House safety monitoring method and system based on machine learning

Abstract

The invention discloses a house safety monitoring method and system based on machine learning, and belongs to the technical field of machine learning.The house safety monitoring method and system based on machine learning comprise the following specific steps that firstly, a plurality of sensors are deployed at key parts of a house, and the sensors comprise a strain sensor, a displacement sensor and an acceleration sensor; collecting physical parameter data of the house structure in real time, and storing the data according to a time sequence; step 2, carrying out cleaning and normalization processing on the collected data; 3, key features reflecting the health state of the house structure are mined from the preprocessed data. By means of the powerful data processing capacity of a machine learning algorithm, the system can carry out deep analysis on data from various high-precision sensors, hidden danger omission caused by human negligence or detection means limitation is effectively avoided, and omnibearing and dead-corner-free guarantee is provided for house safety.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to a housing safety monitoring method and system based on machine learning. Background Art

[0002] As urbanization is advancing rapidly, housing construction is springing up like mushrooms after a rain, and its safety has become an increasingly important priority. Traditional housing safety monitoring methods mostly rely on manual regular inspections, which not only requires a lot of manpower and material costs, but also has significant shortcomings in terms of timeliness. Many potential safety hazards, such as the quiet extension of initial fine cracks in the wall and the slow and hidden settlement of the foundation, are extremely difficult to be detected by manual inspections in the early stages. At the same time, although some existing monitoring equipment has certain data collection capabilities, it lacks intelligent data in-depth analysis capabilities, making it difficult to accurately judge the complex and changeable real-time status of houses.

[0003] Therefore, there is an urgent need for an efficient and intelligent housing safety monitoring method. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a house safety monitoring method and system based on machine learning.

[0005] The technical solution adopted to solve the above technical problems is: a housing safety monitoring method based on machine learning, including the following specific steps: Step 1: Deploy several sensors at key locations of the house, including strain sensors, displacement sensors, and acceleration sensors, to collect physical parameter data of the house structure in real time and store them in time series; Step 2: Clean and normalize the collected data; Step 3: Mining key features reflecting the health status of the building structure from the preprocessed data; Step 4: Use the machine learning model to train with the extracted features as input and the actual safety status of the house as output label; Step 5: Input the new data collected and processed in real time into the trained model, issue warnings based on the model output, send alarms to relevant personnel, and upload data to the cloud server.

[0006] Through the above technical solution, the sensor continuously collects various physical parameter data of the house structure at a preset high-frequency sampling rate of 20 times per second, covering the tiny strain of the wall, the displacement changes of beams and columns, the vibration acceleration of the entire building, the pressure difference between indoors and outdoors, and the potential impact of air humidity on building materials, and stores it in a precise time series to provide the original data basis for subsequent analysis.

[0007] Furthermore, the feature extraction in step 3 adopts the following formula: Discrete Wavelet Transform: Where x[n] is the original signal, φj0,k[n] and ψj,k[n] are the scaling function and wavelet function respectively, M and N are the length of the transform and the length of the signal, j0 and j are the decomposition scales, and k is the translation parameter; Use soft threshold to make judgment: Among them, wj,k is the wavelet coefficient, T is the threshold, and sgn is the sign function. By analyzing the transformed coefficients, the time-frequency characteristics of the signal at different scales are extracted. These characteristics can reflect the dynamic response characteristics of the building structure at different frequency components, and then reflect the health status of the structure.

[0008] Furthermore, the data cleaning in step 2 adopts a Butterworth low-pass filter, and appropriate filtering parameters are set according to the characteristics of different sensor data to accurately remove high-frequency noise components and restore the real change trend of the data.

