CN-121978438-A - Building electrical system fault diagnosis method and system

CN121978438ACN 121978438 ACN121978438 ACN 121978438ACN-121978438-A

Abstract

The application provides a building electrical system fault diagnosis method and system, belongs to the field of building electrical system monitoring and fault diagnosis, and is used for solving the problems that in the related technology, the diagnosis dimension is single, a fault dynamic propagation and causal mechanism is difficult to model, and the model self-adaption and generalization capability are insufficient. According to the method, the multi-mode time sequence data are mapped to Riemann manifold, a unified manifold control differential equation model is utilized, node state evolution dynamics and interaction between geometric causes and effects are synchronously modeled internally, and accurate fault detection and positioning are achieved based on Li Daoshu differences and geometric divergence differences. The system correspondingly comprises a data acquisition module, a unified model and a fault judgment module. By combining element learning rapid adaptation, robust training for antagonism and geometric federal learning, the scheme realizes high-precision and interpretable diagnosis of composite faults and has strong self-adaptive capacity and co-evolution potential.

Inventors

WANG XIAO
QIAN XUEFENG
CHEN XIANGCHEN

Assignees

靖江市同润电气有限公司

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. A fault diagnosis method for a building electrical system is characterized by comprising the following steps, Obtaining multi-mode time sequence data of a plurality of monitoring nodes in a building electrical system and physical topological connection relations among the monitoring nodes, Inputting the multi-mode time series data and the physical topological connection relation into a unified diagnosis model for processing, Wherein the unified diagnostic model is configured to synchronously perform node state evolution dynamics modeling and causal interaction modeling between nodes in an abstract mathematical space constructed from system states, And determining a fault diagnosis result of the building electrical system according to the dynamic evolution information and the causal interaction information output by the unified diagnosis model.
2. The method for diagnosing a fault in a building electrical system according to claim 1, wherein, The abstract mathematical space is a Riemann manifold, and the node state evolution dynamics modeling and the causal interaction modeling between nodes are integrally described on the Riemann manifold.
3. The method for diagnosing a fault in a building electrical system according to claim 2, wherein, The integrated description is implemented by a manifold control differential equation, for any node on the Riemann manifold, its state change rate is determined jointly by a first neural network based on the current state of that node, and causal influence contributions from other nodes, The causal influence contribution from any other node is obtained by generating a first influence vector by a second neural network according to the state of the other node, and then transporting the first influence vector in parallel to the manifold position of the current node from the manifold position of the other node according to a dynamic causal attention weight.
4. The method for diagnosing a fault in a building electrical system according to claim 3, wherein, The step of determining the fault diagnosis result comprises the steps of calculating Li Daoshu differences between dynamic evolution of the current state and dynamic evolution of the reference health state, calculating geometrical divergence differences between causal interactions of the current state and causal interactions of the reference health state, and judging that faults occur when a comprehensive index formed by the Li Daoshu differences and the geometrical divergence differences exceeds a threshold value.
5. The method for diagnosing a fault in a building electrical system according to claim 2, wherein, The step of mapping the multi-modal time series data to the Riemann manifold is accomplished by a manifold self-encoder whose training objectives include minimizing errors between the reconstructed data and the raw data and satisfying a preset geometric constraint.
6. The method for diagnosing a fault in a building electrical system according to claim 1 or 2, wherein, The unified diagnosis model adopts a meta-learning strategy to carry out rapid adaptation, specifically, parameters of the model are regarded as points positioned on a model manifold, an initial updating direction is determined on the model manifold according to data characteristics of a new diagnosis task, and model parameters are updated along geodesic wires corresponding to the initial updating direction.
7. The method for diagnosing a fault in a building electrical system according to claim 2, wherein, In training the unified diagnostic model, an antagonistic training is introduced to enhance its robustness, which requires a dynamic causal attention weight in the model, which remains unchanged for any equidistant transformation on the Riemann manifold that keeps the geodesic distance unchanged.
8. The method for diagnosing a fault in a building electrical system according to claim 1 or 2, wherein, The unified diagnostic model is cooperatively trained and updated through a geometric federal learning framework, the geometric federal learning framework performs weighted Freund Xie Te mean calculation on local model parameters of each client on the model manifold to obtain a global model, and the updating directions are aligned by parallel transportation in the calculation process.
9. The method for diagnosing a fault in a building electrical system according to claim 1, wherein, The multi-modal time series data includes at least two of electrical quantity data, temperature data, and partial discharge data.
10. A building electrical system fault diagnosis system, characterized by being used for realizing the building electrical system fault diagnosis method according to any one of claims 1 to 9, the system comprising, A data acquisition module for acquiring multi-mode time series data and physical topological connection relation of a plurality of monitoring nodes in the building electrical system, A unified diagnostic model module configured to synchronously perform node state evolution dynamics modeling and inter-node causal interaction modeling in an abstract mathematical space constructed according to system states, and to output dynamics evolution information and causal interaction information, And the fault judging module is used for determining and outputting a fault diagnosis result according to the dynamic evolution information and the causal interaction information.

