Integrated Control System and Method of Multi-Site Power Distribution Panels Using AI-Based Risk Assessment

Abstract

The present invention relates to a multi-point integrated control system and method for power distribution panels using AI-based risk assessment. More specifically, it relates to a system and method that integrally controls distribution panels installed in multiple buildings or locations using cloud-based artificial intelligence (AI) technology, analyzes the status of each distribution panel with a quantified risk level, and automatically selects and executes an optimal response policy. Specifically, the invention relates to a technology that secures the stability and efficiency of power infrastructure by predicting and scoring the risk of distribution panel data through an AI-based analysis module, and dynamically executing control policies for each location based on the score. According to the present invention, status data of switchgear located at each point is collected in an integrated manner on a cloud server, and by converting this into a risk score using an AI-based parallel analysis module, an active and quantitative judgment system becomes possible, moving beyond conventional passive status monitoring. In particular, a time-series prediction module determines the possibility of load increase after a specific point in time, an event detection module identifies fault-like patterns, and a correlation analysis module determines whether there is an abnormal link between equipment; these results can then be integrated into a single quantitative indicator called a 'Risk Score'. Through this, different judgment results can be compared and weighted, and the most appropriate control policy for the corresponding switchgear can be selected.

Description

Technology Field

[0001] The present invention relates to a multi-point integrated control system and method for power distribution panels using AI-based risk assessment. More specifically, it relates to a system and method that integrally controls distribution panels installed in multiple buildings or locations using cloud-based artificial intelligence (AI) technology, analyzes the status of each distribution panel with a quantified risk level, and automatically selects and executes an optimal response policy. Specifically, the invention relates to a technology that secures the stability and efficiency of power infrastructure by predicting and scoring the risk of distribution panel data through an AI-based analysis module, and dynamically executing control policies for each location based on the score. Background Technology

[0002] Most existing switchgear management technologies are structured to monitor the electrical status of a single point or provide simple alarms.

[0003] For example, Registered Patent No. 10-1476660 determines the possibility of fire by detecting internal temperature, leakage current, smoke, etc., through multiple sensors, but this is a diagnostic function limited to individual distribution panels and does not provide linked control or integrated judgment functions between branches.

[0004] Public Patent No. 10-2018-0003857 also determines whether there is an abnormality in a switchgear based on temperature and voltage and performs simple control functions, but centralized integrated control was not considered.

[0005] Recently registered patent No. 10-2624178 proposes an IoT-based distribution panel control device and implements an automatic shut-off function based on anomaly detection between the earth leakage circuit breaker and the main circuit breaker; however, this also operates at the individual device level and does not incorporate multi-point cooperative control or an AI-based predictive analysis system.

[0006] As such, most conventional technologies are limited to determining the physical status of a single switchgear or outputting alarms, and there is a lack of a prediction-based central decision system capable of actively controlling by integrally considering the situations at multiple points. Prior art literature

[0007] (Patent Document 0001) KR 10-1476660 B1(Patent Document 0002) KR 10-2018-0003857 A(Patent Document 0003) KR 10-2624178 B1 The problem to be solved

[0008] The present invention was devised to solve such problems, and aims to solve the following tasks in order to overcome the limitations of the aforementioned conventional technology and achieve advanced management of power facilities.

[0009] First, the purpose is to enable simultaneous monitoring and response to the entire power infrastructure, rather than individual locations, by integrally controlling switchgear distributed across multiple buildings or sites through a cloud-based central control server.

[0010] Second, another objective is to analyze the status of switchgear not merely based on whether thresholds are exceeded, but through multi-layered information such as time-series forecasting, abnormal event detection, and correlation analysis between equipment, and to integrate this into a quantified Risk Score to guide control policies.

[0011] Third, another objective is to achieve predictive response to abnormal situations and resource optimization while maintaining power quality by comparing and judging the risk level of each switchgear or branch in real time based on these risk scores, and automatically selecting and executing policies according to predefined conditions. means of solving the problem

[0012] To achieve the above objective, the multi-point switchgear integrated control system using AI-based risk judgment according to the present invention comprises: a data collection and preprocessing unit that collects status data including time-series load usage data and equipment event data from sensors installed in each switchgear and performs preprocessing for input into an AI model; a prediction analysis unit that generates load prediction data including a predicted load rate at a future point in time based on the time-series load usage data of each switchgear; an event detection unit that analyzes the equipment event data of each switchgear in real time to generate fault detection data including fault types and event anomaly probabilities; and a correlation analysis unit that stores a normal operation correlation matrix in which mutual operation patterns are defined during normal operation between equipment within the switchgear, compares a real-time correlation coefficient calculated from equipment-specific status information collected in real time among the status data with a reference correlation coefficient extracted from the normal operation correlation matrix, and outputs data analyzing anomalies in inter-equipment relationships including a correlation mismatch index based on the difference between the real-time correlation coefficient and the reference correlation coefficient. A risk scoring unit that calculates an integrated risk score by applying a condition-based weighted amplification method, wherein the predicted load rate included in the load prediction data, the event anomaly probability included in the fault detection data, and the correlation mismatch index are normalized and summed by applying a preset weight to each, and if any one of the predicted load rate, the event anomaly probability, and the correlation mismatch index exceeds a preset threshold for weight amplification, the weight applied to the corresponding item is increased;A control policy determination unit that, when the integrated risk score for each of the above switchgear exceeds a preset threshold for multi-point cooperative control, checks the integrated risk score of an adjacent switchgear connected to the network and the remaining capacity information of an energy storage device (ESS) equipped in the adjacent switchgear, and when the integrated risk score of the adjacent switchgear is less than the preset threshold for multi-point cooperative control and the remaining capacity of the ESS is greater than or equal to a preset standard, generates a multi-point cooperative control signal to control the supply of power from the ESS of the adjacent switchgear to the corresponding switchgear; and a control command transmission unit that transmits a control command including the multi-point cooperative control signal to the corresponding switchgear and the adjacent switchgear.

[0013] The multi-point switchgear integrated control system using the above AI-based risk judgment may further include a monitoring information providing unit that provides data of each control step performed by the switchgear integrated control system to the manager terminal.

[0014] The time-series load usage data that is the input to the prediction analysis unit may include one or more of chronological load rate data, temperature, and external events, and the load prediction data that is the output of the prediction analysis unit may include one or more of a predicted load rate, a predicted temperature change rate, a load increase trend, and a prediction reliability.

