CN117824756B - Building structure monitoring method, equipment and storage medium - Google Patents
Building structure monitoring method, equipment and storage medium Download PDFInfo
- Publication number
- CN117824756B CN117824756B CN202410223955.7A CN202410223955A CN117824756B CN 117824756 B CN117824756 B CN 117824756B CN 202410223955 A CN202410223955 A CN 202410223955A CN 117824756 B CN117824756 B CN 117824756B
- Authority
- CN
- China
- Prior art keywords
- data
- model
- characteristic data
- monitoring
- fusion
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 156
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000003860 storage Methods 0.000 title claims description 9
- 230000010354 integration Effects 0.000 claims abstract description 95
- 230000004927 fusion Effects 0.000 claims abstract description 84
- 238000012549 training Methods 0.000 claims abstract description 62
- 238000007499 fusion processing Methods 0.000 claims abstract description 7
- 230000015654 memory Effects 0.000 claims description 57
- 238000006073 displacement reaction Methods 0.000 claims description 31
- 238000013528 artificial neural network Methods 0.000 claims description 24
- 238000012795 verification Methods 0.000 claims description 24
- 238000005457 optimization Methods 0.000 claims description 21
- 238000013527 convolutional neural network Methods 0.000 claims description 18
- 238000010586 diagram Methods 0.000 claims description 6
- 238000012806 monitoring device Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 20
- 238000012706 support-vector machine Methods 0.000 description 14
- 230000007787 long-term memory Effects 0.000 description 9
- 230000008901 benefit Effects 0.000 description 8
- 238000012545 processing Methods 0.000 description 8
- 238000007405 data analysis Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 230000007613 environmental effect Effects 0.000 description 6
- 230000006403 short-term memory Effects 0.000 description 6
- 238000011156 evaluation Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000036541 health Effects 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 239000012634 fragment Substances 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000004566 building material Substances 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 230000003862 health status Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000004062 sedimentation Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a building structure monitoring method, which comprises the following steps: collecting sensing data of each type of sensor, extracting characteristic data corresponding to the sensing data and storing the characteristic data, wherein each type of sensor is respectively arranged on each key structure of the same building, and the same type of sensor comprises at least one sensor, and fusion processing is carried out on the characteristic data by utilizing a preset multi-mode fusion algorithm to obtain fusion characteristic data; determining a plurality of integration strategies, and constructing a plurality of corresponding monitoring models based on each integration strategy; training and integrating a plurality of corresponding monitoring models based on each integration strategy and the fusion characteristic data to obtain integrated monitoring models; and determining an optimal integrated monitoring model from a plurality of integrated monitoring models corresponding to the plurality of integrated strategies respectively, and monitoring the building structure to be monitored by using the optimal integrated monitoring model.
Description
Technical Field
The invention relates to the technical field of building structure monitoring, in particular to a building structure monitoring method, equipment and a storage medium.
Background
With the current increase of social development speed, buildings in various places are increasing, and the safety of building structures is the most interesting factor, for example, in construction engineering, whether high-rise buildings are safe needs to be monitored to ensure that construction is carried out smoothly.
Currently, the monitoring methods for building structures rely on a single mechanical device, manual inspection and evaluation operations, for example, using theodolites to measure angles, side lengths, etc. of the building, and then evaluate whether the building is deformed based on these data. The method can not effectively and accurately monitor the building structure, and can not timely find the potential safety problem of the building structure.
Disclosure of Invention
The invention provides a building structure monitoring method, equipment and a storage medium, which can monitor a building structure accurately in real time and discover potential safety problems of the building structure in time.
In order to achieve the above object, the present invention provides a building structure monitoring method, the method comprising:
collecting sensing data of each type of sensor, extracting characteristic data corresponding to the sensing data and storing the characteristic data, wherein each type of sensor is respectively arranged on each key structure of the same building, and the same type of sensor comprises at least one sensor;
Carrying out fusion processing on the characteristic data by utilizing a preset multi-mode fusion algorithm to obtain fusion characteristic data;
Determining a plurality of integration strategies, and constructing a plurality of corresponding monitoring models based on each integration strategy;
training and integrating a plurality of corresponding monitoring models based on each integration strategy and the fusion characteristic data to obtain integrated monitoring models;
and determining an optimal integrated monitoring model from a plurality of integrated monitoring models respectively corresponding to a plurality of integrated strategies, and monitoring a building structure to be monitored by using the optimal integrated monitoring model.
Optionally, the step of fusing the feature data by using a predetermined multimodal fusion algorithm to obtain fused feature data includes:
Acquiring comprehensive characteristic data corresponding to characteristic data of the same type, wherein the characteristic data at least comprises strain characteristic data, displacement characteristic data and vibration characteristic data;
defining a linear layer network for each integrated feature data;
inputting each comprehensive characteristic data into a corresponding linear layer network to acquire data output by each linear layer network;
And carrying out weighted fusion on the data output by each linear layer network to obtain the fusion characteristic data.
Optionally, the step of acquiring the comprehensive feature data corresponding to the feature data of the same type includes:
Acquiring the average value of a plurality of strain characteristic data, and calculating the comprehensive stress k of the strain characteristic data according to the average value, wherein the comprehensive stress k is the comprehensive characteristic data of the strain characteristic data;
Obtaining the maximum value of a plurality of displacement characteristic data, and taking the maximum value as the comprehensive displacement q of the displacement characteristic data, wherein the comprehensive displacement q is the comprehensive characteristic data of the displacement characteristic data;
and acquiring main frequencies corresponding to the vibration characteristic data, acquiring amplitude values corresponding to the main frequencies, and taking the maximum value of the amplitude values as comprehensive vibration v of the vibration characteristic data, wherein the comprehensive vibration v is the comprehensive characteristic data of the vibration characteristic data.
Optionally, the step of fusing the feature data by using a predetermined multimodal fusion algorithm to obtain fused feature data includes:
Defining a convolutional neural network for the same type of feature data;
Inputting the characteristic data of the same type into the corresponding convolutional neural networks to obtain characteristic diagrams output by the convolutional neural networks;
and splicing the feature graphs output by the dimensional convolutional neural network, and acquiring the fusion feature data based on the spliced feature graphs.
Optionally, the integration policies include stacking integration policies, and constructing a corresponding plurality of monitoring models based on each integration policy includes:
Constructing a full supervision model, a semi-supervision model and an integration model as a monitoring model corresponding to the stacking integration strategy, or
And constructing a long-short-term memory network and a conditional random field layer, and taking the long-short-term memory network and the conditional random field layer as a monitoring model corresponding to the stacking integration strategy.
Optionally, the step of training and integrating the corresponding multiple monitoring models based on each integration strategy and the fusion feature data to obtain an integrated monitoring model includes:
dividing the fusion characteristic data into a first set, a second set and a third set, and respectively training the full supervision model and the semi-supervision model by using the fusion characteristic data of the first set;
Respectively inputting the fusion characteristic data of the second set into a trained full supervision model and a trained semi-supervision model to obtain first output data output by the trained full supervision model and second output data output by the trained semi-supervision model;
splicing the first output data and the second output data into training sample data;
constructing an integration model, and training the integration model by using the training sample data;
Respectively inputting the fusion characteristic data of the third set into the trained full supervision model and the half supervision model, acquiring first verification data output by the trained full supervision model and second verification data output by the trained half supervision model, and splicing the first verification data and the second verification data into verification sample data;
And verifying the trained integrated model by using the verification sample data, and taking the trained full-supervision model, the half-supervision model and the integrated model passing verification as the integrated monitoring model.
