Excessive power consumption prediction system

Abstract

According to the present invention, an excess power consumption state in which power consumption occurring in a factory exceeds power generation is predicted in advance, and major causes of excess power consumption are analyzed using various operational information, power generation amount information, weather information, and correction information, so that an optimal response strategy can be automatically established and transferred in real time to efficiently manage power consumption and ensure stability of power use, and energy cost reduction and operational efficiency of the factory can be maximized. In addition, the present invention may comprise: a database; a data acquisition unit for collecting usage amount information that is power consumption occurring in a factory, power generation amount information that is an amount of power generated by a power generator, and weather information, and storing the collected information in the database; and a control unit for predicting, on the basis of the information stored in the database, whether an excess power consumption state in which the usage amount information exceeds the power generation amount information occurs, wherein the control unit generates a response strategy for preventing the excess power consumption state if the excess power consumption state is predicted.

Description

Excess power usage prediction system

[0001] The present invention relates to a system for predicting excess power usage, and more specifically, to a power management system that collects power consumption and generation data, analyzes the possibility of excess power usage using a machine learning-based prediction model, establishes an optimal response strategy, and delivers it to the user.

[0002]

[0003] The importance of power management is growing in modern industrial facilities and factories. As industrial facilities expand and automation advances, power consumption continues to rise; consequently, efficient power operation is becoming a key factor in reducing corporate costs and maintaining productivity.

[0004] Furthermore, optimizing the power consumption patterns of large-scale factories is becoming an essential task in terms of energy policy, which must consider power grid stability and sustainability.

[0005] In particular, as the share of renewable energy increases, the imbalance between power production and consumption is emerging as a new problem. Due to the intermittent nature of solar and wind power generation, existing power management systems are struggling to effectively coordinate supply and demand in real time.

[0006] If weather conditions deteriorate or the efficiency of solar panels decreases, the expected power generation drops sharply. This leads factories to consume more electricity than anticipated, which can cause instability in power grid operations.

[0007] Currently used power management systems provide power usage monitoring and limited forecasting capabilities, but they lack the ability to detect and respond to excess power usage in advance.

[0008] Most systems remain limited to performing simple demand forecasting based on historical data or sending warnings when power consumption is detected above a certain level.

[0009] However, this method has the limitation that it can only respond after a power overrun occurs; if proactive measures are not taken, adjusting production schedules becomes difficult, and operating costs may increase due to unnecessary additional power purchases.

[0010] Accordingly, a system is required that can monitor the factory's power consumption and solar power generation in real time, and predict the possibility of power overruns in advance by reflecting weather changes and factory operation patterns.

[0011] Furthermore, there is a growing need to go beyond simple predictions to analyze the causes of power overruns and automatically generate and provide optimal response strategies to users.

[0012] This has been disclosed in Korean Registered Patent No. 10-2227192 and Korean Registered Patent No. 10-2427294.

[0013]

[0014] The present invention has been devised based on the technical background described above. The present invention provides an excess power consumption prediction system that predicts in advance a state of excess power consumption where the power usage generated in a factory exceeds the power generation amount, and analyzes the main causes of the excess power consumption by utilizing various operational information, power generation information, weather information, and correction information; automatically establishes an optimal response strategy and transmits it in real time to efficiently manage power consumption, ensure the stability of power usage, reduce energy costs, and maximize the operational efficiency of the factory.

[0015]

[0016] To achieve the above objectives, the present invention comprises a database, a data acquisition unit that collects and stores in the database usage information, which is the amount of electricity generated in a factory, power generation information, which is the amount of electricity generated by a generator, and weather information, and a control unit that predicts whether a power excess state occurs in which the usage information exceeds the power generation information through the information stored in the database, and if the power excess state is predicted, the control unit generates a response strategy to prevent the power excess state.

[0017] In addition, the control unit may further include a data processing unit that receives a plurality of information stored in the database, removes abnormally appearing outliers, normalizes each of them, and generates preprocessed information in a form that can be analyzed by a machine learning model.

[0018] In addition, the control unit may further include a prediction analysis unit that rapidly generates a prediction result regarding whether the power overload state occurs through a machine learning model that learns by extracting only a portion of the information stored in the database.

[0019] In addition, the machine learning model includes a Convolutional Neural Network (CNN) model, converts multiple pieces of information including usage information, power generation information, and weather information into a two-dimensional form for learning, and can generate the prediction result by analyzing patterns along the time axis.

[0020] In addition, the prediction analysis unit may further include a plurality of verification models for verifying the prediction result generated by the machine learning model, and a prediction correction unit that generates a plurality of comparison values ​​using the plurality of verification models and determines the reliability of the prediction result.