Description

Building electrical system fault diagnosis method and system Technical Field The application relates to the field of building electrical system monitoring and fault diagnosis, in particular to a building electrical system fault diagnosis method and system. Background Along with the improvement of the intelligent level of the building, the building electrical system is increasingly complex, and the safe and stable operation of the building electrical system is vital to the guarantee of production and life. Therefore, the real-time and accurate fault diagnosis is carried out on the electric system, potential hidden dangers are found and positioned in time, and the intelligent building operation and maintenance system becomes one of core requirements. At present, the fault diagnosis technology of the building electrical system mainly comprises methods of alarming based on a threshold value, statistical analysis based on signal spectrum or waveform characteristics, classification by applying a shallow machine learning model and the like. These prior art techniques typically rely on a single type of data (e.g., current, voltage) or simply post-feature-stitching analysis of multiple data. They often take into account the state evolution of the system and the interactions between devices, making it difficult to model the dynamic propagation processes and causal mechanisms of faults in complex electrical networks. However, the prior art has significant drawbacks. First, the diagnostic dimension is single and weak to composite fault recognition caused by multiple factor coupling. Secondly, the model is often based on statistical association of shallow layers, and lack of deep mining on intrinsic dynamics rules and causal structures of the system results in poor interpretation of diagnosis results, and early warning is difficult to realize. Furthermore, diagnostic models are often designed for specific building or fixed conditions, with inadequate generalization and adaptation capabilities, and significant performance degradation in the face of new equipment, new environments, or data islands. Disclosure of Invention The application provides a building electrical system fault diagnosis method and system, which can model the system dynamic and causal relationship in a unified and internal mode to realize accurate, interpretable and self-adaptive diagnosis of complex faults. In a first aspect, the present application provides a method of building electrical system fault diagnosis. A building electrical system fault diagnosis method comprises the following steps of obtaining multi-mode time series data of a plurality of monitoring nodes in a building electrical system and physical topological connection relations among the monitoring nodes, inputting the multi-mode time series data and the physical topological connection relations into a unified diagnosis model for processing, wherein the unified diagnosis model is configured to synchronously perform node state evolution dynamics modeling and inter-node causal interaction modeling in an abstract mathematical space constructed according to a system state, and determining a fault diagnosis result of the building electrical system according to dynamics evolution information and causal interaction information output by the unified diagnosis model. By adopting the technical scheme, the system and the method synchronously describe the state evolution of the system and the causal interaction among the nodes in an abstract mathematical space by constructing a unified diagnosis model. The integrated modeling mode breaks the limitation of separation of dynamics analysis and causal inference in the traditional method, so that the model can learn more essential system operation rules and fault propagation mechanisms from data, thereby laying a theoretical foundation for realizing more accurate and more interpretable fault diagnosis. Further, the abstract mathematical space is a Riemann manifold, and the node state evolution dynamics modeling and the inter-node causal interaction modeling are integrally described on the Riemann manifold. By adopting the technical scheme, the Riemann manifold which is a mathematical space with rich geometric structures is adopted, and a proper framework is provided for describing the dynamic and causal relationship of a complex nonlinear system. The integrated description is carried out on manifold, so that the geometrical consistency and coordination of dynamics and causal models are ensured, and fault analysis can be based on solid geometrical theory. Further, the integrated description is realized through a manifold control differential equation, for any node on the Riemann manifold, the state change rate of the node is determined by a first neural network according to the current state of the node and causal influence contribution amounts from other nodes, wherein the causal influence contribution amounts from any other node are obtained through