Optionally, the semi-supervised model includes a graph neural network, wherein constructing the semi-supervised model includes:
And taking the sensor as a node of the graph neural network, fusing the association data of the node, and constructing an edge of the graph neural network based on the fused association data, wherein the association data comprises the sensing data and the non-sensing data.
Optionally, the fused feature data is time series data, wherein training the semi-supervised model with the first set of fused feature data includes:
Marking part of the fusion characteristic data of the first set by using a label, wherein the label is a building structure state corresponding to the fusion characteristic data at each time point;
Training the graph neural network by using the fusion characteristic data with and without labels.
Optionally, the step of training and integrating the corresponding multiple monitoring models based on each integration strategy and the fusion feature data to obtain an integrated monitoring model includes:
integrating the conditional random field layer to an output layer of the long-short-period memory network to obtain a long-short-period memory and conditional random field model;
dividing the fusion characteristic data into a fourth set and a fifth set, and training the long-short-period memory and the conditional random field model by using the fusion characteristic data of the fourth set;
Optimizing parameters of the long-period memory network and the conditional random field layer according to the training result of the long-period memory and the conditional random field model;
and verifying the long-short-period memory and the conditional random field model by utilizing the fusion characteristic data of the fifth set, and taking the verified long-short-period memory and conditional random field model as the integrated monitoring model.
The embodiment of the invention also provides building structure monitoring equipment, which comprises a memory and a processor, wherein the memory stores computer executable instructions capable of running on the processor, and the computer executable instructions realize the method when being executed by the processor.
Embodiments of the present invention also provide a computer-readable storage medium having stored thereon computer-executable instructions executable by one or more processors to implement the above-described methods.
The method and the system of the embodiment of the invention have the following advantages:
In the embodiment of the invention, firstly, sensing data of different types of sensors deployed in a building are collected, characteristic data are extracted, a multi-mode fusion algorithm is utilized to fuse the characteristic data of multiple types to obtain multi-source or multi-mode fusion characteristic data, then multiple integration strategies are determined, multiple monitoring models of each integration strategy are constructed, the monitoring models are trained by utilizing the fusion characteristic data and integrated, the advantages of the multiple monitoring models can be integrated, an integrated monitoring model is obtained, finally, an optimal integrated monitoring model is determined from the multiple integrated monitoring models, and the optimal integrated monitoring model is deployed to monitor the building structure. The embodiment of the invention fuses multi-source or multi-mode data of the building, can comprehensively and carefully analyze the building structure, can integrate the advantages of a plurality of models by multi-model integration, can accurately analyze complex multi-source or multi-mode data, can timely find potential safety problems of the building structure, and can improve the stability and generalization capability of the model.
Drawings
FIG. 1 is a schematic diagram of a hardware device environment of a building structure monitoring method in one embodiment;
FIG. 2 is a flow chart of a method of building structure monitoring in one embodiment;
Fig. 3 is a schematic diagram of hardware components of a cloud image processing device applied to big data in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
Fig. 1 is a schematic view of a hardware device environment according to an embodiment of the present invention, as shown in fig. 1, including a monitoring device 100, and after each sensor A1-A3, B1-B2, and C1-C1 installed on a building senses data, each sensor transmits the sensed data to the monitoring device 100 in real time through a wireless network, where the sensors A1-A3 are the same type of sensor, the sensors B1-B2 are the same type of sensor, and the sensors C1-C1 are the same type of sensor. A plurality of different types of sensors are installed on a building, and one or more sensors of the same type can be deployed on the building, without limitation. The monitoring device 100 has storage and computing capabilities and is deployed with an integrated monitoring model for performing data analysis processing on the multi-modal sensing data to monitor the state of the building structure.
As shown in fig. 2, an embodiment of the present invention provides a building structure monitoring method, which is applied to the hardware device environment shown in fig. 1, and includes the following steps:
S10, collecting sensing data of each type of sensor, extracting characteristic data corresponding to the sensing data and storing the characteristic data, wherein each type of sensor is respectively arranged on each key structure of the same building, and the same type of sensor comprises at least one sensor;
First, according to the purpose of monitoring the building, suitable sensors are selected, including but not limited to structural sensors and environmental monitoring sensors, wherein the structural sensors include strain gauges, accelerometers, inclinometers, and the like, the environmental monitoring sensors include temperature and humidity sensors, and the like, and all the above sensors, or several of the above sensors, or a combination of several of the above sensors and other suitable sensors, and the like, are not limited herein. One or more sensors of the same type may be deployed, for example, to mount multiple strain gauges on the main beam of a building for monitoring strain of the main beam; a plurality of accelerometers are arranged on a certain wall surface of the building and are used for monitoring vibration of the wall surface in the vertical direction. Various types of sensors are installed on the corresponding key structures of the building.
The strain gauge is used for monitoring stress and deformation of a building structure and can be installed on structures such as beams, columns, foundation bearings and the like of the building; the accelerometer is used for monitoring vibration and acceleration of a building structure, can evaluate earthquake resistance, and can be installed on a main supporting structure of the building, such as a joint of a pillar and a frame; inclinometers are used to monitor the degree of inclination of a building structure, to evaluate foundation stability, and are primarily installed in the foundations and underground structures of the building, as well as in areas that may be affected by sedimentation; the temperature and humidity sensor is used for monitoring the environmental temperature and humidity, can evaluate the influence of the temperature and humidity on building materials, can be installed in different areas inside and outside a building, and particularly is easy to influence by the temperature and humidity, such as an outer wall and a roof.
Illustratively, for monitoring a high-rise office building, strain gauges and accelerometers are mounted on the main spandrel girder and column of the high-rise office building for monitoring the stress and vibration conditions of the structure. Meanwhile, inclinometers are installed on the foundation and the wall body, and possible settlement or inclination is monitored. When monitoring a bridge structure, strain gauges and inclinometers are installed on key supporting points and suspension structures of the bridge so as to monitor the deformation and stability of the bridge under the influence of vehicle load and environmental factors. When the historical building is monitored, a temperature and humidity sensor and an inclinometer are installed at important structural parts of the historical building, the influence of environmental changes on the building structure is monitored, and the sign of structural inclination is found in time.
After the sensors are installed, wireless network configuration is carried out on all the sensors so as to ensure real-time transmission of the sensing data, and the sensing data are transmitted to the monitoring equipment.
And collecting diverse sensing data in advance, preprocessing the sensing data, including denoising, interpolation, standardization and the like, to obtain data with better quality. Features of each type of sensing data, such as peak values of stress, maximum rate of change of displacement, frequency distribution of vibration, etc., are extracted using a predetermined algorithm (e.g., fourier transform, waveform analysis).
In one embodiment, extracting feature data corresponding to the sensed data using a fourier transform includes:
a1, intercepting a data fragment: a small piece of data (e.g., a few seconds) is truncated from the preprocessed data for fourier transformation.
A2, performing Fourier transform: and applying a Fast Fourier Transform (FFT) algorithm to the intercepted data segments to obtain frequency domain information thereof.
