Abnormal power generation detection system using solar power generation tracking based on similar past data and method therefor

Abstract

The present invention relates to an abnormality detection method using solar power generation tracking based on similar past data. More particularly, the abnormality detection method using solar power generation tracking based on similar past data comprises the steps of: on the basis of past data including actually measured weather data and solar position data, extracting variable importance affecting a solar power generation amount; collecting, for an abnormal power generation detection target time point, target data including actually measured weather data, solar position data, and an actually measured power generation amount; calculating a weight for each variable on the basis of the extracted variable importance and calculating a similarity between the target data and the past data by considering the weight for each variable; predicting a power generation amount for the target data by extracting a preset number of pieces of past data in order of highest calculated similarity; and detecting abnormal power generation on the basis of an actually measured power generation amount at the abnormal power generation detection target time point and a power generation amount predicted for target data at a predicted abnormal power generation detection target time point.

Description

Abnormal power generation detection system and method using solar power generation tracking based on similar past data

[0001] The present invention relates to an abnormality detection method using solar power generation tracking based on similar past data, and more specifically, to an abnormality detection method using solar power generation tracking based on similar past data, which calculates the importance of variables affecting power generation based on actually measured past data (weather data, solar position data), and compares the predicted power generation based on similar past measured data and the actual power generation at the time of abnormal power generation detection by considering the importance of variables, thereby detecting whether or not there is abnormal power generation.

[0002] The Virtual Power Plant (VPP) market integrates and manages power plants distributed across multiple regions, and trades the electricity they produce. Solar power, a renewable energy source, is a key power source in the VPP market, accounting for more than half of the renewable energy supply.

[0003] However, solar power generation suffers from greater fluctuations in power generation and greater unpredictability compared to other power generation methods, raising concerns about the stability of transactions for distributed resources. Transactions in the virtual power plant market are conducted by presenting the next day's power generation volume and delivering the power for that time.

[0004] The power exchange then redistributes the power generated during that time and provides it to buyers. Therefore, to ensure the smooth provision of solar-generated power to the virtual power plant market, it is necessary to accurately predict the next 24 hours' power generation and detect any abnormalities in power generation.

[0005] The problem to be solved by the present invention is to provide a system and method for detecting abnormal power generation using solar power generation tracking based on similar past data, which can calculate the importance of variables affecting power generation based on measured past data (weather data, solar position data), and compare the predicted power generation based on similar past measured data and the actual power generation at the time of abnormal power generation detection with the data at the time of abnormal power generation detection by considering the importance of variables.

[0006] The tasks of the present invention are not limited to the tasks mentioned above, and other tasks not mentioned will be clearly understood by those skilled in the art of the present invention from the description below.

[0007] In order to solve the problem described above, in one embodiment of the present invention, a method for detecting abnormal power generation using solar power generation tracking based on similar past data may include (a) a step of extracting variable importance affecting solar power generation based on past data including actual weather data and solar position data, (b) a step of collecting target data including actual weather data, solar position data, and actual power generation for a target time point of abnormal power generation detection, (c) a step of calculating a weight for each variable based on the extracted variable importance, and calculating a similarity between the target data and the past data in consideration of the weight for each variable, (d) a step of extracting a preset number of past data in the order of the calculated high similarity to predict power generation for the target data, and (e) a step of detecting abnormal power generation based on the actual power generation and the predicted power generation for a target time point of abnormal power generation detection.

[0008] Additionally, the above variable importance can be extracted based on one of the Tabnet, Shap value, and tree-based learning models.

[0009] In addition, the above variable importance is extracted for each power plant, and can be extracted by selecting the model with the lowest power generation prediction error rate among the power generation prediction error rates based on the Tabnet, Shap value, and tree-based learning models.

[0010] In addition, the step (c) calculates the Euclidean distance between the target data and each past data, and can adjust the Euclidean distance for each variable by applying a weight according to the importance of the variable.

[0011] In addition, the step (d) above can predict the power generation amount by weighting the past data of a preset number in the order of high similarity calculated above.

[0012] In addition, (e) the step of detecting the above abnormal power generation can be determined as abnormal power generation if the actual power generation amount at the above abnormal power generation detection target time is lower than the power generation amount predicted in step (d).

[0013] Alternatively, (d1) further includes a step of predicting power generation based on past data including the actual weather data and solar position data, and (e) the step of detecting the abnormal power generation may determine that the power generation is abnormal if the difference between the actual power generation at the time of the abnormal power generation detection target and the final predicted power generation is outside a preset range.

