Lightning protection and switching control device through lighting occurrence prediction

Abstract

A lightning protection and switching control device based on lightning occurrence prediction may include a lightning prediction sensor comprising an electric field sensor that detects electrical changes in the atmosphere, a magnetic field sensor that detects changes in the atmosphere's magnetic field, a weather sensor that detects weather conditions in real time, and a radar sensor for detecting the charge distribution of clouds in the atmosphere, and a driving motor and a control unit electrically connected to the lightning prediction sensor. The control unit receives time-series data collected from an electric field sensor, a magnetic field sensor, the radar sensor, and the weather sensor in real time to generate a multidimensional feature vector, and recognizes a lightning strike pattern from the multidimensional feature vector using a pre-trained artificial intelligence model, wherein the artificial intelligence model includes a deep neural network that has learned the correlation between sensor data patterns prior to the time of a past lightning strike and the actual lightning strike, and when the lightning strike probability output by the artificial intelligence model exceeds a dynamically adjusted threshold probability, it predicts the time and intensity of the lightning strike, determines a differentiated protection level according to the predicted lightning intensity, and can adaptively control the power cutoff range and cutoff time.

Description

Technology Field

[0001] The present invention relates to the field of power facility protection and disaster prediction technology.

[0002] More specifically, the invention relates to an intelligent lightning protection and switching control device that analyzes multidimensional time-series data collected from electric field, magnetic field, weather, and radar sensors using an artificial intelligence model to predict the probability, timing, and intensity of lightning strikes in advance, and adaptively controls the connection status of a power system according to the prediction results. Background Technology

[0004] Lightning strikes are one of the major natural disasters that cause massive physical and economic losses to power systems, telecommunications facilities, and key industrial infrastructure. To mitigate such damage, passive protection devices such as lightning arresters and grounding systems have long been widely used.

[0005] However, these devices operate as a reactive measure after lightning has actually reached the equipment, so they have limitations in predicting the occurrence of lightning itself and preemptively preventing damage at the source.

[0006] To overcome these limitations, rudimentary prediction devices have been developed that detect changes in the electric field in the atmosphere and sound an alarm or cut off the power if a specific threshold is exceeded. Some technologies have attempted to improve prediction accuracy by adding weather sensors, such as temperature and humidity, but they have remained at the level of comparing individual values ​​measured by each sensor with independent static thresholds.

[0007] This method had the problem of failing to properly analyze the complex and dynamic interactions of atmospheric conditions generated by lightning strikes, and thus failing to prevent unnecessary equipment shutdowns due to frequent false alarms or serious damage caused by prediction failures. The problem to be solved

[0009] The present invention aims to provide a device that analyzes multidimensional time-series data collected from multiple sensors using an artificial intelligence model to recognize complex precursor patterns of lightning strikes beyond a simple threshold comparison method, and thereby predicts the probability, timing, and intensity of lightning strikes with high accuracy.

[0010] Furthermore, the present invention aims to provide a device that simultaneously ensures the safety of the facility and the continuity of operation by implementing a differentiated protection strategy that adaptively controls the intensity, range, and duration of the protection action, going beyond simply cutting off the power based on the predicted level of lightning risk.

[0011] The present invention aims to overcome the limitations of individual devices operating independently by forming a distributed network with adjacent devices to share data, thereby identifying wide-area lightning strike locations, and continuously enhancing the performance of the overall prediction model through federated learning.

[0012] In addition to the core function of protecting external equipment, the present invention aims to monitor the condition of key components, such as the device's drive motor or electrical contacts, in real time and predict their lifespan to notify the time for replacement in advance. means of solving the problem

[0014] A lightning protection and switching control device based on lightning occurrence prediction may include a lightning prediction sensor comprising an electric field sensor that detects electrical changes in the atmosphere, a magnetic field sensor that detects changes in the atmosphere's magnetic field, a weather sensor that detects weather conditions in real time, and a radar sensor for detecting the charge distribution of clouds in the atmosphere, and a driving motor and a control unit electrically connected to the lightning prediction sensor.

[0015] The control unit receives time-series data collected from an electric field sensor, a magnetic field sensor, the radar sensor, and the weather sensor in real time to generate a multidimensional feature vector, and recognizes a lightning strike pattern from the multidimensional feature vector using a pre-trained artificial intelligence model, wherein the artificial intelligence model includes a deep neural network that has learned the correlation between sensor data patterns prior to the time of a past lightning strike and the actual lightning strike, and when the lightning strike probability output by the artificial intelligence model exceeds a dynamically adjusted threshold probability, it predicts the time and intensity of the lightning strike, determines a differentiated protection level according to the predicted lightning intensity, and can adaptively control the power cutoff range and cutoff time. Effects of the invention

[0017] The lightning protection and switching control device based on lightning occurrence prediction according to the present invention achieves relatively high prediction accuracy compared to existing threshold-based methods and has the effect of minimizing unnecessary false alarms by deeply analyzing the complex time-series patterns of multidimensional sensor data using artificial intelligence.

[0018] In addition, the present invention enables precise analysis and cooperative response to wide-area risk situations through a distributed sensing network and a federated learning system, and can continuously improve the performance of the overall prediction model while protecting the data of individual devices.

[0019] Through reinforcement learning, the device can adapt to changes in the external environment and internal deterioration, and prevent unexpected operational failures caused by failures in advance. Brief explanation of the drawing

[0021] FIG. 1a is a drawing of a state in which a power input part and a movable contact member are in contact, according to one embodiment of the present disclosure. FIGS. 1B and FIGS. 1C are drawings in which a guide member is separated from a power input member so that the power input member and the movable contact member are separated based on the prediction of a lightning strike. FIG. 2 is a schematic diagram of the configuration of a lightning prediction sensor according to one embodiment of the present disclosure. FIG. 3 is a diagram illustrating the configuration of an artificial intelligence model according to one embodiment. Figure 4 is a flowchart illustrating a method for protection against lightning and switching control through lightning occurrence prediction. Figure 5 is a flowchart illustrating a method for protection against lightning and switching control through lightning occurrence prediction. Figure 6 is a flowchart illustrating a method for protection against lightning and switching control through lightning occurrence prediction. Specific details for implementing the invention

[0022] The terms used in the embodiments are for illustrative purposes only and should not be interpreted in a restrictive sense. Singular expressions are understood to include plural forms unless the context clearly indicates otherwise. In this specification, the terms "comprising" or "having" mean that the features, figures, steps, actions, components, parts, or combinations thereof mentioned in the specification are present, and do not preclude the possibility that other features, figures, steps, actions, components, parts, etc. may exist.

[0023] Unless specifically defined otherwise, all terms used herein, including technical or scientific terms, shall be interpreted as having the same meaning as generally accepted in the relevant technical field. Terms defined in commonly used dictionaries shall be interpreted as having meanings consistent with the context of the relevant technology and shall not be interpreted in an overly formal or ideal sense unless specifically defined in this application.

[0024] In addition, when describing with reference to the attached drawings, identical components are given the same reference numerals regardless of the symbols in the drawings, and redundant descriptions are omitted. When describing embodiments, if it is determined that a detailed description of related prior art could unnecessarily complicate the main content of the embodiments, such description may be omitted. Embodiments can be implemented in various forms of products, such as personal computers, laptops, tablets, smartphones, televisions, smart home appliances, intelligent vehicles, kiosks, and wearable devices.

[0026] According to one embodiment of the present disclosure, as illustrated in FIGS. 1a to 1c and FIG. 2,

[0027] A lightning protection control device (100) for preventing damage to power devices and loads due to lightning strikes may be disclosed. The lightning protection control device (100) may include a power input unit (101) connected to a grid supplying commercial AC power, a power output unit (102) connected to a load and transmitting power supplied from the grid, and a grounding unit (103) (e.g., ground) electrically connected to a circuit connecting the power input unit (101) and the power output unit (102). In one embodiment, the lightning protection control device (100) may be configured as a power device that supplies power supplied from the grid to a load, and may be configured to predict a lightning situation and cut off the electrical connection between the power input unit (101) and the power output unit (102).