A3, analyzing a frequency spectrum: the frequency domain information of the signal segment is analyzed, the main frequency component is found, and the frequency, amplitude and phase information of the occurrence of the main frequency component are extracted.
A4, repeating analysis: repeating the steps a1-a3 along a time axis, and carrying out Fourier transformation on a plurality of data fragments to obtain the frequency domain characteristic change of the data in time.
A5, extracting features: based on the spectral information of all segments, features of the signal, such as the main frequency components, the frequency range, and the main energy distribution, etc., are extracted.
A6, storing characteristics: the extracted and stored characteristic data of the sensing data are used for subsequent model input and training analysis of frequency domain characteristics.
Through the above steps, the frequency domain features corresponding to the sensing data can be extracted as the feature data of the present embodiment.
S20, carrying out fusion processing on the characteristic data by utilizing a preset multi-mode fusion algorithm to obtain fusion characteristic data;
In one embodiment, the linear layer network may be used to perform fusion processing on the multi-modal feature data, including the following steps:
b1, acquiring comprehensive characteristic data corresponding to characteristic data of the same type, wherein the characteristic data at least comprises strain characteristic data, displacement characteristic data and vibration characteristic data;
The data sensed by the same type of sensor is the same type of characteristic data, for example, the plurality of strain characteristic data are the same type of characteristic data. The integrated feature data represents data obtained by integrating feature data of the same type.
In an embodiment, acquiring comprehensive feature data corresponding to feature data of the same type includes:
The average value of the strain characteristic data is obtained, the comprehensive stress k of the strain characteristic data is calculated according to the average value, and the comprehensive stress k is the comprehensive characteristic data of the strain characteristic data and can reflect the stress level and the safety margin born by the structure;
The maximum value of the displacement characteristic data is obtained, the maximum value is used as the comprehensive displacement q of the displacement characteristic data, and the comprehensive displacement q is the comprehensive characteristic data of the displacement characteristic data and can reflect the displacement and deformation conditions of the structure;
The main frequencies corresponding to the vibration characteristic data are acquired, the amplitude values corresponding to the main frequencies are acquired, the maximum value in the amplitude values is used as the comprehensive vibration v of the vibration characteristic data, and the comprehensive vibration v is the comprehensive characteristic data of the vibration characteristic data, so that the dynamic characteristics and vibration response of the structure can be reflected.
Illustratively, a bridge is equipped with the following sensors:
1. 4 strain gauges for measuring strain of the main beam;
2. 6 displacement meters for measuring the relative displacement of the wall;
3. 8 accelerometers for measuring wall surface vertical vibration;
then, for a certain period of time, the integrated stress k=e=epsilon, epsilon is the average value of the 4 strain characteristic data, namely the average strain of the main beam, and e is the elastic modulus of the main beam (the object to be measured). The comprehensive displacement q is the maximum value of 6 displacement characteristic data, namely the maximum settlement displacement of the wall body. And analyzing the characteristic data of the 8 accelerometers to obtain a plurality of main frequencies of wall surface vibration, obtaining the amplitude values corresponding to the main frequencies, and taking the maximum value corresponding to the main frequencies and the amplitude values as the comprehensive displacement q.
B2, defining a linear layer network for each comprehensive characteristic data;
Where a linear layer network is defined primarily to determine the number of neurons, each of which is connected to all neurons of the previous layer (e.g., the input layer) and functions to effect a linear combination or linear transformation of the previous layer.
B3, inputting each comprehensive characteristic data into a corresponding linear layer network to acquire data output by each linear layer network;
And inputting each comprehensive characteristic data into a corresponding linear layer network, and transmitting the comprehensive characteristic data forward to obtain corresponding output.
And b4, carrying out weighted fusion on the data output by each linear layer network to obtain the fusion characteristic data.
Optionally, after the weighted fusion, the weight proportion can be further adjusted to realize the optimal fusion of the data. The fused feature data is a feature vector that combines features of different modality data. The different mode data are fused into the fused characteristic data, so that the performance of the structure can be reflected from different aspects.
For example, a building is monitored to obtain stress feature data, displacement feature data and vibration feature data, and then after the stress feature data, the displacement feature data and the vibration feature data are fused through a linear layer network, a feature vector f=w 1*K+w2*Q+w3 ×v is obtained, wherein w 1、w2 and w 3 are weight coefficients, K represents a fused stress component, Q represents a fused displacement component, V represents a fused vibration component, and a feature vector F is fused feature data.
In another embodiment, the convolution layer may be used to perform fusion processing on the multi-modal feature data, including the following steps:
c1, defining a convolutional neural network for the characteristic data of the same type;
Preferably, the convolutional neural network may be a one-dimensional convolutional neural network. A one-dimensional convolutional neural network is a variant of a convolutional neural network that can better handle local relationships in sequence data. Of course, convolutional neural networks may also be used to fuse multi-modal features with other neural networks.
C2, inputting the characteristic data of the same type into the corresponding convolutional neural networks to obtain characteristic diagrams output by the convolutional neural networks;
Setting a convolution kernel parameter, inputting the characteristic data into a corresponding convolution neural network, and extracting a characteristic diagram by utilizing the convolution kernel to serve as output of the convolution neural network.
And c3, splicing the feature graphs output by the convolutional neural networks, and acquiring the fusion feature data based on the spliced feature graphs.
After the feature graphs output by the convolutional neural networks are spliced, a fusion convolutional layer can be constructed, the spliced feature graphs are taken as input, the association between the feature graphs is learned through a convolutional kernel, then a full-connection layer is constructed, the feature graphs are classified or regressed by using the full-connection layer, and the output fusion feature graphs, namely fusion feature data, are obtained, and the fusion feature graphs represent the data by learning the relationship and the mode among different modal features. Compared with the embodiment that the linear layer network is used for fusing the multi-mode characteristic data, the embodiment uses the convolution layer for fusing the multi-mode characteristic data, and better fusing effect can be obtained by learning the association between the characteristics.
S30, determining a plurality of integration strategies, and constructing a plurality of corresponding monitoring models based on each integration strategy;
In this embodiment, a monitoring model is built according to the selected integration strategy, and the monitoring models corresponding to different integration strategies are different. Preferably, the integration strategies include, but are not limited to, bagging (Bagging) integration, boosting (Boosting) integration, stacking (Stacking) integration. Bagging (Bagging) integration is an integration method that improves stability and accuracy by constructing a collection of multiple models, by constructing multiple data sets from random samples that are put back in the original data set, and using these data sets to train the models independently, and then fusing the model results by voting or the like. Boosting (Boosting) integration builds a strong model by combining sequentially weaker models, each new model is trained by analyzing the errors of the last model, and finally the predictions of the various models are linearly combined according to their weights. Stacking (Stacking) integration is a layered integrated learning technique, multiple models are trained on the bottom layer, then the generated new features are used to train the models on the top layer, the models on the bottom layer are trained in parallel, and the models on the top layer are fused with the features on the bottom layer.
In this embodiment, a plurality of integration strategies are selected, for example, the 3 integration strategies may be selected, or two of the 3 integration strategies may be selected, or the 3 integration strategies may be combined with other integration strategies, which are not limited herein.