[0021] In addition, if the prediction correction unit determines that the reliability of the prediction result is low, it may change the information used by the machine learning model to generate multiple correction values ​​and use the average of the generated multiple correction values ​​to re-determine whether a power overload state has occurred.

[0022] In addition, the control unit may further include a response strategy formulation unit that analyzes the cause of the power overload condition and formulates a response strategy.

[0023] In addition, the data acquisition unit further acquires correction information including the process operating rate of the factory, operational information regarding the usage status of the equipment, the number of employees who have reported to work, and the cleaning status of the generator, and the control unit patterns the power generation information based on the operational information and correction information to predict the power excess state.

[0024]

[0025] An excess power usage prediction system according to one embodiment of the present invention can systematically manage the power usage patterns of a factory and predict the possibility of power excess in advance, thereby increasing the stability of energy usage and optimizing operations.

[0026] An excess power consumption prediction system according to one embodiment of the present invention supports a machine learning model to learn more refined data, thereby improving prediction accuracy and providing highly reliable analysis results.

[0027] An excess power consumption prediction system according to one embodiment of the present invention can more effectively learn the relationship between power usage patterns and weather changes, convert time series data into a two-dimensional structure to increase the accuracy of the prediction, and implement a machine learning model capable of real-time analysis.

[0028] An excess power consumption prediction system according to one embodiment of the present invention evaluates the reliability of the prediction result and, if the reliability is low, generates a correction value to improve the final prediction result, thereby enabling a more reliable power prediction.

[0029] An excess power consumption prediction system according to one embodiment of the present invention can establish a response strategy, thereby maximizing the efficiency of energy management and minimizing unnecessary power waste.

[0030]

[0031] FIG. 1 is a configuration diagram of an excess power consumption prediction system according to one embodiment of the present invention.

[0032] FIG. 2 is an operation diagram of a data acquisition unit according to one embodiment of the present invention.

[0033] FIG. 3 is a configuration diagram of a control unit according to one embodiment of the present invention.

[0034] FIG. 4 is an operation diagram of a data processing unit according to one embodiment of the present invention.

[0035] FIG. 5 is an operation diagram of a prediction analysis unit according to one embodiment of the present invention.

[0036] FIG. 6 is a diagram showing the form of preprocessing information according to one embodiment of the present invention.

[0037] FIG. 7 is an operation diagram of a short-term analysis unit according to one embodiment of the present invention.

[0038] FIG. 8 is an operation diagram of a prediction correction unit according to one embodiment of the present invention.

[0039] FIG. 9 is an operation diagram of a response strategy establishment unit and an information transmission unit according to an embodiment of the present invention.

[0040]

[0041] Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings.

[0042] The advantages and features of the present invention and the method for achieving them will become clear by referring to the embodiments described in detail below together with the accompanying drawings.

[0043] However, the present invention is not limited by the embodiments disclosed below but may be implemented in various different forms, and these embodiments are provided merely to make the disclosure of the present invention complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims.

[0044] In addition, in describing the present invention, if it is determined that related known technologies, etc., may obscure the essence of the present invention, a detailed explanation thereof will be omitted.

[0045]

[0046] FIG. 1 is a configuration diagram of an excess power consumption prediction system according to one embodiment of the present invention.

[0047] FIG. 2 is an operation diagram of a data acquisition unit according to one embodiment of the present invention.

[0048] FIG. 3 is a configuration diagram of a control unit according to one embodiment of the present invention.

[0049] FIG. 4 is an operation diagram of a data processing unit according to one embodiment of the present invention.

[0050] FIG. 5 is an operation diagram of a prediction analysis unit according to one embodiment of the present invention.

[0051] FIG. 6 is a diagram showing the form of preprocessing information according to one embodiment of the present invention.

[0052] FIG. 7 is an operation diagram of a short-term analysis unit according to one embodiment of the present invention.

[0053] FIG. 8 is an operation diagram of a prediction correction unit according to one embodiment of the present invention.

[0054] FIG. 9 is an operation diagram of a response strategy establishment unit and an information transmission unit according to an embodiment of the present invention.

[0055]

[0056] Referring to FIG. 1, the excess power usage prediction system (1) is configured to include a data acquisition unit (300), a control unit (100), and a database (500).

[0057] The data acquisition unit (300) performs the role of collecting data from the factory (F) and the generator (G), and includes a consumption measurement unit (310), an operation information management unit (330), a power generation measurement unit (350), a weather information management unit (370), and a correction information acquisition unit (390).

[0058] The consumption measurement unit (310) measures the power usage of the factory (F) to generate consumption information (311), and the operation information management unit (330) obtains operation information (331) including the process operation rate of the factory and the usage status of the equipment.

[0059] The power generation measurement unit (350) measures the power production of the generator (G) to generate power generation information (351), and the weather information management unit (370) obtains weather information (371) including temperature, humidity, solar radiation, etc.