[0009] Through the above technical solution, data accuracy can be improved and additional interference can be avoided. Furthermore, the machine learning in step 4 adopts a neural network model training formula. The details are as follows: Linear combinations: In housing security monitoring, the input data includes data collected from various sensors and image acquisition devices. After preprocessing and feature extraction, the data will serve as the input of the neural network a0= x; For house safety monitoring, different types of data will have different feature dimensions, which will be linearly combined through the weight matrix wl and bias vector bl of the first layer; Then activate it through the ReLU function: ReLU(z) = max(0, z) The ReLU function is linear on the positive semi-axis, which helps to speed up convergence and avoid gradient disappearance; in, is the total number of samples, representing the number of monitored houses; is the total number of categories, representing the safety level classification of the house, It is The samples belong to The true label of the category, if The true category of the sample is Class, then ;otherwise ; The model predicts The samples belong to The probability of the category, this probability value is calculated by the model through forward propagation and satisfies Inner summation The cross entropy loss is calculated for a single sample, and the outer layer is summed The cross entropy losses of all samples are added up and finally divided by the total number of samples to get the average cross entropy loss, which can more fairly measure the performance of the model on the entire dataset; the house safety is divided into multiple categories including safety, mild hidden dangers, moderate hidden dangers, and severe hidden dangers. The cross entropy loss is used. When the actual state of the house is a moderate hidden danger and the probability distribution predicted by the model is inaccurate, the cross entropy loss will be larger, which will drive the model to update parameters and make the prediction results closer to the actual state.

[0010] Furthermore, it includes a sensor network, a data processing center, an early warning module, a communication module and a cloud server. The sensor network is composed of multiple sensor nodes distributed throughout the house, which are responsible for collecting physical parameter data of the house. Each sensor node has low power consumption, high precision, and self-organizing networking capabilities, and can stably and reliably transmit data to the data processing center.

[0011] Furthermore, the data processing center is responsible for being equipped with high-performance processors, undertaking the core tasks of data preprocessing, feature extraction, and model operation, and having a built-in data storage unit for storing raw data, preprocessed data, and intermediate data during model training.

[0012] Furthermore, the early warning module is connected to the data processing center. According to the housing safety status output by the model, when an early warning is needed, the corresponding alarm method is triggered according to the preset rules to ensure that the information is conveyed in a timely manner; The communication module includes wired communication and wireless communication components to ensure high-speed data transmission between the sensor network, data processing center, early warning module and cloud server, and realize remote monitoring and management; The cloud server is used to store massive housing monitoring historical data and support big data analysis. On the one hand, it can provide more data support for local model training, and on the other hand, it can be used by different regions. Remote access by different domains and management departments can be realized to realize regional and coordinated management of housing safety monitoring.

[0013] The beneficial effects of the present invention are as follows: By leveraging the powerful data processing capabilities of machine learning algorithms, the system can perform in-depth analysis of data from a variety of high-precision sensors. The features automatically extracted by the deep learning model cover a wide range of information, from subtle structural deformation to the development of exterior cracks. Compared with the traditional method of relying solely on manual inspections or simple sensor measurements, the accuracy of identifying potential safety hazards has been greatly improved. Whether it is a tiny stress change hidden in the internal structure of the building or a subtle crack expansion that is not easily detected on the exterior wall, it can be captured in a timely and accurate manner, effectively avoiding the omission of hidden dangers due to human negligence or limitations in detection methods, and providing all-round, no-dead-angle protection for housing safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] FIG1 is a monitoring flow chart of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0016] As shown in FIG1 , a machine learning-based housing safety monitoring method and system of this embodiment first deploys sensors: according to the type of house (such as residential, commercial building, industrial plant, etc.) and structural characteristics (brick-concrete, frame, steel structure, etc.), the installation positions of sensors in key parts of the house are scientifically planned, with a focus on key stress-bearing parts such as load-bearing walls, beam-column joints, and foundations to ensure that the collected data is representative.

[0017] Then the system is initialized: when used for the first time, the sensor network and data processing The center and early warning modules are debugged, the initial machine learning model is loaded, and preliminary training is carried out using a small amount of house monitoring data accumulated in the early stage to optimize the model parameters to adapt it to the monitoring needs of specific houses.

[0018] During the daily operation stage, when the system enters the stable operation stage, it continuously collects and processes data, and regularly updates the model to adapt to changes in the structural characteristics of the house over time and environmental changes (such as seasonal changes and the impact of surrounding construction), and continuously improve the accuracy and reliability of monitoring.

