Building automatic safety monitoring method and early warning device

Abstract

The invention relates to an automatic safety monitoring method and early warning device for a building, and the method comprises the steps: determining an arrangement optimization point location of a sensor through employing a finite element analysis method based on a building structure diagram and historical monitoring data; arranging sensors according to the optimized point locations, and collecting monitoring data of the building; the monitoring data is preprocessed; training a deep residual network by using the preprocessed monitoring data to obtain a prediction model; and monitoring the building in real time by using the prediction model, obtaining dangerous building state probability distribution, and triggering early warning. According to the invention, intelligent, precise and efficient building monitoring can be realized, and the system is of great significance to guarantee building safety and social stability.

Description

Technical Field

[0001] The present invention relates to the technical field of building safety monitoring, and in particular to an automated building safety monitoring method and an early warning device. Background Art

[0002] Building safety monitoring is a critical component of building security. Traditional building monitoring methods rely heavily on manual inspections or simple sensor deployments, resulting in limited coverage, inefficient data analysis, and insufficient early warning accuracy. These methods often struggle to cope with complex building structures and dynamic environmental changes, leading to unreliable monitoring results and insensitive early warning mechanisms. In building safety monitoring, sensor placement, data feature selection, model parameter optimization, and early warning threshold setting are key technical factors influencing system performance. Due to the complexity of building structures and the diversity of monitored data, it is difficult to cover critical areas with a limited number of sensor locations, resulting in insufficient data representativeness. Furthermore, massive amounts of monitoring data contain a large amount of redundant information, and inappropriate feature selection can reduce model efficiency. Furthermore, neural network training is prone to local optima and struggles to adapt to the complex nonlinear characteristics of building conditions. Failure to balance the risk of missed alerts and false alarms in early warning threshold setting will directly impact system reliability. These technical challenges collectively hinder the development of intelligent and precise building monitoring systems. Therefore, optimizing sensor placement, selecting key features, improving neural network training, and setting appropriate early warning thresholds to achieve efficient and reliable building safety monitoring remain key challenges that urgently need to be addressed. Summary of the Invention

[0003] The purpose of the present invention is to provide a building automation safety monitoring method and early warning device to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for automated building safety monitoring, comprising:

[0006] Based on the building structure diagram and historical monitoring data, the finite element analysis method is used to determine the optimal placement of sensors;

[0007] Arrange sensors at optimized locations to collect monitoring data of the building; wherein the monitoring data includes stress, displacement, and vibration frequency;

[0008] Preprocessing the monitoring data;

[0009] Using the preprocessed monitoring data, the deep residual network is trained to obtain the prediction model;

[0010] The prediction model is used to monitor buildings in real time, obtain the probability distribution of dangerous building status, and trigger early warning.

[0011] Optionally, determining the optimal placement of sensors includes:

[0012] Obtain geometric information and material properties from the building structure drawing, construct a finite element analysis model, and obtain an initial analysis model;

[0013] Process historical monitoring data to extract deformation trends and stress distribution characteristics, and determine reference stress values and deformation thresholds;

[0014] Finite element analysis is used to input reference stress values and deformation thresholds into the initial analysis model, calculate the structural stress distribution and deformation sensitive points, and obtain the coordinates of stress concentration areas and sensitive points.

[0015] If the stress value in the stress concentration area exceeds the preset threshold, a preliminary sensor layout plan is generated based on the coordinates of the sensitive points to determine the candidate monitoring points;

[0016] According to the candidate monitoring points and building structure diagram, the genetic algorithm is used to optimize the sensor positions, calculate the coverage rate and monitoring efficiency between points, and obtain the optimized sensor positions.

[0017] Optionally, preprocessing the monitoring data includes:

[0018] Using principal component analysis to extract high-variance feature vectors from the monitoring data, eliminating redundant dimensions, and obtaining an initial feature set;

[0019] If the dimension of the initial feature set is lower than the preset threshold, the feature importance is evaluated by the random forest algorithm, the preset key features are supplemented, and the final feature set is determined.

[0020] Optionally, using the preprocessed monitoring data, training a deep residual network includes:

[0021] The particle swarm optimization algorithm is used to adjust the hyperparameters of the deep residual network based on the preprocessed monitoring data to obtain the optimized parameter configuration;

[0022] By optimizing the parameter configuration, initializing the deep residual network, and inputting the preprocessed monitoring data for model training, the original prediction model is obtained;

[0023] Based on the original prediction model, obtain the prediction results of the model on the preset validation set;

[0024] If the error of the prediction result is higher than the preset threshold, the learning rate is adjusted and the model is retrained to obtain an updated prediction model;

[0025] For the updated prediction model, extract the weight distribution information of the model, and use the weight distribution information to remove low-contribution neurons to obtain a streamlined network structure;

[0026] A streamlined network structure is used, combined with preprocessed monitoring data, to perform final model training and obtain the final prediction model.

