A Machine Learning-Based Method and System for Predicting and Early Warning of Micro-vibrations at Construction Sites

Abstract

The application discloses a kind of based on machine learning's construction site microvibration prediction early warning method and system, the method includes establishing site characteristic parameter data;Unified time stamp is carried out to the microvibration response data of collection, denoising, reduce data scale amount;Microvibration response data time domain data, frequency domain data, time-frequency domain data are respectively input with site characteristic parameter by characteristic representation model and output the microvibration time domain, frequency domain and time-frequency domain representation graph of current collection period;The microvibration time domain, frequency domain and time-frequency domain representation graph of next period are predicted according to microvibration time domain, frequency domain and time-frequency domain representation graph by regression model;Frequency domain and time-frequency domain representation graph are respectively reconstructed time domain data, and the final prediction of the time domain data of the microvibration response data of next period is realized by three time domain data weighted average;Time domain data that exceeds warning line sends early warning signal.The application can monitor the microvibration response data of the region of sensor that has not been deployed.

Description

Technical Field

[0001] This invention relates to the field of civil engineering technology, and in particular to a method and system for predicting and warning micro-vibrations at construction sites based on machine learning. Background Technology

[0002] Currently, with the increasing construction area of ​​high-precision factory buildings such as semiconductor plants, it is necessary to monitor the micro-vibrations generated during construction to avoid their impact on semiconductor equipment in surrounding buildings. Comprehensive monitoring of micro-vibrations at the construction site requires deploying a large number of sensors, which significantly increases monitoring costs. Furthermore, some areas within construction sites or existing factory buildings are difficult to sensorize. Without sensors, the micro-vibration situation in these areas cannot be effectively monitored, leading to gaps in micro-vibration monitoring. This negatively impacts overall micro-vibration monitoring and early warning systems for construction sites. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for predicting and warning micro-vibrations at construction sites based on machine learning, in order to solve the problem of missing micro-vibration response data monitoring in areas of construction sites where sensors are not deployed.

[0004] To address the aforementioned technical problems, this invention provides a machine learning-based method for predicting and warning of micro-vibrations at construction sites, comprising:

[0005] Step S101: Establish site characteristic parameter data for areas of the construction site where no sensors are deployed. The site characteristic parameters are parameters that can reflect the physical and mechanical properties of the soil.

[0006] Step S102: The multidimensional micro-vibration response data collected by deployed sensors near the area where no sensors are deployed at the construction site during the current collection period are unified with timestamps, denoised, and the data size is reduced.

[0007] Step S103: The time-domain data of micro-vibration response data with unified timestamp, noise reduction, and reduced data size, and the site feature parameters are taken as input and the feature representation model with the first neural network model as the skeleton is used to output the time-domain representation map of micro-vibration in the area where no sensors are deployed during the current collection period.

[0008] Step S104: Convert the time domain data of the micro-vibration response data into frequency domain data through Fourier transform, and output the micro-vibration frequency domain characterization map of the area without deployed sensors in the current acquisition period through the feature characterization model with the second neural network model as the skeleton, using the frequency domain data and site feature parameters as input.

[0009] Step S105: Convert the time-domain data of the micro-vibration response data into time-frequency domain data through continuous wavelet transform. Take the time-frequency domain data and site feature parameters as inputs and output the micro-vibration time-frequency domain representation map of the area without deployed sensors in the current acquisition period through the feature representation model with the visual neural network model as the skeleton.

[0010] Step S106: Using a regression model, the time-domain, frequency-domain, and time-frequency-domain representations of micro-vibration in the sensorless area during the data collection period are analyzed to preliminarily predict the output of the time-domain, frequency-domain, and time-frequency-domain representations of micro-vibration in the sensorless area during the next time period. Inverse Fourier transform is used to reconstruct the frequency-domain data in the predicted output of the sensorless area's frequency-domain representation of micro-vibration in the next time period into time-domain data. Inverse continuous wavelet transform is used to reconstruct the time-frequency-domain data in the predicted output of the sensorless area's time-frequency-domain representation of micro-vibration in the next time period into time-domain data. Finally, a weighted average is taken of the time-domain data in the predicted output of the sensorless area's time-domain representation of micro-vibration in the next time period to obtain the time-domain data of the predicted sensorless area's micro-vibration response in the next time period.