[0015] The equipment event data, which is the input of the above event detection unit, may include one or more of instantaneous current, leakage current, internal temperature rise rate, and temperature sensor distribution imbalance, and the fault detection data, which is the output of the above event detection unit, may include one or more of fault type classification, similarity score, detection grade, event anomaly probability, and event response request flag.

[0016] The inter-equipment correlation and individual equipment real-time data, which are inputs to the correlation analysis unit, include one or more of a normal operation correlation matrix, real-time current amount time series data, real-time power consumption, real-time operation rate / ON-OFF event log, and time reference information, and the inter-equipment relationship anomaly analysis data, which is the output of the correlation analysis unit, may include one or more of a current correlation coefficient, reference correlation coefficient, correlation mismatch index, relationship anomaly detection flag, list of major anomaly equipment pairs, and relationship anomaly reliability.

[0018] According to another aspect of the present invention, a multi-point switchgear integrated control system using AI-based risk judgment and a method for performing multi-point switchgear integrated control using AI-based risk judgment comprises: (a) a step in which a data collection and preprocessing unit collects state data including time-series load usage data and equipment event data from sensors installed in each switchgear and performs preprocessing for input into an AI model; (b) a step in which a prediction analysis unit generates load prediction data including a predicted load rate at a future point in time based on the time-series load usage data of each switchgear; and (c) a step in which an event detection unit analyzes the equipment event data of each switchgear in real time and generates fault detection data including a fault type and an event anomaly probability. (d) A correlation analysis unit stores a normal operation correlation matrix in which mutual operation patterns are defined during normal operation between the equipment within the switchgear, and compares a real-time correlation coefficient calculated from equipment-specific status information collected in real-time among the status data with a reference correlation coefficient extracted from the normal operation correlation matrix, and outputs data for analyzing abnormal relationships between equipment including a correlation mismatch index based on the difference between the real-time correlation coefficient and the reference correlation coefficient; (e) A risk scoring unit calculates an integrated risk score by applying a condition-based weighted amplification method to normalize the predicted load rate, the event abnormality probability, and the correlation mismatch index, and summing them by applying a preset weight to each, wherein if any one of the predicted load rate, the event abnormality probability, and the correlation mismatch index exceeds a preset threshold for weight amplification, the weight applied to the corresponding item is increased;(f) a control policy determination unit, when the integrated risk score for each switchgear exceeds a preset threshold for multi-point cooperative control, checks the integrated risk score of an adjacent switchgear connected to the network and the remaining capacity information of an energy storage device (ESS) equipped in the adjacent switchgear, and when the integrated risk score of the adjacent switchgear is less than the preset threshold for multi-point cooperative control and the remaining capacity of the ESS is greater than or equal to a preset standard, generates a multi-point cooperative control signal to control the supply of power from the ESS of the adjacent switchgear to the switchgear; and (g) a control command transmission unit, transmits a control command including the multi-point cooperative control signal to the switchgear and the adjacent switchgear.;

[0019] The time-series load usage data that is the input to the prediction analysis unit may include one or more of chronological load rate data, temperature, and external events, and the load prediction data that is the output of the prediction analysis unit may include one or more of a predicted load rate, a predicted temperature change rate, a load increase trend, and a prediction reliability.

[0020] The equipment event data, which is the input of the above event detection unit, may include one or more of instantaneous current, leakage current, internal temperature rise rate, and temperature sensor distribution imbalance, and the fault detection data, which is the output of the above event detection unit, may include one or more of fault type classification, similarity score, detection grade, event anomaly probability, and event response request flag.

[0021] The inter-equipment correlation and individual equipment real-time data, which are inputs to the correlation analysis unit, include one or more of a normal operation correlation matrix, real-time current amount time series data, real-time power consumption, real-time operation rate / ON-OFF event log, and time reference information, and the inter-equipment relationship anomaly analysis data, which is the output of the correlation analysis unit, may include one or more of a current correlation coefficient, reference correlation coefficient, correlation mismatch index, relationship anomaly detection flag, list of major anomaly equipment pairs, and relationship anomaly reliability.

[0023] According to another aspect of the present invention, a computer program stored in a computer-readable non-transient storage medium for performing integrated control of multi-point switchgear using AI-based risk judgment is stored in the non-transient storage medium and, by a processor, comprises the steps of: (a) collecting status data including time-series load usage data and equipment event data from sensors installed in each switchgear and performing preprocessing for input into an AI model; (b) generating load prediction data including a predicted load rate at a future point in time based on the time-series load usage data of each switchgear; and (c) analyzing the equipment event data of each switchgear in real time to generate fault detection data including a fault type and an event anomaly probability. (d) storing a normal operation correlation matrix in which mutual operation patterns are defined during normal operation among the equipment within the switchgear, and comparing a real-time correlation coefficient calculated from equipment-specific status information collected in real-time among the status data with a reference correlation coefficient extracted from the normal operation correlation matrix, and outputting data for analyzing abnormal relationships between equipment including a correlation mismatch index based on the difference between the real-time correlation coefficient and the reference correlation coefficient; (e) normalizing the predicted load rate, the event abnormality probability, and the correlation mismatch index, applying a preset weight to each, summing them, and applying a condition-based weighted amplification method in which the weight applied to the item is increased if any one of the predicted load rate, the event abnormality probability, and the correlation mismatch index exceeds a preset threshold for weight amplification, thereby calculating an integrated risk score;(f) a step of generating a multi-point cooperative control signal to control the supply of power from the ESS of the adjacent switchgear to the switchgear when the integrated risk score for each switchgear exceeds a preset threshold for multi-point cooperative control, checking the integrated risk score of an adjacent switchgear connected to the network and the remaining capacity information of the energy storage device (ESS) equipped in the adjacent switchgear, and when the integrated risk score of the adjacent switchgear is less than the preset threshold for multi-point cooperative control and the remaining capacity of the ESS is greater than or equal to a preset standard; and (g) a command to execute the step of transmitting a control command including the multi-point cooperative control signal to the switchgear and the adjacent switchgear.; Effects of the invention

[0024] According to the present invention, by integrating and collecting status data of switchgear located at each point into a cloud server and converting it into a risk score by an artificial intelligence-based parallel analysis module, an active and quantitative judgment system becomes possible, going beyond conventional passive status monitoring.