[0014] At this time, the final predicted power generation amount may be the average of the predicted power generation amount based on similarity predicted in step (d) and the predicted power generation amount based on past data predicted in step (d1).

[0015] In addition, the above-mentioned actual weather data may include temperature, relative humidity, dew point, wind direction, wind speed, gust speed, ultraviolet index, visibility, cloud cover, cloud height, barometric pressure, perceived temperature, wet bulb temperature, precipitation, and snowfall.

[0016] Additionally, the above solar position data may be the altitude and azimuth of the sun.

[0017] In addition, an abnormal power generation detection system utilizing solar power generation tracking based on similar past data according to an embodiment of the present invention may include a variable importance extraction unit that extracts variable importance affecting solar power generation based on past data including actual weather data and solar position data, a similarity calculation unit that calculates a weight for each variable based on the extracted variable importance and calculates a similarity between target data and the past data in consideration of the weight for each variable, a similarity-based prediction unit that extracts a preset number of past data in the order of the calculated high similarity and predicts solar power generation for prediction target data, and an abnormal power detection unit that detects abnormal power generation based on the actual power generation at the target time of abnormal power generation detection and the predicted power generation at the target time of abnormal power generation detection predicted by the similarity-based prediction unit.

[0018] Additionally, the above variable importance can be extracted based on one of the Tabnet, Shap value, and tree-based learning models.

[0019] In addition, the above variable importance is extracted for each power plant, and can be extracted by selecting the model with the lowest power generation prediction error rate among the power generation prediction error rates based on the Tabnet, Shap value, and tree-based learning models.

[0020] In addition, the similarity calculation unit calculates the Euclidean distance between the target data and each past data, and can adjust the Euclidean distance for each variable by applying a weight according to the importance of the variable.

[0021] In addition, the similarity-based prediction unit can predict the power generation amount by weighting and averaging a preset number of past data in order of high similarity.

[0022] In addition, the above-mentioned abnormal power generation detection unit can determine that abnormal power generation occurs if the difference between the actual power generation amount at the time of the above-mentioned abnormal power generation detection target and the power generation amount predicted by the similarity-based prediction unit exceeds a preset range.

[0023] Alternatively, the method further includes a past data-based prediction unit that predicts power generation based on past data including the actual weather data and solar position data, and the abnormal power generation detection unit can determine that the actual power generation at the time of the abnormal power generation detection target is lower than the power generation predicted by the similarity-based prediction unit.

[0024] At this time, the final predicted power generation amount may be the average of the similarity-based predicted power generation amount predicted by the similarity-based prediction unit and the past data-based predicted power generation amount predicted by the past data-based prediction unit.

[0025] In addition, the above-mentioned actual weather data may include temperature, relative humidity, dew point, wind direction, wind speed, gust speed, ultraviolet index, visibility, cloud cover, cloud height, barometric pressure, perceived temperature, wet bulb temperature, precipitation, and snowfall.

[0026] Additionally, the above solar position data may be the altitude and azimuth of the sun.

[0027] Specific details of other embodiments are included in the detailed description and drawings.

[0028] According to one embodiment of the present invention, a system and method for detecting abnormal power generation using solar power generation tracking based on past similar data calculates the importance of variables affecting power generation based on measured past data (weather data, solar position data), and compares the predicted power generation based on past actual data similar to the data at the time of abnormal power generation detection with the actual power generation in consideration of the importance of variables, thereby detecting whether or not there is abnormal power generation.

[0029] Accordingly, the power generation prediction rate can be improved based on similar past data with highly reliable similar power generation conditions.

[0030] In addition, the accuracy of abnormal power generation detection can be improved by tracking the predicted power generation and comparing it with the actual power generation at the time of abnormal power generation detection.

[0031] Additionally, as actual historical data and solar power location data accumulate, weather factors, weather types, and solar power location data become more diverse, enabling more accurate power generation predictions.

[0032] The effects according to the present invention are not limited to those exemplified above, and other effects can be clearly understood by those skilled in the art from the description of the following specification.

[0033] FIG. 1 is a block diagram schematically illustrating the configuration of an abnormal power generation detection system utilizing solar power generation tracking based on similar past data according to one embodiment of the present invention.

[0034] FIG. 2 is a block diagram schematically illustrating the configuration of an abnormal power generation detection system utilizing solar power generation tracking based on similar past data according to another embodiment of the present invention.