[0028] In one embodiment, referring to FIGS. 1a to 1c, a lightning protection control device (100) may include a plurality of movable contact members (110). The movable contact members (110) may be configured to electrically connect a power input unit (101) and a power output unit (102). The movable contact members (110) may include a first contact unit (111) and a second contact unit (112). The first contact unit (111) may face the power input unit (101) and may come into contact with the power input unit (101). The second contact unit (112) is electrically connected to the power output unit (102). In one embodiment, the second contact unit (112) and the power output unit (102) may be electrically / physically connected through a spring member (120) formed of a conductive material. As described below, the spring member (120) can be elastically deformed based on the movement of the guide member (130) and the movable contact member (110) according to the driving of the drive motor (140).

[0029] In one embodiment, referring to FIGS. 1a to 1c, the movable contact member (110) can electrically transmit or cut off power from the system to the load. For example, the movable contact member (110) can move relative to the power input unit (101) via a driving motor (140) as described below. In one embodiment, when the movable contact member (110) is in contact with the power input unit (101), power can be supplied to the load via the system - power input unit (101) - movable contact member (110) - power output unit (102). In one embodiment, the control unit (300) of the lightning protection control device (100) controls the drive motor (140) based on the prediction of lightning occurrence through the lightning prediction sensor (200) to move the movable contact member (110) away from the power input unit (101), thereby separating the movable contact member (110) from the power input unit (101). Accordingly, the electrical connection between the power input unit (101) and the power output unit (102) through the movable contact member (110) can be severed.

[0030] In one embodiment, referring to FIGS. 1a through 1c, a guide member (130) may be disposed between a plurality of movable contact members (110). A plurality of movable contact members (110) may be fixed by being coupled to the guide member (130). In one embodiment, the guide member (130) may be coupled to a part of a drive motor (140). For example, the guide member (130) may be coupled to a linear part (141) of the drive motor (140). In one embodiment, the drive motor (140) may be an actuator that moves the linear part (141) in a linear direction. In one embodiment, the drive motor (140) may be a linear motor. In one embodiment, the linear part (141) may be linearly moved in a first direction or a second direction based on the driving of the drive motor (140). In one embodiment, the first direction may be a direction from the power output part (102) toward the power input part (101). The second direction is the opposite direction of the first direction and may be a direction from the power input section (101) toward the power output section (102).

[0031] In one embodiment, the driving motor (140) moves the guide member (130) to make contact between the first contact portion of the movable contact member and the power input portion, or to separate the contact if necessary.

[0032] According to one embodiment of the present disclosure, as illustrated in FIGS. 1a to 1c and FIG. 2, a lightning protection control device (100) may include a lightning prediction sensor (200). In one embodiment, the lightning prediction sensor (200) may include an electric field sensor (201), a magnetic field sensor (202), a weather sensor (204), and a radar sensor (203). Each sensor may detect changes in electrical, magnetic, and weather conditions in the atmosphere in real time. In one embodiment, the electric field sensor (201) may detect electrical changes in the atmosphere. The electric field sensor (201) may detect a phenomenon in which the voltage difference increases before a lightning strike occurs. Generally, before a lightning strike occurs, the electric field may exhibit a pattern of rapid increase. The electric field sensor (201) may include a sensitivity electric field measurement sensor and a signal processing device. The electric field sensor (201) may monitor data in real time and transmit the data to a control unit (300). In one embodiment, the control unit (300) may be a personal computer (PC) that is communicationally connected to the lightning protection control device (100). In one embodiment, the control unit (300) may be part of the lightning protection control device (100). In one embodiment, the magnetic field sensor (202) may detect changes in the magnetic field in the atmosphere. Before a lightning strike, the current in the atmosphere increases, and accordingly, the magnetic field may also be strengthened. The magnetic field sensor (202) may detect this change and predict the possibility of a lightning strike. In one embodiment, the radar sensor (203) may detect the charge distribution of clouds in the atmosphere. The control unit (300) may predict the possibility of a lightning strike by tracking the charge distribution inside the clouds through the radar sensor (203). In one embodiment, the control unit (203) may calculate the location and time where a lightning strike is likely to occur through the radar sensor (203). In one embodiment, the weather sensor (204) can detect weather conditions such as atmospheric pressure, temperature, and humidity in real time.Sensors detect lightning precursors occurring in the atmosphere and transmit the detected data to the control unit (300). The control unit (300) can analyze the data transmitted from the lightning prediction sensor (200). The control unit (300) can determine whether data such as electric field changes, magnetic field changes, weather conditions, and charge distribution within clouds exceed a preset threshold. The control unit (300) determines that lightning is predicted to occur as the collected data exceeds the preset threshold, and controls the drive motor (140) to move the guide member (130) in a second direction, thereby separating the first contact part (111) of the movable contact member (110) from the power input part (101). In one embodiment, the control unit (300) can move the guide member (130) in a second direction relative to the power input part (101) so that the movable contact member (110) contacts and / or is electrically connected to the ground part (103). The control unit (300) can detect that the movable contact member (110) is in contact with and / or electrically connected to the grounding unit (103). The control unit (300) can control the drive motor (140) so that the guide member (130) stops when the movable contact member (110) is detected to be in contact with and / or electrically connected to the grounding unit (103). This prevents overvoltage or overcurrent caused by lightning that may occur in the system or lightning protection control device (100) (e.g., power device) from being transmitted to the load by cutting off the power supply. In one embodiment, the control unit (300) can control the drive motor (140) based on the probability of lightning occurrence falling below a preset value to move the guide member (130) in a first direction, thereby connecting the first contact part (111) of the movable contact member (110) to the power input part (101). Accordingly, power can be supplied to the load via the system - power input section (101) - movable contact member (110) - power output section (102).

[0033] In one embodiment, the control unit (300) may be designed such that the algorithm for predicting lightning strikes processes data collected from the lightning prediction sensor (200) in real time and warns of the possibility of lightning strikes when a specific threshold is exceeded. In one embodiment, the electric field sensor (201) measures the magnitude of the electric field in the atmosphere in real time and the rate of change of the electric field over a specific period ( E) can be calculated. The magnetic field sensor (202) measures changes in the magnetic field in the atmosphere and can calculate the rate of change of the magnetic field. The weather sensor (204) can monitor the amount of change in weather conditions by collecting data such as atmospheric pressure, temperature, humidity, and wind in real time.

[0034] In one embodiment, threshold values ​​may be pre-entered into the control unit (300). For example, threshold values ​​may be stored in the memory of the control unit (300). The threshold values ​​may include an electric field change rate threshold, a magnetic field change rate threshold, a charge amount threshold, a threshold for the atmospheric pressure change rate, a threshold for the temperature change rate, and a threshold for the humidity change rate.

[0035] In one embodiment, the control unit (300) can calculate the probability of a lightning strike based on the fact that the rate of change in the electric field collected from the electric field sensor (201), the rate of change in the magnetic field collected from the magnetic field sensor (202), and the amount of charge collected from the radar sensor (203) each exceed a preset threshold value, and at least two of the pressure change, temperature change, and humidity change detected by the weather sensor (204) exceed a preset threshold value. The control unit (300) can determine that a lightning strike is predicted based on the fact that the probability of a lightning strike exceeds the threshold probability, and can control the drive motor (140) so that the movable contact member (110) is separated from the power input unit (101).

[0036] In one embodiment, the measurable maximum electric field change rate, measurable maximum magnetic field change rate, measurable maximum charge amount, measurable atmospheric pressure change rate, measurable maximum temperature change rate, and measurable maximum humidity change rate may be values ​​predetermined through past data collected by the control unit (300) through the lightning prediction sensor (200).

