Search

CN-122022490-A - Building risk monitoring and evaluating method and system based on neural network

CN122022490ACN 122022490 ACN122022490 ACN 122022490ACN-122022490-A

Abstract

The invention discloses a building risk monitoring and evaluating method and system based on a neural network, and relates to the technical field of building safety, wherein the method comprises the steps of utilizing a sensor network deployed on a building structure to collect multi-source time sequence data in real time; the method comprises the steps of inputting collected multi-source time sequence data into a pre-trained cooperative attention neural network encoder, outputting a building state coding sequence enhanced by physical information and a space risk weight matrix, predicting building comprehensive risk indexes in a specific time window in the future based on the building state coding sequence, and generating a safety evaluation report according to a prediction result of the building comprehensive risk indexes and the space risk weight matrix. According to the invention, the multi-source sensor data and the physical priori knowledge are deeply fused through the cooperative attention mechanism, so that the accuracy of risk assessment is remarkably improved, and the problem that the traditional method cannot trace the source is solved.

Inventors

  • ZHANG LIANG

Assignees

  • 北京华邑建设集团有限公司

Dates

Publication Date
20260512
Application Date
20260319

Claims (10)

  1. 1. A neural network-based building risk monitoring and assessment method, comprising: Step1, acquiring multi-source time sequence data in real time by using a sensor network deployed on a building structure; Step2, inputting the acquired multi-source time sequence data into a pre-trained cooperative attention neural network encoder, and outputting a building state coding sequence enhanced by physical information and a space risk weight matrix; Step3, predicting a building comprehensive risk index in a future specific time window based on the building state coding sequence; step4, generating a safety evaluation report according to the prediction result of the building comprehensive risk index and the space risk weight matrix.
  2. 2. The building risk monitoring and evaluating method based on the neural network according to claim 1, wherein the multi-source time sequence data is collected in real time by using a sensor network deployed on a building structure, and specifically comprises the following sub-steps: distributing a unique space coordinate identifier for each sensor node; synchronously acquiring sensing data of all nodes in a sensor network, and transmitting the sensing data to an edge computing gateway in real time; The edge computing gateway processes the received original data into multi-source time sequence data packets and forwards the multi-source time sequence data packets to the central processing server.
  3. 3. The building risk monitoring and evaluating method based on the neural network according to claim 2, wherein the cooperative attention neural network encoder is a three-layer architecture comprising a space-time feature extraction layer, a physical information weighting layer and a state coding layer, and the specific data processing flow is as follows: Grouping the received multi-source time sequence data according to the sensor type, and respectively inputting the multi-source time sequence data into a space-time feature extraction layer to obtain data feature vectors; injecting physical priori knowledge into a physical information weighting layer, and executing a physical enhancement flow of the data feature vector; Compressing the enhanced data feature vector into a final building state code by using a state code layer; Repeatedly executing the steps on the multi-source time sequence data in the continuous time window to obtain a building state coding sequence, and outputting the building state coding sequence and a space risk weight matrix generated in a physical enhancement process.
  4. 4. The building risk monitoring and evaluating method based on the neural network according to claim 3, wherein the processing flow of the space-time feature extraction layer is as follows: Converting each node data sequence under the current type into an embedded matrix, and obtaining the characteristic vector of each node after average pooling; calculating importance weights of all nodes by combining the historical failure frequencies of the sensors; and weighting and summing the feature vectors of all the nodes by using the importance weights to obtain the data feature vector of the current sensor type.
  5. 5. The building risk monitoring and evaluating method based on the neural network according to claim 3, wherein the physical enhancement flow of the physical information weighting layer is as follows: calculating the integrity characteristics, material performance characteristics and environmental action characteristics of each building structure according to the physical priori knowledge; the features are fused into a physical context vector of a structural level through a full connection layer; calculating the attention weight of the physical context vector of the structure to each sensor characteristic on the structure; Weighting and fusing the sensor characteristics on the structure by using the attention weight to obtain a data characteristic vector with the physical information of the structure level enhanced; and splicing all the enhanced feature vectors, and then aggregating through a full connection layer to obtain the final data feature vector enhanced by the global physical information.
  6. 6. The method for monitoring and evaluating building risks based on neural network according to claim 5, wherein during the physical enhancement process, the physical information weighting layer further records the attention weight of each structure in real time, and forms a global spatial risk weight matrix after summarizing and normalizing.
  7. 7. The method for monitoring and evaluating building risk based on neural network according to claim 3, wherein the model for predicting the building integrated risk index is formed by mixing a time sequence convolution module and a gating circulation unit.
  8. 8. The method for monitoring and evaluating the risk of the building based on the neural network according to claim 3, wherein the safety evaluation report is generated according to the prediction result of the comprehensive risk index of the building and the space risk weight matrix, and the method comprises the following steps: Dividing risk grades according to the predicted building comprehensive risk indexes; performing risk tracing positioning based on the space risk weight matrix; Diagnosing a risk evolution mode by combining a historical building comprehensive risk index sequence; and filling the security assessment report by utilizing the content.
  9. 9. The building risk monitoring and evaluating system based on the neural network is characterized by comprising a sensing data acquisition module, a sensing data coding module, a comprehensive risk prediction module and an evaluation report generation module; The sensing data acquisition module is used for acquiring multi-source time sequence data in real time by utilizing a sensor network deployed on a building structure; The sensing data coding module is used for inputting the acquired multi-source time sequence data into the pre-trained cooperative attention neural network coder and outputting a building state coding sequence enhanced by physical information and a space risk weight matrix; the comprehensive risk prediction module is used for predicting a building comprehensive risk index in a future specific time window based on the building state coding sequence; And the evaluation report generation module is used for generating a safety evaluation report according to the prediction result of the building comprehensive risk index and the space risk weight matrix.
  10. 10. A computer storage medium includes at least one memory and at least one processor; a memory for storing one or more program instructions; A processor for executing one or more program instructions for performing a neural network-based building risk monitoring and assessment method as claimed in any one of claims 1 to 8.