[0035] FIG. 3 is a flowchart illustrating a method for detecting abnormal power generation using solar power generation tracking based on similar past data according to one embodiment of the present invention.

[0036] Figure 4 is a graph for explaining abnormal development detection according to various embodiments of the present invention.

[0037] * Explanation of major symbols in the drawing *

[0038] 100: Collection and Preprocessing Unit

[0039] 200: Variable Importance Extraction Section

[0040] 300: Similarity calculation unit

[0041] 400: Similarity-based prediction unit

[0042] 450: Prediction Department Based on Past Data

[0043] 500: Abnormal power generation detection unit

[0044] The following merely exemplifies the principles of the invention. Therefore, those skilled in the art will be able to implement the principles of the invention and invent various devices within the scope and spirit of the invention, even if not explicitly described or illustrated herein. Furthermore, all conditional terms and embodiments listed herein are expressly intended, in principle, to facilitate understanding of the invention, and should be understood as being in no way limited to the specifically listed embodiments and conditions.

[0045] Additionally, in the following description, ordinal expressions such as first, second, etc. are intended to describe objects that are equal and independent of each other, and should be understood as having no meaning in terms of main / sub or master / slave.

[0046] The above-described purposes, features and advantages will become clearer through the following detailed description with reference to the attached drawings, so that a person having ordinary skill in the art to which the invention pertains can easily practice the technical idea of ​​the invention.

[0047] The individual features of the various embodiments of the present invention can be partially or wholly combined or combined with each other, and as can be fully understood by those skilled in the art, various technical connections and operations are possible, and each embodiment can be implemented independently of each other or can be implemented together in a related relationship.

[0048] Hereinafter, various embodiments of the present invention will be described in detail with reference to the attached drawings.

[0049] In general, power generation predictions and abnormal generation detection accuracy can be improved by accurately measuring and utilizing meteorological data, such as through environmental sensors installed within power plants. However, considering that not all power plants have environmental sensors installed and that third-party management of the plant may not utilize these sensors, actual meteorological data collected from meteorological observatories near the power plants targeted for abnormal generation detection can be utilized.

[0050] In the abnormal power generation detection system (hereinafter referred to as the “system”) utilizing similar past data-based solar power generation tracking according to various embodiments of the present invention, the accuracy of predicted power generation and abnormal power generation detection is increased by extracting actual weather data and solar power location data that are as similar as possible to the target data at the time of abnormal power generation detection.

[0051] FIG. 1 is a block diagram schematically illustrating the configuration of an abnormal power generation detection system utilizing solar power generation tracking based on similar past data according to one embodiment of the present invention.

[0052] An abnormal power generation detection system utilizing solar power generation tracking based on similar past data according to one embodiment of the present invention may include a collection and preprocessing unit (100), a variable importance extraction unit (200), a similarity calculation unit (300), a similarity-based prediction unit (400), and an abnormal power generation detection unit (500).

[0053] The collection and preprocessing unit (100) can collect meteorological data, solar position data, etc. measured at an observatory at a preset cycle at the same time, and can perform preprocessing as data for learning such as variable importance extraction, power generation prediction, and abnormal power generation detection. That is, a multidimensional vector with each variable as a dimension can be created as an input variable and used for variable importance extraction, power generation prediction, and abnormal power generation detection.

[0054] Here, meteorological data refers to actual data observed at a meteorological observatory, and can include temperature, relative humidity, dew point, wind direction, wind speed, gust speed, UV index, visibility, cloud cover, cloud height, air pressure, perceived temperature, wet bulb temperature, precipitation, and snowfall.

[0055] Additionally, solar position data can be the solar altitude and azimuth.

[0056] The variable importance extraction unit (200) can extract variable importance affecting solar power generation based on historical data including actual weather data and solar position data. Variable importance can be extracted for each power plant, and can also be extracted for each power plant group according to the administrator's settings.

[0057] The variable importance extraction unit (200) can extract importance variables based on one of the Tabnet, Shap value, and tree-based learning models for each power plant, i.e., using the past data of the power plant as learning data. At this time, for each power plant, the model with the lowest power generation prediction error rate among the power generation prediction error rates based on the Tabnet, Shap value, and tree-based learning models can be selected and used.

[0058] At this time, the input variables can be actual historical data, such as actual weather data and solar power location data, and the output variable can be the actual power generation amount at that point in time. At this time, variable importance can be calculated during the learning process of each learning model.