[0011] Step S107: Using a classification model, issue an early warning signal for the micro-vibration response data in the undeployed sensor area in the next time period whose time-domain data exceeds the warning line.

[0012] Furthermore, in step S102, the machine learning-based method for predicting and warning micro-vibrations at construction sites provides a method for denoising multidimensional micro-vibration response data collected by deployed sensors through discrete wavelet transform.

[0013] Furthermore, the machine learning-based micro-vibration prediction and early warning method for construction sites provided by the present invention reduces the data size in step S102 by downsampling the multi-dimensional micro-vibration response data collected by the deployed sensors.

[0014] Furthermore, in step S102, the machine learning-based method for predicting and warning micro-vibrations at construction sites provided by this invention uses the nearest neighbor method to unify the timestamps of multidimensional data.

[0015] Furthermore, in the machine learning-based method for predicting and warning micro-vibrations at construction sites provided by the present invention, in step S103, the first neural network model includes one of a recurrent neural network, an improved recurrent neural network with gated recurrent units, a long short-term memory neural network, and a Transformer-type neural network with self-attention mechanism as its core.

[0016] Furthermore, in the machine learning-based micro-vibration prediction and early warning method for construction sites provided by the present invention, in step S104, the second neural network model is one of a convolutional neural network, a recurrent neural network, and a self-attention mechanism neural network.

[0017] Furthermore, in the machine learning-based method for predicting and warning micro-vibrations at construction sites provided by the present invention, in step S106, the regression model is an autoregressive model or a decision tree regression model.

[0018] Furthermore, in the machine learning-based method for predicting and warning micro-vibrations at construction sites provided by the present invention, the classification model in step S107 is a fully connected neural network or a support vector machine.

[0019] To address the aforementioned technical problems, this invention also provides a machine learning-based micro-vibration prediction and early warning system for construction sites, comprising:

[0020] The multimodal data module includes a multidimensional micro-vibration response data submodule and a site characteristic parameter submodule, wherein:

[0021] The multidimensional micro-vibration response data submodule is used to unify the timestamps, denoise, and reduce the data size of multidimensional micro-vibration response data collected by deployed sensors in the vicinity of the construction site where no sensors are deployed during the current data collection period.

[0022] The site characteristic parameter submodule is used to establish site characteristic parameter data for areas of the construction site where no sensors have been deployed.

[0023] The joint feature representation module includes a time-domain representation submodule, a frequency-domain representation submodule, and a time-frequency-domain representation submodule, wherein:

[0024] The time-domain characterization submodule is used to take the time-domain data of micro-vibration response data with unified timestamps, noise reduction, and reduced data size, and the site feature parameters as input, and output the micro-vibration time-domain characterization map of the area without deployed sensors in the current acquisition period through the feature characterization model with the first neural network model as the skeleton.

[0025] The frequency domain characterization submodule is used to convert the time domain data of micro-vibration response data into frequency domain data through Fourier transform. It takes the frequency domain data and site feature parameters as input and outputs the micro-vibration frequency domain characterization map of the area without deployed sensors in the current acquisition period through the feature characterization model with the second neural network model as the skeleton.

[0026] The time-frequency domain characterization submodule is used to convert the time-domain data of micro-vibration response data into time-frequency domain data through continuous wavelet transform. It takes the time-frequency domain data and site feature parameters as input and outputs the micro-vibration time-frequency domain characterization map of the area without deployed sensors in the current acquisition period through a feature characterization model with a visual neural network model as the skeleton.