[0025] In particular, the time series prediction module determines the likelihood of load increase after a specific point in time, the event detection module identifies fault-like patterns, and the correlation analysis module determines the correlation between equipment, allowing these results to be integrated into a single quantitative indicator called the 'Risk Score'. Through this process, different judgment results can be compared and weighted to select the most appropriate control policy for the corresponding switchgear.

[0026] This integrated structure goes beyond a simple parallel decision-making system to enable a mutual feedback-type integrated decision-making system that organically monitors the status of the entire power grid from a central server and allows high-risk situations at specific points to be directly reflected in the status and control decisions of adjacent points.

[0027] Furthermore, the risk score-based policy branching method overcomes the limitations of existing manual control systems and realizes policy automation and multi-point synchronous control, thereby enabling various industrial benefits such as efficient energy use, prevention of power failures, and reduced repair response times.

[0028] Therefore, going beyond simple power monitoring technology, the present invention has high industrial utility value as a next-generation smart power infrastructure system capable of intelligent power control and predictive management. Brief explanation of the drawing

[0029] FIG. 1 is a diagram showing a network configuration for performing multi-point switchgear integrated control using AI-based risk judgment of the present invention. FIG. 2 is a diagram showing the configuration of a multi-point switchgear integrated control system using AI-based risk judgment according to the present invention. FIG. 3 is a flowchart for performing integrated control of multi-point switchgear using AI-based risk judgment of the present invention. FIG. 4 is a diagram illustrating an example of the input and output of a predictive analysis AI model operating in a predictive analysis unit. FIG. 5 is a diagram illustrating an example of the input and output of an event detection AI model operating in an event detection unit. FIG. 6 is a diagram illustrating an example of the input and output of a correlation analysis AI model operating in a correlation analysis unit. Specific details for implementing the invention

[0030] Preferred embodiments of the present invention will be described in detail below with reference to the attached drawings. Prior to this, terms and words used in this specification and claims should not be interpreted as being limited to their ordinary or dictionary meanings, but should be interpreted in a meaning and concept consistent with the technical spirit of the present invention, based on the principle that the inventor can appropriately define the concept of the terms to best describe his invention. Accordingly, the embodiments described in this specification and the configurations illustrated in the drawings are merely one preferred embodiment of the present invention and do not represent all aspects of the technical spirit of the present invention; therefore, it should be understood that various equivalents and modifications capable of replacing them may exist at the time of filing this application.

[0032] FIG. 1 is a diagram showing a network configuration for performing multi-point switchgear integrated control using AI-based risk judgment of the present invention.

[0033] The network for performing multi-point power distribution panel integrated control using AI-based risk judgment of the present invention is configured such that power distribution panels (PD: Power Distribution panel) (100) installed at multiple buildings or multiple points of a single building (hereinafter collectively referred to as 'multi-point') are interconnected through a communication network with a central integrated control server, the 'multi-point power distribution panel integrated control system using AI-based risk judgment' (hereinafter referred to as the 'power distribution panel integrated control system') (1000).

[0034] Each power distribution panel (100.1, ..., 100.n) is installed in a building unit or an independent control section within a factory, which is a power demand location, and includes a load circuit, a circuit breaker, a sensor module, a control interface, etc. These measure the power status at each point, collect information such as temperature, current, load amount, and abnormal events locally, and then transmit it in real time to the power distribution panel integrated control system (1000).

[0035] The switchgear integrated control system (1000) collects heterogeneous data received from a plurality of switchgear (100) in this manner, evaluates the status using an AI-based analysis module, establishes an optimized control policy for each point or all points, and controls the switchgear of each point accordingly.

[0036] In particular, the switchgear integrated control system (1000) of the present invention is configured to go beyond simple monitoring to enable state correlation between points, dynamic estimation of expected load rate, event-based anomaly evaluation, and establishment of multi-point cooperative control policies, and consequently enables predictive control of the entire power infrastructure.

[0037] For example, if the predicted load value at point A increases rapidly and an abnormal event occurs simultaneously at adjacent point B, the switchgear integrated control system (1000) can evaluate the risk based on the past correlation between the two points and, if necessary, order the automatic shutdown of non-essential loads at point B or the pre-charging of the battery at point A.

[0038] As such, FIG. 1 clearly demonstrates the overall structural concept of the present invention as a multi-point cooperative control system capable of integrated situational awareness and intelligent control execution, going beyond simple single-point response.

[0039] Hereinafter, with reference to FIGS. 2 to 6, the configuration of the switchgear integrated control system (1000) of the present invention and the method by which the switchgear integrated control system (1000) performs multi-point switchgear integrated control using the AI-based risk judgment of the present invention will be described in detail.

[0041] FIG. 2 is a diagram showing the configuration of a multi-point switchgear integrated control system (1000) using AI-based risk judgment of the present invention, FIG. 3 is a flowchart showing the multi-point switchgear integrated control system (1000) of the present invention performing AI-based risk judgment of the multi-point switchgear integrated control, FIG. 4 is a diagram showing an example of the input and output of a prediction analysis AI model operating in a prediction analysis unit (1200), FIG. 5 is a diagram showing an example of the input and output of an event detection AI model operating in an event detection unit (1300), and FIG. 6 is a diagram showing an example of the input and output of a correlation analysis AI model operating in a correlation analysis unit (1400).

[0042] The data collection / preprocessing unit (1100) of the switchgear integrated control system (1000) collects time-series measurements in real time, for example, in units of 1 to 5 seconds, from sensors (temperature sensor, current sensor, voltmeter, leakage current sensor, etc.) installed in each switchgear (S301). This data is used as initial data for detecting risks in each switchgear, and the collected data is stored in the data storage unit (1800). Preprocessing operations are performed on the collected data to input it into each AI model described later, such as noise removal, outlier filtering, and normalization (S302). Here, the AI ​​model refers to the prediction analysis AI model provided in the prediction analysis unit (1200), the event detection AI model provided in the event detection unit (1300), and the correlation analysis AI model provided in the correlation analysis unit (1400).

