Search

KR-102961441-B1 - Method and system for managing an integrated distribution frame with AI-based energy reduction using a recall-maximizing feedback loop

KR102961441B1KR 102961441 B1KR102961441 B1KR 102961441B1KR-102961441-B1

Abstract

In the AI energy-saving integrated wiring panel management method based on a feedback loop maximizing recall rate according to the present invention, the method comprises the steps of: a service providing server collecting basic information on the integrated wiring panel and energy consumption information from network equipment; a service providing server preprocessing the collected basic information on the integrated wiring panel and energy consumption information to generate processed energy data; a service providing server generating or improving an energy prediction algorithm that predicts energy overconsumption and inefficiency based on processed energy data and change data; a service providing server initiating the prediction of energy overconsumption and inefficiency based on the energy prediction algorithm; a service providing server determining whether the prediction of energy overconsumption and inefficiency is successful; and a service providing server calculating the prediction accuracy if the prediction of energy overconsumption and inefficiency is successful. The processed energy data includes at least one of power consumption efficiency per traffic, power increase rate relative to temperature change rate, and optimal operating range deviation rate.

Inventors

  • 박준민
  • 박지윤

Assignees

  • 국제텔레시스(주)

Dates

Publication Date
20260507
Application Date
20251230

Claims (7)

  1. In an AI energy-saving integrated wiring panel management method based on a feedback loop that maximizes recall, A step in which a service provider server collects basic information on the integrated wiring board and energy consumption information from network equipment; A step of generating energy processing data by preprocessing basic integrated wiring panel information and energy consumption information collected by the service provider server; A step in which a service providing server generates or improves an energy prediction algorithm that predicts energy overconsumption and inefficiency based on energy processing data and change data; A step in which the service provider server starts predicting energy overconsumption and inefficiency based on an energy prediction algorithm; A step of determining whether the service provider server succeeds in predicting energy overconsumption and inefficiency; and If the service provider server succeeds in predicting energy overconsumption and inefficiency, it includes a step of calculating prediction accuracy, and The above energy processing data is, It includes at least one of power consumption efficiency per traffic, power growth rate relative to temperature change rate, and optimal operating range deviation rate, and In the step of determining whether the above service providing server succeeds in predicting energy overconsumption and inefficiency, A step in which the service provider server changes the risk weight if the service provider server fails to predict energy overconsumption and inefficiency; Step of the service provider server changing the prediction threshold; and The service providing server further includes the step of generating or improving an energy prediction algorithm that predicts energy overconsumption and inefficiency based on energy processing data and change data including the above-mentioned risk weights and prediction thresholds, and An AI energy-saving integrated wiring panel management method based on a recall-maximizing feedback loop, characterized in that the above prediction threshold is calculated by the following formula. At this time is the calculated prediction threshold, is the existing prediction threshold, β is the threshold adjustment coefficient, It refers to the proportion of cases where energy overconsumption or inefficiency occurred but was not predicted.
  2. delete
  3. In paragraph 1, The above energy prediction algorithm is, Using at least one of the above power consumption efficiency per traffic, power increase rate relative to temperature change rate, and optimal operating range deviation rate as an input value, Each of the above input values is compared with a reference value that is pre-set or adjusted according to the above change data to determine whether there is an anomaly for each indicator, and A method for managing an AI energy-saving integrated wiring panel based on a feedback loop that maximizes recall, characterized by determining a state with a high probability of energy overconsumption or inefficiency occurring when at least one of the above indicators is satisfied or when multiple energy processing data simultaneously deviate from a reference value, and performing a prediction.
  4. delete
  5. A service providing server that collects basic information on integrated wiring panels and energy consumption information from network equipment, preprocesses the collected basic information on integrated wiring panels and energy consumption information to generate processed energy data, generates or improves an energy prediction algorithm that predicts energy overconsumption and inefficiency based on processed energy data and modified data including risk weights and prediction thresholds, starts predicting energy overconsumption and inefficiency based on the energy prediction algorithm, determines whether the prediction of energy overconsumption and inefficiency is successful, calculates the prediction accuracy if the prediction of overconsumption and inefficiency is successful, and if the prediction of energy overconsumption and inefficiency is not successful, changes the risk weights and changes the prediction thresholds; and It includes network equipment that provides real-time data to the service provider server for predicting energy overconsumption and inefficiency, and The above energy processing data is, It includes at least one of power consumption efficiency per traffic, power growth rate relative to temperature change rate, and optimal operating range deviation rate, and An AI energy-saving integrated wiring panel management system based on a recall-maximizing feedback loop, characterized in that the above prediction threshold is calculated by the following formula. At this time is the calculated prediction threshold, is the existing prediction threshold, β is the threshold adjustment coefficient, It refers to the proportion of cases where energy overconsumption or inefficiency occurred but was not predicted.
  6. In paragraph 5, The above energy prediction algorithm is, Using at least one of the above power consumption efficiency per traffic, power increase rate relative to temperature change rate, and optimal operating range deviation rate as an input value, Each of the above input values is compared with a reference value that is pre-set or adjusted according to the above change data to determine whether there is an anomaly for each indicator, and An AI energy-saving integrated wiring panel management system based on a feedback loop maximizing recall rate, characterized by performing a prediction by determining a state with a high probability of energy overconsumption or inefficiency occurring when at least one of the above indicator abnormalities is satisfied or when multiple energy processing data simultaneously deviate from a reference value.
  7. delete