[0060] In addition, the correction information acquisition unit (390) further acquires correction information (391), including the number of employees who have come to work and the cleaning status of the generator, to support more precise prediction.

[0061] The consumption information (311), operation information (331), power generation information (351), weather information (371), and correction information (391) obtained from the data acquisition unit (300) are stored in the database (500).

[0062] The database (500) stores all information transmitted from the data acquisition unit (300) and provides it so that the control unit (100) can analyze it.

[0063] The control unit (100) includes a data processing unit (110), a prediction analysis unit (130), a response strategy establishment unit (150), and an information transmission unit (170).

[0064] The data processing unit (110) performs the role of processing data collected from the data acquisition unit (300) into an analyzable form and is composed of a synchronization unit (111), a filtering unit (113), a normalization unit (115), a feature extraction unit (117), and a preprocessing storage unit (119).

[0065] The synchronization unit (111) synchronizes the collected data in time, the filtering unit (113) removes outliers, and the normalization unit (115) converts the data into a certain range.

[0066] The feature extraction unit (117) generates key features so that the machine learning model (133a) can learn effectively, and the preprocessing storage unit (119) stores preprocessing information (119a) in a form that can be analyzed by the machine learning model (133a) so that it can be used for subsequent analysis.

[0067] The prediction analysis unit (130) is composed of an input optimization unit (131), a short-term analysis unit (133), a prediction correction unit (135), and a cyclic learning unit (137), and predicts a power overload state using a machine learning model (133a).

[0068] The input optimization unit (131) selects the optimal input data for the machine learning model (133a) to learn, and the short-term analysis unit (133) analyzes the data to generate a prediction result (133b).

[0069] The prediction correction unit (135) evaluates the reliability of the prediction result (133b) using the verification model (135a), generates a comparison value (135b) and a correction value (135c), and derives a correction result (135d).

[0070] At this time, the verification model (135a) may include ARIMA (AutoRegressive Integrated Moving Average) and SARIMA (Seasonal ARIMA) which learn past patterns based on time series data to predict future values, exponential smoothing which predicts by weighting recent data, LSTM (Long Short-Term Memory) which is a recurrent neural network model that learns long-term patterns of time series data, and an Autoencoder which learns typical power consumption patterns and evaluates reliability by analyzing the difference from the prediction result.

[0071] The cyclic learning unit (137) continuously improves the performance of the machine learning model (133a) by reflecting the corrected data.

[0072] The response strategy formulation unit (150) is composed of a cause analysis unit (151) and a response strategy generation unit (153), analyzes the cause of power overrun (151a), and derives an optimal response strategy (153a) accordingly.

[0073] Response strategies may include measures such as load adjustment of the factory (F), utilization of ESS (energy storage system), and purchase of external power.

[0074] The information transmission unit (170) is composed of a result transmission unit (171) and a condition-based warning unit (173), and performs the role of transmitting information generated by the prediction analysis unit (130) and the response strategy establishment unit (150) to the manager.

[0075] The result delivery unit (171) provides prediction results and response strategies, and the condition-based warning unit (173) can notify the manager when specific conditions set by the manager are met.

[0076] In this way, the excess power usage prediction system (1) of the present invention stores various information collected from the data acquisition unit (300) in the database (500).

[0077] Power usage in the factory can be optimized by effectively predicting the power overload condition and establishing a response strategy through the data processing unit (110), prediction analysis unit (130), response strategy establishment unit (150), and information transmission unit (170) of the control unit (100).

[0078] Referring to FIG. 2, the data acquisition unit (300) includes a consumption measurement unit (310), an operation information management unit (330), a power generation measurement unit (350), a weather information management unit (370), and a correction information acquisition unit (390), and each component collects data through specific sensors and systems.

[0079] The consumption measurement unit (310) performs the role of measuring the power usage of the factory and may include a power meter, a smart meter, a current sensor, etc.

[0080] The consumption information (311) obtained from the consumption measurement unit (310) includes real-time power usage, voltage, current, power factor, and peak load data for the entire factory and individual facilities.

[0081] This information can serve as a standard for identifying power usage patterns, predicting peak loads, and defining power overload conditions. It also enables energy management optimization by detecting the overload status of specific equipment in real time.

[0082] The Operation Information Management Department (330) collects information related to factory operations and may include a PLC (Programmable Logic Controller), SCADA (Supervisory Control and Data Acquisition) system, temperature and humidity sensors, etc.

[0083] The operational information (331) obtained from the operational information management department (330) includes process operating rate, production schedule, equipment operating status, maintenance history and internal factory environment data.

[0084] Operation information (331) can be used to predict power consumption in a specific process and to generate a response strategy (153a).