[0027] The present invention also provides the following solution:

[0028] An automated building safety monitoring and early warning system, the system comprising:

[0029] The point determination module is used to determine the optimal point locations for sensor layout based on the building structure diagram and historical monitoring data using the finite element analysis method;

[0030] A data acquisition module is used to arrange sensors according to the optimized points to collect monitoring data of the building; wherein the monitoring data includes: stress, displacement, and vibration frequency;

[0031] A preprocessing module, used for preprocessing the monitoring data;

[0032] The model training module is used to train the deep residual network using the preprocessed monitoring data to obtain the prediction model;

[0033] The real-time monitoring module is used to use the prediction model to monitor buildings in real time, obtain the probability distribution of dangerous building status, and trigger early warning.

[0034] Optionally, the point determination module includes:

[0035] A construction unit is used to obtain geometric information and material properties from the building structure drawing, construct a finite element analysis model, and obtain an initial analysis model;

[0036] An extraction unit is used to process historical monitoring data, extract deformation trends and stress distribution characteristics, and determine reference stress values and deformation thresholds;

[0037] A calculation unit is used to use the finite element analysis method to input the reference stress value and deformation threshold for the initial analysis model, calculate the structural stress distribution and deformation sensitive points, and obtain the stress concentration area and sensitive point coordinates;

[0038] A judgment unit is used to generate a preliminary sensor layout plan and determine candidate monitoring points based on the coordinates of the sensitive points if the stress value of the stress concentration area exceeds a preset threshold;

[0039] The optimization unit is used to optimize the sensor position according to the candidate monitoring points and the building structure diagram using a genetic algorithm, calculate the coverage rate and monitoring efficiency between points, and obtain the optimized sensor position.

[0040] Optionally, the preprocessing module includes:

[0041] A principal component analysis unit is used to extract high-variance feature vectors from the monitoring data using a principal component analysis method, eliminate redundant dimensions, and obtain an initial feature set;

[0042] The evaluation unit is used to evaluate the feature importance by using a random forest algorithm if the dimension of the initial feature set is lower than a preset threshold, supplement the preset key features, and determine the final feature set.

[0043] Optionally, the model training module includes:

[0044] A parameter acquisition unit is used to adjust the hyperparameters of the deep residual network based on the preprocessed monitoring data using a particle swarm optimization algorithm to obtain an optimized parameter configuration;

[0045] The first training unit is used to initialize the deep residual network by optimizing parameter configuration, input preprocessed monitoring data for model training, and obtain the original prediction model;

[0046] The verification unit is used to obtain the prediction results of the model on the preset verification set based on the original prediction model;

[0047] The second training unit is used to adjust the learning rate and retrain the model to obtain an updated prediction model if the error of the prediction result is higher than a preset threshold;

[0048] The model simplification unit is used to extract the weight distribution information of the updated prediction model, and remove low-contribution neurons based on the weight distribution information to obtain a streamlined network structure;

[0049] The third training unit is used to use a streamlined network structure and combine it with the preprocessed monitoring data to perform final model training and obtain the final prediction model.

[0050] The beneficial effects of the present invention are:

[0051] The present invention optimizes the sensor layout through finite element analysis and genetic algorithms, accurately covers key areas of buildings, and improves the representativeness and efficiency of monitoring data. After real-time monitoring data is processed to ensure data quality, it is then subjected to principal component analysis and random forest algorithm to extract key features, ensuring that the data is streamlined and the information is complete, thereby improving the efficiency of model training. By optimizing and adjusting the hyperparameters of the deep residual network using the particle swarm optimization algorithm, a more suitable hyperparameter combination can be found, improving the training efficiency and convergence speed of the model, and enabling the model to reach a better performance level more quickly. By training the deep residual network with optimized hyperparameters, a stable prediction model is obtained, which can more accurately predict the state probability distribution of the building, provide a more reliable basis for subsequent early warning and risk assessment, and effectively improve the intelligence level and prediction accuracy of the entire monitoring system. Overall, the present invention can realize intelligent, precise and efficient building monitoring, which is of great significance to ensuring building safety and social stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 The figure is a flow chart of a method for automated building safety monitoring according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] Example 1

[0057] like Figure 1 As shown, this embodiment proposes a building automation safety monitoring method, including:

[0058] Based on the building structure diagram and historical monitoring data, the finite element analysis method is used to determine the optimal placement of sensors;

[0059] Arrange sensors at optimized locations to collect building monitoring data, including stress, displacement, and vibration frequency.