[0043] Examples of collected data include current: [103.2A, 104.8A, 106.7A…], temperature: [27.3℃, 27.5℃, 28.0℃…], leakage current: [2.3mA, 2.5mA, 2.9mA…], log data of the time of failure, power system topology information (connection information between facilities), etc.

[0044] Thus, data from multiple switchboards (100) is time-synchronized and integrated into a switchboard integrated control system (1000), and the received data is supplied to AI models such as a prediction analysis unit (1200), an event detection unit (1300), and a correlation analysis unit (1400) through a preprocessing operation.

[0045] The data collection / preprocessing unit (1100) supports both the training of the AI ​​model and current inference using the AI ​​model by simultaneously collecting historical data and real-time data.

[0047] The prediction analysis unit (1200) performs the function of outputting load prediction data based on power usage data that changes over time, i.e., time-series load usage data, for each branch distribution panel (100) (S303). The load prediction data includes, for example, a predicted load rate, a predicted temperature change rate, information indicating a load increase trend, and information indicating the reliability of the prediction.

[0048] The predicted load rate is data representing, for example, the power load rate at a certain point in the future. This is not a simple average prediction, but is based on a deep learning-based load prediction algorithm that reflects time series characteristics and abnormal patterns.

[0049] Responding to abnormalities in power systems is not sufficient with 'real-time monitoring' alone. Many accidents occur when an overload or instability has already been gradually progressing for minutes or tens of minutes. Therefore, response based on prior prediction is critically important in protecting power facilities rather than response after the fact. The prediction analysis unit (1200) serves as the core axis for such future-oriented judgment. Hereinafter, 'facility' refers to the electrical equipment connected to each switchboard (100.n).

[0050] To this end, the prediction analysis unit (1200) receives time series load usage data.

[0051] In other words, for example, it performs the function of predicting the power load rate at a certain point in the future based on power usage data at each point that changes over time. This is not a simple average prediction, but a prediction based on a load prediction algorithm using a deep learning-based AI model that reflects time series characteristics and abnormal patterns.

[0052] In one embodiment, the prediction analysis unit (1200) can use a neural network structure specialized for time series, such as LSTM (Long Short-Term Memory) or Transformer, to receive inputs such as minute-by-minute load data for the past 3 days to 2 weeks, time zone, day of the week, temperature, humidity, work schedule, and external related variables, as well as history of load surges before and after past failures / events, and calculate the load rate at an expected time (e.g., 30 minutes later) in % units.

[0053] For example, if a pattern of the load surging on average immediately after lunchtime repeats in circuit 1 of building A over the past two days, the prediction analysis unit can predict the load rate to be 93% after 30 minutes, which is determined to exceed a set 'boundary threshold (e.g., 90%)'. This value functions as a signal to induce measures such as preemptive shutdown or battery charging.

[0054] The prediction analysis unit (1200) performs the role of detecting, in particular, future high-load situations or abnormal surge conditions in advance, and enabling a response based on this. This enables preemptive risk mitigation measures for power system operation and provides basic information that allows for the preparatory operation of UPS or ESS.

[0055] FIG. 4 illustrates an example of the input and output of a prediction analysis AI model of a prediction analysis unit (1200). The input can be collectively referred to as time-series load usage data, and the output as load prediction data, respectively. Further details regarding such input and output data are shown below. Furthermore, the examples of the input and output are not limited to the items exemplified in FIG. 4 or below, and it is understood that various data included in the technical concept presented in the overall description of the present invention may also be included.

[0056] - Input (Time series load-enabled data):

[0057] Hourly Load Rate Data: [76%, 78%, 80%, 85%, 89%, … ] (5-minute intervals, past 48 hours)

[0058] Temperature: [21.2℃, 21.4℃, 21.8℃, … ]

[0059] Day / Time / Shift Schedule (Qualitative variables are one-hot encoded)

[0060] External Events: Equipment Replacement Schedule, Process Change Time

[0061] - Output (Load prediction data):

[0062] Predicted load rate (after 30 minutes): 92.4%

[0063] Predicted rate of temperature change (after 1 hour): +3.1℃

[0064] Load increase tendency: Rapid rise (Classification type output)

[0065] Prediction confidence: 0.89 (softmax-based confidence)

[0067] This output is used as a prediction-based pre-warning coefficient in the subsequent risk scoring unit (1500).

[0069] The event detection unit (1300) receives equipment event data for each switchboard (100) and outputs fault detection data that quantitatively evaluates how similar the current sensor signal is to patterns that appeared during past faults (S304). The key is how dangerous the current signal indicates a dangerous state.

[0070] Not all risks occurring in the switchboard (100) appear in a form that gradually accumulates. Electrical equipment failures, sparks, leakage currents, abnormal currents, etc., often occur as sudden and rapidly changing events within a short period. This detection requires immediacy and precision rather than prediction, and is made possible through learning patterns of similar accidents in the past.

[0071] The event detection unit (1300) is implemented as an AI classification model for detecting abnormal events, and analyzes how similar various sensor values ​​currently measured in the switchgear are to past failure cases, and outputs a classification result of failure similarity or risk events.

[0072] The event detection unit (1300) can use XGBoost, Random Forest, or CNN-based sequence-classification models, which learn pattern features immediately before a failure to perform similarity evaluation and classification decisions for new inputs, and in particular, immediately determine sudden abnormal phenomena such as arcs, sparks, rapid temperature rises, and short circuits, and provide a function to realign the urgency and priority of risk judgments.

[0073] For example, if the internal temperature of the UPS equipment at point B rises by 12 degrees in 3 minutes, and at the same time the leakage current surges, and shows a high similarity to a fault precursor signal that occurred in the same UPS unit in the past, the event detection unit calculates a similarity of 0.86. This value becomes a condition that induces 'immediate shutdown'.

[0074] FIG. 5 illustrates an example of the input and output of an event detection AI model of an event detection unit (1300). The input can be collectively referred to as equipment event data, and the output as fault detection data, respectively; further detailed data regarding such input and output is shown below. Furthermore, the examples of the input and output are not limited to the items exemplified in FIG. 5 or below, and it is understood that various data included in the technical concept presented in the overall description of the present invention may also be included.