In an embodiment, a stacking integration strategy is selected, and a full supervision model, a semi-supervision model and an integration model can be constructed to be used as a monitoring model.
In another embodiment, a stack integration strategy is selected, and a long-short term memory network and a conditional random field layer can be constructed as a monitoring model.
It can be understood that if a bagged integration, an improved integration or other integration strategies are selected, a monitoring model corresponding to the strategy can be constructed.
S40, training and integrating a plurality of corresponding monitoring models based on each integration strategy and the fusion characteristic data to obtain integrated monitoring models;
After selecting various integration strategies, determining corresponding monitoring models, and performing model training by using the fusion characteristic data or combining the fusion characteristic data with other processed sensing data according to actual monitoring purposes, such as temperature, humidity and the like, without limitation.
It will be appreciated that one or more integrated monitoring models may be trained under each integration strategy.
In an embodiment, if a full supervision model, a semi-supervision model and an integration model are built as monitoring models, training and integrating the corresponding plurality of monitoring models based on each integration strategy (stacking integration strategy) and the fusion feature data to obtain an integrated monitoring model includes the following steps:
d1, dividing the fusion characteristic data into a first set, a second set and a third set, and respectively training the full supervision model and the semi-supervision model by using the fusion characteristic data of the first set;
it can be understood that the full-supervision model and the semi-supervision model can be trained by using the labeled data, so that a corresponding label can be marked on a part of the fusion characteristic data in advance, wherein the label is a building structure state corresponding to the fusion characteristic data at each time point, for example, the label is normal, sinking, damage and the like. The feature data are fused in the first set, the second set and the third set, so that the data containing the labels can be contained according to actual needs, and the labels in each set are ensured to have diversity, so that the generalization capability of the model is enhanced.
Training a full supervision model:
1. The full supervision model selects the support vector machine, although other suitable full supervision models may be selected.
2. And selecting a proper support vector machine variant, such as a support vector machine classifier, a support vector machine regressor and the like, according to the monitoring purpose of the monitoring task, wherein the support vector machine classifier is used for classifying, and the support vector machine regressor is used for regressing or predicting.
3. A suitable kernel function, such as a linear kernel, a polynomial kernel, or a radial basis function, is selected as the kernel function of the support vector machine.
4. The support vector machine is subjected to parameter tuning, so that the accuracy of the model is improved, meanwhile, over-fitting is prevented, and optimal parameters such as penalty parameters C and kernel function parameters can be selected through methods such as cross-validation.
5. The support vector machine is trained using the partial labeled data of the first set.
6. The training effect of the support vector machine is evaluated by using the label data in the rest of the first set, and indexes such as the accuracy and recall rate of the support vector machine can be calculated so as to ensure that the model has higher accuracy in practical application.
7. And adjusting the support vector machine according to the evaluation result, and adjusting the kernel function type, parameters or rebalancing data for training and the like to optimize the support vector machine, so as to finally obtain a trained model.
In the training of the full-supervision model, the support vector machine is used as a better full-supervision learning method, can be used for processing and analyzing the feature vector data of the building structure, is suitable for high-dimensional data, can better analyze complex fusion feature data, and finally can provide strong monitoring capability for the building structure through optimization.
Training a semi-supervised model:
1. the semi-supervised model selects the neural network, although other suitable semi-supervised models may be selected.
2. And taking the sensor as a node of the graph neural network, fusing the association data of the node, and constructing an edge of the graph neural network based on the fused association data.
The association data are data with a logical association relationship, each node comprises sensing data and non-sensing data, each node is provided with a data set, before the association data are fused, the data sets only comprise the sensing data of the sensors of the nodes, and after the association data are fused, the data sets also comprise a plurality of other association data. It should be noted that, after the associated data are fused, they can be properly processed into frequency domain data for subsequent model training.
In a building structure, in one embodiment, for a certain sensor, sensing data of other sensors may be fused into a data set of corresponding nodes of the sensor based on association relations such as a functional relation and a physical position of the sensor, for example, based on the identity of functions, data of all strain gauges with the same function in a certain building are fused so that the data set of each strain gauge corresponding node contains sensing data of all strain gauges, based on the similarity of functions, data of a temperature sensor and a humidity sensor in a certain building are fused so that the data set of corresponding nodes of the temperature sensor/the humidity sensor contains sensing data of the humidity sensor/the temperature sensor, and based on the physical position, data of all types of sensors installed on the same wall are fused. Further, the above-mentioned sensing data may be further disassembled by a feature engineering and then fused, for example, strain rate, peak strain and other data may be extracted from strain gauge data; for accelerometers, data may be extracted for amplitude, frequency, etc. Then, the disassembled data are fused, so that the fused data have more information.
In another embodiment, for a certain sensor, non-sensor data may be fused into a data set of a corresponding node of the sensor based on a spatial relationship of the sensor, context information of a building, and the like, for example, position information of a sensor installed within a predetermined distance around the certain sensor is fused into a data set of a corresponding node of the sensor based on a spatial distance, and history data or trend information of a building, overall design, and the like, non-sensor data is fused into a data set of a corresponding node of the corresponding sensor based on context information of a building. It will be appreciated that the non-sensor data with associated relationships may be further mined and is not exhaustive herein.
After the association data are fused, in the graph neural network, two nodes with the association data are fused are represented by edges, and the edges indicate which association data and the values of the association data are fused between the two nodes.
3. The selection of an appropriate graph neural network, such as a graph convolution network, a graph annotation network, etc., may be used for classification or prediction, depending on the monitoring purpose of the monitoring task.
4. Training the graph neural network by using the fusion characteristic data with the labels, and learning the initial representation of the nodes.
5. The method comprises the steps of constructing a loss function, such as a classification loss function, retraining a graph neural network by using the fused characteristic data with labels and without labels, and performing characteristic propagation and aggregation by the nodes through edges so as to learn high-level characteristic representation of each node and simultaneously minimize the loss function.
6. Cross-validation is used to adjust and optimize model parameters such as learning rate, hidden layer size and layer number, etc.
7. And evaluating the performance of the graph neural network, calculating the accuracy and evaluating the accuracy.
8. And adjusting the model according to the evaluation result, including modifying the structure of the graph neural network, adjusting the model architecture or rebalancing training data and the like, so as to optimize the graph neural network and finally obtain a trained model.
In training a semi-supervised model, the graph neural network is well suited for use in performing correlation data analysis on building structures. By constructing the graph structure with the monitoring points as nodes, the graph neural network can effectively capture the relationships and interactions between the nodes, thereby providing deeper data analysis. In addition, the graph neural network can effectively utilize a large amount of unlabeled data, and the accuracy of the model in complex data analysis is improved.
D2, respectively inputting the fusion characteristic data of the second set into a trained full supervision model and a trained semi-supervision model to obtain first output data output by the trained full supervision model and second output data output by the trained semi-supervision model;
In this embodiment, new data, that is, the fusion feature data of the second set, is input into the trained full-supervision model and the half-supervision model to obtain corresponding output data respectively.
The values are described in terms of the fully supervised model and the semi-supervised model being consistent for purposes of data analysis processing, e.g., both for prediction or both for classification, and both for prediction results or classification results.
D3, splicing the first output data and the second output data into training sample data;
and splicing the first output data and the second output data as new characteristics into training sample data for training the integrated model.