[0059] For example, Tabnet is divided into an input section and Steps 1 through N, each of which consists of a feature transformer, an attentive transformer, and a feature masking unit. The Split block splits the representation from the feature transformer into two, processing one ReLU input and sending it as the final output, while the other is passed on to the next attentive transformer. The Mask block, which selects features, can provide insight into how features work at each step, and the Agg (regate) block ultimately determines which features are important.

[0060] At this time, the Mask block can apply masking to each variable so that it can be used for learning sequentially. Specifically, masking can be applied to all variables except the first variable, allowing only the first variable to be used for learning. Then, masking can be applied to all variables except the second variable, allowing only the second variable to be used for learning. This method allows for the same learning up to the Nth variable, and during this process, the importance of variables affecting power generation can be calculated.

[0061] In the present invention, for each variable of weather data and solar position data, variable importance calculated based on Tabnet can be extracted and utilized for similarity-based prediction of the present invention.

[0062] Additionally, as another example, the SHAp (SHapley Additive exPlanations) value is a value obtained by combining various variables to obtain the importance of a single variable and calculating the average change in the presence or absence of that variable. Based on this, the overall range of the dataset can be interpreted.

[0063] In the present invention, for each variable of weather data and solar position data, variable importance calculated based on the shape value can be extracted and utilized for the similarity-based prediction of the present invention.

[0064] Also, as another example, tree-based learning models can be LightGBM, XGBoost, RandomForest, etc. Tree-based learning models can learn by sorting each variable based on a certain variable, calculating entropy for all possible branching points, comparing it to before the branching to investigate information gain, and selecting the variable with the greatest information gain, that is, the lowest entropy, and setting it as the classification criterion for the node.

[0065] At this time, the importance of each variable in the power generation when acquiring information for each variable in a tree-based learning model can be determined. In this case, the present invention can extract and utilize the importance of each variable. In other words, variable importance can be determined by sorting them in order of highest information acquisition.

[0066] In the present invention, for each variable of weather data and solar position data, variable importance calculated based on a tree-based learning model can be extracted and utilized for similarity-based prediction of the present invention.

[0067] The similarity calculation unit (300) calculates a weight for each variable based on the importance of the variables extracted from the variable extraction unit (200), and can calculate the similarity between the target data and the past data by considering the weight for each variable.

[0068] Here, variable-specific weights can be assigned to each variable based on its importance, such that the sum of the weights equals 1. Furthermore, the target data can be data (actual weather data and solar power location data) at the time of detection of abnormal power generation.

[0069] The similarity calculation unit (300) calculates the Euclidean distance between the target data and each historical data, and adjusts the Euclidean distance for each variable by applying a weight based on the importance of the variable, thereby extracting the historical data most similar to the target data. At this time, the number of similar data extracted can be set by the designer, and the closer the Euclidean distance, the higher the similarity.

[0070] Specifically, the target data and each past data can be a multidimensional vector including each variable, and an adjusted Euclidean distance can be calculated by applying a weight to the Euclidean distance between the two vectors.

[0071] The similarity-based prediction unit (400) can predict the amount of solar power generation for the prediction target data by extracting a preset number of past data in order of high similarity calculated by the similarity calculation unit (300).

[0072] Power generation forecast ( ) can be calculated as shown in the following mathematical formula 1.

[0073] [Mathematical Formula 1]

[0074]

[0075] Here, , d is the distance , x can be the prediction target data vector, i can be the data index, v can be the past data vector, and g can be the actual past power generation.

[0076] Specifically, the similarity-based prediction unit (400) weights the past power generation corresponding to a preset number of past data in the order of high similarity calculated by the similarity calculation unit (300) to produce a prediction value for the target data ( ) can be produced. Accordingly, by reflecting the weights according to the importance of the variables and using the historical data with high similarity among the historical data, the prediction accuracy can be improved.

[0077] The abnormal power generation detection unit (500) can detect abnormal power generation based on the actual power generation amount at the time of the abnormal power generation detection target and the predicted power generation amount at the time of the abnormal power generation detection target predicted through the similarity-based prediction unit.

[0078] Specifically, the abnormal power generation detection unit (500) may determine that the actual power generation at the abnormal power generation detection target point in time is lower than the power generation predicted by the similarity-based prediction unit (400). Alternatively, if the decrease in the actual power generation compared to the predicted power generation falls outside a preset range, the abnormal power generation may be determined.

