.

Figure 13 shows an example SHAP summary plot. The x-axis in Figure 13 represents the SHAP value indicating the contribution of the input variable. The y-axis represents the input variables, and the order of the input variables is sorted in descending order according to their contribution.

Additionally, each dot in Figure 13 is an instance within an input variable, and the color of the dot indicates the size of the instance value. Therefore, red dots correspond to higher values, and blue dots correspond to lower value instances. Additionally, the clustering of dots shows a dense distribution of SHAP values.

Accordingly, Figure 13 is a SHAP summary plot for temperature (temprt), humidity (humdt), cloud amount (cloud), and precipitation (raisn) among a plurality of input variables. Additionally, the temperature at the top has the greatest contribution to power generation, and precipitation has the lowest contribution.

Figure 14 shows an example of a decision plot. The decision plot shows the path leading to the prediction. The path proceeds from the bottom to the top of the graph. At this time, the contribution is expressed in the direction of movement of the line and the distance. The direction to the right from the center of the graph indicates positive contribution, and the direction to the left indicates negative contribution.

The distance traveled indicates the size of the contribution. Additionally, the color of the line is specified according to the quantile of the instance within the prediction results of all instances. Therefore, the temperature (temprt) with the longest moving distance of the line has the largest contribution, and the precipitation (rain) with the shortest moving distance of the line has the lowest contribution.

Figure 15 shows an example of a clustering SHAP plot. The clustering SHAP plot shows the amount of change in prediction results for the entire input dataset.

The x-axis represents the instance of the data, and instances are sorted in descending order of prediction results. The y-axis represents the prediction result and the size of contribution. The prediction result of each instance corresponds to the point where red and blue meet. Red indicates positive contribution, and blue indicates negative contribution. Therefore, if the sum of contributions of all input variables of an instance is positive, the prediction result appears higher than the reference value.

In addition, as shown in FIG. 12, the power generation prediction method according to the embodiment not only generates a fifth graph based on contribution, but also creates a sixth graph using the dependence between input variables and power generation prediction results. .

Specifically, in the dependency derivation step (S430) between input variables and prediction results, after assuming that the conditions of all other input variables (marginal feature set) are the same, a change in one input variable (selected feature set) affects the prediction result. Dependency is derived from the marginal effect. And, in the sixth graph derivation step (S440), a sixth graph is generated using the dependence between the input variables and the power generation prediction result. The sixth graph can be derived as shown in FIG. 16.

Figure 16 shows an example of an ICE (Individual Conditional Expectation) plot. The ICE plot shows the change in power generation prediction result for all instances when the value of the input variable changes. Therefore, Figure 16 is an example of an ICE plot when the input variables are power, wind, humidity, irradiance, and temperature.

In an ICE plot, dependency refers to the change relationship between input variable values and predicted results. Therefore, the dependencies are shown using lines corresponding to the dependencies of each instance.

The bold line represents the average dependence of all instances. Average dependence shows the same results as PDP (Partial Dependence Plot). Additionally, in the ICE plot, the x-axis represents the values of input variables, and the y-axis represents the power generation prediction result. In addition, the dependencies of multiple models can be color-coded and displayed on a single ICE plot.

Figures 17 to 19 are exemplary diagrams of power generation prediction results according to an embodiment.

Figure 17 is a graph showing the actual power generation value to be predicted and the power generation prediction value predicted using the power generation prediction device and method according to the embodiment. The x-axis represents the time axis (epoch), and the y-axis represents solar power generation.

L1 represents the predicted solar power generation amount predicted using all input variables. L2 represents the solar power generation value that should actually be predicted. L3 and L4 represent the solar power generation predicted values predicted by changing the hyperparameter settings.

Figures 18 and 19 are graphs for checking bias-variance tradeoff. The x-axis represents error, bias, or variance, and the y-axis represents loss. L5 represents bias and L6 represents variance. You can check the bias-dispersion trade-off by analyzing overfitting and underfitting according to time (epoch).

Additionally, the total error can be defined as Equation 1 below.

[Equation 1]

Total error = bias ² + Variance + Irreducible Error

One embodiment of the present invention may also be implemented in the form of a recording medium containing instructions executable by a computer, such as program modules executed by a computer. Computer-readable media can be any available media that can be accessed by a computer and includes both volatile and non-volatile media, removable and non-removable media. Additionally, computer-readable media may include computer storage media. Computer storage media includes both volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data.

Although the methods and systems of the present invention have been described with respect to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.

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

The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention. do.