[0037] In one embodiment, the control unit (300) determines that the occurrence of lightning is predicted based on the probability of lightning occurrence exceeding a threshold probability, and controls the drive motor (140) to move the guide member (130) and the movable contact member (110) in a second direction relative to the power input unit (101). In one embodiment, the threshold probability may be set to 90% and may be changed based on past lightning patterns.

[0038] According to one embodiment of the present disclosure, a lightning protection control device (100) can minimize damage to a power device (e.g., a lightning protection control device (100)) and a load by cutting off the power supply in advance before lightning strikes occur. Additionally, a grounding part (103) can safely discharge electrical shocks caused by lightning strikes to the ground. In one embodiment, when lightning strikes are predicted, a movable contact member (110) is electrically connected to the grounding part (103), thereby effectively preventing damage to the lightning protection control device (100) caused by lightning strikes.

[0039] Specifically, the operation process of the present invention is as follows. Sensors (e.g., lightning prediction sensor (200)) that detect changes in the electric and magnetic fields in the atmosphere, weather conditions, and the amount of charge within the cloud collect data in real time and transmit it to the control unit (300). The control unit (300) analyzes the possibility of lightning strikes based on this data, and if the data exceeds a set threshold, it can immediately operate the drive motor (140) to separate the power input unit (101) and the movable contact member (110). By doing so, it is possible to block electrical shock caused by lightning strikes that could be transmitted from the grid to the load.

[0040] In addition, the control unit (300) can predict the possibility of a lightning strike by communicating with the meteorological agency through a network. If the probability of a lightning strike occurring at the location where the lightning protection control device (100) is installed, calculated by the meteorological agency, exceeds a threshold probability, the drive motor (140) can be controlled to separate the movable contact member (110) from the power input unit (101).

[0042] FIG. 3 is a diagram illustrating the configuration of an artificial intelligence model according to one embodiment.

[0043] According to one embodiment, an artificial intelligence model may be composed of an input layer, a hidden layer, and an output layer.

[0044] The input layer is the layer associated with the values ​​input into the artificial intelligence model. In the hidden layer, feature maps can be generated by performing multiply-accumulate (MAC) operations and activation operations on the input values. The MAC operation is a process of multiplying the input values ​​by their corresponding weights and then summing these multiplied values. The activation operation is a process of inputting the results of the MAC operation into an activation function to output a final result, and various types of activation functions can be used. For example, activation functions may include, but are not limited to, the sigmoid function, tangent function, ReLU function, Leaky ReLU function, Max Out function, and ELU function.

[0045] A hidden layer may consist of at least one layer, for example, divided into a first hidden layer and a second hidden layer. In this case, the first hidden layer generates a feature map by performing MAC operations and activation operations based on the input values ​​of the input layer, and this feature map can be used as the input value of the second hidden layer. The second hidden layer can perform MAC operations and activation operations based on the feature map generated as a result of the first hidden layer.

[0046] The output layer may be a layer related to the result of an operation performed in the hidden layer.

[0047] In one embodiment, a learning model learns syllable (character) patterns that are frequently used together in a given corpus to automatically identify the boundaries between compound words and named entities, and integrates object information from a first UI source with object information rendered in a browser to generate a learning object information file. Using this learning object information file, data for training a deep learning network is generated, and based on data received from various domains, the data is standardized into an integrated format according to one or more standardization methods suitable for each domain. Subsequently, data of a specific domain is trained and inferred, information necessary for standardization in that domain is determined, and post-processing can be performed on the data received from various domains.

[0048] The first UI source includes an XML file, and the training object information file includes an input JSON file for feature learning and an output JSON file used as label data during training. This output JSON file is a file containing HTML DOM Tree information implemented in compliance with web standards. Various domains include at least one of a RAN (radio access network), a transport, or a core, and the post-processing process may include a correlation function.

[0049] Artificial intelligence (AI) systems are computer systems that mimic human intelligence; unlike traditional rule-based smart systems, they are equipped with the ability for machines to learn and make decisions autonomously. As the use of these AI systems increases, their cognitive capabilities improve and they gain a more accurate understanding of individual user preferences, leading to the gradual replacement of existing rule-based systems with deep learning-based AI.

[0050] Artificial intelligence technology consists of machine learning and various component technologies based on it. Machine learning includes algorithmic technologies that autonomously analyze and classify the characteristics of input data, while component technologies are techniques that mimic human cognitive and judgment functions by utilizing machine learning algorithms, such as deep learning, and encompass fields such as linguistic understanding, visual understanding, reasoning and prediction, knowledge representation, and motion control.

[0051] Artificial intelligence technology is being applied in various fields; linguistic understanding encompasses the technology of recognizing and processing human language, while visual understanding encompasses the technology of recognizing and processing objects. Reasoning and prediction technologies are used to analyze information and make logical judgments, and knowledge representation technology performs the function of automating human experience into data. Motion control is the technology that manages the movements of autonomous vehicles or robots.

[0052] Generally, applying machine learning algorithms to real-world environments involves a trial-and-error approach. In particular, deep learning requires numerous iterations; since execution in actual environments is difficult, learning is achieved through simulations created in a virtual environment on a computer.

[0053] In this invention, artificial intelligence refers to a technology that mimics human learning, reasoning, and perceptual abilities and implements them on a computer, and may include concepts such as machine learning and symbolic logic. Machine learning is an algorithmic technology that independently classifies or learns the characteristics of input data; various artificial intelligence technologies utilize such machine learning algorithms to analyze input data and perform judgments or predictions based on the results.

[0054] Through machine learning, computer software can enhance its data processing capabilities, and neural network models are constructed by modeling the correlations between data. In this process, neural network models extract and analyze features from input data to derive relationships, and optimize parameters through this iterative process. For example, neural network models can learn the relationship between input and output for a given data pair, or learn relationships by identifying regularities solely from the input data.

[0055] Artificial intelligence learning models can be designed to computerize the structure of the human brain, with multiple network nodes simulating the signal transmission of neurons. These nodes are located in layers of different depths to exchange data and can take the form of artificial neural networks, convolutional neural networks (CNNs), and the like. Such models can be trained using methods such as supervised learning, unsupervised learning, and reinforcement learning, and various machine learning algorithms can be applied.

[0056] In particular, CNNs are a type of multilayer perceptron that requires minimal preprocessing and consist of convolutional layers and general neural network layers. CNNs can effectively process 2D input data and demonstrate superior performance in image and speech processing compared to other deep learning architectures. CNNs can be trained using the standard backpropagation method and have the advantage of being easily learned with a small number of parameters.

[0057] Convolutional networks are neural networks composed of nodes that share parameters, and the performance of computer vision tasks has been significantly improved through the combination of large amounts of training data and computational power. As the volume of currently available datasets increases, increasing the size of networks contributes to improving test accuracy. Along with this, the optimal utilization of computing resources has become a critical factor, and distributed implementations of deep neural networks can be used.

[0059] Figure 4 is a flowchart illustrating a method for protection against lightning and switching control through lightning occurrence prediction.

[0060] In the flowchart of FIG. 4, processes, methods, algorithms, etc. are described sequentially, but they may be configured to operate in any suitable order. In other words, the steps of the processes, methods, and algorithms described in various embodiments of the present invention do not need to be performed in the order described in the present invention. Even if some steps are described as being performed asynchronously, in other embodiments, such steps may be performed simultaneously. The processes illustrated in the drawings are not to exclude other changes or modifications, and none of the illustrated processes or steps are essential to one or more of the various embodiments of the present invention.

