CN-121980517-A - Intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning

CN121980517ACN 121980517 ACN121980517 ACN 121980517ACN-121980517-A

Abstract

The invention discloses an intelligent diagnosis method for leakage risk of an overhanging curtain wall based on machine learning, which relates to the technical field of machine learning and comprises a feature fusion step, a feature analysis step and a feature analysis step, wherein the feature fusion step is used for fusing original monitoring features in real-time monitoring data with physical mechanism features to construct fusion feature vectors of each evaluation unit; and a primary diagnosis step of inputting the fusion feature vector into a trained first diagnosis model, outputting the primary leakage risk probability of each evaluation unit, and calculating the uncertainty measure of the primary leakage risk probability. According to the invention, the primary diagnosis result and the optimized diagnosis result are fused through the Bayesian inference framework, and the dominant risk factor set is generated, so that not only is the lower diagnosis risk probability obtained, but also key physical or monitoring factors and association relations thereof causing risks can be automatically traced, clear and operable decision basis is provided for operation and maintenance personnel, and the crossing from early warning to attribution is realized.

Inventors

YUE SHUCAI
TANG XIAOQING
CHEN XIAOJUN
PAN JIE

Assignees

江苏恒尚节能科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260206

Claims (8)

1. The intelligent diagnosis method for the leakage risk of the overhanging curtain wall based on machine learning is characterized by comprising the following steps of: S1, acquiring real-time monitoring data of a plurality of evaluation units of a target overhanging curtain wall; s2, processing the real-time monitoring data based on a physical model of building leakage to extract physical mechanism characteristics, wherein the physical mechanism characteristics at least comprise dynamic wind pressure distribution characteristics of the surface of a curtain wall, water film thickness and flow velocity characteristics of a drainage path and dynamic opening quantity characteristics of joints; s3, fusing original monitoring features in the real-time monitoring data with the physical mechanism features to construct fused feature vectors of each evaluation unit; S4, a primary diagnosis step, namely inputting the fusion feature vector into a trained first diagnosis model, outputting the primary leakage risk probability of each evaluation unit, and calculating the uncertainty measure of the primary leakage risk probability; S5, judging a diagnosis path, namely judging according to the primary leakage risk probability and the uncertainty measure thereof: If the primary leakage risk probability is higher than the first threshold value and the uncertainty measure is lower than the second threshold value, determining that the primary leakage risk probability is a high-confidence risk unit, and turning to the step S8; otherwise, judging the unit to be optimized and converting into a step S6; S6, constructing context information containing high uncertainty characteristics and adjacent unit data of the unit to be diagnosed, inputting the fusion characteristic vector and the context information into a second diagnosis model together, and outputting risk probability after optimization; S7, confirming and fusing, namely fusing the primary leakage risk probability and the optimized risk probability by adopting a Bayesian inference framework based on consistency verification of the optimized risk probability and the historical diagnosis result to generate a confirmed risk probability and a dominant risk factor set; and S8, generating a report, namely generating a diagnosis report containing the diagnosis risk probability, the dominant risk factor set and the physical association relation thereof according to the result of the step S5 or the step S7.
2. The intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning according to claim 1, characterized in that in step S4, the first diagnosis model is a multitask learning neural network based on attention mechanism, which performs two tasks simultaneously: predicting leakage risk probability based on the fusion feature vector regression; Outputting contribution degree of each feature in the fusion feature vector to risk prediction; The uncertainty measure is obtained by starting Monte Carlo dropouout in the reasoning process of the first diagnosis model, performing forward propagation for the same fusion feature vector for a plurality of times, and calculating standard deviations of a plurality of risk probabilities as the uncertainty measure.
3. The intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning according to claim 1, wherein in step S5, the diagnosis path judging step specifically comprises: Setting a basic high risk threshold T base and a dynamic adjustment coefficient alpha positively correlated to the uncertainty measure U; Calculating a dynamic high risk threshold T high =T base +α×u; If the average value of the primary leakage risk probability of a certain evaluation unit is greater than T high , the evaluation unit is determined to be a high-confidence risk unit.
4. The intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning according to claim 1, wherein in S5, the method further comprises low risk determination logic: setting a fixed low risk threshold T low ; extracting the features of M bits before the contribution degree ranking output by the first diagnosis model as high contribution degree features; If the average value of the primary leakage risk probabilities of a certain evaluation unit is smaller than T low and the proportion of the high contribution degree features belonging to the physical mechanism features exceeds a preset proportion threshold value, the evaluation unit is judged to be a low risk unit.
5. The intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning according to claim 1, wherein in step S6, the second diagnosis model adopts a memory-based meta learning strategy, and the reasoning process is as follows: according to the context information, K historical cases which are most similar to the current diagnosis unit to be optimized are retrieved from a historical case library; Training a lightweight adaptation model by utilizing the characteristic data of the K historical cases and known risk labels; And calculating the current diagnosis unit to be optimized through the adaptive model, and outputting the risk probability after optimization.
6. The intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning according to claim 1, wherein in step S7, the consistency check is specifically: Calculating the deviation between the optimized risk probability and the sliding average value of all risk probabilities of the same evaluation unit in a historical preset time window; if the deviation exceeds a preset deviation threshold, marking the optimized risk probability as 'in doubt', and triggering a data review flow.
7. The intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning according to claim 1, wherein in step S7, the fusion using bayesian inference framework is specifically: Taking the mean value and the variance of the primary leakage risk probability as prior distribution parameters; Constructing a likelihood function by taking the optimized risk probability as observation data; and calculating posterior distribution according to a Bayesian formula, and taking the mean value of the posterior distribution as the diagnosis risk probability.
8. The intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning as claimed in claim 7, wherein the generation method of the dominant risk factor set is as follows: screening out physical mechanism features and key original monitoring features from the high contribution features output by the first diagnosis model to form a first feature subset; Extracting common risk influence factors from K similar historical cases retrieved from the second diagnosis model to form a second feature subset; and taking the intersection of the first feature subset and the second feature subset, sorting according to the influence weights of the factors in Bayesian posterior probability calculation, and selecting the first N factors to form the dominant risk factor set.