[0079] FIG. 2 is a block diagram schematically illustrating a configuration of an abnormal power generation detection system utilizing solar power generation tracking based on similar past data according to another embodiment of the present invention. The system illustrated in FIG. 2 may further include a past data-based prediction unit (450) in addition to the system configuration of FIG. 1. That is, the system illustrated in FIG. 2 may include a collection and preprocessing unit (100), a variable importance extraction unit (200), a similarity calculation unit (300), a similarity-based prediction unit (400), and an abnormal power generation detection unit (500).

[0080] At this time, the collection and preprocessing unit (100), variable importance extraction unit (200), similarity calculation unit (300), and similarity-based prediction unit (400) are identical to the functions and operations of each component of Fig. 1, and thus are omitted below.

[0081] The past data-based prediction unit (450) can predict power generation using only actual past data at the time of anomaly detection, without considering variable importance and similarity. In other words, the past data-based prediction unit (450) is a prediction model that predicts power generation based on actual measured weather data and solar power location data, and can be trained based on algorithms such as LGBM.

[0082] At this time, the final predicted power generation amount at the time of the abnormal power generation detection target may be the average of the predicted power generation amount based on similarity predicted by the similarity-based prediction unit (400) and the predicted power generation amount based on past data predicted by the past data-based prediction unit (450).

[0083] The abnormal power generation detection unit (500) can determine that the actual power generation at the time of abnormal power generation detection is lower than the final predicted power generation.

[0084] Figure 3 is a flowchart illustrating a method for detecting abnormal power generation using solar power generation tracking based on similar past data, according to one embodiment of the present invention. The method of Figure 3 can be performed by each component of the systems of Figures 1 and 2. Accordingly, the following description will be made with reference to Figures 1 and 2.

[0085] First, the collection and preprocessing unit (100) can collect meteorological data, solar position data, etc. measured at an observatory at a preset cycle at the same time, and can perform preprocessing as data for learning such as variable importance extraction, power generation prediction, and abnormal power generation detection. That is, a multidimensional vector with each variable as a dimension can be created as an input variable and used for variable importance extraction, power generation prediction, and abnormal power generation detection, etc.

[0086] Here, meteorological data refers to actual data observed at a weather station, such as temperature, relative humidity, dew point, wind direction, wind speed, gust speed, UV index, visibility, cloud cover, cloud height, air pressure, perceived temperature, wet-bulb temperature, precipitation, and snowfall. Additionally, solar position data may include the sun's altitude and azimuth.

[0087] Next, the variable importance extraction unit (200) can extract variable importance affecting solar power generation based on past data including actual weather data and solar position data (S110).

[0088] Variable importance can be extracted based on one of the following models: Tabnet, Shap value, and tree-based learning models. Specifically, variable importance is extracted for each power plant. The model with the lowest power generation prediction error rate among the Tabnet, Shap value, and tree-based learning models is selected, and variable importance can be extracted based on the selected model for the corresponding power plant data.

[0089] Next, the similarity calculation unit (300) or the similarity-based prediction unit (400) or the abnormal power generation detection unit (500) can extract target data including actual weather data, solar power location data, and actual power generation amount for the target time of abnormal power generation detection (S200).

[0090] Next, the similarity calculation unit (300) can calculate the weights for each variable based on the variable importance extracted in step S100, and calculate the similarity between the target data and past data by considering the weights for each variable (S300).

[0091] Specifically, similarity can be calculated by calculating the Euclidean distance between the target data vector and each historical data vector. At this time, weights based on variable importance can be applied to adjust the Euclidean distance for each variable, thereby extracting historical data similar to the target data. Similarity can be judged to be higher when the Euclidean distance between the target data vector and the historical data vector is close.

[0092] Next, the similarity-based prediction unit (400) can extract a preset number of past data in order of high similarity calculated in step S300 to predict the power generation for the target data (S400). At this time, the power generation can be predicted by taking a weighted average of the power generation corresponding to the preset number of past data in order of high similarity calculated.

[0093] Next, the abnormal power generation detection unit (500) can detect abnormal power generation based on the actual power generation at the point in time for detecting abnormal power generation extracted in step S200 and the predicted power generation at step S400. Specifically, if the actual power generation at the point in time for detecting abnormal power generation is lower than the predicted power generation at step S400, it can be determined as abnormal power generation. Alternatively, if the actual power generation at the point in time for detecting abnormal power generation is lower than the predicted power generation at step S400, it can be determined as abnormal power generation if the decrease value is outside a preset range.