[0060] Preprocessing of monitoring data;

[0061] Using the preprocessed monitoring data, the deep residual network is trained to obtain the prediction model;

[0062] Using the prediction model, buildings are monitored in real time, the probability distribution of dangerous building status is obtained, and early warnings are triggered.

[0063] Furthermore, determining the optimal placement of sensors includes:

[0064] Obtain geometric information and material properties from the building structure drawing, construct a finite element analysis model, and obtain an initial analysis model;

[0065] Process historical monitoring data to extract deformation trends and stress distribution characteristics, and determine reference stress values and deformation thresholds;

[0066] Finite element analysis is used to input reference stress values and deformation thresholds into the initial analysis model, calculate the structural stress distribution and deformation sensitive points, and obtain the coordinates of stress concentration areas and sensitive points.

[0067] If the stress value in the stress concentration area exceeds the preset threshold, a preliminary sensor layout plan is generated based on the coordinates of the sensitive points to determine the candidate monitoring points;

[0068] According to the candidate monitoring points and building structure diagram, the genetic algorithm is used to optimize the sensor positions, calculate the coverage rate and monitoring efficiency between points, and obtain the optimized sensor positions.

[0069] More specifically, in this embodiment, optimizing the sensor positions using a genetic algorithm includes:

[0070] Set algorithm parameters: determine genetic algorithm parameters such as population size, crossover probability, and mutation probability;

[0071] Initialize the population: Generate an initial population based on the candidate monitoring points, where each individual represents a sensor layout scheme;

[0072] Calculate the fitness function: Construct a fitness function with the coverage rate and monitoring efficiency between points as key indicators to evaluate the pros and cons of each individual (sensor layout scheme);

[0073] Screening excellent individuals: Screening out better individuals based on fitness values to form the parent population;

[0074] Crossover and mutation operations: Perform crossover and mutation operations on the parent population to generate a new offspring population;

[0075] Repeated iteration: Repeat the process of calculating fitness, screening and generating new populations until the number of iterations or fitness convergence conditions are met;

[0076] Select the best individual: Select the individual with the highest fitness from all iteratively generated populations and determine the optimized sensor position.

[0077] This embodiment uses finite element analysis and genetic algorithms to precisely determine the optimal sensor placement. This ensures maximum coverage of critical building areas, such as stress concentration zones and deformation-sensitive points, with a limited number of sensors. This improves the representativeness of monitoring data and lays the foundation for subsequent accurate assessment of building conditions. This avoids placing too many sensors in non-critical areas, reduces redundant monitoring points, and reduces the burden of data collection and processing, thereby improving the efficiency of the entire monitoring system and saving resources and costs.

[0078] Specifically, in this embodiment, collecting monitoring data of a building includes:

[0079] Real-time data streams are acquired from sensor locations to extract stress values, displacements, and vibration frequencies to generate an initial monitoring dataset. Data stream processing techniques are used to perform time series analysis on the initial monitoring dataset to determine the dynamic trends in key areas. If the dynamic trends exceed preset thresholds, the distribution of abnormal points is calculated for the stress values, displacements, and vibration frequencies in the key areas to determine the coordinates of the abnormal areas. Based on the coordinates of the abnormal areas and the sensor locations, a clustering algorithm is used to adjust the optimal point layout and generate an updated sensor layout. Using the updated sensor layout, new real-time data streams are acquired, a coverage dataset is extracted, and the regional coverage rate is determined.

[0080] According to the optimized sensor arrangement scheme, this embodiment can obtain key data such as stress, displacement, vibration frequency of dangerous buildings in real time and accurately, forming a complete real-time monitoring data stream, and providing data support for timely detection of abnormal conditions of dangerous buildings.

[0081] Furthermore, preprocessing of monitoring data includes:

[0082] For the monitoring data, the principal component analysis method is used to extract high-variance feature vectors, eliminate redundant dimensions, and obtain the initial feature set;

[0083] If the dimension of the initial feature set is lower than the preset threshold, the feature importance is evaluated through the random forest algorithm, the preset key features are supplemented, and the final feature set is determined.