[0027] The feature application module includes a micro-vibration prediction submodule and a micro-vibration early warning submodule, wherein:

[0028] The micro-vibration prediction submodule is used to analyze the time-domain, frequency-domain, and time-frequency-domain micro-vibration representation maps of the sensorless area during the data collection period using a regression model, and initially predict the output of the same maps for the next time period. Then, it uses inverse Fourier transform to reconstruct the frequency-domain data from the predicted output of the sensorless area's micro-vibration frequency-domain representation map into time-domain data, and uses inverse continuous wavelet transform to reconstruct the time-frequency-domain data from the predicted output of the sensorless area's micro-vibration time-frequency-domain representation map into time-domain data. Finally, it calculates a weighted average of the time-domain data from the predicted output of the sensorless area's micro-vibration time-domain representation map and the two reconstructed time-domain data to obtain the predicted time-domain data of the sensorless area's micro-vibration response in the next time period.

[0029] The micro-vibration early warning submodule is used to issue early warning signals for micro-vibration response data in the un-deployed sensor area that exceed the warning line in the time domain data of the micro-vibration response data in the next time period, based on the classification model.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] The present invention provides a machine learning-based method and system for predicting and warning micro-vibrations at construction sites. This method combines multi-dimensional micro-vibration response data collected by deployed sensors near un-deployed areas of the construction site during the current acquisition period with site characteristic parameters of the un-deployed areas. The system uses Fourier transform and continuous wavelet transform to convert the multi-dimensional micro-vibration response data collected during the current acquisition period into time-domain, frequency-domain, and time-frequency-domain micro-vibration data. A feature representation model with a neural network model as its framework outputs time-domain, frequency-domain, and time-frequency-domain micro-vibration representation maps for the current acquisition period. Inverse Fourier transform and inverse continuous wavelet transform are used to reconstruct time-domain data from the frequency-domain and time-frequency-domain micro-vibration representation maps for the current acquisition period. Combining the time-domain data from the current acquisition period's micro-vibration time-domain representation map, a weighted average of the three time-domain data is calculated to predict the time-domain data of the micro-vibration response data for the un-deployed areas in the next time period. Warnings are issued based on whether the predicted time-domain data of the micro-vibration response data for the un-deployed areas in the next time period exceeds a warning threshold. By introducing site characteristic parameters, the accuracy of micro-vibration response data prediction, as well as the accuracy and timeliness of early warning, are improved. Combined with the monitoring of micro-vibration response data in the areas where sensors have been deployed, it is possible to monitor and warn of micro-vibration response data across the entire construction site, thus solving the problem of missing micro-vibration response data in sensorless locations on the site. Attached Figure Description

[0032] Figure 1 This is a flowchart of a machine learning-based method for predicting and warning micro-vibrations at construction sites.

[0033] Figure 2 This is a block diagram of the structural components of a machine learning-based micro-vibration prediction and early warning system for construction sites.

[0034] As shown in the figure:

[0035] 100. Machine learning-based micro-vibration prediction and early warning system for construction sites;

[0036] 110. Multimodal data module; 111. Multidimensional micro-vibration response data submodule; 112. Site characteristic parameter submodule;

[0037] 120. Feature Joint Representation Module; 121. Time Domain Representation Submodule; 122. Frequency Domain Representation Submodule; 123. Time-Frequency Domain Representation Submodule;

[0038] 130. Feature Application Module; 131. Micro-vibration Prediction Submodule; 132. Micro-vibration Early Warning Submodule. Detailed Implementation

[0039] The present invention will now be described in detail with reference to the accompanying drawings. The advantages and features of the present invention will become clearer from the following description. It should be noted that the drawings are all in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0040] Please refer to Figure 1 This invention provides a machine learning-based method for predicting and warning of micro-vibrations at construction sites, which may include the following steps:

[0041] Step S101: Establish site characteristic parameter data for areas of the construction site where no sensors are deployed. These site characteristic parameters are those that reflect the physical and mechanical properties of the soil layers, including but not limited to site elevation, soil layer information, depth of each soil layer, pressure modulus of each soil layer, pressure shear modulus of each soil layer, horizontal subgrade reaction coefficient, initial pressure, plastic pressure, ultimate pressure, and undrained shear strength.