[0019] The specific steps include: Step 1: Deploy several sensors at key locations of the house, including strain sensors, displacement sensors, and acceleration sensors, to collect physical parameter data of the house structure in real time and store them in time series; Step 2: Clean and normalize the collected data; Step 3: Mining key features reflecting the health status of the building structure from the preprocessed data; Step 4: Use the machine learning model to train with the extracted features as input and the actual safety status of the house as output label; Step 5: Input the new data collected and processed in real time into the trained model, issue warnings based on the model output, send alarms to relevant personnel, and upload data to the cloud server.

[0020] The feature extraction in step 3 adopts the following formula: Discrete wavelet transform: Where x[n] is the original signal, φj0,k[n] and ψj,k[n] are the scaling function and wavelet function respectively, M and N are the length of the transform and the length of the signal, j0 and j are the decomposition scales, and k is the translation parameter; Use soft threshold to make judgment: Among them, wj,k is the wavelet coefficient, T is the threshold, and sgn is the sign function. By analyzing the transformed coefficients, the time-frequency characteristics of the signal at different scales are extracted. These characteristics can reflect the dynamic response characteristics of the building structure at different frequency components, and then reflect the health status of the structure.

[0021] The data cleaning in step 2 adopts a Butterworth low-pass filter, sets appropriate filtering parameters according to the characteristics of different sensor data, accurately removes high-frequency noise components, and restores the real change trend of the data.

[0022] The machine learning in step 4 adopts the neural network model training formula, which is as follows: Linear combination: In housing security monitoring, the input data includes data collected from various sensors and image acquisition devices. After preprocessing and feature extraction, the data will be used as the input of the neural network a0 = x; For house safety monitoring, different types of data will have different feature dimensions, which will be linearly combined through the weight matrix wl and bias vector bl of the first layer; Then activate it through the ReLU function: ReLU(z) = max(0, z) The ReLU function is linear on the positive semi-axis, which helps to speed up convergence and avoid gradient disappearance; in, is the total number of samples, representing the number of monitored houses; is the total number of categories, representing the safety level classification of the house, It is The samples belong to The true label of the category, if The true category of the sample is Class, then ;otherwise ; The model predicts The samples belong to The probability of the category, this probability value is calculated by the model through forward propagation and satisfies Inner summation The cross entropy loss is calculated for a single sample, and the outer layer is summed The cross entropy losses of all samples are added up and finally divided by the total number of samples to get the average cross entropy loss, which can more fairly measure the performance of the model on the entire dataset; the house safety is divided into multiple categories: safe, mild hidden dangers, moderate hidden dangers, and severe hidden dangers. When the real state of the house is a moderate hidden danger and the probability distribution predicted by the model is inaccurate, the cross entropy loss will be larger, which will drive the model to update parameters and make the prediction results closer to the real state.

[0023] It includes a sensor network, a data processing center, an early warning module, a communication module and a cloud server. The sensor network is composed of multiple sensor nodes distributed throughout the house, which are responsible for collecting physical parameter data of the house. Each sensor node has low power consumption, high precision, and self-organizing networking capabilities, and can transmit data to the data processing center stably and reliably.

[0024] The data processing center is equipped with high-performance processors and is responsible for the core tasks of data preprocessing, feature extraction, and model operation. It has a built-in data storage unit for storing raw data, preprocessed data, and intermediate data during model training.

[0025] The early warning module is connected to the data processing center. According to the housing safety status output by the model, when an early warning is needed, the corresponding alarm method is triggered according to the preset rules to ensure that the information is conveyed in a timely manner; The communication module includes wired communication and wireless communication components to ensure high-speed data transmission between the sensor network, data processing center, early warning module and cloud server, and realize remote monitoring and management; The cloud server is used to store massive amounts of historical housing monitoring data and supports big data analysis. On the one hand, it can provide more data support for local model training. On the other hand, it can be remotely accessed by different regions and management departments to achieve regionalized and coordinated management of housing safety monitoring.