[0084] This example uses principal component analysis to extract high-variance feature vectors from a large amount of monitoring data, removing redundant dimensions to obtain a streamlined feature set. This not only reduces the amount of data and lowers data storage and transmission costs, but also improves the efficiency of subsequent data processing and model training. It also retains the most important information in the data, helping to more accurately identify the key features of dangerous buildings. This dimensionality-reduced data effectively addresses the "curse of dimensionality" problem associated with high-dimensional data, making subsequent model training more efficient, improving the model's performance in predicting and assessing dangerous building conditions, reducing the risk of overfitting, and enhancing the model's generalization capabilities.

[0085] When the streamlined feature set lacks sufficient dimensions, a random forest algorithm is used to assess feature importance, supplement key features, and further refine the feature set to ensure it comprehensively and accurately describes the status characteristics of dilapidated buildings, avoiding inaccurate model predictions due to missing features. After multiple rounds of screening and verification, low-importance features are removed while highly stable features are retained. This ensures that the final feature set performs well across different situations and data samples, improving its reliability and stability and providing higher-quality input for subsequent model training and prediction.

[0086] Furthermore, using the preprocessed monitoring data, training the deep residual network includes:

[0087] The particle swarm optimization algorithm is used to adjust the hyperparameters of the deep residual network based on the preprocessed monitoring data to obtain the optimized parameter configuration;

[0088] By optimizing the parameter configuration, initializing the deep residual network, and inputting the preprocessed monitoring data for model training, the original prediction model is obtained;

[0089] Based on the original prediction model, obtain the prediction results of the model on the preset validation set;

[0090] If the error of the prediction result is higher than the preset threshold, the learning rate is adjusted and the model is retrained to obtain an updated prediction model;

[0091] For the updated prediction model, extract the weight distribution information of the model, and use the weight distribution information to remove low-contribution neurons to obtain a streamlined network structure;

[0092] A streamlined network structure is used, combined with preprocessed monitoring data, to perform final model training and obtain the final prediction model.

[0093] Specifically, in this embodiment, based on the final feature set, the particle swarm optimization algorithm is used to optimize the hyperparameters of the deep residual network to improve model performance. First, the particle swarm size is initialized to 30, the maximum number of iterations is set to 100, and the search space is defined including the learning rate (0.001 to 0.1), batch size (32 to 256), and the number of hidden layer neurons (64 to 512). The fitness value of each particle is calculated by the fitness function, which uses the mean square error (MSE) as the evaluation index, and the initial fitness value is 0.05. During the iteration process, the speed and position of the particle are updated, and the inertia weight linear decrease strategy is adopted, with an initial weight of 0.9 and a final weight of 0.4, to ensure that the search process gradually shifts from global exploration to local optimization. After 50 iterations, the optimal hyperparameter combination is found: the learning rate is 0.01, the batch size is 128, and the number of hidden layer neurons is 256, at which time the fitness value drops to 0.02. Subsequently, a deep residual network was constructed based on the optimized hyperparameters. The network structure consists of five residual blocks, each containing two convolutional layers and one skip connection. Reluctant Unit (ReLU) was used as the activation function, and cross-entropy was chosen as the loss function. During training, the Adam optimizer was used, with an initial learning rate of 0.01 and 50 training rounds. The accuracy on the validation set was calculated after each round of training. The final model achieved an accuracy of 96.5% on the validation set, demonstrating high predictive stability.

[0094] Specifically, in this embodiment, the probability distribution of the dangerous building status is obtained through the prediction model to obtain risk probability data. If the risk probability data exceeds the preset dynamic threshold, the Bayesian update method is used to adjust the threshold to obtain the updated threshold parameters. According to the updated threshold parameters, it is determined whether the dangerous building status meets the warning trigger conditions, and the warning status is determined. Through the warning status, the time series data of the dangerous building status is obtained to obtain the status change trend. With respect to the status change trend, the sliding window method is used to extract the trend features to obtain the trend feature set. Through the trend feature set, the pre-trained classification model is input to determine the future risk level of the dangerous building status and obtain the risk level result. According to the risk level result, the monitoring frequency of the dangerous building status is updated, and the monitoring adjustment parameters are determined.

[0095] This embodiment uses a particle swarm optimization algorithm to optimize and adjust the hyperparameters of the deep residual network. This algorithm can find a more appropriate hyperparameter combination, improve the model's training efficiency and convergence speed, and enable the model to reach a good performance level more quickly. By training the deep residual network with optimized hyperparameters, a stable prediction model is obtained. This model can more accurately predict the state probability distribution of dangerous buildings, providing a more reliable basis for subsequent early warning and risk assessment, effectively improving the intelligence level and prediction accuracy of the entire monitoring system.