[0042] Step S102 involves unifying the timestamps, denoising, and reducing the data size of the multidimensional micro-vibration response data collected by deployed sensors near the un-deployed areas of the construction site during the current data collection period. This can be achieved by using discrete wavelet transform to denoise the data, downsampling the data to reduce its size, and using the nearest neighbor method to unify the timestamps. Reducing the data size decreases the amount of data processing required in subsequent steps.

[0043] Step S103 involves taking the time-domain data of the micro-vibration response data (with unified timestamps, denoised, and reduced data size) and site feature parameters as input, and outputting a feature representation model with a first neural network model as its backbone to output a time-domain representation map of the micro-vibration in the sensorless area during the current data acquisition period. The first neural network model includes, but is not limited to, RNN (Recurrent Neural Network), an improved version of RNN called GRU (Gated Recurrent Unit Network), LSTM (Long Short-Term Memory Neural Network), and Transformer-type neural networks with a self-attention mechanism at their core. The time-domain representation map of micro-vibration can reflect the amplitude and duration-related characteristics of the micro-vibration response data, i.e., the time-domain parameters.

[0044] Step S104: The time-domain data of the micro-vibration response data is converted into frequency-domain data using Fourier transform. The frequency-domain data and site feature parameters are used as inputs, and a feature representation model with a second neural network model as its backbone outputs a frequency-domain representation map of the micro-vibration in the area without deployed sensors during the current data acquisition period. The second neural network model includes, but is not limited to, convolutional neural networks, recurrent neural networks, and self-attention mechanism neural networks. The frequency-domain representation map of the micro-vibration response data reflects the frequency-related characteristics of the micro-vibration response data, i.e., the frequency parameters.

[0045] Step S105: The time-domain data of the micro-vibration response data is converted into time-frequency domain data through continuous wavelet transform. Using the time-frequency domain data and site feature parameters as input, a feature representation model with a visual neural network model as its skeleton outputs a time-frequency domain representation map of the micro-vibration in the area without deployed sensors during the current acquisition period. This time-frequency domain representation map reflects the time-domain and frequency parameters of the micro-vibration response data. The corresponding time-domain, frequency-domain, and time-frequency domain data in the time-domain, frequency-domain, and time-frequency domain representation maps are consistent with the dimensions of the site feature parameters. The dimension of the micro-vibration response is the number of monitoring points monitoring the micro-vibration response. For example, if there are 10 monitoring points, the micro-vibration response is 10-dimensional, and the dimension of the site feature parameters is also 10-dimensional, facilitating the extraction of corresponding features by the subsequent joint representation model.

[0046] Step S106: Using a regression model, the time-domain, frequency-domain, and time-frequency-domain representations of micro-vibration in the sensorless area during the data collection period are analyzed and preliminarily predicted to output the same representations for the next time period. An inverse Fourier transform is used to reconstruct the frequency-domain data from the predicted output of the micro-vibration frequency-domain representation for the sensorless area in the next time period into time-domain data. Similarly, an inverse continuous wavelet transform is used to reconstruct the time-frequency-domain data from the predicted output of the micro-vibration time-frequency representation for the sensorless area in the next time period into time-domain data. Finally, a weighted average is calculated between the predicted time-domain data from the predicted output of the micro-vibration time-domain representation for the sensorless area in the next time period and the two reconstructed time-domain data to obtain the predicted time-domain data of the micro-vibration response for the sensorless area in the next time period. The regression model can be an autoregressive model or a decision tree regression model.

[0047] Step S107 involves issuing a warning signal based on the predicted micro-vibration response data of the un-deployed sensor area in the next time period, where the temporal data exceeds the warning line, using a classification model. The classification model can be a fully connected neural network or a support vector machine. Specifically, the outputs of each feature representation model are used as input, and whether a warning is issued is used as the label. A dataset is created, the classification model is trained, and the parameters in the classification model are adjusted to establish a correlation between the outputs of each feature representation model and whether a warning is issued. In application, the actual collected micro-vibration response and length feature parameters, after being processed by the feature representation and classification models, can determine whether a warning signal needs to be issued.