[0061] In operation 410, a lightning protection control device (e.g., the lightning protection control device (100) of FIG. 1a) can generate a multidimensional feature vector by acquiring time-series data from a lightning prediction sensor (200) under the control of a control unit (e.g., the control unit (300) of FIG. 1a). This means collecting different types of data, such as electric fields, magnetic fields, weather conditions, and cloud charge distributions, continuously over time, going beyond simply reading the values ​​of individual sensors. The control unit (300) can synchronize and normalize the heterogeneous data streams collected in this way to convert them into a multidimensional feature vector, which is a single standardized data structure that comprehensively represents the atmospheric state at a specific point in time.

[0062] In operation 420, the control unit (300) inputs the generated multidimensional feature vector stream into a pre-trained artificial intelligence model to analyze complex precursor patterns of the data and, as a result, can calculate a multi-stage threat index. The artificial intelligence model used at this time is a type of deep neural network and may have learned subtle temporal change patterns of feature vectors that appeared in numerous past lightning strike cases. The model determines the probability of a lightning strike by comprehensively analyzing not only the simple size of the currently input data but also its rate of change (speed) and acceleration of change (suddenness of change). The analysis result can be output as a threat index divided into multiple stages such as 'caution', 'warning', and 'danger', which enables a much more sophisticated and situation-appropriate judgment than a single threshold method, thereby reducing false alarms and increasing the reliability of the prediction.

[0063] In operation 430, the control unit (300) can adaptively perform differentiated switching protection actions in conjunction with the calculated multi-stage threat index. This means applying different levels of protection measures depending on the stage of the threat index. For example, in the 'caution' stage, the monitoring cycle is simply shortened and data is recorded, while in the 'alert' stage, an alert is issued to the manager or the power to non-critical equipment is cut off. Finally, when it is determined that the 'danger' stage has been reached, the control unit (300) can perform the strongest protection action by sending an immediate disconnection command to the drive motor (140) to physically disconnect the movable contact member from the power input. This differential response has the effect of finding an optimal balance point that ensures operational efficiency by minimizing unnecessary shutdown of the entire facility, while providing reliable protection in actual dangerous situations. The stage classification is merely an example and may vary depending on the settings.

[0064] A control unit (300) of a lightning protection control device (100) according to one embodiment can generate a multidimensional feature vector by receiving time-series data in real time from an electric field sensor, a magnetic field sensor, a weather sensor, and a radar sensor constituting a lightning prediction sensor (200). The control unit (300) can recognize a lightning occurrence pattern from the multidimensional feature vector using a pre-trained artificial intelligence model, wherein the artificial intelligence model may include a deep neural network that has learned the correlation between sensor data patterns prior to the time of a past lightning occurrence and the actual lightning occurrence. The control unit (300) can predict the time and intensity of a lightning occurrence when the lightning occurrence probability output by the artificial intelligence model exceeds a threshold probability that is dynamically adjusted. The control unit (300) can adaptively control the power cut-off range and cut-off time by determining a differentiated protection level according to the predicted lightning intensity.

[0065] In one embodiment, the control unit (300) can calculate the electric field change acceleration by differentiating the electric field measurement value collected from the electric field sensor along the time axis to calculate the rate of change, which is the first derivative, and the acceleration, which is the second derivative. Additionally, the control unit (300) can estimate the current density distribution in the atmosphere by analyzing the rotational component of the magnetic field vector collected from the magnetic field sensor, which can be done by deriving the current density from the rotational value of the magnetic field by applying Ampere's law of Maxwell's equations. The control unit (300) can track the movement path and speed of the charge center by generating three-dimensional spatial mapping data of the charge distribution within the cloud collected from the radar sensor, and in this process, it can identify areas with high charge density and track changes in the position of the charge center over consecutive time frames by analyzing the intensity and phase changes of the radar reflection signal. The control unit (300) can calculate the atmospheric instability index by collecting temperature, humidity, and atmospheric pressure data by altitude from the weather sensor. Specifically, the collected temperature and humidity profiles are integrated to calculate the Convective Available Potential Energy (CAPE), the Richardson number is calculated using the temperature gradient and wind speed difference by altitude, and the two evaluation values ​​are combined to quantify the final atmospheric instability index so that the higher the CAPE value and the lower the Richardson number, the higher the value. The control unit (300) can issue a lightning imminent warning when a composite condition is satisfied in which the calculated electric field change acceleration, current density, charge center approach distance, and atmospheric instability index each exceed their respective threshold values. The control unit (300) can set a three-stage protection level including pre-warning, warning, and danger depending on the degree to which the composite condition is satisfied, and can perform stepwise protection actions such as cutting off sensitive equipment, disconnecting the main power system, and grounding the entire facility according to each level. The start time of the protection action can be determined based on the expected time of lightning arrival predicted by the artificial intelligence model.After the lightning strike ends, if the AI ​​model recognizes a pattern where the sensor measurements return to the normal range, power can be restored in reverse order; additionally, if an error occurs when comparing the actual occurrence of lightning with the prediction results, the corresponding data can be saved as additional training data to retrain the AI ​​model online.

[0066] In one embodiment, the control unit (300) can calculate the electric field change acceleration by differentiating the electric field measurement value collected from the electric field sensor along the time axis to calculate the rate of change, which is the first derivative, and the acceleration, which is the second derivative. This is intended to determine not only the strength of the electric field but also how quickly the strength changes, and a high acceleration value indicates that the atmospheric condition is rapidly becoming unstable and can be interpreted as a strong precursor to an imminent lightning strike. Additionally, the control unit (300) can estimate the current density distribution in the atmosphere by analyzing the rotational component of the magnetic field vector collected from the magnetic field sensor, which may utilize the physical principle that the spatial curl of the magnetic field induces current density according to Ampere's law of Maxwell's equations. This allows for the detection of invisible electric charge flow in the atmosphere, thereby providing the effect of preemptively identifying the process of forming a lightning discharge path.

[0067] According to one embodiment, the control unit (300) can generate three-dimensional spatial mapping data of the charge distribution within the cloud collected from a radar sensor to track the movement path and speed of the charge center. In this process, the control unit (300) can analyze the intensity and phase changes of the radar reflection signal to identify areas with high charge density in three dimensions, and calculate a speed vector by tracking the change in the position of the center point of the area over a continuous time frame. This can provide important grounds for evaluating the direction and urgency of the threat by determining whether the source of the threat is approaching or moving away from the protected facility.

[0068] According to one embodiment, the control unit (300) can calculate an atmospheric instability index by collecting temperature, humidity, and atmospheric pressure data at different altitudes from a weather sensor. Specifically, the potential for convective phenomena, i.e., the energy source of a storm, can be evaluated by integrating the collected temperature and humidity profiles to calculate the Convective Available Potential Energy (CAPE). At the same time, the mechanical stability of the atmosphere can be evaluated by calculating the Richardson number using the temperature gradient and wind speed difference at different altitudes. The control unit (300) can quantify the final atmospheric instability index by combining the two evaluation values ​​to have a higher value when the Convective Available Potential Energy higher than a specified level and the Richardson number lower than a specified level (=unstable atmospheric state) are combined.

[0069] According to one embodiment, the control unit (300) can issue a lightning imminent warning only when a composite condition is satisfied in which the independently calculated electric field change acceleration, current density, charge center approach distance, and atmospheric instability index each exceed their respective threshold values ​​simultaneously. This has the effect of minimizing false alarms caused by the limitations of a single indicator and maximizing the reliability of the prediction. The control unit (300) can set three protection levels, including pre-warning, warning, and danger, depending on the degree to which the composite condition is satisfied, and can perform stepwise protection actions such as disconnecting sensitive equipment, isolating the main power system, and grounding the entire facility according to each level. The start time of the protection action is determined based on the expected time of lightning arrival predicted by an artificial intelligence model, so that protection measures can be executed at an optimal timing that is neither too early nor too late.