D4, constructing an integration model, and training the integration model by using the training sample data;
In this embodiment, the integration model is a Meta model, such as a logistic regression model, and the integration model is trained using training sample data.
The integration model is used for integrating the output of the basic model, namely the full supervision model and the semi-supervision model to obtain a final output result, and the optimal output result of the comprehensive information of all the basic models can be obtained through the integration model.
D5, respectively inputting the fusion characteristic data of the third set into the trained full supervision model and the trained half supervision model, obtaining first verification data output by the trained full supervision model and second verification data output by the trained half supervision model, and splicing the first verification data and the second verification data into verification sample data;
In this embodiment, new data, that is, the fusion feature data of the third set, is input into the trained full-supervision model and the half-supervision model to obtain corresponding output data respectively, the integrated model is verified by using the spliced output data, the accuracy of the integrated model is calculated, and the integrated model is optimized.
And d6, verifying the trained integrated model by using the verification sample data, and taking the trained full-supervision model, the half-supervision model and the integrated model passing verification as the integrated monitoring model.
In this embodiment, the integrated model, the previous full-supervision model and the semi-supervision model are combined into a two-level Stacking (Stacking) model, and the integrated model integrates and makes decisions on information from different base models, so that performance and accuracy of the whole integrated monitoring model are improved.
In another embodiment, if a long-term and short-term memory network and a conditional random field layer are constructed as monitoring models, training and integrating a plurality of corresponding monitoring models based on each integration strategy (stacking integration strategy) and fusion characteristic data to obtain an integrated monitoring model includes the following steps:
e1, integrating the conditional random field layer to an output layer of the long-short-period memory network to obtain a long-short-period memory and conditional random field model;
The long-period memory network can capture time sequence relativity and dynamic change of the fusion characteristic data, and the conditional random field layer can effectively label and classify the data after integrating the conditional random field layer to an output layer of the long-period memory network.
E2, dividing the fusion characteristic data into a fourth set and a fifth set, and training the long-period memory and the conditional random field model by using the fusion characteristic data of the fourth set;
Training long-short term memory and conditional random field model includes:
1. Initializing long-short-period memory and a conditional random field model, and creating a long-short-period memory and a conditional random field model instance;
2. and compiling a long-term memory and short-term memory and conditional random field model, and constructing a proper loss function so that the conditional random field can be marked and classified through the loss function.
3. An optimizer (e.g., adam or SGD) is selected for the long-term memory and conditional random field model and its learning rate is set.
4. And preparing fusion characteristic data with labels in the fourth set for long-term memory and supervised learning of the conditional random field model.
5. The fit method or similar interface is used to start training long and short term memory and conditional random field models and set the appropriate batch size and iteration number.
6. The training process is monitored using a callback function or monitoring tool, such as TensorBoard, to monitor loss functions and accuracy changes during the training process. If over-fitting occurs, an early stop strategy may be used or a regularization term may be added to prevent over-fitting from occurring.
Illustratively, the following is a training of a long-short term memory and conditional random field model for monitoring the health of a high-rise building and predicting whether damage will occur:
1. Initializing long-short-period memory and a conditional random field model, and creating a long-short-period memory and a conditional random field model instance;
2. compiling a long-term memory and short-term memory and a conditional random field model, and constructing a loss function;
3. An Adam optimizer was selected and the learning rate was set to 0.001.
4. Defining long-term and short-term memory and input and output of a conditional random field model: inputting time series data which is fusion characteristic data; the structural health status label (e.g., normal, minor, major) is output for each time point.
5. Model training: starting training a long-short term memory and a conditional random field model with the batch size of 32 and the iteration number of 100, and using 80% of the stored fusion characteristic data as a training set and 20% as a verification set.
6. Monitoring a training process: the loss function and accuracy are monitored using TensorBoard.
7. And (5) verifying long-term and short-term memory and a conditional random field model, and monitoring whether fitting occurs. If the loss does not drop significantly over 10 consecutive iterations, training is stopped and regularization terms are added to the model to prevent overfitting.
8. After training is completed, the model is saved for subsequent use.
The long-term memory and the conditional random field model trained by the above examples can monitor the health condition of a high-rise building structure in real time. The model can identify potential structural damage caused by continuous load, provide basis for maintenance and reinforcement, and discover potential safety problems through continuous monitoring and timely feedback, so that risks are greatly reduced.
E3, optimizing parameters of the long-period memory network and the conditional random field layer according to the training result of the long-period memory and the conditional random field model;
in this embodiment, if the training result shows that the long-short term memory and the conditional random field model are to be optimized, parameters of the long-short term memory network and the conditional random field layer, including but not limited to learning rate, network layer number and hidden layer node number, may be adjusted.
And e4, verifying the long-short term memory and the conditional random field model by utilizing the fusion characteristic data of the fifth set, and taking the verified long-short term memory and conditional random field model as the integrated monitoring model.
In this embodiment, the verification of the fusion feature data of the fifth set is used to verify the long-short term memory and the indexes such as the accuracy, recall, and F1 score of the conditional random field model, where F1 score=2 (accuracy×recall)/(accuracy+recall). And verifying the passed long-short period memory and the conditional random field model to achieve the corresponding expected long-short period memory and conditional random field model for various indexes.
S50, determining an optimal integrated monitoring model from a plurality of integrated monitoring models respectively corresponding to a plurality of integrated strategies, and monitoring a building structure to be monitored by using the optimal integrated monitoring model.
In this embodiment, a plurality of integration strategies are selected to obtain a plurality of corresponding integration monitoring models, and an optimal integration monitoring model is selected for practical deployment application by optimizing the plurality of integration monitoring models.
Further, the integrated monitoring model may be optimized according to one or more predetermined optimization rules, and an optimal integrated monitoring model may be determined from the optimized integrated monitoring model, including but not limited to parameter optimization, weight optimization, and integration policy optimization.
The parameter optimization comprises single model parameter optimization and integrated strategy parameter optimization, wherein the single model parameter optimization can be fine parameter adjustment of each single model in the integrated monitoring model, and comprises learning rate, regularization coefficient and the like so as to improve the independent performance of the single model; the integration policy parameter optimization includes optimization of weights in the integration policy, e.g., weight optimization of learners that enhance the integration policy.
In one embodiment, the weight optimization of learner to improve integration policy includes:
1. A base learner, such as a CART decision tree, defining an integration policy.
2. Initializing sample weights, and initializing the weights of all training sample data to the same value.
3. The weight update formula, i.e. the weight of how to update the misclassified samples in each round, may be determined using an exponential or linear function as the update formula.
4. The number of iterations is planned, for example, 10-100.
5. And training a model, namely training a monitoring model corresponding to the improved integration strategy by using a weight updating formula.
6. The effect of the different update formulas and iteration numbers is tested to evaluate the model.
7. Comparing the performance of the model, selecting the optimal configuration of the weight update formula and the iteration number.
8. Retraining the model, and retraining the monitoring model corresponding to the integration strategy by using the selected weight updating formula and the iteration times.
In the above embodiment, by adjusting the weight updating formula and the iteration number, the integration effect of the integration strategy can be optimized and improved, and the generalization performance of the model can be improved.
The weight optimization is performance-based weight adjustment, namely the weight of each monitoring model in the integration is adjusted according to the performance of the monitoring model on the verification set, and the monitoring model with better performance is larger in weight.