Description

Intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning Technical Field The invention relates to the technical field of machine learning, in particular to an intelligent diagnosis method for leakage risk of an overhanging curtain wall based on machine learning. Background With the development of modern buildings towards high-rise and complicated modeling, large-area overhanging glass curtain walls are widely applied due to the transparent visual effect and flexible design possibility. However, such curtain wall systems are exposed to complex environmental loads such as wind and rain, temperature differences, wind pressure alternation for a long period of time, and the seam sealing materials thereof are prone to aging, and drainage paths may be blocked, resulting in a significant increase in rainwater penetration risk. Leakage not only damages the interior decoration and equipment of the building, influences the use function and comfort level, but also can corrode metal components, influence structural durability and even threaten safety after long-term moisture invasion, so that the leakage risk of the overhanging curtain wall is subjected to efficient and accurate early diagnosis and early warning, and the leakage risk is an urgent and important requirement in the field of building operation and maintenance. At present, the diagnosis technology for the leakage risk of the building curtain wall mainly develops along two directions, namely a physical monitoring method based on a sensor, an environment and response data are collected by arranging wind pressure, displacement, temperature and humidity or humidity sensors at key parts of the curtain wall, and abnormal judgment is carried out by relying on threshold comparison or simple statistical analysis, and part of research begins to try to apply a machine learning algorithm, such as a support vector machine or a traditional neural network, to the classification of monitoring data along with the progress of artificial intelligence so as to identify a potential leakage mode. These techniques improve the level of automation of monitoring to some extent, providing a rudimentary tool for finding problems from massive data. However, prior art solutions still face a series of key challenges and development bottlenecks. Firstly, most methods only carry out shallow statistics on original monitoring data or directly input the model, and fail to deeply fuse physical mechanisms of building leakage, such as wind-rain-structure multi-field coupling action, so that characteristic engineering and actual problems are disjointed, the model has poor interpretation, and a diagnosis conclusion lacks clear physical causal support. Secondly, most of common diagnostic models are static and single judgment systems, inherent noise, uncertainty and individual case specificity in monitoring data cannot be effectively processed, and a prediction result with low confidence degree lacks further optimization and verification mechanisms, so that false alarm or missing alarm is easily caused. Moreover, the existing method often outputs a single risk probability value, and fails to automatically identify and quantify the leading factors and interactions thereof causing risks, so that operation and maintenance personnel can not easily formulate accurate intervention measures. Therefore, it is necessary to invent an intelligent diagnosis method for leakage risk of overhanging curtain wall based on machine learning to solve the above problems. Disclosure of Invention The invention aims to provide an intelligent diagnosis method for leakage risk of an overhanging curtain wall based on machine learning, which aims to solve the problems in the background technology. In order to achieve the purpose, the invention provides the following technical scheme that the intelligent diagnosis method for the leakage risk of the overhanging curtain wall based on machine learning specifically comprises the following steps: S1, acquiring real-time monitoring data of a plurality of evaluation units of a target overhanging curtain wall; s2, processing the real-time monitoring data based on a physical model of building leakage to extract physical mechanism characteristics, wherein the physical mechanism characteristics at least comprise dynamic wind pressure distribution characteristics of the surface of a curtain wall, water film thickness and flow velocity characteristics of a drainage path and dynamic opening quantity characteristics of joints; s3, fusing original monitoring features in the real-time monitoring data with the physical mechanism features to construct fused feature vectors of each evaluation unit; S4, a primary diagnosis step, namely inputting the fusion feature vector into a trained first diagnosis model, outputting the primary leakage risk probability of each evaluation unit, and calculating the uncertainty measure of the primary leakage risk probabilit