[0085] Unnecessary power consumption can be reduced by adjusting the production schedule based on the operation information (331). In addition, the energy efficiency of the equipment can be predicted based on the maintenance history, thereby improving long-term energy efficiency.

[0086] The power generation measurement unit (350) performs the role of monitoring the power generation of a solar power generator (G) used in a factory and may include an inverter, a power generation measurement sensor, and a solar panel monitoring system.

[0087] The power generation information (351) obtained from the power generation measurement unit (350) includes solar power generation amount, solar panel efficiency, generator output voltage and ESS (energy storage system) charge status.

[0088] This power generation information (351) is used not only for training the machine learning model (133a), but also allows real-time power generation to be monitored and compared with power consumption, and enables immediate action to be taken if solar power generation efficiency decreases.

[0089] In addition, energy storage and release strategies can be optimized by identifying the ESS charge status.

[0090] The weather information management department (370) is responsible for measuring weather conditions outside the factory and may include a temperature sensor, a humidity sensor, a wind speed sensor, a pressure sensor, and a solar radiation sensor.

[0091] The weather information (371) obtained from the weather information management department (370) includes temperature, humidity, solar radiation, wind speed, atmospheric pressure, precipitation, and air pollution levels.

[0092] Weather information (371) is essential for predicting solar power generation and can reflect patterns of increased power consumption under specific weather conditions, and can make more precise predictions by taking into account the possibility of factors that hinder power generation under specific weather conditions.

[0093] For example, power consumption of the cooling system may increase during high temperatures, or the power generation efficiency of solar panels may change during strong winds, so compensation for this is necessary.

[0094] The correction information acquisition unit (390) collects additional information to increase the accuracy of power consumption and power generation, and may include an access control system, an equipment diagnostic sensor, a pollution level measurement sensor, etc.

[0095] The correction information (391) obtained from the correction information acquisition unit (390) includes the number of employees at work, the contamination status of the solar panels, the maintenance schedule of the generator, and the operating temperature of the equipment.

[0096] Correction information (391) does not measure the direct power consumption of the factory (F), but helps improve the accuracy of the prediction model by analyzing the correlation between factors that can affect power consumption.

[0097] For example, if the number of people going to work increases, the likelihood of increased power consumption increases, and if the probability of solar panels being contaminated increases, the amount of power generated may decrease, so the machine learning model (133a) can operate more precisely by considering the possibility of this.

[0098] Referring to FIG. 3, the control unit (100) includes a data processing unit (110), a prediction analysis unit (130), a response strategy establishment unit (150), and an information transmission unit (170), and each component performs the role of predicting power consumption and establishing a response strategy.

[0099] The data processing unit (110) processes the collected data to make it into an analyzable form. The synchronization unit (111) aligns the data collected from various sensors and data sources by unifying the time units to create a data set of how much value was measured per consistent hour.

[0100] The filtering unit (113) detects and removes outliers in the data to prevent errors during the model's learning and analysis process.

[0101] The normalization unit (115) converts the collected data into a fixed unit, correcting values ​​measured in different units to the same unit so that the machine learning model (133a) can effectively process various data formats.

[0102] The feature extraction unit (117) derives key feature values ​​for the machine learning model (133a) to efficiently perform analysis, thereby improving the performance of the prediction analysis.

[0103] The preprocessing storage unit (119) stores and manages the preprocessed preprocessing information (119a) so that it can be used for subsequent analysis. The data processing unit (110) is a component for increasing the accuracy of the entire system.

[0104] The prediction analysis unit (130) predicts whether there will be an excess power condition in which the amount of power to be produced by the generator (G) exceeds the power required by the factory (F).

[0105] The input optimization unit (131) can select the optimal information for the machine learning model to learn from among a large number of pieces of information in the preprocessing information (119a), such that the current state of the factory (F) and the state of the generator (G) are similar through past cases. This reduces computational costs caused by unnecessary information.

[0106] The short-term analysis unit (133) performs the role of predicting changes in power consumption within a specific time period using a machine learning model (133a) such as a CNN (Convolutional Neural Network).

[0107] The prediction correction unit (135) uses multiple verification models (135c) to generate a comparison value (135b) and evaluate reliability in order to verify the prediction result (133b) of the short-term analysis unit (133). If reliability is low, a correction value (135c) is derived to correct the prediction result (133b).

[0108] The cyclic learning unit (137) plays a role in improving the performance of the machine learning model (133a) by continuously reflecting the prediction results (133b) and correction results (135d). The prediction analysis unit (130) enables the identification of the possibility of an overpowered state occurring in advance and the response thereof.