[0096] Example 2

[0097] This embodiment provides an automated building safety monitoring and early warning device, including:

[0098] The point determination module is used to determine the optimal point locations for sensor layout based on the building structure diagram and historical monitoring data using the finite element analysis method;

[0099] The data acquisition module is used to arrange sensors according to the optimized points to collect monitoring data of the building; the monitoring data includes stress, displacement, and vibration frequency;

[0100] Preprocessing module, used for preprocessing monitoring data;

[0101] The model training module is used to train the deep residual network using the preprocessed monitoring data to obtain the prediction model;

[0102] The real-time monitoring module is used to use the prediction model to monitor buildings in real time, obtain the probability distribution of dangerous building status, and trigger early warning.

[0103] Furthermore, the point determination module includes:

[0104] A construction unit is used to obtain geometric information and material properties from the building structure drawing, construct a finite element analysis model, and obtain an initial analysis model;

[0105] An extraction unit is used to process historical monitoring data, extract deformation trends and stress distribution characteristics, and determine reference stress values and deformation thresholds;

[0106] A calculation unit is used to use the finite element analysis method to input the reference stress value and deformation threshold for the initial analysis model, calculate the structural stress distribution and deformation sensitive points, and obtain the stress concentration area and sensitive point coordinates;

[0107] A judgment unit is used to generate a preliminary sensor layout plan and determine candidate monitoring points based on the coordinates of the sensitive points if the stress value of the stress concentration area exceeds a preset threshold;

[0108] The optimization unit is used to optimize the sensor position according to the candidate monitoring points and the building structure diagram using a genetic algorithm, calculate the coverage rate and monitoring efficiency between points, and obtain the optimized sensor position.

[0109] Furthermore, the preprocessing module includes:

[0110] The principal component analysis unit is used to extract high-variance feature vectors from the monitoring data using the principal component analysis method, eliminate redundant dimensions, and obtain the initial feature set;

[0111] The evaluation unit is used to evaluate the feature importance through the random forest algorithm if the dimension of the initial feature set is lower than the preset threshold, supplement the preset key features, and determine the final feature set.

[0112] Furthermore, the model training module includes:

[0113] A parameter acquisition unit is used to adjust the hyperparameters of the deep residual network based on the preprocessed monitoring data using a particle swarm optimization algorithm to obtain an optimized parameter configuration;

[0114] The first training unit is used to initialize the deep residual network by optimizing parameter configuration, input preprocessed monitoring data for model training, and obtain the original prediction model;

[0115] The verification unit is used to obtain the prediction results of the model on the preset verification set based on the original prediction model;

[0116] The second training unit is used to adjust the learning rate and retrain the model to obtain an updated prediction model if the error of the prediction result is higher than a preset threshold;

[0117] The model simplification unit is used to extract the weight distribution information of the updated prediction model, and remove low-contribution neurons based on the weight distribution information to obtain a streamlined network structure;

[0118] The third training unit is used to use a streamlined network structure and combine it with the preprocessed monitoring data to perform final model training and obtain the final prediction model.

[0119] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for automated building safety monitoring, characterized in that: include: Based on the building structure diagram and historical monitoring data, the finite element analysis method is used to determine the optimal placement of sensors; Arrange sensors at optimized locations to collect monitoring data of the building; wherein the monitoring data includes stress, displacement, and vibration frequency; Preprocessing the monitoring data; Using the preprocessed monitoring data, the deep residual network is trained to obtain the prediction model; The prediction model is used to monitor buildings in real time, obtain the probability distribution of dangerous building status, and trigger early warning.

2. The building automation safety monitoring method according to claim 1, characterized in that: Determining the optimal placement of sensors includes: Obtain geometric information and material properties from the building structure drawing, construct a finite element analysis model, and obtain an initial analysis model; Process historical monitoring data to extract deformation trends and stress distribution characteristics, and determine reference stress values and deformation thresholds; Finite element analysis is used to input reference stress values and deformation thresholds into the initial analysis model, calculate the structural stress distribution and deformation sensitive points, and obtain the coordinates of stress concentration areas and sensitive points. If the stress value in the stress concentration area exceeds the preset threshold, a preliminary sensor layout plan is generated based on the coordinates of the sensitive points to determine the candidate monitoring points; According to the candidate monitoring points and building structure diagram, the genetic algorithm is used to optimize the sensor positions, calculate the coverage rate and monitoring efficiency between points, and obtain the optimized sensor positions.