[0075] - Input (Facility Event Data):

[0076] Instant current: 123A → 142A (15% increase within 3 seconds)

[0077] Leakage current: 3.8mA → 8.9mA

[0078] Internal temperature rise rate: +5.5℃ / 2 minutes

[0079] Camera-based busbar discoloration rate: 17%

[0080] Temperature sensor distribution imbalance: Temperature deviation of the top 10% range = 4.8℃

[0081] - Output :(Fault detection data)

[0082] Fault Type Classification: ["UPS Overheating", "Wiring Short Circuit", "Leakage", "Normal"]

[0083] Similarity score: 0.87

[0084] Detection Level: High-Risk (class 3 of 4)

[0085] Event Anomaly Probability: 0.8

[0086] Event response request flag: True

[0088] This output serves as a key input for determining whether immediate control intervention is required in a subsequent step, and receives a high weight in the risk scoring calculation of the risk scoring unit (1500).

[0090] The correlation analysis unit (1400) receives real-time data on the correlation between equipment and individual equipment for a plurality of equipment connected to the respective switchboard (100.n) from each switchboard (100), and outputs data on the relationship between equipment abnormalities that determines the possibility of structural abnormalities by analyzing the consistency of the operation patterns between the equipment (S305).

[0091] The correlation analysis unit (1400) is a system consistency evaluation module based on the interrelationship between equipment. That is, the correlation analysis unit (1400) does not evaluate the state of individual elements such as a single piece of equipment failure or a sensor malfunction, but rather performs the role of quantitatively detecting relationship-based abnormal signs by comparing them with the way equipment connected to each switchboard (100.n) typically operates together. This is a device for preemptively identifying structural synchronization errors or out-of-control, which cannot be captured by time-series prediction or event-based abnormal detection modules.

[0092] The correlation analysis unit (1400) is an AI-based judgment module that learns the normal operation correlation between multiple facilities connected to the switchgear (100.n), receives real-time operation data as input, calculates the deviation from the reference correlation matrix, and performs the function of outputting whether there is a structural abnormality based on the result. The AI ​​judgment module can be implemented as a GNN-based structural relationship learning model, an Autoencoder-based abnormality restoration error evaluation model, or other time series linkage models.

[0093] To this end, the correlation analysis unit (1400) uses two main types of inputs.

[0094] The first is the Normal Correlation Matrix, generated based on the operational correlations observed between each piece of equipment in a past steady state, which is reference data that quantifies the standard relationships of interactions between each piece of equipment within the switchgear.

[0095] The second is Real-time Operation Data, which includes time-series operation data such as real-time power consumption, current consumption, and operating rate of each facility collected at the current time.

[0096] The correlation analysis unit (1400) compares the reference correlation coefficient extracted from the normal operation correlation matrix with the real-time correlation coefficient calculated from the real-time operation data to calculate the correlation mismatch index (Δr) for each pair of equipment, and if the deviation exceeds a preset threshold, activates a relationship anomaly detection flag.

[0097] In this case, if there are n facilities, the correlation analysis unit is n(n-1) / 2(= n C2) Performs a diagnosis for each pair of equipment pairs, and outputs the reference correlation coefficient (r_normal), real-time correlation coefficient (r_now), correlation mismatch index (Δr), and relationship anomaly detection flag (AnomalyFlag) for each equipment pair.

[0098] The AI ​​model of the correlation analysis unit (1400) can typically be implemented as a multivariate statistical model based on Pearson correlation coefficients, or a Graph Neural Network (GNN) capable of learning relationship structures, or an attention-based anomaly detection model.

[0099] In one embodiment, three pieces of equipment—a cooling unit (A), lighting (B), and UPS (C)—are connected to a power distribution panel. In the past, when normal conditions were met, the cooling unit and lighting mostly operated simultaneously and the correlation coefficient was maintained at 0.85 or higher. However, if the real-time correlation coefficient of these two pieces of equipment drops sharply to 0.30 or lower at the current time, Δr = 0.55, and a relationship anomaly detection flag is activated. At this time, even if the relationship with the UPS is normal, if there is an anomaly in the relationship of a single pair of devices (AB), it is determined that there is a problem with the structural integrity of the entire power distribution panel, and this is reflected as a structural risk factor in the risk scoring module of the present invention.

[0100] These judgment results are integrated with results from other diagnostic modules, such as predicted load rate and event anomaly probability, in the risk scoring module to construct a risk score. This score is then used as input to the policy selection module, leading to specific control policy decisions for the corresponding switchgear, such as tripping, load balancing, and backup circuit switching.

[0101] In this way, the correlation analysis unit (1400) can identify potential risks such as synchronization failure between equipment and systemic malfunction in advance without relying on simple fault detection through anomaly judgment based on the relationship consistency of the internal structure of the switchboard, and realizes a relationship-based consistency diagnosis function that was not considered in the prior art.

[0102] FIG. 6 illustrates an example of the input and output of the correlation analysis AI model of the correlation analysis unit (1300). The input can be collectively referred to as correlation between facilities and real-time data of individual facilities, and the output as data on anomalies in relationships between facilities, respectively. Further details regarding such input and output data are shown below. Furthermore, the examples of input and output are not limited to the items exemplified in FIG. 6 or below, and it is understood that various data included in the technical concept presented in the overall description of the present invention may also be included.

[0103] - Input (Inter-facility correlation and individual facility real-time data):

[0104] 1) Normal Correlation Matrix

[0105] As a reference matrix for the operational correlation coefficients between equipment learned in the past steady state, for example, it is as follows.

[0106] A (air conditioner) B (Lighting) C(UPS) A 1.00 0.85 0.65 B 0.85 1.00 0.58 C 0.65 0.58 1.00

[0108] 2) Real-time current time series data

[0109] Current usage for each piece of equipment (e.g., average current per minute), e.g., A: [145, 143, 146], B: [10, 12, 9] (Unit: A)

[0110] 3) Real-time power consumption (P)

[0111] Active power consumption (W or kW), for example, A: 8.5 kW, B: 0.6 kW

[0112] 4) Real-time Operation Rate / ON-OFF Event Log

[0113] As equipment operating status (Binary or Ratio), for example, A: ON, B: ON

[0114] 5) Time-based information :

[0115] As the standard time interval for correlation calculation, for example, based on the average value of the last 5-minute interval

[0117] - Output (Data on anomaly analysis of inter-equipment relationships):

[0118] 1) Current correlation coefficient

[0119] As a correlation coefficient between equipment pairs based on real-time measurements, for example AB: 0.32, AC: 0.62

[0120] 2) Standard correlation coefficient

[0121] As a reference value extracted from the normal correlation matrix, for example, AB: 0.85

[0122] 3) Correlation discrepancy index (Δr)

[0123] It is the difference between the current correlation coefficient and the reference correlation coefficient.