[0109] The response strategy formulation department (150) is responsible for formulating an optimal response strategy when power overrun occurs. The cause analysis department (151) analyzes the cause of power overrun and identifies the cause of the overrun (151a) by considering the factory's operating status and external environmental factors.

[0110] The response strategy generation unit (153) generates optimal response strategies, such as power load adjustment, utilization of ESS (energy storage system), and external power purchase, based on the excess cause (151a) derived from the cause analysis unit (151).

[0111] The information delivery unit (170) serves to deliver prediction results and response strategies to the user. The result delivery unit (171) provides the administrator with the predicted power usage and possible power overrun status, enabling the user to take action based on this.

[0112] The condition-based warning unit (173) sends an alert to the manager in real time when a specific condition occurs, enabling a quick response.

[0113] At this time, specific conditions may include a situation where a power load exceeding a set threshold occurs, a situation where the operating time of a specific facility continues for longer than a preset time, and a situation where thermal management is required because the temperature of the generator rises above a preset temperature.

[0114] The information delivery unit (170) plays a role in effectively delivering the results of the system to the user so that the prediction and response strategies can be actually executed.

[0115] The control unit (100) operates by refining data through the data processing unit (110), predicting the possibility of power overrun using the prediction analysis unit (130), deriving an appropriate response strategy through the response strategy establishment unit (150), and then delivering the results to the user using the information delivery unit (170).

[0116] This allows for the optimization of energy usage in factories and generators, improved prediction accuracy, and prevention of cost increases resulting from increased power consumption.

[0117] Referring to FIG. 4, the data processing unit (110) is composed of a synchronization unit (111), a filtering unit (113), a normalization unit (115), a feature extraction unit (117), and a preprocessing storage unit (119), and each component performs the role of refining the collected data and processing it into a form that is easy to analyze.

[0118] The synchronization unit (111) performs temporal alignment of data and aligns the data collected from various sensors in the factory to the same time axis when the collection times of the data are different.

[0119] For example, if the consumption measurement unit (310) provides data in 1-minute intervals, while the weather information management unit (370) provides data in 1-hour intervals, the synchronization unit (111) creates a consistent data set by matching them to the same standard.

[0120] This allows the machine learning model (133a) to learn with consistent data over time.

[0121] The filtering unit (113) performs the role of removing outliers within the data. For example, if a specific sensor in a factory records an abnormally high power consumption due to a temporary error, the filtering unit (113) detects the data and removes or corrects it.

[0122] This prevents machine learning models from being distorted by error data.

[0123] Additionally, the normalization unit (115) converts the data units so that the model can analyze them effectively. For example, while temperature data is provided in Celsius, weather forecast data may be provided in Fahrenheit (μ).

[0124] In this case, the normalization unit (115) converts the data into the same unit and normalizes the power consumption data so that it can be processed within a certain range. This increases the learning speed of the model and improves performance.

[0125] The feature extraction unit (117) plays the role of extracting meaningful features from the data so that the machine learning model (133a) can learn.

[0126] For example, instead of simply inputting power usage data as is, additional features such as average consumption over a specific period, peak load occurrence times, and peak power usage patterns are extracted to enable the model to perform more precise predictions.

[0127] The preprocessing storage unit (119) stores the preprocessed data in the form of preprocessing information (119a) in a pre-set format and links with the database (500) so that it can be utilized in the subsequent analysis process.

[0128] The information stored in the database (500) is stored in the format of a single fixed preprocessing information (119a), making it easy to retrieve and compare data from a specific point in the past, and enabling long-term analysis while maintaining data quality.

[0129] Referring to FIG. 5, the prediction analysis unit (130) is composed of an input optimization unit (131), a short-term analysis unit (133), a prediction correction unit (135), and a cyclic learning unit (137).

[0130] The input optimization unit (131) plays the role of selecting and optimizing the data necessary to predict whether the power is exceeded.

[0131] The preprocessed information (119a) and historical information (510) are analyzed to determine the essential features that the machine learning model (133a) must learn.

[0132] For example, if specific weather conditions have a significant impact on solar power generation, the optimal combination of input data is selected to give significant weight to that weather data. This eliminates unnecessary data, thereby improving analysis speed and increasing prediction accuracy.

[0133] The short-term analysis unit (133) uses a Convolutional Neural Network (CNN) model to analyze power usage patterns and quickly performs short-term predictions.

[0134] By reflecting factors such as real-time changes in power usage and weather changes through the short-term analysis unit (133), it is possible to predict the possibility of power overruns occurring within the next few hours. This enables proactive response to power overruns.

[0135] The prediction correction unit (135) performs the role of verifying and correcting the prediction result (133b) generated by the short-term analysis unit (133).

[0136] The prediction correction unit (135) evaluates the reliability of the prediction result (133b) using multiple verification models (135c), and if the reliability is low, generates multiple correction values ​​(135c) to derive a more accurate correction result (135d).