[0070] After the lightning strike ends, if the AI ​​model recognizes a pattern in which the sensor measurements return to the normal range, power can be safely restored in reverse order. Furthermore, if an error occurs when comparing the actual occurrence of lightning with the prediction results, the sensor data pattern at that point in time is automatically saved as new training data to retrain the AI ​​model online, thereby implementing an adaptive learning system that autonomously improves prediction accuracy as the device operates.

[0071] The types of protection levels, including preliminary boundaries, boundaries, and risks mentioned in this specification, and the measures for each stage are merely examples and are not limited thereto; they may vary depending on the characteristics of the equipment to be protected and the operating environment settings.

[0072] In one embodiment, the control unit (300) can generate a frequency spectrum by performing a Fast Fourier Transform on electromagnetic field data and calculate the power spectrum density of each component within a predetermined frequency range from low frequency to high frequency to track changes in energy distribution over time. The control unit (300) can construct a polyhedron of minimum volume that encloses a high-charge density region using three-dimensional cloud charge distribution data and calculate the center point to derive the approach velocity vector of the charge-dense region. Additionally, the control unit (300) can calculate the critical electric field strength required for the breakdown of the insulation properties of air in real time based on weather data, and define and monitor the lightning critical proximity by comparing this with the actual measured electric field strength. The control unit (300) can calculate a membership value for each section by substituting multiple numerical input values, such as the energy growth rate of the frequency spectrum, the approach velocity of the charge-dense region, and the lightning critical proximity, into a membership function composed of multiple predefined sections. Subsequently, a set of rules configured to output a risk level as a conclusion by conditionally applying various combinations of the membership values ​​of each input variable can be applied to infer an overall risk level, and the results can be weighted averaged to generate a final lightning risk score expressed as a single quantitative number. The control unit (300) can dynamically adjust a threshold value that serves as a criterion for initiating protection action based on this risk score; the sensitivity can be controlled by lowering the threshold value to react more sensitively when the hourly growth rate of the risk score is high, and by raising the threshold value to prevent overprotection when the score is on a decreasing trend. The control unit (300) can detect an inflection point where the risk score rises rapidly to estimate the expected time of lightning strike occurrence, and can exponentially increase the intensity of the protection action in inverse proportion to the remaining time until the expected time.

[0073] In one embodiment, the control unit (300) may use a combination of several independent analysis techniques to increase the accuracy of lightning prediction. The control unit (300) can convert the change in the signal over time into a frequency-specific energy distribution by performing a Fast Fourier Transform on electromagnetic field data collected from electric and magnetic field sensors. This is intended to capture electromagnetic wave signals of specific low or high frequency bands emitted in the atmosphere before a lightning strike occurs, and by calculating the power spectral density of each frequency component, it has the effect of preemptively detecting minute energy changes associated with lightning precursor phenomena.

[0074] According to one embodiment, the control unit (300) may apply a Convex Hull algorithm to three-dimensional cloud charge distribution data obtained from a radar sensor. This may mean a process of mathematically constructing the smallest convex polyhedron that includes all points with high charge density to objectively determine the overall size (volume) and center of gravity (center point) of a threatening charge mass. By tracking the movement of the center point over time, the control unit (300) can accurately derive the direction and speed of the charge-dense area approaching the protected facility as a vector, thereby analyzing the spatiotemporal movement path of the threat.

[0075] According to one embodiment, the control unit (300) can calculate the dielectric strength of air in real time under current atmospheric conditions based on weather sensor data. This utilizes the physical principle that the density and mean free path of air molecules change depending on atmospheric pressure, temperature, and humidity, and accordingly, the strength of the electric field required for dielectric breakdown changes. The control unit (300) can calculate this critical electric field strength in real time and define the lightning critical proximity as the ratio of the current electric field strength to the currently measured electric field strength. This indicator can serve as a very direct and reliable risk measure indicating how physically close the current state is to the point of discharge.

[0076] According to one embodiment, the control unit (300) can make a final decision by synthesizing multiple data such as the energy growth rate of the frequency spectrum derived in different ways, the approach speed of the charge density region, and the lightning threshold proximity. To this end, each numerical input value can be substituted into a membership function composed of multiple predefined intervals such as 'low', 'medium', 'high', etc., to calculate the degree of membership value for each interval. Subsequently, a set of multiple condition-conclusion type rules configured to output a risk grade corresponding to the result as a conclusion by conditionally applying various combinations of the degree of membership values ​​of each input variable can be applied to infer an overall risk level, and the results can be weighted averaged to generate a final lightning risk score expressed as a single quantitative number. This method is effective for deriving a rational conclusion similar to human reasoning by synthesizing various factors in complex situations where clear mathematical modeling is difficult.

[0077] According to one embodiment, the control unit (300) can dynamically adjust the threshold value, which serves as the criterion for initiating protection action based on the calculated lightning risk score, depending on the situation. For example, the sensitivity of the system can be controlled by lowering the threshold value when the hourly increase rate of the risk score is high to react sensitively to small changes in risk, and by raising the threshold value when the score is on a decreasing trend to prevent unnecessary overprotection action. The control unit (300) can mathematically detect the point where the risk score increases gradually and then begins to skyrocket, i.e., the inflection point, and estimate this as the most critical time of imminent lightning strike. In this case, the intensity of the protection action can be increased exponentially in inverse proportion to the remaining time until the expected time, thereby enabling the execution of stronger protection measures as time approaches.

[0078] The specific forms of the Fast Fourier Transform, Paschen law-based computation, or membership functions and rule sets mentioned in this specification are merely examples for implementing the invention and may be replaced by other signal processing, geometric analysis, physical modeling, or inference methods that produce equivalent technical effects, and the scope of the invention is not limited thereto.

[0081] Figure 5 is a flowchart illustrating a method for protection against lightning and switching control through lightning occurrence prediction.

[0082] In the flowchart of FIG. 5, processes, methods, algorithms, etc. are described sequentially, but they may be configured to operate in any suitable order. In other words, the steps of the processes, methods, and algorithms described in various embodiments of the present invention do not need to be performed in the order described in the present invention. Even if some steps are described as being performed asynchronously, in other embodiments, such steps may be performed simultaneously. The processes illustrated in the drawings are not to exclude other changes or modifications, and none of the illustrated processes or steps are essential to one or more of the various embodiments of the present invention.

[0083] In operation 510, a lightning protection control device (e.g., the lightning protection control device (100) of FIG. 1a) can form a distributed sensing network and share data in real time by being connected via a wireless communication network with other adjacent lightning protection control devices (100) under the control of a control unit (e.g., the control unit (300) of FIG. 1a). At this time, each device acts as an individual node and may have a mesh network structure that transmits data by bypassing through other nodes even if communication of a specific node is interrupted. Through the mesh network structure, it is possible to comprehensively understand the wide-area standby state beyond the observation range of a single device and ensure the continuity of data collection even in the event of communication failure of some devices.

[0084] In operation 520, the control unit (300) can estimate the three-dimensional coordinates of a point where lightning is expected to occur by using sensor data shared through a distributed sensing network and a triangulation technique. Specifically, it can precisely measure the time difference of arrival (TDOA) of electromagnetic waves generated by lightning precursors reaching at least three other devices within the network. The control unit (300) can apply this time difference data to a hyperbolic positioning algorithm to calculate the point where each hyperbolic trajectory intersects and correct the error using the least squares method, thereby specifying the expected location of lightning in three dimensions. This function has the effect of enabling a more precise response by accurately identifying the direction and location of the threat.

[0085] In operation 530, the control unit (300) can calculate the physical distance between the estimated expected lightning strike point and each device within the network and coordinate sequential protection operations by differentially assigning protection priorities according to the distance. For example, the highest priority can be assigned to the device closest to the expected lightning strike point, and the priority can be lowered as the distance increases. Accordingly, by initiating protection operations in sequence starting from the equipment most directly exposed to danger, the most efficient defense strategy can be executed within a limited time.