[0124] 4) Relationship Anomaly Detection Flag

[0125] Set to True if Δr exceeds the threshold (e.g., 0.4).

[0126] 5) List of major abnormal equipment pairs

[0127] As a pair of facilities with detected relationship anomalies, for example (A,B)

[0128] 6) Relationship reliability

[0129] As the output of a GNN-based model, for example, AB: 0.91

[0131] For example, above, AB usually works together with r=0.85.

[0132] Currently A is ON and B is OFF → r = 0.32 → Δr = 0.53 → Threshold exceeding 0.4 → Relationship anomaly detection flag activated: This result is reflected in the risk score of the switchgear unit and contributes to policy decisions.

[0134] The risk scoring unit (1500) integrates the results of analyzing the condition of multiple facilities connected to each switchboard (100.n) from various diagnostic perspectives (S303 to S305) into a single quantitative indicator to calculate data that quantifies the power operation risk for each switchboard (hereinafter referred to as the 'Risk Score') (S306).

[0135] The risk scoring unit (1500) is differentiated from existing technology in that it deviates from the existing simple conditional branching judgment structure and converts multidimensional diagnostic results into a single axis, thereby simultaneously securing the conciseness of the policy selection logic and the quantitativeness of the policy priority judgment.

[0136] The risk scoring unit (1500) receives as input independent diagnostic results transmitted from the prediction analysis unit (1200), the event anomaly detection unit (1300), the correlation analysis unit (1400), etc. For example, specifically, normalized judgment values ​​such as the expected load rate based on time series prediction, the anomaly probability based on real-time event detection, and the correlation mismatch index based on the results of the alignment analysis between facilities are input, and these items reflect different perspectives on risk.

[0137] The risk scoring unit (1500) calculates an integrated risk score (Risk Score) by applying pre-set weights (α, β, γ, etc.) considering the risk contribution of each diagnostic item, and then using a weighted sum method. The calculation formula at this time is expressed as follows.

[0138] Risk_Score = α × Prediction Load + β × Event Anomaly Probability + γ × Correlation Discrepancy Index

[0139] At this time, the expected load rate is an example of the main output value of the load prediction as core data among the load prediction data output of the prediction analysis unit (1200), the event anomaly probability is an example of the main output value of the fault detection as core data among the fault detection data output of the event detection unit (1300), and the correlation mismatch index is an example of the main output value of the fault analysis between facilities as core data among the fault analysis data output of the correlation analysis unit (1300). The main output values ​​of each of these models may also be set to other appropriate data.

[0140] Here, each item is normalized to a value between 0 and 1, and the weights are designed to be adjustable according to failure history, equipment characteristics, operating scenarios, etc. For example, if failures follow a pre-predicted pattern, the weight of the predicted load rate item can be set high, and for equipment with frequent sudden failures, adjustments can be made by emphasizing the contribution of the event anomaly probability.

[0141] This structure is designed as a flexible integration method that reflects the chronological order of actual risk occurrences and causal contributions, rather than a simple weighted average. Furthermore, sensitivity to complex risk scenarios is enhanced by additionally applying a condition-based weighted amplification method that temporarily increases the weight of an item when it exceeds a threshold.

[0142] For example, let's assume the following judgment result was derived for switchboard A.

[0143] - Estimated Load: 92% (Normalized value 0.92)

[0144] - Anomaly detection probability: 0.7

[0145] - Correlation Discord Index: 0.6

[0146] - Weight settings: α=0.4, β=0.4, γ=0.2

[0147] In this case, the final risk score is,

[0148] Risk_Score = 0.4×0.92 + 0.4×0.7 + 0.2×0.6 = 0.368 + 0.28 + 0.12 = 0.768

[0149] It is calculated as such, and the switchgear integrated control system (1000) selects a response policy of at least an intermediate level, such as pre-load distribution, partial circuit suspension, and manager warning, for the corresponding switchgear (100.n) based on this.

[0150] This integrated scoring method has technical benefits in that it converts heterogeneous judgment results generated from multiple diagnostic modules into a single quantitative axis, thereby resolving the logic explosion problem of conditional branching that inevitably occurred in existing technologies, and enabling flexible responses to various risk scenarios while maintaining a concise control policy decision structure.

[0151] In addition, the risk scoring unit (1500) is not limited to a statistical calculation method and can be expanded into a self-adaptive AI risk assessment engine that continuously improves the correlation between the diagnosis result and the actual failure by optimizing each weight through AI learning based on actual operation log data or by replacing the score calculation logic with a non-linear regression or neural network-based evaluation formula.

[0152] Ultimately, the risk scoring unit (1500) of the present invention functions as an intelligent risk judgment hub that enables single numerical-based policy branching by integrating diagnostic information calculated in parallel from various AI models, namely the prediction analysis AI model of the prediction analysis unit (1200), the event detection AI model of the event detection unit (1300), and the correlation analysis AI model of the correlation analysis unit (1400), into a single formula in real time, thereby ensuring both the quantitative nature of the judgment criteria and the consistency of the policy structure, thus providing clear advancement and scalability compared to the prior art.

[0154] The control policy determination unit (1600) receives the integrated risk score calculated by the risk scoring unit (1500) of each switchboard (100.n) as input, and automatically selects and executes an appropriate control policy according to the state of the switchboard unit (S307). At this time, the control policy determination unit (1600) determines a control policy for cooperative control between each switchboard for each switchboard, which will be explained in detail below.

[0155] The control policy decision unit (1600) does not make a judgment based only on the isolated state of a single switchboard (100.n), but rather realizes a distributed cooperative control policy based on inter-point cooperation by comparing and adjusting the risk level and resource status between each point where switchboards are installed within the entire power system in which multiple switchboards are connected.

[0156] The control policy decision unit (1600) is designed so that the control command of a specific switchgear reflects the status of other points by considering not only the risk score of each switchgear (100.n) as input, but also the score of adjacent points, the ESS holding status, and the resource reserve rate.