[0137] For example, if the current CNN model has a large error under specific weather conditions, when generating the correction value (135c), other information excluding the weather information (371) is used to generate the correction value. This can increase the reliability of the prediction system.

[0138] The cyclic learning unit (137) plays the role of continuously training the machine learning model by comparing the predicted result with the actual power usage. If the predicted result differs from the actual result, the cause is analyzed and the training data is updated so that the model is gradually improved.

[0139] For example, if power consumption changes as the operating pattern of a specific factory changes, the cyclic learning unit (137) updates the prediction model to reflect this. Through this, the accuracy of the prediction is continuously improved over time.

[0140] By utilizing the components of such a prediction analysis unit (130), real-time data and past data can be comprehensively analyzed to predict whether the power is exceeded more precisely.

[0141] In addition, the performance of the prediction model can be continuously improved through the calibration and learning process, thereby enabling the optimization of power operation of the factory (F).

[0142] Referring to FIGS. 6 and FIGS. 7, in the present invention, the preprocessed information (119a) refined in the data processing unit (110) is organized in a table format so that the machine learning model (133a) can learn effectively.

[0143] Preprocessing information (119a) includes weather information (371) including weather elements (temperature, humidity, solar radiation) according to time (t, t-1, t-2, ..., tn), operation information (331), correction information (391), and power consumption information (311).

[0144] This tabular data is arranged along a time axis and is recorded at regular time intervals from past data (tn) to the latest data.

[0145] This enables the machine learning model (133a) to learn the relationship between power usage patterns and other weather information (371), operational information (331), and correction information (391) based on time series data.

[0146] The machine learning model (133a) of the present invention includes a Convolutional Neural Network (CNN) model, and the CNN model can be used to analyze not only image data but also time series data by converting it into a two-dimensional matrix form.

[0147] CNN models have the ability to automatically detect important patterns in input data and learn the correlations between adjacent data using specific filters (kernels).

[0148] In the present invention, a machine learning model (133a) is used as a CNN model to learn the relationship between power consumption and weather conditions, and based on this, it is possible to predict whether the power is exceeded.

[0149] The reason table-format information is needed is that CNN models accept data in the form of a two-dimensional array. While other machine learning models (e.g., regression analysis, LSTM, etc.) learn individual time-series data as is, CNN models can analyze local patterns by spatially interpreting the data.

[0150] For example, if there is a pattern of increased power consumption under specific temperature and humidity conditions, a CNN model can effectively detect it. To achieve this, the data is converted into a tabular format containing the time axis and meteorological elements, and optimized so that the CNN model can learn from it.

[0151] The short-term analysis unit (133) uses a CNN model to analyze power usage patterns and perform short-term predictions. For the CNN model to operate effectively, it is very important to determine which information to select for learning.

[0152] For example, if there is a pattern of surging power consumption during a specific season (summer or winter), the CNN model needs to focus on learning that data.

[0153] In the present invention, key data that the CNN model needs to learn is selected through the input optimization unit (131), thereby increasing prediction accuracy and reducing computational costs caused by unnecessary data.

[0154] One of the features of CNN models is that they can detect local patterns through kernel operations.

[0155] In the present invention, the kernel size is set to (3,3) to enable learning the relationship between adjacent data (t-1, t-2, etc.) and meteorological elements (temperature, humidity, solar radiation, etc.) based on time (t).

[0156] This allows for the effective detection of patterns where power consumption changes rapidly under specific weather conditions.

[0157] When the CNN model is trained, some data is selected from the preprocessed information (119a) and used as training data.

[0158] For example, a CNN model can be trained by selecting a small subset containing weather information and power usage data for the past 24 hours. This allows for more precise predictions to be made by utilizing only important information while reducing the computational burden on the training model.

[0159] Model performance is heavily influenced by the selection of data, and training with unnecessary data can actually degrade prediction performance. Therefore, it is crucial to carefully structure the input data and incorporate essential features.

[0160] After the CNN model is trained, it finally derives a prediction result (133b), and this result is verified by the prediction correction unit (135), and if the reliability is low, an additional correction process is performed.

[0161] The excess power usage prediction system (1) of the present invention applies a feature engineering technique to improve the accuracy of power usage prediction and generates new features.

[0162] To this end, past consumption information (311), weather information (371), and power generation information (351) of a specific factory (F) are collected, and a clustering model is applied using this information to reflect seasonal patterns based on the weather information (371).

[0163] First, weather observation data from the past N years can be analyzed and classified into 3 to 4 seasonal clusters, and clusters such as early summer, autumn, winter, and spring can be generated.

[0164] Subsequently, based on weather forecast data for the T+1 hour period, a pre-trained clustering model is utilized to determine which cluster the weather condition at that time belongs to.