1: Power generation prediction device 100: Generator
200: database 300: memory
400: processor 500: terminal

Claims (18)

Generating a first graph representing the correlation between any one selection variable data among a plurality of input variable data and a plurality of other input variable data;
Classifying the input variable data into a plurality of feature data sets based on the shape of the first graph, and
Predicting power generation by inputting the input variable data into a power generation prediction model learned using the plurality of feature data sets.
Including, a method for predicting power generation. According to paragraph 1,
The step of creating a graph for the input variable data is.
Generating a second graph using one or more of the plurality of input variable data as an independent variable and/or a dependent variable,
Deriving a correlation between the independent variable and the dependent variable using the second graph,
Selecting a third graph whose correlation is greater than or equal to a preset reference value, and
Generating the first graph using one or more of the input variables corresponding to the third graph.
Including, a method for predicting power generation. According to paragraph 2,
The step of classifying the input variable data into a plurality of feature data sets is:
Classifying the input variable data into the feature data set using one or more of the maximum value, minimum value, location of the inflection point, number of inflection points, slope, distribution, and average value of the first graph.
Including, a method for predicting power generation. According to paragraph 3,
The step of predicting the power generation amount is,
Creating a fourth graph in which one or more of the plurality of input variable data is set as an independent variable and the power generation prediction result is set as a dependent variable.
Including, a method for predicting power generation. According to paragraph 4,
The step of generating the fourth graph is,
Deriving a contribution between the input variable data and the power generation prediction result and generating a fifth graph using the contribution.
Including, a method for predicting power generation. According to clause 5,
The step of generating the fourth graph is,
Deriving a dependency between the input variable data and the power generation prediction result and generating a sixth graph using the dependency.
A method for predicting power generation, further comprising: According to claim 3 or 6,
The step of predicting the power generation amount is,
Learning an ensemble model using the plurality of feature data sets, and deriving a power generation prediction result using the ensemble model.
A method for predicting power generation, further comprising: In clause 7,
The step of predicting the power generation amount is,
Performing learning and power generation prediction of the feature data set using one or more of RNN (Recurrent Neural Networks), LSTM (Long Short Term Memory), and DENSE structures.
A method for predicting power generation, further comprising: According to clause 8,
The step of predicting the power generation amount is,
Performing preprocessing of the feature data set by performing one or more of missing value processing, Yeo-Johnson Transformation, and min-max scaling on the feature data set. step
A method for predicting power generation, further comprising: A memory that stores the power generation prediction program, and
It includes a processor that executes a power generation prediction program stored in the memory,
The processor executes the power generation prediction program to generate a first graph representing the correlation between any one selection variable data among the plurality of input variable data and the other plurality of input variable data, and determines the form of the first graph. A power generation prediction device that classifies the input variable data into a plurality of feature data sets as a standard, and predicts power generation by inputting the input variable data into a power generation prediction model learned using the plurality of feature data sets. According to clause 10,
The processor executes the power generation prediction program to generate a second graph using any one or more of the plurality of input variables as an independent variable and/or a dependent variable, and uses the second graph to determine the independent variable and the dependent variable. A power generation prediction device that derives the correlation of variables, selects a third graph whose correlation is greater than or equal to a preset reference value, and generates the first graph using one or more of the input variables corresponding to the third graph. . According to clause 11,
The processor executes the power generation prediction program and converts the input variable data into the feature data set using one or more of the maximum value, minimum value, location of the inflection point, number of inflection points, slope, distribution, and average value of the first graph. A classification and power generation prediction device. According to clause 12,
The processor executes the power generation prediction program to generate a fourth graph in which any one or more of the plurality of input variable data is set as an independent variable and the power generation prediction result is set as a dependent variable. According to clause 13,
The processor executes the power generation prediction program, derives a contribution between the input variable and the power generation prediction result, and generates a fifth graph using the contribution. According to clause 14,
The processor executes the power generation prediction program, derives a dependency between the input variable and the power generation prediction result, and generates a sixth graph using the dependency. According to claim 12 or 15,
The processor executes the power generation prediction program, learns an ensemble model using the plurality of feature data sets, and derives a power generation prediction result using the ensemble model. According to clause 16,
The power generation prediction program is,
A power generation prediction device that includes one or more of RNN (Recurrent Neural Networks), LSTM (Long Short Term Memory), and DENSE structures. According to clause 17,
The processor executes the power generation prediction program and performs one or more of missing value processing, Yeo-Johnson Transformation, and min-max scaling on the feature data set. A power generation prediction device that performs preprocessing of the feature data set.

Images