[0048] Please refer to Figure 2Embodiment 2 of the present invention also provides a micro-vibration prediction and early warning system 100 for construction sites based on machine learning, which adopts the above method and includes a multimodal data module 110, a feature joint representation module 120, and a feature application module 130, wherein:

[0049] The multimodal data module 110 includes a multidimensional micro-vibration response data submodule 111 and a site characteristic parameter submodule 112, wherein:

[0050] The multidimensional micro-vibration response data submodule 111 is used to unify the timestamps, denoise, and reduce the data size of the multidimensional micro-vibration response data collected by deployed sensors in the area near the construction site where no sensors are deployed during the current collection period.

[0051] Site characteristic parameter submodule 112 is used to establish site characteristic parameter data for areas of the construction site where no sensors are deployed.

[0052] The feature joint representation module 120 includes a time-domain representation submodule 121, a frequency-domain representation submodule 122, and a time-frequency-domain representation submodule 123, wherein:

[0053] The time-domain characterization submodule 121 is used to take the time-domain data of micro-vibration response data with unified timestamp, noise reduction, and reduced data size, and the site characteristic parameters as input, and output the micro-vibration time-domain characterization map of the area without deployed sensors in the current acquisition period through the feature characterization model with the first neural network model as the skeleton.

[0054] The frequency domain characterization submodule 122 is used to convert the time domain data of the micro-vibration response data into frequency domain data through Fourier transform. It takes the frequency domain data and site feature parameters as input and outputs the micro-vibration frequency domain characterization map of the area without deployed sensors in the current acquisition period through the feature characterization model with the second neural network model as the skeleton.

[0055] The time-frequency domain characterization submodule 123 is used to convert the time-domain data of micro-vibration response data into time-frequency domain data through continuous wavelet transform. It takes the time-frequency domain data and site feature parameters as inputs and outputs the micro-vibration time-frequency domain characterization map of the area without deployed sensors in the current acquisition period through a feature characterization model with a visual neural network model as the skeleton.

[0056] Feature application module 130 includes micro-vibration prediction submodule 131 and micro-vibration early warning submodule 132, wherein:

[0057] The micro-vibration prediction submodule 131 is used to analyze the time-domain, frequency-domain, and time-frequency-domain micro-vibration characterization maps of the sensorless area during the data collection period using a regression model, and preliminarily predict the output of the time-domain, frequency-domain, and time-frequency-domain micro-vibration characterization maps of the sensorless area in the next time period. It then uses inverse Fourier transform to reconstruct the frequency-domain data in the predicted output of the sensorless area's frequency-domain micro-vibration characterization map for the next time period into time-domain data, and uses inverse continuous wavelet transform to reconstruct the time-frequency-domain data in the predicted output of the sensorless area's time-frequency-domain micro-vibration characterization map for the next time period into time-domain data. Finally, it obtains the predicted time-domain data of the sensorless area's micro-vibration response data for the next time period by weighted averaging the predicted output of the sensorless area's time-domain micro-vibration characterization map for the next time period.

[0058] The micro-vibration early warning submodule 132 is used to issue an early warning signal for micro-vibration response data in the undeployed sensor area in the next time period that exceeds the warning line, based on a classification model.

[0059] The machine learning-based micro-vibration prediction and early warning method and system 100 for construction sites provided in this invention combines multi-dimensional micro-vibration response data collected by deployed sensors near the un-deployed area of ​​the construction site during the current acquisition period with site characteristic parameters of the un-deployed area. The system uses Fourier transform and continuous wavelet transform to convert the time-domain, frequency-domain, and time-frequency-domain data of the micro-vibration, along with the two dimensions of the site characteristic parameters of the un-deployed area, into a feature representation model with a neural network model as its skeleton. This model outputs a time-domain representation map, a frequency-domain representation map, and a time-frequency-domain representation map of the micro-vibration during the current acquisition period. Furthermore, inverse Fourier transform and inverse continuous wavelet transform are used to reconstruct the time-domain data from the frequency-domain representation map and the time-frequency-domain representation map of the micro-vibration during the current acquisition period. Combining the time-domain data from the current acquisition period's time-domain representation map, a weighted average of the three time-domain data is applied to predict the time-domain data of the micro-vibration response data of the un-deployed area in the next time period. An early warning is issued based on whether the predicted time-domain data of the micro-vibration response data of the un-deployed area in the next time period exceeds a warning line. It can realize the monitoring and early warning of micro-vibration across the entire site based on a sparse monitoring network. By introducing site characteristic parameters, the accuracy of micro-vibration response data prediction, as well as the accuracy and timeliness of early warning, are improved. Combined with the monitoring of micro-vibration response data in the areas where sensors have been deployed, it is possible to monitor and warn of micro-vibration response data across the entire construction site, solving the problem of missing micro-vibration response data in sensorless locations on the site.