[0026] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. A housing safety monitoring method based on machine learning, characterized in that: The specific steps include: Step 1: Deploy several sensors at key locations of the house, including strain sensors, displacement sensors, and acceleration sensors, to collect physical parameter data of the house structure in real time and store them in time series; Step 2: Clean and normalize the collected data; Step 3: Mining key features reflecting the health status of the building structure from the preprocessed data; Step 4: Use the machine learning model to train with the extracted features as input and the actual safety status of the house as output label; Step 5: Input the new data collected and processed in real time into the trained model, issue warnings based on the model output, send alerts to relevant personnel, and upload data to the cloud server; The feature extraction in step 3 adopts the following formula: Discrete Wavelet Transform: Where x[n] is the original signal, φj0,k[n] and ψj,k[n] are the scaling function and wavelet function respectively, M and N are the length of the transform and the length of the signal, j0 and j are the decomposition scales, and k is the translation parameter; Use soft threshold to make judgment: Among them, wj,k is the wavelet coefficient, T is the threshold, and sgn is the sign function. By analyzing the transformed coefficients, the time-frequency characteristics of the signal at different scales are extracted. These characteristics can reflect the dynamic response characteristics of the building structure at different frequency components, and then reflect the health status of the structure; The data cleaning in step 2 adopts a Butterworth low-pass filter, sets appropriate filtering parameters according to the characteristics of different sensor data, accurately removes high-frequency noise components, and restores the real change trend of the data; The machine learning in step 4 adopts the neural network model training formula, which is as follows: Linear combinations: In housing security monitoring, the input data includes data collected from various sensors and image acquisition devices. After preprocessing and feature extraction, the data will be used as the input of the neural network a0 = x; For house safety monitoring, different types of data will have different feature dimensions, which will be linearly combined through the weight matrix wl and bias vector bl of the first layer; Then activate it through the ReLU function: ReLU(z) = max(0, z) The ReLU function is linear on the positive semi-axis, which helps to speed up the convergence and avoid the vanishing gradient; in, is the total number of samples, representing the number of monitored houses; is the total number of categories, representing the safety level classification of the house, It is The samples belong to The true label of the category, if The true category of the sample is Class, then ;otherwise ; The model predicts The samples belong to The probability of the category, this probability value is calculated by the model through forward propagation and satisfies Inner summation The cross entropy loss is calculated for a single sample, and the outer layer is summed The cross entropy losses of all samples are added up and finally divided by the total number of samples to get the average cross entropy loss, which can more fairly measure the performance of the model on the entire dataset; the house safety is divided into multiple categories including safety, mild hidden dangers, moderate hidden dangers, and severe hidden dangers. The cross entropy loss is used. When the actual state of the house is a moderate hidden danger and the probability distribution predicted by the model is inaccurate, the cross entropy loss will be larger, which will drive the model to update parameters and make the prediction results closer to the actual state.

2. A housing safety monitoring system according to the housing safety monitoring method based on machine learning as claimed in claim 1, characterized in that: It includes a sensor network, a data processing center, an early warning module, a communication module and a cloud server. The sensor network is composed of multiple sensor nodes distributed throughout the house, which are responsible for collecting physical parameter data of the house. Each sensor node has low power consumption, high precision, and self-organizing networking capabilities, and can transmit data to the data processing center stably and reliably.

3. A housing safety monitoring system according to the housing safety monitoring method based on machine learning as claimed in claim 2, characterized in that: The data processing center is equipped with high-performance processors and is responsible for the core tasks of data preprocessing, feature extraction, and model operation. It has a built-in data storage unit for storing original data, preprocessed data, and intermediate data during model training.

4. A housing safety monitoring system according to the housing safety monitoring method based on machine learning as claimed in claim 3, characterized in that: The early warning module is connected to the data processing center. According to the housing safety status output by the model, when an early warning is needed, the corresponding alarm method is triggered according to the preset rules to ensure that the information is conveyed in a timely manner; The communication module includes wired communication and wireless communication components to ensure high-speed data transmission between the sensor network, data processing center, early warning module and cloud server, and realize remote monitoring and management; The cloud server is used to store massive historical housing monitoring data and support big data analysis. On the one hand, it provides more data support for local model training, and on the other hand, it provides remote access to different regions and management departments to achieve regionalized and coordinated management of housing safety monitoring.

Images