[0165] By adding the average power consumption and average power generation values ​​belonging to the identified cluster as new features and utilizing them as input data for the machine learning model (133a), it is possible to predict power consumption more precisely.

[0166] This allows for more accurate modeling of the relationship between weather conditions and power consumption, and maximizes prediction accuracy by reflecting various seasonal characteristics.

[0167] Through this, the prediction system of the present invention can generate highly reliable prediction results by reflecting real-time changing power usage patterns and weather conditions.

[0168] Referring to FIG. 8, the prediction correction unit (135) performs the role of verifying the prediction result (133b) of the machine learning model (133a), and generating a correction value (135c) when the reliability is low to derive the final correction result (135d).

[0169] First, a machine learning model (133a) predicts whether the power is exceeded and generates a prediction result (133b). This prediction result (133b) is compared with multiple comparison values ​​(135b) by a verification model (135a) capable of predicting whether the power is exceeded in multiple ways.

[0170] The comparison value (135b) may include results derived from existing past patterns, other machine learning models, statistical prediction models, etc. This allows for evaluating how well the current prediction value matches the existing data pattern.

[0171] The verification model (135a) analyzes the difference between the prediction result (133b) and the comparison value (135b) to determine reliability. If the difference between the prediction result (133b) and the comparison value (135b) is large, it is determined that the reliability of the prediction result is low, and a correction process is performed.

[0172] At this time, the prediction result (133b) may be information regarding whether or not a power overload will occur. That is, determining the numerical amount of the shortage may make it difficult to draw a quick and immediate conclusion because the computational process of the machine learning model (133a) becomes lengthy.

[0173] However, if the prediction result (133b) is simply about whether or not a power overload will occur, the result can be obtained more quickly.

[0174] At this time, multiple comparison values ​​(135b) also simply determine whether a power overload state has occurred, and when the number of comparison values ​​(135b) predicted in the same way as the prediction result (133b) is greater than or equal to a preset number, the prediction result (133b) is determined to be valid.

[0175] In the correction process, multiple correction values ​​(135c) are generated. The correction values ​​(135c) are values ​​for modifying the prediction result (133b) and can be generated by retraining the machine learning model (133a) based on a different data combination or by reflecting data patterns from similar past situations.

[0176] The machine learning model (133a) generates multiple correction values ​​(135c), each of which represents a modified prediction value to complement the existing prediction.

[0177] Finally, the average of the generated multiple correction values ​​(135c) is calculated to derive a correction result (135d). The correction result (135d) may differ from, but may also be the same as, the prediction result (133b) initially predicted by the machine learning model (133a).

[0178] The configuration of the prediction correction unit (135) improves prediction accuracy by evaluating the prediction reliability of the machine learning model in real time and automatically correcting it when necessary to generate a correction result (135d).

[0179] In addition, it supports the gradual improvement of the learning model through a continuous calibration process, which can increase the reliability of power consumption predictions and reduce errors.

[0180] Referring to Fig. 9, when a power overload condition is predicted, a process is performed to analyze the cause, generate an appropriate response strategy, and deliver it to the user.

[0181] First, operational information (331), power generation information (351), weather information (371), and correction information (391) stored in the database (500) are provided to the response strategy establishment unit (150).

[0182] The response strategy formulation unit (150) is composed of a cause analysis unit (151) and a response strategy generation unit (153), each of which performs the role of analyzing the cause of the power overload state and formulating an optimal response strategy.

[0183] The cause analysis unit (151) is responsible for identifying the main causes of the power overload. To this end, it analyzes whether power usage increases rapidly in a specific process or facility based on operation information (331), and identifies the point in time when solar power generation decreases using power generation information (351).

[0184] Additionally, the impact of weather changes on power consumption is evaluated through weather information (371), and additional environmental factors are considered using correction information (391). As a result of the analysis, if a specific cause that causes a power overload is identified, that cause is defined as the cause of the overload (151a).

[0185] Afterwards, the response strategy generation unit (153) derives an appropriate response strategy (153a) based on the excess cause (151a).

[0186] Response strategies (153a) include power load adjustment, strategies to alleviate peak load by adjusting power consumption of specific processes, utilization of energy storage systems (ESS), strategies to release stored energy during periods of excess power consumption by utilizing ESS batteries, optimization of power purchase, strategies to secure additional power by utilizing KEPCO and external power supply networks, optimization of factory operations, and strategies to adjust process schedules to reduce power consumption during specific periods.

[0187] The derived response strategy (153a) is transmitted to the information transmission unit (170), which includes a result transmission unit (171) and a condition-based warning unit (173). The result transmission unit (171) performs the role of providing the analyzed prediction results and response strategy to the manager.