The integration strategy optimization comprises the steps of determining a better integration strategy according to the monitoring purpose and the effect of the integrated monitoring model, and increasing the diversity of the monitoring model, so as to ensure that the integrated monitoring model has different types and/or configurations, thereby reducing overfitting and improving generalization capability.
It can be seen that optimizing the integrated monitoring model is an iterative process involving fine tuning of individual models, selection and adjustment of integration strategies, and continuous optimization based on practical application feedback. Through optimization, the performance of the integrated monitoring model is optimal when the monitoring data of the complex building is processed, the monitoring accuracy of the model is improved, and the stability of the model in a changeable environment is enhanced. In addition, the method for integrating the models can obtain reliable safety evaluation and timely risk early warning in the application of building structure monitoring based on the accuracy of monitoring and the stability of the system by combining the advantages of different models, and is suitable for processing multi-mode data and can effectively process multiple types of data by using different models.
And finally, deploying the optimized integrated monitoring model into actual monitoring equipment to perform real-time data analysis.
Illustratively, when monitoring the safety of a high-rise building structure, the integrated monitoring model is used to analyze stress, displacement and vibration data to predict the safety condition of the building structure. The method can be used for accurately predicting the labeled data by utilizing the full supervision model, and meanwhile, the method is combined with the semi-supervision model to process wider unlabeled data. When evaluating the health state of the bridge, the integrated monitoring model can be used for evaluation, so that potential structural problems can be found in advance. When the historical building is evaluated, the integrated monitoring model can be used for evaluating by combining stress, displacement, vibration data, environmental change and other data, and the service life and the potential risk of the historical building are predicted.
In one embodiment, feedback and pre-warning is automatically performed when an abnormal condition of the building structure is detected, such as generating a report including risk areas, types of problems, suggested inspection or maintenance measures, etc.
It will be appreciated that the data in all of the embodiments described above may be suitably pre-processed, e.g. normalized, etc., prior to input of the model, to accommodate the input requirements of the model.
The method of the embodiment of the invention has the following advantages:
In the embodiment of the invention, a building structure monitoring method is provided, firstly, sensing data of sensors of different types deployed in a building are collected, characteristic data are extracted, a multi-mode fusion algorithm is utilized to fuse the characteristic data of multiple types to obtain multi-source or multi-mode fusion characteristic data, then multiple integration strategies are determined, multiple monitoring models of each integration strategy are constructed, the multiple monitoring models are trained and integrated by utilizing the fusion characteristic data, the advantages of the multiple monitoring models can be integrated to obtain an integrated monitoring model, finally, an optimal integrated monitoring model is determined from the multiple integrated monitoring models, and the building structure is monitored by deploying the optimal integrated monitoring model. The embodiment of the invention fuses multi-source or multi-mode data of the building, can comprehensively and carefully analyze the building structure, can integrate the advantages of a plurality of models by multi-model integration, can accurately analyze complex multi-source or multi-mode data, can timely find potential safety problems of the building structure, and can improve the stability and generalization capability of the model.
As shown in fig. 3, an embodiment of the present invention further provides a building structure monitoring device 1000, which includes a memory 101 and a processor 102, where the memory stores computer executable instructions, and the processor executes the computer executable instructions on the memory to implement the following steps:
collecting sensing data of each type of sensor, extracting characteristic data corresponding to the sensing data and storing the characteristic data, wherein each type of sensor is respectively arranged on each key structure of the same building, and the same type of sensor comprises at least one sensor;
Carrying out fusion processing on the characteristic data by utilizing a preset multi-mode fusion algorithm to obtain fusion characteristic data;
Determining a plurality of integration strategies, and constructing a plurality of corresponding monitoring models based on each integration strategy;
training and integrating a plurality of corresponding monitoring models based on each integration strategy and the fusion characteristic data to obtain integrated monitoring models;
and determining an optimal integrated monitoring model from a plurality of integrated monitoring models respectively corresponding to a plurality of integrated strategies, and monitoring a building structure to be monitored by using the optimal integrated monitoring model.
In practical applications, the monitoring device may also include other necessary elements, including but not limited to any number of input 103/output devices 104, processors, controllers, memories, etc., and all systems that can implement the big data management method of the embodiments of the present application are within the scope of the present application.
The memory includes, but is not limited to, random access memory (random access memory, RAM), read-only memory (ROM), erasable programmable read-only memory (erasable programmable read only memory, EPROM), or portable read-only memory (compact disc read to only memory, CD to ROM) for the associated instructions and data.
The input means is for inputting data and/or signals and the output means is for outputting data and/or signals. The output device and the input device may be separate devices or may be a single device.
The processor may include one or more processors, including for example one or more central processing units (central processing unit, CPU), which in the case of a CPU, may be a single-core CPU or a multi-core CPU. The processor may also include one or more special purpose processors, which may include GPUs, FPGAs, etc., for acceleration processing.
The memory is used to store program codes and data for the network device.
The processor is used to call the program code and data in the memory to perform the steps of the method embodiments described above. Reference may be made specifically to the description of the method embodiments, and no further description is given here.
In the several embodiments provided by the present application, it should be understood that the disclosed systems and methods may be implemented in other ways. For example, the division of the unit is merely a logic function division, and there may be another division manner when actually implemented, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted or not performed. The coupling or direct coupling or communication connection shown or discussed with each other may be through some interface, system or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the application, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable system. The computer instructions may be stored in or transmitted across a computer-readable storage medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, digital subscriber line (digital subscriber line, DSL)), or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a read-only memory (ROM), or a random-access memory (random access memory, RAM), or a magnetic medium such as a floppy disk, a hard disk, a magnetic tape, a magnetic disk, or an optical medium such as a digital versatile disk (DIGITAL VERSATILE DISC, DVD), or a semiconductor medium such as a Solid State Disk (SSD), or the like.
The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any equivalent modifications or substitutions will be apparent to those skilled in the art within the scope of the present application, and are intended to be included within the scope of the present application. Therefore, the protection scope of the application is subject to the protection scope of the claims.