[0086] In operation 540, the control unit (300) can continuously improve the global prediction model through federated learning and ensure data integrity through a blockchain. Each device can independently train a local artificial intelligence model based on collected data and then transmit only the weight change amount, which is the learning result, to the central server, rather than the original data. The central server can aggregate the weight change amounts collected from multiple devices to create a global model with superior performance and distribute it back to each device. This federated learning method has the effect of collectively improving the prediction performance of the entire network while protecting the data privacy of each device. At the same time, all prediction and operation history is recorded on a blockchain-based distributed ledger, which fundamentally prevents data tampering and guarantees the reliability of all records.

[0087] In one embodiment, the control unit (300) may configure a distributed sensing network that is connected to a plurality of adjacent lightning protection devices via a wireless communication network to share sensing data from each device in real time. This network may have a network structure in which each device autonomously relays data packets to ensure the continuity of data transmission. The control unit (300) may estimate the three-dimensional coordinates of the expected lightning strike location by analyzing the difference in arrival times of electromagnetic waves received from multiple devices within the network. The control unit (300) may coordinate protection operations to be performed starting from the device closest to the estimated location by assigning a differentiated protection priority based on the estimated location and the distance to each device. The control unit (300) may enable reliable history tracking by adding the data of each device to a data block linked to a previous record using an encryption hash function, verifying its validity through a consensus algorithm among the distributed devices, and recording it in a distributed ledger that is impossible to tamper with. If the sensor of a specific device exhibits abnormal operation, the control unit (300) may estimate and supplement the data of that device using a statistical interpolation method that utilizes data from adjacent devices. Additionally, the control unit (300) can derive a final lightning occurrence probability reflecting past patterns by combining a statistical prior probability calculated from the frequency of lightning occurrences by past seasons and time periods with the current probability of occurrence output in real time by an artificial intelligence model based on Bayes' theorem. The control unit (300) can continuously improve overall prediction performance while protecting data privacy through a federated learning method in which only the weight change amount of the artificial intelligence model trained by each local device is transmitted to a central server, and the central server aggregates this to create a global model and then distributes it back to each local device.

[0089] In one embodiment, the control unit (300) may perform a sequential decision-making process to learn an optimal switching policy that maximizes long-term cumulative rewards. The control unit (300) may recognize the current situation by combining the current lightning threat index and the power connection status, and may decide one action among maintaining, disconnecting, or reconnecting the power connection based on the recognized state. After performing the determined action, the control unit (300) may obtain a reward, which is a numerical score from a pre-designed reward function according to the result. This reward function may be designed to provide the highest positive reward when the power is successfully cut off during the danger phase when an actual lightning strike occurs, a negative reward proportional to the duration in the case of unnecessary cutoff, and the largest negative reward proportional to the scale of damage when equipment damage occurs due to a failure of protection. The artificial intelligence model may include a value neural network that predicts the value, which is the total sum of rewards to be obtained in the future when a specific action is taken in a specific state, and a policy neural network that determines the probability of selecting the optimal action in each state. The control unit (300) can learn an optimal action policy that balances protection success rate and operational efficiency by progressively improving the policy neural network using the predicted value of the value neural network. The control unit (300) can ensure data efficiency and stability of the learning process by storing sequential data of states, actions, and rewards experienced in the past in a memory buffer, then randomly extracting them and using them for re-learning.

[0090] In one embodiment, the control unit (300) can learn an optimal switching policy by applying a reinforcement learning technique that goes beyond simply operating according to a set rule and finds the optimal course of action on its own through experience. This means that the device performs a sequential decision-making process to select the most advantageous action in the current situation to achieve a long-term goal, much like an intelligent agent. In this process, the control unit (300) can recognize a 'State' representing the current phase by synthesizing the current lightning threat index, power connection status, and past sensor data flow. Based on this recognized state, the control unit (300) can determine one 'Action' from among possible options, such as 'maintaining the current power connection,' 'immediately disconnecting the power,' or 'reconnecting the disconnected power.'

[0091] According to one embodiment, the control unit (300) can obtain a 'Reward,' which is a numerical score from a pre-designed reward function, based on the result of the action after performing a determined action. This reward function is a key part that defines the learning goal of the device. For example, the highest positive reward may be given if the power is successfully cut off during a risk stage where actual lightning strikes occur, and a negative reward (penalty) proportional to the duration of the cutoff may be given in the case of an unnecessary cutoff where the power was cut off based on risk prediction but no actual lightning strike occurred. In particular, it may be designed to give the largest negative reward proportional to the scale of damage if actual damage occurs to the equipment due to a failure to take protective measures. This reward system has the effect of inducing the control unit (300) to find an optimal balance point between two potentially conflicting goals—protection success rate and operational efficiency—beyond simply avoiding risk.

[0092] According to one embodiment, the control unit (300) can sequentially store sequential data of 'state-behavior-reward' experienced in the past in a memory buffer. Subsequently, when learning with a new experience, past experiences can be randomly extracted from this buffer and used together for re-learning. This prevents excessive bias toward specific experiences and allows for the repeated learning of important but rarely occurring events, thereby having the effect of significantly improving overall data efficiency and the stability of the learning process.

[0093] The specific structures of the value neural network and policy neural network or the design methods of the reward functions mentioned in this specification are merely examples for implementing the invention and may be replaced with other reinforcement learning algorithms and reward systems that produce equivalent technical effects, and the scope of the invention is not limited thereto.

[0095] Figure 6 is a flowchart illustrating a method for protection against lightning and switching control through lightning occurrence prediction.

[0096] In the flowchart of FIG. 6, processes, methods, algorithms, etc. are described sequentially, but they may be configured to operate in any suitable order. In other words, the steps of the processes, methods, and algorithms described in various embodiments of the present invention do not need to be performed in the order described in the present invention. Even if some steps are described as being performed asynchronously, in other embodiments, such steps may be performed simultaneously. The processes illustrated in the drawings are not excluded from other changes or modifications, and none of the illustrated processes or steps are essential to one or more of the various embodiments of the present invention.

[0097] In operation 610, a lightning protection control device (e.g., the lightning protection control device (100) of FIG. 1a) can quantitatively calculate the total energy to be emitted by an expected lightning strike through sensor data fusion under the control of a control unit (e.g., the control unit (300) of FIG. 1a). This can be achieved by combining information on the charge distribution and volume of clouds obtained from a radar sensor with information on the potential gradient of the atmosphere obtained from an electric field sensor. The control unit (300) can apply these data to a physical model to pre-calculate the total amount of energy that may be emitted when a lightning strike occurs. This process has the effect of generating an objective and quantitative risk indicator for subsequent decision-making by converting a vague possibility of risk into a concrete physical quantity.

[0098] In operation 620, the control unit (300) can calculate the estimated damage cost based on the calculated lightning energy and the loss cost resulting from the protection measures based on operational data. The estimated damage cost can be calculated by applying the impact of the calculated energy on the equipment to be protected into a predefined damage model and converting the possible repair or replacement costs into monetary value. The operational loss cost can be calculated by referring to economic parameters, such as the hourly production value of the equipment stored in a database, and calculating the opportunity cost that would occur if the power were cut off.

[0099] In operation 630, the control unit (300) can compare and analyze the previously calculated estimated damage costs and operating loss costs through a decision model. This decision model may be based on economic rationality analysis techniques such as utility functions and aims to minimize net economic losses beyond simply avoiding physical risks.

[0100] In operation 640, the control unit (300) may select and execute a first protection mode corresponding to total shutdown or a second protection mode corresponding to partial shutdown based on the results of a comparative analysis. If the decision model determines that the expected damage cost is significantly greater than the operating loss cost, it may execute a first protection mode that immediately shuts off the power and grounds the entire facility to protect it. Conversely, if it determines that the expected damage cost is less than the operating loss cost, it may execute a second protection mode that partially shuts off only non-core facilities while maintaining core facilities, taking into account the continuity of operation more importantly.