[0060] The micro-vibration prediction and early warning method and system 100 based on machine learning for construction sites provided in this embodiment of the invention extracts time-domain, frequency-domain and other features separately through various feature representation models and shares the features with regression models and classification models. This solves the problem of having to repeatedly train all models when the input is the same but the output is different, thus avoiding repeated training.

[0061] This invention is not limited to the specific embodiments described above. Obviously, the embodiments described above are only a part of the embodiments of this invention, not all of them. All other embodiments obtained by those skilled in the art based on the described embodiments of this invention are within the scope of protection of this invention. Those skilled in the art can make other modifications and variations to this invention. Therefore, if these modifications and variations of this invention fall within the scope of the claims of this invention, then this invention also intends to include these modifications and variations.

Claims

1.A machine learning-based micro-vibration prediction and early warning method for a construction site, characterized in that, The method comprises the following steps: Step S101, establish the site characteristic parameter data of the area where no sensor is deployed in the construction site, wherein the site characteristic parameter is a parameter capable of reflecting the physical and mechanical properties of the soil layer; Step S102, unify the time stamp, denoise and reduce the data scale of the multi-dimensional micro-vibration response data collected by the deployed sensor near the area where no sensor is deployed in the construction site in the current collection period; Step S103, input the time domain data of the micro-vibration response data and the site characteristic parameter into a feature representation model with a first neural network model as a skeleton to output a micro-vibration time domain representation graph of the area where no sensor is deployed in the current collection period; Step S104, convert the time domain data of the micro-vibration response data into frequency domain data by Fourier transform, input the frequency domain data and the site characteristic parameter into a feature representation model with a second neural network model as a skeleton to output a micro-vibration frequency domain representation graph of the area where no sensor is deployed in the current collection period; Step S105, convert the time domain data of the micro-vibration response data into time-frequency domain data by continuous wavelet transform, input the time-frequency domain data and the site characteristic parameter into a feature representation model with a visual neural network model as a skeleton to output a micro-vibration time-frequency domain representation graph of the area where no sensor is deployed in the current collection period; Step S106, analyze the micro-vibration time domain representation graph, the micro-vibration frequency domain representation graph and the micro-vibration time-frequency domain representation graph of the area where no sensor is deployed in the collection period by a regression model to preliminarily predict and output the micro-vibration time domain representation graph, the micro-vibration frequency domain representation graph and the micro-vibration time-frequency domain representation graph of the area where no sensor is deployed in the next period; inverse Fourier transform is used to reconstruct the frequency domain data in the predicted micro-vibration frequency domain representation graph of the area where no sensor is deployed in the next period into time domain data, inverse continuous wavelet transform is used to reconstruct the time-frequency domain data in the predicted micro-vibration time-frequency representation graph of the area where no sensor is deployed in the next period into time domain data, and the time domain data in the predicted micro-vibration time domain representation graph of the area where no sensor is deployed in the next period and the two reconstructed time domain data are weighted and averaged to obtain the final predicted time domain data of the micro-vibration response data of the area where no sensor is deployed in the next period; Step S107, the classification model is used to issue a warning signal for the micro-vibration response data whose time domain data in the final predicted micro-vibration response data of the area where no sensor is deployed in the next period exceeds the warning line. 2.The machine learning-based construction site micro-vibration prediction and early warning method according to claim 1, characterized in that, In step S102, the multi-dimensional micro-vibration response data collected by the deployed sensor is denoised by discrete wavelet transform. 3.The machine learning-based construction site micro-vibration prediction and early warning method of claim 1, wherein, In step S102, the data scale of the multi-dimensional micro-vibration response data collected by the deployed sensor is reduced by downsampling. 4.The machine learning based construction site micro-vibration prediction and warning method of claim 1, wherein, In step S102, the time stamps of the multi-dimensional data are unified by the nearest neighbor method. 