[0094] Alternatively, in another embodiment, between steps S400 and S500, a past data-based prediction unit (450) may further include a step of predicting the amount of power generation for a target point in time for detecting abnormal power generation based on past data including actual weather data and solar position data.

[0095] At this time, the final predicted power generation can be the average of the predicted power generation based on similarity predicted in step S400 and the predicted power generation based on past data.

[0096] In this case, in step S500, the abnormal power generation detection unit (500) can determine that the actual power generation at the time of the abnormal power generation detection target extracted in step S200 is lower than the final predicted power generation, or can determine that the abnormal power generation is if the decrease in the actual power generation is outside the preset range.

[0097] FIG. 4 is a graph illustrating abnormal power generation detection according to various embodiments of the present invention. Referring to FIG. 4, (a) represents actual power generation, (b) represents predicted power generation based on past data using only the past data-based prediction unit (450), (c) represents predicted power generation based on similarity-based prediction based on variable importance and similarity calculation, and (d) represents predicted power generation based on the average of (b) and (c). In this case, the power generation can be power generation per unit area relative to capacity.

[0098] For example, when determining abnormal development at the point in time (A) when detecting an abnormal development target, if the power generation amount is predicted using a prediction model based on past data as in (b), it may be determined as abnormal development, but in the case of the similarity-based prediction model of (c) and the ensemble model of (d), it can be confirmed that normal development has occurred.

[0099] As a result, the abnormal power generation detection system and method utilizing solar power generation tracking based on similar past data according to various embodiments of the present invention can detect abnormal power generation by calculating the importance of variables affecting power generation based on measured past data (weather data, solar position data), and comparing the predicted power generation based on similar past measured data and the actual power generation at the time of abnormal power generation detection by considering the importance of variables.

[0100] Accordingly, power generation predictions can be improved based on similar past data with highly reliable power generation conditions. Furthermore, by tracking predicted power generation and comparing it with actual power generation at the target point in time for abnormal power generation detection, the accuracy of abnormal power generation detection can be improved.

[0101] Additionally, as actual historical data and solar power location data accumulate, weather factors, weather types, and solar power location data become more diverse, enabling more accurate power generation predictions.

[0102] Although the embodiments of the present invention have been described in more detail with reference to the attached drawings, the present invention is not necessarily limited to these embodiments, and various modifications may be implemented without departing from the technical spirit of the present invention. Therefore, the embodiments disclosed in the present invention are not intended to limit the technical spirit of the present invention, but to explain it, and the scope of the technical spirit of the present invention is not limited by these embodiments. Therefore, it should be understood that the embodiments described above are illustrative in all aspects and not restrictive. The protection scope of the present invention should be interpreted by the claims below, and all technical ideas within a scope equivalent thereto should be interpreted as being included in the scope of the rights of the present invention.

Claims

1. (a) A step of extracting the importance of variables affecting solar power generation based on past data including actual weather data and solar position data; (b) A step of extracting target data including actual weather data, solar power location data, and actual power generation amount for the target time of abnormal power generation detection; (c) a step of calculating a weight for each variable based on the importance of the extracted variables, and calculating the similarity between the target data and the past data by considering the weight for each variable; (d) a step of extracting a preset number of past data in the order of high similarity calculated above and predicting the power generation amount for the target data; and (e) A method for detecting abnormal power generation using solar power generation tracking based on similar past data, including a step of detecting abnormal power generation based on the actual power generation amount and the predicted power generation amount at the time of detection of abnormal power generation.

2. In paragraph 1, The above variable importance is extracted based on one of the models of Tabnet, Shap value, and tree-based learning model, and is a method for detecting abnormal power generation using solar power generation tracking based on similar past data.

3. In paragraph 2, The above variable importance is extracted for each power plant, and a method for detecting abnormal power generation using solar power generation tracking based on similar past data is extracted by selecting a model with the lowest power generation prediction error rate among the power generation prediction error rates based on the Tabnet, Shap value, and tree-based learning models.

4. In paragraph 1, Step (c) above, A method for detecting abnormal power generation using solar power generation tracking based on similar past data, which calculates the Euclidean distance between the above target data and each past data, and adjusts the Euclidean distance for each variable by applying a weight according to the importance of the variable.