[0188] For example, if a power overload is anticipated, the timing and cause are explained in detail, and a response strategy is communicated to the manager. This enables the manager to take appropriate action.

[0189] The condition-based warning unit (173) generates an alarm in real time when a specific condition is met. For example, if a power overload condition exceeding a specific threshold is expected or if the remaining ESS battery level is low, it automatically provides a notification to the manager to induce an immediate response.

[0190] In addition, if power generation decreases due to severe contamination of the solar panels, an alarm can be generated to notify the manager that maintenance is required.

[0191] The operation method of the response strategy establishment unit (150) and the information transmission unit (170) contributes to maximizing the energy efficiency of the factory by predicting the power overload condition in advance and providing the optimal response plan.

[0192] In addition, the real-time alert system supports immediate response in the event of an emergency, thereby enabling the optimization of power usage.

[0193]

[0194] Although the present invention has been described above with reference to the embodiment(s) illustrated in the drawings, this is merely illustrative, and those skilled in the art will understand that various modifications may be made therefrom, and that all or part of the described embodiment(s) may be optionally combined. Accordingly, the true technical scope of protection of the present invention should be determined by the technical spirit of the appended claims.

[0195] [Explanation of the symbol]

[0196] 1: Prediction System

[0197] 100: Control unit 110: Data processing unit

[0198] 111: Synchronization unit 113: Filtering unit

[0199] 115: Normalization Unit 117: Feature Extraction Unit

[0200] 119: Preprocessing Storage Unit 119a: Preprocessing Information

[0201] 130: Predictive Analysis Unit 131: Input Optimization Unit

[0202] 133: Short-term Analysis Department 133a: Machine Learning Model

[0203] 133b: Prediction result 135: Prediction correction unit

[0204] 135a: Validation model 135b: Comparison value

[0205] 135c: Correction value 135d: Correction result

[0206] 137: Rotating Learning Department 150: Response Strategy Formulation Department

[0207] 151: Cause Analysis Department 151a: Excess Cause

[0208] 153: Response Strategy Generation Unit 153a: Response Strategy

[0209] 170: Information Transmission Section 171: Result Transmission Section

[0210] 173: Condition-based warning section

[0211] 300: Data Acquisition Unit 310: Consumption Measurement Unit

[0212] 311: Consumption Information 330: Operations Information Management Department

[0213] 331: Operation Information 350: Power Generation Measurement Unit

[0214] 351: Power Generation Information 370: Weather Information Management Department

[0215] 371: Weather Information 390: Correction Information Acquisition Unit

[0216] 391: Correction Information

[0217] 500: Database 510: Historical Information

[0218] G: Generator F: Factory

Claims

1. Database; A data acquisition unit that collects usage information, which is the amount of electricity generated in a factory, power generation information, which is the amount of electricity generated by a generator, and weather information, and stores them in the above database; An excess power consumption prediction system comprising: a control unit that predicts whether a power excess state occurs in which the usage information exceeds the power generation information through information stored in the database.

2. In Paragraph 1, The above control unit is, An excess power consumption prediction system characterized by further including a prediction analysis unit that generates a prediction result regarding whether the excess power state occurs through a machine learning model that learns by extracting one or more of the information among the plurality of information stored in the database.

3. In Paragraph 2, The above machine learning model is, An excess power consumption prediction system characterized by including a CNN (Convolutional Neural Network) model, converting multiple pieces of information including usage information, power generation information, and weather information into a two-dimensional form for learning, and analyzing patterns along a time axis to generate the prediction result.

4. In Paragraph 3, The above control unit is, An excess power consumption prediction system characterized by further including a data processing unit that receives multiple pieces of information stored in the above database and generates preprocessed information in a form that can be analyzed by the above CNN model.

5. In Paragraph 2, The above prediction analysis unit is, A verification model for verifying the prediction result generated by the machine learning model is included, and An excess power consumption prediction system characterized by further including a prediction correction unit that generates multiple comparison values ​​using the above verification model and determines the reliability of the above prediction result.

6. In Paragraph 5, The above prediction correction unit is, An excess power consumption prediction system characterized by, if it is determined that the reliability of the above prediction result is lower than a preset level, changing the plurality of information used by the machine learning model to generate the above prediction result to generate a plurality of correction values, and using the average of the generated plurality of correction values ​​to re-determine whether an excess power state occurs.

7. In Paragraph 1, The above data acquisition unit is, Operational information regarding the process operating rate and equipment usage status of the above-mentioned factory; and Further obtaining correction information including the number of employees who reported to work and the cleaning status of the generator; and An excess power consumption prediction system characterized by the above-described control unit patterning the power generation information based on the above-described operation information and correction information to predict the above-described excess power state.

Images