[0101] In operation 650, the control unit (300) can automatically update the decision model by receiving feedback on whether actual damage occurred after the lightning event ended. For example, it can receive information on the actual amount of damage that occurred through maintenance records or asset management data, and reflect this in the existing damage model and economic parameters to train the model to make more accurate cost predictions for the next decision.

[0102] In one embodiment, the control unit (300) fuses radar sensor and electric field sensor data to estimate the discharge path and charge amount of the expected lightning strike, and thereby can quantitatively calculate the total energy expected to be emitted by the lightning strike. The control unit (300) can input the calculated energy into a pre-stored equipment damage model to convert the expected physical damage into an expected damage cost, which is a monetary value. Additionally, the control unit (300) can retrieve economic parameters, such as the hourly production loss cost of the equipment to be protected, from a database to calculate the operating loss cost that will occur when performing the protection operation. The control unit (300) can derive an optimal protection strategy that minimizes net economic loss through a decision-making model that compares and analyzes the expected damage cost and the operating loss cost. If, as a result of the analysis, it is determined that the expected damage cost is greater than the operating loss cost by a specified level, a first protection mode can be executed to prevent damage at the source by cutting off the power to the entire equipment and grounding it. Conversely, if it is determined that the expected damage cost is less than the operating loss cost, a second protection mode can be executed to selectively cut off only the remaining loads while maintaining the power to the core loads. After the lightning event ends, the control unit (300) can receive feedback on whether actual damage occurred and update the damage model and economic parameters to continuously improve the accuracy of the decision model.

[0104] In one embodiment, the control unit (300) can perform a maintenance function to ensure the physical reliability of the device. To determine the soundness of the drive motor, the control unit (300) can continuously collect operational data including accumulated operating time, current consumption patterns, and time to reach a target position by measuring the motor's rotational speed with a Hall sensor and the drive current with a current sensor in real time. Additionally, to monitor the deterioration state of the electrical contact, the control unit (300) can use a four-terminal measurement method that excludes the influence of resistance of the measurement wire to precisely detect minute changes in resistance, and can evaluate the degree of erosion on the contact surface by analyzing the temporal increasing trend of the measured resistance value. The control unit (300) can input the collected operational data and resistance data into an anomaly detection model based on a long-short-term memory neural network that has learned only normal state data patterns. This model calculates a reconstruction error, which is the difference between the result of restoring the current data to a normal pattern and the actual measured value, and if the reconstruction error exceeds a preset standard, it can be determined that the performance degradation of the component has begun. The control unit (300) can probabilistically predict the remaining useful life through reliability analysis based on the Weibull distribution, which statistically models the life distribution of the parts based on the determined performance degradation trend. If the predicted remaining useful life is shortened to below a critical period or the failure probability exceeds an allowable level, the control unit (300) can generate and send a notification to the manager including the name of the part that needs replacement, the expected time of failure, and the recommended time for maintenance.

[0105] In one embodiment, the control unit (300), in addition to its primary function of protecting the equipment from external lightning strike risks, can ensure the physical reliability of the device by performing a function of self-diagnosing whether the core components of the device itself are maintaining an optimal state and predicting future failures. This aims to fundamentally improve the reliability of the entire protection system by preventing the protection device from malfunctioning at critical moments. To this end, the control unit (300) can continuously monitor the health of the drive motor, which directly performs the physical connection and disconnection of the power supply unit. Specifically, by measuring the precise rotational speed of the motor through a Hall sensor and measuring the drive current according to the load in real time through a current sensor, it can collect and store operational data in a time series, including accumulated operating time, current consumption patterns, and time to reach the target position. This data can serve as basic data for determining performance degradation due to the aging of the motor.

[0106] According to one embodiment, the control unit (300) can precisely monitor the condition of an electrical contact that may deteriorate as the switching operation is repeated. This can be achieved by periodically measuring the contact resistance of a movable contact member while the power is connected. In particular, a four-terminal measurement method can be used to precisely detect minute resistance changes in the micro-ohm (μΩ) range by fundamentally eliminating errors caused by the resistance of the measuring wire itself. By analyzing the trend of the resistance value measured in this way increasing over time, the control unit (300) can quantitatively evaluate the degree of oxidation and erosion of the contact surface caused by the occurrence of an electrical arc.

[0107] According to one embodiment, the control unit (300) may input the motor operation data and contact resistance data collected in this manner into an anomaly detection model based on Long Short-Term Memory (LSTM) that has been pre-learned only the normal state data patterns. This model may undergo a process of compressing the input data and then restoring it to its original form to determine how similar the input current data is to past normal patterns. At this time, the 'reconstruction error,' which is the difference between the original data and the restored data, is calculated, and if this error exceeds a preset standard, it can be determined that performance degradation deviating from the normal pattern has begun.

[0108] According to one embodiment, when a performance degradation trend is detected, the control unit (300) can probabilistically predict the remaining useful life (RUL) of a component. This can be achieved through reliability analysis based on a Weibull distribution suitable for statistically modeling the life distribution of a component, and predicts the likelihood of future failure by considering both the operation history to date and surrounding environmental conditions. If the predicted remaining useful life is shortened to below a preset threshold period or the probability of failure exceeds an acceptable level, the control unit (300) can generate and transmit a detailed notification to the manager, rather than a simple warning, which includes the specific name of the part requiring replacement, the expected time of failure, and the recommended time for maintenance.

[0109] Hall sensors, four-terminal measurement methods, long short-term memory neural networks, or reliability models based on Weibull distributions mentioned in this specification are merely examples for implementing functions, and may be replaced with other types of sensors, measurement methods, anomaly detection algorithms, or life prediction models that produce equivalent technical effects, and the scope of the invention is not limited thereto.

[0110] According to one embodiment, the control unit (300) may further include an analysis function that utilizes environmental acoustic data to complement electromagnetic analysis. To this end, the lightning protection control device (100) may include a microphone array sensitive to a specific frequency band, and the control unit (300) may analyze acoustic signals collected therefrom in real time. The control unit (300) generates a frequency spectrum of the acoustic signal through a Fast Fourier Transform and can identify specific acoustic patterns, such as low-frequency noise of strong winds accompanying thundercloud development, collision sounds of hail or heavy rain, or characteristic high-frequency noise of corona discharge occurring just before lightning discharge. The acoustic features identified in this way are used as input feature vectors for an artificial intelligence model along with existing sensor data, thereby increasing the accuracy of the prediction by considering even physical phenomena that are difficult to capture electromagnetically.

[0111] According to one embodiment, the control unit (300) can perform damage prediction based on virtual simulation based on the predicted trajectory and intensity of lightning strikes to increase the reliability of the protection operation. The control unit (300) may store a three-dimensional digital twin model of the equipment to be protected in memory. When a lightning strike is predicted, the control unit (300) can apply the predicted lightning energy and predicted impact point data to the digital twin model to simulate the electromagnetic shock propagation and potential damage range in the case where no protection measures are taken and in the case where protection measures are taken at each stage. The results of this simulation can be used not only to select the optimal protection mode but also to provide intuitive visual information about the predicted damage to the manager, thereby supporting rapid decision-making.

[0112] The acoustic data, space weather data, or digital twin-based simulations mentioned in this specification are merely examples for implementing the invention, and other types of auxiliary data or analysis techniques that produce equivalent technical effects may be additionally utilized to enhance the prediction accuracy and reliability of the artificial intelligence model, and the scope of the invention is not limited thereto.