3. The building automation safety monitoring method according to claim 1, characterized in that: Preprocessing the monitoring data includes: Using principal component analysis to extract high-variance feature vectors from the monitoring data, eliminating redundant dimensions, and obtaining an initial feature set; If the dimension of the initial feature set is lower than the preset threshold, the feature importance is evaluated by the random forest algorithm, the preset key features are supplemented, and the final feature set is determined.

4. The building automation safety monitoring method according to claim 1, characterized in that: Using the preprocessed monitoring data, training the deep residual network includes: The particle swarm optimization algorithm is used to adjust the hyperparameters of the deep residual network based on the preprocessed monitoring data to obtain the optimized parameter configuration; By optimizing the parameter configuration, initializing the deep residual network, and inputting the preprocessed monitoring data for model training, the original prediction model is obtained; Based on the original prediction model, obtain the prediction results of the model on the preset validation set; If the error of the prediction result is higher than the preset threshold, the learning rate is adjusted and the model is retrained to obtain an updated prediction model; For the updated prediction model, extract the weight distribution information of the model, and use the weight distribution information to remove low-contribution neurons to obtain a streamlined network structure; A streamlined network structure is used, combined with preprocessed monitoring data, to perform final model training and obtain the final prediction model.

5. A building automation safety monitoring and early warning device, characterized in that: For implementing the building automation safety monitoring method according to any one of claims 1 to 4, the device comprises: The point determination module is used to determine the optimal point locations for sensor layout based on the building structure diagram and historical monitoring data using the finite element analysis method; A data acquisition module is used to arrange sensors according to the optimized points to collect monitoring data of the building; wherein the monitoring data includes: stress, displacement, and vibration frequency; A preprocessing module, used for preprocessing the monitoring data; The model training module is used to train the deep residual network using the preprocessed monitoring data to obtain the prediction model; The real-time monitoring module is used to use the prediction model to monitor buildings in real time, obtain the probability distribution of dangerous building status, and trigger early warning.

6. The building automation safety monitoring and early warning device according to claim 5, characterized in that: The point determination module includes: A construction unit is used to obtain geometric information and material properties from the building structure drawing, construct a finite element analysis model, and obtain an initial analysis model; An extraction unit is used to process historical monitoring data, extract deformation trends and stress distribution characteristics, and determine reference stress values and deformation thresholds; A calculation unit is used to use the finite element analysis method to input the reference stress value and deformation threshold for the initial analysis model, calculate the structural stress distribution and deformation sensitive points, and obtain the stress concentration area and sensitive point coordinates; A judgment unit is used to generate a preliminary sensor layout plan and determine candidate monitoring points based on the coordinates of the sensitive points if the stress value of the stress concentration area exceeds a preset threshold; The optimization unit is used to optimize the sensor position according to the candidate monitoring points and the building structure diagram using a genetic algorithm, calculate the coverage rate and monitoring efficiency between points, and obtain the optimized sensor position.

7. The building automation safety monitoring and early warning device according to claim 5, characterized in that: The pre-processing module comprises: A principal component analysis unit is used to extract high-variance feature vectors from the monitoring data using a principal component analysis method, eliminate redundant dimensions, and obtain an initial feature set; The evaluation unit is used to evaluate the feature importance by using a random forest algorithm if the dimension of the initial feature set is lower than a preset threshold, supplement the preset key features, and determine the final feature set.

8. The building automation safety monitoring and early warning device according to claim 5, characterized in that: The model training module includes: A parameter acquisition unit is used to adjust the hyperparameters of the deep residual network based on the preprocessed monitoring data using a particle swarm optimization algorithm to obtain an optimized parameter configuration; The first training unit is used to initialize the deep residual network by optimizing parameter configuration, input preprocessed monitoring data for model training, and obtain the original prediction model; The verification unit is used to obtain the prediction results of the model on the preset verification set based on the original prediction model; The second training unit is used to adjust the learning rate and retrain the model to obtain an updated prediction model if the error of the prediction result is higher than a preset threshold; The model simplification unit is used to extract the weight distribution information of the updated prediction model, and remove low-contribution neurons based on the weight distribution information to obtain a streamlined network structure; The third training unit is used to use a streamlined network structure and combine it with the preprocessed monitoring data to perform final model training and obtain the final prediction model.

Images