[0157] As an example, when a battery switch is required at one point due to an excessive load, a distributed resource allocation policy that utilizes spare resources from adjacent points, or a mutually complementary control policy that enables supplementary supply from other points when a specific circuit is temporarily cut off, may be automatically selected.

[0158] For example, if the switchgear at point A is in a high-risk state with a Risk Score of 0.88, and the adjacent point B is in a relatively stable state with a Risk Score of 0.42 and the remaining ESS capacity is sufficient, the control policy decision unit may select a distributed power sharing policy that distributes the battery from point B to A instead of cutting off the emergency load at point A. At the same time, if the correlation analysis unit activates a correlation abnormality flag due to a synchronous abnormality between the air conditioner and lighting at point C and calculates a Risk Score of 0.65, C selects to hold a backup circuit and notify the manager rather than self-cutting, and if necessary, a synchronous distribution policy is linked to temporarily take over a portion of the air conditioner load from A or B.

[0159] While this judgment is based on a single numerical value, it is configured to dynamically coordinate within the network structure rather than processing control policies in parallel independently, by considering the relative priority and linkage relationships between branches. This is a key differentiator that distinguishes it from conventional techniques relying on simple threshold branching.

[0160] For this purpose, the control policy decision unit (1600) may embed a multi-point state map and a policy table, or a GNN-based distributed policy judgment algorithm, and may calculate an optimized control policy split by receiving a risk vector of each point as input. As a result, the control command is transmitted in parallel not only to the corresponding switchgear but also to power resource controllers, relays, ESS, or BMS at other connected points, thereby executing a prediction-diagnosis-control integrated strategy at the entire network level.

[0161] In this way, the control policy decision unit (1600) performs the function of automatically selecting and distributing cooperative control policies among multiple switchboards (100) by considering not only the risk level of a single switchboard (100.n) but also the status of other adjacent or connected points, the remaining battery resource rate, and controllable margin. As a result, the policy selection of each switchboard (100.n) is not made independently of one another, but is coordinated based on a linkage and complementary structure within the entire system, so that the high-risk state of a single point can be mitigated through the utilization of resources from adjacent points.

[0162] As such, the control policy decision unit (1600) is a data-based autonomous control engine that combines risk score-based policy branching and inter-branch cooperative response strategies, rather than a simple fault response structure, and provides clear technical advancement and system scalability compared to existing technology.

[0164] The control command transmitting unit (1700) transmits the policy decision result determined by the control policy decision unit (1600) as a control command to each switchboard (100.n) (S308), thereby enabling actual control of each switchboard (100.n).

[0166] Meanwhile, the monitoring information providing unit (1900) provides the manager terminal (200) with the data required at each step performed by the switchgear integrated control system (1000) at each of the aforementioned steps (S309). Explanation of the symbols

[0167] 100: Switchboard (common name) 100.1, ..., 100.N : Individual branch switchboards 200: Administrator Terminal 1000: Multi-point Switchgear Integrated Control System Using AI-Based Risk Assessment 1100: Data Collection / Preprocessing Unit 1200: Predictive Analytics Department 1300: Event Detection Unit 1400: Correlation Analysis Department 1500: Risk Scoring Section 1600: Control Policy Decision Unit 1700: Control Order Transmitter 1800: Data storage unit 1900: Monitoring Information Provision Department