[0114] The embodiments described above may be implemented as hardware components, software components, or a combination thereof. For example, the described devices, methods, and components may be implemented using one or more general-purpose or special-purpose computers, such as processors, controllers, arithmetic logic units (ALUs), digital signal processors (DSPs), microcomputers, field programmable gate arrays (FPGAs), programmable logic units (PLUs), microprocessors, etc. The processing unit may execute an operating system (OS) and various software applications running on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to software execution.

[0115] For the sake of understanding, a processing unit is sometimes described as a single unit, but those generally familiar with the relevant technical field will know that a processing unit can include multiple processing elements and / or various types of processing elements. For example, a processing unit may include multiple processors or one processor and one controller, and other processing configurations, such as parallel processors, are also possible.

[0116] Software may include computer programs, code, instructions, or a combination thereof, and may configure a processing unit to perform desired functions or control the processing unit independently or in combination. Software and / or data may be permanently or temporarily implemented on various machines, components, physical devices, virtual devices, computer storage media, or transmission signals to be interpreted by the processing unit or to provide instructions or data. Software may be stored or executed in a distributed manner on computer systems connected to a network, and may be stored on one or more computer-readable recording media.

[0117] Although embodiments have been described through specific drawings, a person skilled in the art can apply various technical modifications and variations based thereon. For example, appropriate results can be obtained even if the described technology is performed in a different order, or if the components of the described system, structure, device, circuit, etc. are combined in a different form, or if they are replaced with other components or equivalents.

Claims

Claim 1 A lightning protection and switching control device through lightning occurrence prediction includes a lightning prediction sensor comprising an electric field sensor that detects electrical changes in the atmosphere, a magnetic field sensor that detects changes in the magnetic field in the atmosphere, a weather sensor that detects weather conditions in real time, and a radar sensor for detecting the charge distribution of clouds in the atmosphere; and a control unit electrically connected to the lightning prediction sensor; wherein the control unit receives time-series data collected from the electric field sensor, the magnetic field sensor, the radar sensor, and the weather sensor in real time to generate a multidimensional feature vector, and recognizes a lightning occurrence pattern from the multidimensional feature vector using a pre-trained artificial intelligence model, wherein the artificial intelligence model includes a deep neural network that has learned the correlation between sensor data patterns prior to the time of past lightning occurrence and actual lightning occurrence; wherein the control unit generates a frequency spectrum by performing a Fast Fourier Transform on electromagnetic field data collected from the electric field sensor and the magnetic field sensor, calculates the power spectral density of each component in a predetermined frequency range from low frequency to high frequency to track changes in energy distribution over time, constructs a polyhedron of minimum volume that encloses an area where the charge amount exceeds a preset charge amount threshold using 3D cloud charge distribution data collected from the radar sensor, calculates the center point thereof to derive an approach velocity vector of the area where the charge amount exceeds the preset charge amount threshold, and based on weather data collected from the weather sensor, a critical electric field required for the breakdown of the insulation properties of air Calculate the intensity in real time and define the lightning critical proximity by comparing it with the actual measured electric field strength, and calculate the membership value for each segment by substituting a plurality of numerical input values, including the energy growth rate in the power spectrum density, the approach speed of the charge-dense region, and the lightning critical proximity, into a membership function composed of a plurality of predefined segments.A lightning protection and switching control device through lightning occurrence prediction, which infers an overall risk level by applying a rule set configured to output a risk grade as a conclusion based on a combination of membership values ​​of each input variable conditionally, generates a final lightning risk score expressed as a single quantitative number by weighting the result, predicts the timing and intensity of lightning occurrence when the lightning occurrence probability output by the artificial intelligence model exceeds a dynamically adjusted threshold probability, determines a differentiated protection level according to the predicted lightning intensity to adaptively control the power cut-off range and cut-off time, and controls sensitivity by lowering the threshold probability to react sensitively when the hourly increase rate of the final lightning risk score is high, and raising the threshold probability to prevent overprotection when the final lightning risk score is on a decreasing trend. Claim 2 delete Claim 3 In claim 1, the control unit performs a process of calculating the electric field change acceleration by differentiating the electric field measurement value collected from the electric field sensor with respect to the time axis to calculate the rate of change, which is the first derivative, and the acceleration, which is the second derivative; performs a process of calculating the electric field change acceleration by calculating the time derivative of the electric field change rate collected from the electric field sensor and the spatial gradient, wherein the electric field measurement value is differentiated with respect to the time axis to calculate the rate of change, which is the first derivative, and the acceleration, which is the second derivative; performs a process of estimating the atmospheric current density distribution by analyzing the rotational component of the magnetic field vector collected from the magnetic field sensor, wherein the current density is derived from the rotational value of the magnetic field by applying Ampere's law of Maxwell's equations; and performs a process of generating 3D spatial mapping data of the charge distribution within the cloud collected from the radar sensor to track the movement path and velocity of the charge center, wherein the control unit analyzes the intensity and phase change of the radar reflection signal to identify the area where the charge amount exceeds the preset charge amount threshold and tracks the position change of the charge center in a continuous time frame, and the A process is performed to calculate an atmospheric instability index by collecting temperature, humidity, and pressure data at different altitudes from weather sensors; the collected temperature and humidity profiles are integrated to calculate the Convective Available Potential Energy (CAPE), which represents the potential for vertical uplift due to energy imbalance between the upper and lower layers of the atmosphere; mechanical instability is evaluated by calculating the Richardson number, which is the ratio between atmospheric stability and shear stress, using the temperature gradient and wind speed difference at different altitudes; a predefined weight is applied to ensure a higher value when the calculated Convective Available Potential Energy value is higher, and a predefined weight is applied to ensure a higher value when the calculated Richardson number is lower, thereby quantifying the final atmospheric instability index.A lightning imminent warning is issued when a composite condition is satisfied in which the electric field change acceleration exceeds a first threshold, the current density exceeds a second threshold, the charge center approaches within a set distance from the protected equipment, and the atmospheric instability index exceeds a third threshold; a three-stage protection level including pre-warning, warning, and danger is established based on the degree to which the composite condition is satisfied, and phased protection actions are performed in which power to pre-set sensitive equipment is cut off at the pre-warning stage, the main power grid is disconnected at the warning stage, and the entire system is connected to the ground at the danger stage; for each protection level, the timing for preemptive cutoff is determined based on the estimated time of lightning arrival predicted by the AI ​​model; power is restored in reverse order when the AI ​​model recognizes a pattern in which sensor measurements return to the normal range after the lightning ends; prediction accuracy is evaluated by comparing the actual occurrence of lightning with the AI ​​model's prediction results, and if an error occurs, the sensor data pattern at that time is saved as additional training data to retrain the AI ​​model online; and to determine the soundness of the drive motor, the motor's rotational speed is measured using a Hall sensor and the drive current is measured in real-time using a current sensor for cumulative operation Continuously collecting operational data including time, current consumption patterns, and time to reach a target position; measuring contact resistance using a four-terminal measurement method capable of precisely detecting minute resistance changes by excluding the resistance influence of the measurement wire to monitor the deterioration status of electrical contacts; evaluating the degree of erosion on the contact surface by analyzing the temporal increasing trend of the measured resistance values; inputting the collected operational data and resistance data into an anomaly detection model based on a Long Short-Term Memory neural network that has learned only normal state data patterns; and calculating the reconstruction error, which is the difference between the result of the anomaly detection model restoring the current data to a normal pattern and the actual measured value,A lightning protection and switching control device through lightning occurrence prediction, which determines that performance degradation of a component has begun if the above-mentioned reconstruction error exceeds a preset standard, probabilistically predicts the remaining useful life through reliability analysis based on a Weibull distribution that statistically models the life distribution of the component based on the determined performance degradation trend, and generates and transmits an alert to an administrator including the name of the component requiring replacement, the expected time of failure, and the recommended time for maintenance if the predicted remaining useful life is shortened below a critical period or the failure probability exceeds an allowable level.

Images