Description

Building risk monitoring and evaluating method and system based on neural network Technical Field The invention relates to the technical field of building safety, in particular to a building risk monitoring and evaluating method and system based on a neural network. Background Building structures, particularly large public buildings, historic buildings, and buildings in complex geological environments or extreme climatic regions, are subject to multiple risks of structural aging, material property degradation, natural disasters (such as earthquakes, strong winds, storms), sustained loads, and incidental actions over their entire life cycle. The real-time and accurate monitoring and prospective evaluation of the risks are key technical requirements for guaranteeing the life and property safety of people and maintaining social stability. Currently, the mainstream technologies and methods in the field of building risk monitoring and assessment mainly depend on the following modes, but they all have significant limitations: 1. the traditional manual regular inspection and visual inspection method is highly dependent on experience and subjective judgment of detection personnel, lacks unified and quantitative judgment standards, is difficult to find for small, concealed or early-stage damage (such as internal cracks and steel bar initial rust), is easy to cause missed inspection, and often has serious consequences when the damage is obvious; 2. an automatic monitoring system based on a traditional sensor is characterized in that a plurality of sensors (such as strain gauges, accelerometers and inclinometers) are usually deployed in the system, massive real-time data can be acquired, however, the data are often isolated and unilateral, only the local response (such as strain or vibration at a certain point) of a structure is reflected, the system lacks effective data fusion and comprehensive analysis capability, and comprehensive evaluation indexes reflecting the overall safety state of the structure are difficult to extract from multi-source heterogeneous data; 3. Some researches try to apply traditional machine learning algorithms (such as a support vector machine and a decision tree) or statistical analysis to carry out damage identification based on a simple machine learning model or a statistical method, however, the methods are generally limited in capability of processing complex, high-dimensional and nonlinear data, deep and abstract characteristic modes in mass monitoring data are difficult to fully mine, and the prediction precision and generalization capability of the risk evolution rule of a complex system of a building under the multi-factor coupling effect often cannot meet the requirements of engineering application. In summary, the prior art generally has core pain points such as insufficient intelligence level, low evaluation precision and reliability, and incapability of realizing intelligent crossing from data to decision. The field of building safety management is in urgent need of a novel monitoring and evaluating method capable of automatically, accurately fusing multi-source information in real time and having strong nonlinear mapping and prediction capability so as to realize the fundamental transition from passive response to active early warning. Disclosure of Invention The invention provides a building risk monitoring and evaluating method based on a neural network, which comprises the following steps: Step1, acquiring multi-source time sequence data in real time by using a sensor network deployed on a building structure; Step2, inputting the acquired multi-source time sequence data into a pre-trained cooperative attention neural network encoder, and outputting a building state coding sequence enhanced by physical information and a space risk weight matrix; Step3, predicting a building comprehensive risk index in a future specific time window based on the building state coding sequence; step4, generating a safety evaluation report according to the prediction result of the building comprehensive risk index and the space risk weight matrix. The building risk monitoring and evaluating method based on the neural network, which is described above, acquires multi-source time sequence data in real time by using a sensor network deployed on a building structure, and specifically comprises the following sub-steps: distributing a unique space coordinate identifier for each sensor node; synchronously acquiring sensing data of all nodes in a sensor network, and transmitting the sensing data to an edge computing gateway in real time; The edge computing gateway processes the received original data into multi-source time sequence data packets and forwards the multi-source time sequence data packets to the central processing server. The building risk monitoring and evaluating method based on the neural network, wherein the cooperative attention neural network encoder is a three-layer architecture compri