Claims

Claim 1 A multi-point switchgear integrated control system utilizing AI-based risk judgment, comprising: a data collection and preprocessing unit that collects status data including time-series load usage data and equipment event data from sensors installed in each switchgear and performs preprocessing for input into an AI model; a prediction analysis unit that generates load prediction data including a predicted load rate at a future point in time based on the time-series load usage data of each switchgear; an event detection unit that analyzes the equipment event data of each switchgear in real time to generate fault detection data including fault types and event anomaly probabilities; a correlation analysis unit that stores a normal operation correlation matrix in which mutual operation patterns are defined during normal operation between equipment within the switchgear, compares a real-time correlation coefficient calculated from equipment-specific status information collected in real time among the status data with a reference correlation coefficient extracted from the normal operation correlation matrix, and outputs equipment relationship anomaly analysis data including a correlation mismatch index based on the difference between the real-time correlation coefficient and the reference correlation coefficient; and normalizes the predicted load rate included in the load prediction data, the event anomaly probability included in the fault detection data, and the correlation mismatch index, and respectively A risk scoring unit that calculates an integrated risk score by applying a condition-based weighted amplification method, wherein the weight applied to an item is increased when any one of the prediction load rate, the event anomaly probability, and the correlation mismatch index exceeds a preset threshold for weight amplification;A multi-point switchgear integrated control system using AI-based risk judgment, comprising: a control policy determination unit that, when the integrated risk score for each switchgear exceeds a preset threshold for multi-point cooperative control, checks the integrated risk score of a network-connected adjacent switchgear and the remaining capacity information of an energy storage device (ESS) equipped in the adjacent switchgear, and when the integrated risk score of the adjacent switchgear is less than the preset threshold for multi-point cooperative control and the remaining capacity of the ESS is greater than or equal to a preset standard, generates a multi-point cooperative control signal to control the supply of power from the ESS of the adjacent switchgear to the switchgear; and a control command transmission unit that transmits a control command including the multi-point cooperative control signal to the switchgear and the adjacent switchgear. Claim 2 A multi-point switchgear integrated control system using AI-based risk judgment, characterized in that, in claim 1, it further includes a monitoring information providing unit that provides data of each control step performed by the switchgear integrated control system to the manager terminal. Claim 3 A multi-point switchgear integrated control system using AI-based risk judgment according to claim 1, wherein the time-series load usage data, which is the input of the prediction analysis unit, includes one or more of chronological load rate data, temperature, and external events, and the load prediction data, which is the output of the prediction analysis unit, includes one or more of a predicted load rate, a predicted temperature change rate, a load increase trend, and a prediction reliability. Claim 4 A multi-point switchgear integrated control system using AI-based risk judgment according to claim 1, wherein the equipment event data, which is the input of the event detection unit, includes one or more of instantaneous current, leakage current, internal temperature rise rate, and temperature sensor distribution imbalance, and the fault detection data, which is the output of the event detection unit, includes one or more of fault type classification, similarity score, detection grade, event anomaly probability, and event response request flag. Claim 5 A multi-point switchgear integrated control system using AI-based risk judgment according to claim 1, wherein the inter-equipment correlation and individual equipment real-time data, which are inputs to the correlation analysis unit, include one or more of a normal operation correlation matrix, real-time current amount time series data, real-time power consumption, real-time operation rate / ON-OFF event log, and time reference information, and the inter-equipment relationship anomaly analysis data, which is output of the correlation analysis unit, includes one or more of a current correlation coefficient, reference correlation coefficient, correlation mismatch index, relationship anomaly detection flag, list of major anomaly equipment pairs, and relationship anomaly reliability. Claim 6 delete Claim 7 A multi-point switchgear integrated control system using AI-based risk judgment according to claim 1, as a method for performing multi-point switchgear integrated control using AI-based risk judgment, comprising: (a) a step in which a data collection and preprocessing unit collects state data including time-series load usage data and equipment event data from sensors installed in each switchgear and performs preprocessing for input into an AI model; (b) a step in which a prediction analysis unit generates load prediction data including a predicted load rate at a future point in time based on the time-series load usage data of each switchgear; (c) a step in which an event detection unit analyzes the equipment event data of each switchgear in real time and generates fault detection data including a fault type and an event anomaly probability; (d) a correlation analysis unit stores a normal operation correlation matrix in which mutual operation patterns are defined between equipment within the switchgear during normal operation, and compares a real-time correlation coefficient calculated from equipment-specific state information collected in real time among the state data with a reference correlation coefficient extracted from the normal operation correlation matrix, and calculates a correlation mismatch index based on the difference between the real-time correlation coefficient and the reference correlation coefficient. A step of outputting data for analyzing abnormal relationships between facilities including; (e) a step in which a risk scoring unit normalizes the predicted load rate, the event abnormality probability, and the correlation mismatch index, applies a preset weight to each, sums them up, and calculates an integrated risk score by applying a condition-based weighted amplification method in which, if any one of the predicted load rate, the event abnormality probability, and the correlation mismatch index exceeds a preset threshold for weight amplification, the weight applied to the corresponding item is increased;(f) a control policy decision unit, when the integrated risk score for each switchgear exceeds a preset threshold for multi-point cooperative control, checks the integrated risk score of an adjacent switchgear connected to a network and the remaining capacity information of an energy storage device (ESS) equipped in the adjacent switchgear, and when the integrated risk score of the adjacent switchgear is less than the preset threshold for multi-point cooperative control and the remaining capacity of the ESS is greater than or equal to a preset standard, generates a multi-point cooperative control signal to control the supply of power from the ESS of the adjacent switchgear to the switchgear; and, (g) a control command transmission unit, transmits a control command including the multi-point cooperative control signal to the switchgear and the adjacent switchgear, comprising a method for performing multi-point switchgear integrated control using AI-based risk judgment. Claim 8 A method for performing integrated control of multi-point switchgear using AI-based risk judgment, wherein, in claim 7, the time-series load usage data input to the prediction analysis unit includes one or more of chronological load rate data, temperature, and external events, and the load prediction data output to the prediction analysis unit includes one or more of a predicted load rate, a predicted temperature change rate, a load increase trend, and a prediction reliability. Claim 9 A method for performing integrated control of multi-point switchgear using AI-based risk judgment according to claim 7, wherein the equipment event data, which is the input of the event detection unit, includes one or more of instantaneous current, leakage current, internal temperature rise rate, and temperature sensor distribution imbalance, and the fault detection data, which is the output of the event detection unit, includes one or more of fault type classification, similarity score, detection grade, event anomaly probability, and event response request flag. Claim 10 A method for performing integrated control of multi-point switchgear using AI-based risk judgment according to claim 7, wherein the inter-equipment correlation and individual equipment real-time data, which are inputs to the correlation analysis unit, include one or more of a normal operation correlation matrix, real-time current amount time series data, real-time power consumption, real-time operation rate / ON-OFF event log, and time reference information, and the inter-equipment relationship anomaly analysis data, which is output of the correlation analysis unit, includes one or more of a current correlation coefficient, reference correlation coefficient, correlation mismatch index, relationship anomaly detection flag, list of major anomaly equipment pairs, and relationship anomaly reliability. Claim 11 delete Claim 12 A computer program stored on a computer-readable, non-transient storage medium for performing integrated control of multi-point switchgear using AI-based risk judgment, which is stored on the non-transient storage medium and, by means of a processor, comprises: (a) collecting status data including time-series load usage data and equipment event data from sensors installed in each switchgear and performing preprocessing for input into an AI model; (b) generating load prediction data including a predicted load rate at a future point in time based on the time-series load usage data of each switchgear; (c) generating fault detection data including fault types and event anomaly probabilities by analyzing the equipment event data of each switchgear in real time; (d) storing a normal operation correlation matrix in which mutual operation patterns are defined during normal operation between equipment within the switchgear, comparing a real-time correlation coefficient calculated from equipment-specific status information collected in real time among the status data with a reference correlation coefficient extracted from the normal operation correlation matrix, and outputting data for analyzing anomalies in inter-equipment relationships including a correlation mismatch index based on the difference between the real-time correlation coefficient and the reference correlation coefficient. Step; (e) normalizing the prediction load rate, the event anomaly probability, and the correlation mismatch index, applying preset weights to each, and summing them, wherein if any one of the prediction load rate, the event anomaly probability, and the correlation mismatch index exceeds a preset threshold for weight amplification, a condition-based weighted amplification method is applied to increase the weight applied to the corresponding item to calculate an integrated risk score;(f) a step of generating a multi-point cooperative control signal to control the supply of power from the ESS of the adjacent switchgear to the switchgear, wherein if the integrated risk score for each switchgear exceeds a preset threshold for multi-point cooperative control, the integrated risk score of the adjacent switchgear connected to the network and the remaining capacity information of the energy storage device (ESS) equipped in the adjacent switchgear are checked, and if the integrated risk score of the adjacent switchgear is less than the preset threshold for multi-point cooperative control and the remaining capacity of the ESS is greater than or equal to a preset standard; and (g) a command to execute the step of transmitting a control command including the multi-point cooperative control signal to the switchgear and the adjacent switchgear, wherein the computer program stored on a computer-readable non-transient storage medium for performing multi-point switchgear integrated control using AI-based risk judgment.

Images