5.The machine learning based construction site micro-vibration prediction and warning method of claim 1, wherein, In step S103, the first neural network model comprises one of a recurrent neural network, an improved gated recurrent unit network of the recurrent neural network, a long short-term memory neural network and a Transformer type neural network with a self-attention mechanism as the core. 6.The machine learning based construction site micro-vibration prediction and warning method of claim 1, wherein, In step S104, the second neural network model is one of a convolutional neural network, a recurrent neural network, and a self-attention mechanism neural network. 7.The machine learning based construction site micro-vibration prediction and warning method of claim 1, wherein, In step S106, the regression model is an autoregressive model or a decision tree regression model. 8.The machine learning based construction site micro-vibration prediction and warning method of claim 1, wherein, In step S107, the classification model is a fully connected neural network or a support vector machine. 9.A machine learning based construction site micro-vibration prediction and early warning system, characterized in that, Comprise: The multi-modal data module comprises a multi-dimensional micro-vibration response data submodule and a site feature parameter submodule, wherein: The multi-dimensional micro-vibration response data submodule is configured to perform unified timestamping, denoising, and data size reduction on the multi-dimensional micro-vibration response data collected by the deployed sensors near the area without deployed sensors in the current collection period; The site feature parameter submodule is configured to establish site feature parameter data for the area without deployed sensors in the construction site; The feature joint representation module comprises a time domain representation submodule, a frequency domain representation submodule, and a time-frequency domain representation submodule, wherein: The time domain representation submodule is configured to input the time domain data of the micro-vibration response data subjected to unified timestamping, denoising, and data size reduction and the site feature parameters into a feature representation model with a first neural network model as a skeleton to output a micro-vibration time domain representation map of the area without deployed sensors in the current collection period; The frequency domain representation submodule is configured to convert the time domain data of the micro-vibration response data into frequency domain data by Fourier transform, input the frequency domain data and the site feature parameters into a feature representation model with a second neural network model as a skeleton to output a micro-vibration frequency domain representation map of the area without deployed sensors in the current collection period; The time-frequency domain representation submodule is configured to convert the time domain data of the micro-vibration response data into time-frequency domain data by continuous wavelet transform, input the time-frequency domain data and the site feature parameters into a feature representation model with a visual neural network model as a skeleton to output a micro-vibration time-frequency domain representation map of the area without deployed sensors in the current collection period; The feature application module comprises a micro-vibration prediction submodule and a micro-vibration early warning submodule, wherein: The micro-vibration prediction submodule is configured to analyze the micro-vibration time domain representation map, the micro-vibration frequency domain representation map, and the micro-vibration time-frequency domain representation map of the area without deployed sensors in the collection period by a regression model to preliminarily predict and output the micro-vibration time domain representation map, the micro-vibration frequency domain representation map, and the micro-vibration time-frequency domain representation map of the area without deployed sensors in the next period; inverse Fourier transform is used to reconstruct the frequency domain data in the predicted micro-vibration frequency domain representation map of the area without deployed sensors in the next period into time domain data, inverse continuous wavelet transform is used to reconstruct the time-frequency domain data in the predicted micro-vibration time-frequency representation map of the area without deployed sensors in the next period into time domain data, and the time domain data in the predicted micro-vibration time domain representation map of the area without deployed sensors in the next period and the two reconstructed time domain data are weighted and averaged to obtain the final predicted time domain data of the micro-vibration response data of the area without deployed sensors in the next period. A micro-vibration early warning sub-module is configured to issue an early warning signal for the micro-vibration response data of the final predicted micro-vibration response data of the non-deployed sensor region in the time domain data of the next period exceeding the warning line through the classification model.

Images