KR-20260067768-A - An algorithms with multiple analysis models and a method of RUL(Remaining Useful life) prediction for electrical facilities

Abstract

The present invention relates to a technology for predicting the remaining lifespan of electrical equipment by applying a multi-analysis model. Specifically, electrical safety data such as inspection, diagnosis, safety management, history, environment, and operation of electrical equipment is structured and preprocessed, and state intervals are derived through basic statistical analysis. Next, a primary remaining lifespan prediction result is derived through a multi-analysis model, and then a final remaining lifespan prediction result is derived through retraining in which only important variables are extracted from a validation model to prevent overfitting and improve prediction reliability.

Inventors

• 박지만
• 문영채
• 전정채

Assignees

• 한국전기안전공사

Dates

Publication Date

20260513

Application Date

20241106

Claims (1)

1. A data storage/missing value/outlier processing unit that reviews an input data set and processes missing values and outliers; Basic statistical analysis unit that calculates state intervals based on electrical equipment usage data; A state interval analysis model verification and application unit that applies the optimal analysis model to each interval based on the suitability evaluation results of the analysis models for each state interval output from the basic statistical analysis unit, and outputs the first remaining life prediction results; and An electrical equipment remaining life prediction system applying a multi-analysis model, characterized by comprising a final remaining life prediction result output unit through retraining that selects and retrains important variables to prevent overfitting and enhance the reliability of the prediction result, and outputs the final remaining life prediction result.

Description

System and method for predicting remaining useful life of electrical facilities using multiple analysis models The present invention relates to a technology for predicting the remaining lifespan of electrical facilities by applying a multi-analysis model. More specifically, it involves structuring and preprocessing electrical safety data regarding inspection, diagnosis, safety management, history, environment, and operation of electrical facilities, and deriving state intervals through basic statistical analysis. Next, after deriving a primary remaining lifespan prediction result through a multi-analysis model, the invention relates to an algorithm and method for deriving a final remaining lifespan prediction result through retraining by extracting only important variables from a validation model to prevent overfitting and improve prediction reliability. Private electrical facilities encompass all facilities excluding general and industrial use, such as major domestic industrial sites and public utility facilities. While the specific impact varies depending on the type and purpose of the facility, operational shutdowns caused by power outages or facility accidents generally result in socioeconomic losses. Consequently, there is a growing global movement to apply online electrical safety and asset management technologies to electrical facilities. Leading international companies, such as ABB and Siemens, are developing cutting-edge technologies based on superior technical capabilities and big data related to electrical safety, and have begun commercializing them. However, the reality is that the gap between domestic companies, which remain in the development or early stages of applying asset management technologies, and the aforementioned foreign companies is bound to continue widening. The primary purpose of applying asset management technology to electrical facilities is to operate facility assets economically through preemptive maintenance and replacement based on risk and soundness assessment results. Quantitatively assessing the current condition of the facilities is crucial for taking measures, such as determining the timing for maintenance and replacement in a timely manner; to achieve this, the reliability of risk and soundness assessments must be enhanced. Evaluating the risks, soundness, and remaining lifespan of electrical facilities requires various types of online and offline electrical safety data as well as appropriate AI algorithms, and technological development is currently actively underway, primarily led by heavy electrical equipment manufacturers. However, given the scarcity of data and the fact that the full-scale acquisition of online monitoring data is still in its early stages, it is nearly impossible to significantly improve the reliability of lifespan assessment results in the short term. Ultimately, maximizing the use of existing offline rather than online data is the only way to enhance the reliability of electrical facility evaluations at this point. To achieve this, technology is required that collects big data related to electrical facility safety, performs statistical analysis, and predicts reliable remaining lifespans through multi-learning models. FIG. 1 is a detailed configuration diagram of one embodiment. Figure 2 is a detailed configuration diagram of the data processing unit of Figure 1. Figures 3 to 5 are figures showing the output waveforms of each part. One embodiment of the present invention is broadly composed of a data storage/missing value/outlier processing unit, a basic statistical analysis unit, a state-by-state analysis model verification/application unit, and a final remaining life prediction result output unit through retraining. The data storage/missing value/outlier processing unit reviews an input data set and processes missing values and outliers. The basic statistical analysis unit calculates state intervals based on electrical equipment usage data, and the state-by-state analysis model verification/application unit applies the optimal analysis model to each state interval output by the basic statistical analysis unit through the suitability evaluation results of the analysis models and outputs the first remaining life prediction result. The final remaining life prediction result output unit through retraining selects important variables to prevent overfitting and improve the reliability of the prediction result, retrains them, and outputs the final remaining life prediction result. (1) Data storage/missing/outlier processing unit The data storage unit stores offline data regarding private electrical facilities, such as inspections and diagnoses. Preprocessing is essential to perform statistical analysis using the data; in particular, the efficient handling of missing values and outliers is directly linked to data quality and the reliability of the analysis. As shown in Fig. 2, the missing value/outlier handling unit performs an analysis of the input da