Claims (8)
1. A method of building structure monitoring, the method comprising:
collecting sensing data of each type of sensor, extracting characteristic data corresponding to the sensing data and storing the characteristic data, wherein each type of sensor is respectively arranged on each key structure of the same building, and the same type of sensor comprises at least one sensor;
Carrying out fusion processing on the characteristic data by utilizing a preset multi-mode fusion algorithm to obtain fusion characteristic data;
Determining a plurality of integration strategies, and constructing a plurality of corresponding monitoring models based on each integration strategy;
training and integrating a plurality of corresponding monitoring models based on each integration strategy and the fusion characteristic data to obtain integrated monitoring models;
Determining an optimal integrated monitoring model from a plurality of integrated monitoring models corresponding to a plurality of integrated strategies respectively, and monitoring a building structure to be monitored by using the optimal integrated monitoring model;
The step of fusing the feature data by using a preset multi-mode fusion algorithm to obtain fused feature data comprises the following steps:
Acquiring comprehensive characteristic data corresponding to characteristic data of the same type, wherein the characteristic data at least comprises strain characteristic data, displacement characteristic data and vibration characteristic data;
defining a linear layer network for each integrated feature data;
inputting each comprehensive characteristic data into a corresponding linear layer network to acquire data output by each linear layer network;
carrying out weighted fusion on the data output by each linear layer network to obtain the fusion characteristic data;
the step of acquiring comprehensive feature data corresponding to the feature data of the same type comprises the following steps:
Acquiring the average value of a plurality of strain characteristic data, and calculating the comprehensive stress k of the strain characteristic data according to the average value, wherein the comprehensive stress k is the comprehensive characteristic data of the strain characteristic data;
Obtaining the maximum value of a plurality of displacement characteristic data, and taking the maximum value as the comprehensive displacement q of the displacement characteristic data, wherein the comprehensive displacement q is the comprehensive characteristic data of the displacement characteristic data;
Acquiring main frequencies corresponding to a plurality of vibration characteristic data, acquiring amplitude values corresponding to the plurality of main frequencies, and taking the maximum value of the amplitude values as comprehensive vibration v of the vibration characteristic data, wherein the comprehensive vibration v is comprehensive characteristic data of the vibration characteristic data;
The integration strategy comprises stacking integration strategies, and the construction of a plurality of corresponding monitoring models based on each integration strategy comprises the following steps:
Constructing a full supervision model, a semi-supervision model and an integration model as monitoring models corresponding to the stacking integration strategy, or constructing a long-short-term memory network and a conditional random field layer as monitoring models corresponding to the stacking integration strategy;
wherein, from a plurality of integrated monitoring models that a plurality of integrated strategies respectively correspond, confirm the optimal integrated monitoring model, include:
And optimizing the integrated monitoring model according to one or more preset optimization rules, and determining an optimal integrated monitoring model from the optimized integrated monitoring model, wherein the optimal integrated monitoring model comprises parameter optimization, weight optimization and integration strategy optimization.
2. The method of claim 1, wherein the step of fusing the feature data using a predetermined multi-modal fusion algorithm to obtain fused feature data comprises:
Defining a convolutional neural network for the same type of feature data;
Inputting the characteristic data of the same type into the corresponding convolutional neural networks to obtain characteristic diagrams output by the convolutional neural networks;
and splicing the feature graphs output by the dimensional convolutional neural network, and acquiring the fusion feature data based on the spliced feature graphs.
3. The method of claim 1, wherein the step of training and integrating the corresponding plurality of monitoring models based on each integration strategy and the fused feature data to obtain an integrated monitoring model comprises:
dividing the fusion characteristic data into a first set, a second set and a third set, and respectively training the full supervision model and the semi-supervision model by using the fusion characteristic data of the first set;
Respectively inputting the fusion characteristic data of the second set into a trained full supervision model and a trained semi-supervision model to obtain first output data output by the trained full supervision model and second output data output by the trained semi-supervision model;
splicing the first output data and the second output data into training sample data;
constructing an integration model, and training the integration model by using the training sample data;
Respectively inputting the fusion characteristic data of the third set into the trained full supervision model and the half supervision model, acquiring first verification data output by the trained full supervision model and second verification data output by the trained half supervision model, and splicing the first verification data and the second verification data into verification sample data;
And verifying the trained integrated model by using the verification sample data, and taking the trained full-supervision model, the half-supervision model and the integrated model passing verification as the integrated monitoring model.
4. A method according to claim 3, wherein the semi-supervised model comprises a graph neural network, wherein constructing the semi-supervised model comprises:
And taking the sensor as a node of the graph neural network, fusing the association data of the node, and constructing an edge of the graph neural network based on the fused association data, wherein the association data comprises the sensing data and the non-sensing data.
5. The method of claim 4, wherein the fused feature data is time series data, wherein training the semi-supervised model with the first set of fused feature data comprises:
Marking part of the fusion characteristic data of the first set by using a label, wherein the label is a building structure state corresponding to the fusion characteristic data at each time point;
Training the graph neural network by using the fusion characteristic data with and without labels.
6. The method of claim 1, wherein the step of training and integrating the corresponding plurality of monitoring models based on each integration strategy and the fused feature data to obtain an integrated monitoring model comprises:
integrating the conditional random field layer to an output layer of the long-short-period memory network to obtain a long-short-period memory and conditional random field model;
dividing the fusion characteristic data into a fourth set and a fifth set, and training the long-short-period memory and the conditional random field model by using the fusion characteristic data of the fourth set;
Optimizing parameters of the long-period memory network and the conditional random field layer according to the training result of the long-period memory and the conditional random field model;
and verifying the long-short-period memory and the conditional random field model by utilizing the fusion characteristic data of the fifth set, and taking the verified long-short-period memory and conditional random field model as the integrated monitoring model.
7. A building structure monitoring device comprising a memory and a processor, the memory having stored thereon computer executable instructions executable on the processor, which when executed by the processor, implement the method of any of claims 1-6.
8. A computer-readable storage medium having stored thereon computer-executable instructions executable by one or more processors to implement the method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410223955.7A CN117824756B (en) | 2024-02-29 | 2024-02-29 | Building structure monitoring method, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410223955.7A CN117824756B (en) | 2024-02-29 | 2024-02-29 | Building structure monitoring method, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117824756A CN117824756A (en) | 2024-04-05 |