Description

Method and system for managing an integrated distribution frame with AI-based energy reduction using a recall-maximizing feedback loop The present invention relates to an AI energy-saving integrated wiring panel management method and system based on a recall-maximizing feedback loop, which predicts whether energy overconsumption and inefficiency occur based on power consumption information, traffic information, and environmental information collected from the integrated wiring panel, and improves energy efficiency by dynamically adjusting a prediction algorithm and a prediction threshold through a feedback loop to maximize the recall of the prediction results. Artificial intelligence technology has continuously advanced since the late 2010s, accompanied by dramatic performance improvements in deep learning technology. In particular, with the commercialization of generative AI technology based on large-scale language models in late 2022, the practical adoption of AI across industries expanded rapidly. Since 2023, the utilization of AI technology has intensified not only in generative AI but also in real-time prediction, control, and optimization in industrial settings. Consequently, the convergence of existing equipment management, energy management, and infrastructure operation technologies with AI is becoming an inevitable technological trend. Meanwhile, although integrated distribution boards were initially utilized as devices performing simple electrical wiring organization and cable aggregation functions, they have evolved into complex infrastructure devices capable of integrally housing and managing various systems—such as power lines, data communication lines (LAN), optical cables, CCTV, firefighting equipment, and Internet of Things (IoT) sensors—within a single rack structure, driven by the advancement of information and communication technology and the increasing complexity of network infrastructure. In particular, as the installation of integrated distribution boards has become essential in 5G base stations, edge data centers, smart buildings, and large-scale enterprise network environments, power consumption at the integrated distribution board level has increased from tens of watts to hundreds of watts. Consequently, the proportion of integrated distribution boards in the total energy consumption of data centers or building networks is also gradually expanding. However, conventional integrated distribution board management technologies have been limited to threshold-based alarm methods using temperature, humidity, and current sensors, remote status monitoring and fault notifications via SNMP or IoT platforms, and simple power usage recording and inquiry functions. While these methods can contribute to reactively recognizing the current status of the integrated distribution board, they have limitations in predicting the potential for energy overconsumption or inefficiency before it occurs, or in actively reducing energy consumption by reflecting network traffic patterns, seasonal changes, and operational characteristics by time of day. Recently, there have been some attempts to apply artificial intelligence to integrated distribution board management, but most have adopted existing prediction models applied to general server facilities or HVAC systems, using only traditional fault indicators such as CPU usage and temperature as inputs, or setting prediction accuracy as a primary performance indicator. This has resulted in frequent instances of False Negatives, where actual situations of energy overconsumption or inefficiency are missed. In particular, technology has not been sufficiently presented to systematically define energy efficiency-specific indicators—such as power consumption efficiency relative to network traffic volume, abnormal power increase patterns relative to the rate of temperature change, and optimal operating intervals by equipment and time period, which are unique characteristics of integrated distribution panels—and to utilize them for prediction. As a prior art patent, there is Korean Registered Patent No. 10-2793436 (Method and System for Managing Integrated Wiring Boards Based on Artificial Intelligence), but it merely includes the steps of: a service providing server collecting basic integrated wiring board information from network equipment; a service providing server preprocessing the collected basic integrated wiring board information to generate processed data; a service providing server generating or improving a prediction algorithm that predicts failures and faults according to a prediction classification based on the processed data or analysis data; a service providing server initiating failure and fault prediction according to the prediction classification based on the prediction algorithm; a service providing server determining whether the failure and fault prediction is successful; and a service providing server calculating the prediction accuracy if the failure