5. In paragraph 1, Step (d) above, A method for detecting abnormal power generation using solar power generation tracking based on similar past data, which predicts the power generation amount for the target data by calculating a weighted average of the power generation amounts corresponding to a preset number of past data in order of high similarity calculated above.

6. In paragraph 1, (e) The step of detecting the above abnormal development is: A method for detecting abnormal power generation using solar power generation tracking based on similar past data, which determines that abnormal power generation is present if the actual power generation amount at the above-mentioned abnormal power generation detection target time is lower than the power generation amount predicted in step (d).

7. In paragraph 1, (d1) further includes a step of predicting power generation based on past data including the above-mentioned actual weather data and solar position data, (e) The step of detecting the above abnormal development is: If the actual power generation at the above abnormal power generation detection target time is lower than the final predicted power generation, it is judged as abnormal power generation. The above final predicted power generation amount is an average of the predicted power generation amount based on the similarity predicted in the above step (d) and the predicted power generation amount based on the past data predicted in the above step (d1), and is a method for detecting abnormal power generation using solar power generation tracking based on similar past data.

8. In paragraph 1, The above actual weather data includes temperature, relative humidity, dew point, wind direction, wind speed, gust speed, UV index, visibility, cloud cover, cloud height, air pressure, perceived temperature, wet bulb temperature, precipitation, and snowfall. The above solar power location data is a method for detecting abnormal power generation using solar power generation tracking based on similar past data, which is the altitude and azimuth of the sun.

9. A variable importance extraction unit that extracts the importance of variables affecting solar power generation based on past data including actual weather data and solar position data; A similarity calculation unit that calculates a weight for each variable based on the importance of the extracted variables and calculates a similarity between the target data and the past data by considering the weight for each variable; A similarity-based prediction unit that extracts a preset number of past data in the order of high similarity calculated above and predicts the amount of solar power generation for the prediction target data; and An abnormal power generation detection system utilizing solar power generation tracking based on similar past data, including an abnormal power generation detection unit that detects abnormal power generation based on the actual power generation amount at the time of abnormal power generation detection target and the predicted power generation amount at the time of abnormal power generation detection target predicted through the similarity-based prediction unit.

10. In paragraph 9, The above variable importance is extracted based on one of the models of Tabnet, Shap value, and tree-based learning model, and is an abnormal power generation detection system utilizing solar power generation tracking based on similar past data.

11. In paragraph 10, The above variable importance is extracted for each power plant, and the model with the lowest power generation prediction error rate is selected among the power generation prediction error rates based on the Tabnet, Shap value, and tree-based learning models, and an abnormal power generation detection system utilizing solar power generation tracking based on similar past data is extracted.

12. In paragraph 10, The above similarity calculation unit is, An abnormal power generation detection system utilizing solar power generation tracking based on similar past data, which calculates the Euclidean distance between the above target data and each past data, and adjusts the Euclidean distance for each variable by applying a weight according to the importance of the variable.

13. In paragraph 9, The above similarity-based prediction unit is, An abnormal power generation detection system utilizing solar power generation tracking based on similar past data, which predicts the power generation amount for the target data by calculating the weighted average of the power generation amounts corresponding to a preset number of past data in order of high similarity calculated above.

14. In paragraph 10, The above abnormal power generation detection unit is, An abnormal power generation detection system utilizing solar power generation tracking based on similar past data, which determines that abnormal power generation is present if the actual power generation amount at the above abnormal power generation detection target time is lower than the power generation amount predicted by the similarity-based prediction unit.

15. In paragraph 9, It further includes a past data-based prediction unit that predicts power generation based on past data including the above-mentioned actual weather data and solar position data, The above abnormal power generation detection unit determines that the actual power generation at the above abnormal power generation detection target point is lower than the final predicted power generation, The above final predicted power generation amount is an average of the predicted power generation amount based on similarity predicted by the similarity-based prediction unit and the predicted power generation amount based on past data predicted by the past data-based prediction unit, and is an abnormal power generation detection system utilizing similar past data-based solar power generation tracking.

16. In paragraph 9, The above actual weather data includes temperature, relative humidity, dew point, wind direction, wind speed, gust speed, UV index, visibility, cloud cover, cloud height, air pressure, perceived temperature, wet bulb temperature, precipitation, and snowfall. The above solar power location data is an abnormal power generation detection system that utilizes solar power generation tracking based on similar past data, which is the altitude and azimuth of the sun.

Images