| CN117824756B true CN117824756B (en) | 2024-05-28 |
Family
ID=90524362
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410223955.7A Active CN117824756B (en) | 2024-02-29 | 2024-02-29 | Building structure monitoring method, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117824756B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118013268A (en) * | 2024-04-08 | 2024-05-10 | 中大智能科技股份有限公司 | Bridge support monitoring system design method and device |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105865522A (en) * | 2016-04-05 | 2016-08-17 | 江苏道亿智能科技有限公司 | Bridge structure monitoring system |
| CN106525129A (en) * | 2016-11-21 | 2017-03-22 | 深圳市道桥维修中心桥梁检测站 | Intelligent bridge monitoring system based on multi-source detection fusion means |
| CN110135058A (en) * | 2019-05-14 | 2019-08-16 | 北京工业大学 | Mill load parameter prediction method based on multi-modal feature subset selection integrated moulding |
| CN112997223A (en) * | 2018-10-29 | 2021-06-18 | 赫克斯冈技术中心 | Facility monitoring system and method |
| WO2022000528A1 (en) * | 2020-06-29 | 2022-01-06 | 浙江工业大学 | Bridge scouring multi-source monitoring system and monitoring method, and scouring depth evaluation method therefor |
| KR20220053333A (en) * | 2020-10-22 | 2022-04-29 | 중앙대학교 산학협력단 | Bridge Displacement Estimation Using a Co-Located Acceleration and Strain |
| CN115146230A (en) * | 2022-07-06 | 2022-10-04 | 河南大学 | Ancient building health monitoring system, method and equipment |
| CN115808324A (en) * | 2023-01-30 | 2023-03-17 | 湖南东数交通科技有限公司 | Lightweight safety management monitoring method and system for small and medium-span bridges |
| CN116304835A (en) * | 2023-03-31 | 2023-06-23 | 北京博华信智科技股份有限公司 | AI-based dynamic equipment working condition monitoring management method, system, equipment and medium |
| CN116337158A (en) * | 2023-05-24 | 2023-06-27 | 中国建筑科学研究院有限公司 | Building health monitoring device and method |
| CN116561670A (en) * | 2023-07-12 | 2023-08-08 | 森特士兴集团股份有限公司 | Metal roof health state identification and alarm method |
| CN116957156A (en) * | 2023-07-26 | 2023-10-27 | 深圳市丰锦晟建筑工程管理有限公司 | Building engineering monitoring management system and optimization method based on BIM technology |
| CN117076928A (en) * | 2023-08-25 | 2023-11-17 | 中交路桥科技有限公司 | Bridge health state monitoring method, device and system and electronic equipment |
| CN117171842A (en) * | 2023-08-04 | 2023-12-05 | 华南理工大学 | Urban slow-moving bridge health monitoring and digital twin system |
| CN117236930A (en) * | 2023-10-08 | 2023-12-15 | 中交长大桥隧技术有限公司 | 5G and artificial intelligence based highway bridge comprehensive monitoring system |
| CN117540174A (en) * | 2024-01-09 | 2024-02-09 | 江苏省建筑工程质量检测中心有限公司 | Building structure multi-source heterogeneous data intelligent analysis system and method based on neural network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11210570B2 (en) * | 2018-01-23 | 2021-12-28 | Intelligent Fusion Technology, Inc. | Methods, systems and media for joint manifold learning based heterogenous sensor data fusion |
| JP2024519533A (en) * | 2021-05-06 | 2024-05-15 | ストロング フォース アイオーティ ポートフォリオ ２０１６，エルエルシー | Quantum, Biological, Computer Vision and Neural Network Systems for the Industrial Internet of Things |
-
2024
- 2024-02-29 CN CN202410223955.7A patent/CN117824756B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105865522A (en) * | 2016-04-05 | 2016-08-17 | 江苏道亿智能科技有限公司 | Bridge structure monitoring system |
| CN106525129A (en) * | 2016-11-21 | 2017-03-22 | 深圳市道桥维修中心桥梁检测站 | Intelligent bridge monitoring system based on multi-source detection fusion means |
| CN112997223A (en) * | 2018-10-29 | 2021-06-18 | 赫克斯冈技术中心 | Facility monitoring system and method |
| CN110135058A (en) * | 2019-05-14 | 2019-08-16 | 北京工业大学 | Mill load parameter prediction method based on multi-modal feature subset selection integrated moulding |
| WO2022000528A1 (en) * | 2020-06-29 | 2022-01-06 | 浙江工业大学 | Bridge scouring multi-source monitoring system and monitoring method, and scouring depth evaluation method therefor |
| KR20220053333A (en) * | 2020-10-22 | 2022-04-29 | 중앙대학교 산학협력단 | Bridge Displacement Estimation Using a Co-Located Acceleration and Strain |
| CN115146230A (en) * | 2022-07-06 | 2022-10-04 | 河南大学 | Ancient building health monitoring system, method and equipment |
| CN115808324A (en) * | 2023-01-30 | 2023-03-17 | 湖南东数交通科技有限公司 | Lightweight safety management monitoring method and system for small and medium-span bridges |
| CN116304835A (en) * | 2023-03-31 | 2023-06-23 | 北京博华信智科技股份有限公司 | AI-based dynamic equipment working condition monitoring management method, system, equipment and medium |
| CN116337158A (en) * | 2023-05-24 | 2023-06-27 | 中国建筑科学研究院有限公司 | Building health monitoring device and method |
| CN116561670A (en) * | 2023-07-12 | 2023-08-08 | 森特士兴集团股份有限公司 | Metal roof health state identification and alarm method |
| CN116957156A (en) * | 2023-07-26 | 2023-10-27 | 深圳市丰锦晟建筑工程管理有限公司 | Building engineering monitoring management system and optimization method based on BIM technology |
| CN117171842A (en) * | 2023-08-04 | 2023-12-05 | 华南理工大学 | Urban slow-moving bridge health monitoring and digital twin system |
| CN117076928A (en) * | 2023-08-25 | 2023-11-17 | 中交路桥科技有限公司 | Bridge health state monitoring method, device and system and electronic equipment |
| CN117236930A (en) * | 2023-10-08 | 2023-12-15 | 中交长大桥隧技术有限公司 | 5G and artificial intelligence based highway bridge comprehensive monitoring system |
| CN117540174A (en) * | 2024-01-09 | 2024-02-09 | 江苏省建筑工程质量检测中心有限公司 | Building structure multi-source heterogeneous data intelligent analysis system and method based on neural network |
Non-Patent Citations (2)
| Title |
|---|
| 基于光纤传感的桥梁病害识别技术探究;侯跃敏 等;《大众标准化》;20240131(第2期);第61-63页 * |
| 大型桥梁结构阻尼特性及识别方法研究综述;陈政清 等;《中国公路学报》;20230731;第36卷(第7期);第1-15页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117824756A (en) | 2024-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Avci et al. | A review of vibration-based damage detection in civil structures: From traditional methods to Machine Learning and Deep Learning applications | |
| Bao et al. | Computer vision and deep learning–based data anomaly detection method for structural health monitoring | |
| CN117824756B (en) | Building structure monitoring method, equipment and storage medium | |
| Guo et al. | Structural health monitoring by using a sparse coding-based deep learning algorithm with wireless sensor networks | |
| Laory et al. | Combined model-free data-interpretation methodologies for damage detection during continuous monitoring of structures | |
| Sarmadi et al. | Application of supervised learning to validation of damage detection | |
| CN108363886A (en) | Deformation prediction method and system based on deep learning | |
| Ding et al. | Structural damage identification by sparse deep belief network using uncertain and limited data | |
| CN112528365B (en) | Method for predicting healthy evolution trend of underground infrastructure structure | |
| Tian et al. | Relationship modeling between vehicle‐induced girder vertical deflection and cable tension by BiLSTM using field monitoring data of a cable‐stayed bridge | |
| CN112116184A (en) | Factory risk estimation using historical inspection data | |
| Dang et al. | Structural damage detection framework based on graph convolutional network directly using vibration data | |
| Luleci et al. | Generative adversarial networks for labeled acceleration data augmentation for structural damage detection | |
| CN114563150B (en) | Bridge health online detection module generation method, detection method, tool box and device | |
| Lei et al. | Displacement response estimation of a cable-stayed bridge subjected to various loading conditions with one-dimensional residual convolutional autoencoder method | |
| Garcia-Sanchez et al. | Bearing assessment tool for longitudinal bridge performance | |
| Ye et al. | A deep learning-based method for automatic abnormal data detection: Case study for bridge structural health monitoring | |
| Seventekidis et al. | Model error effects in supervised damage identification of structures with numerically trained classifiers | |
| Anaissi et al. | Multi-objective variational autoencoder: an application for smart infrastructure maintenance | |
| Shi et al. | A novel unsupervised real‐time damage detection method for structural health monitoring using machine learning | |
| US20220101198A1 (en) | Automated generation of a machine learning model from computational simulation data | |
| Luckey et al. | Explainable artificial intelligence to advance structural health monitoring | |
| Lin et al. | Damage detection in a benchmark structure using long short-term memory networks | |
| Wang et al. | Adaptive change detection for long-term machinery monitoring using incremental sliding-window | |
| Chen et al. | A vibration-based 1DCNN-BiLSTM model for structural state recognition of RC beams |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |