Construction engineering quality safety data identification system based on quality safety spectrum embedded operator

Abstract

The invention discloses a construction engineering quality safety data identification system based on a quality safety spectrum embedded operator, and relates to the technical field of construction quality supervision. The objective of the invention is to solve the problem that the existing construction engineering safety data identification method is low in identification accuracy. The method comprises the following steps: inputting a quality-safety tensor, and obtaining a quality-safety feature fusion tensor; establishing a quality-safety spectrum embedding operator, inputting a quality-safety feature fusion variable tensor, and outputting a quality-safety embedding feature matrix; establishing a quality-safety multi-scale feature extraction operator, inputting a quality-safety embedded feature matrix, and outputting a quality-safety multi-scale feature vector; processing the quality and safety training data set by using a quality and safety multi-scale feature extraction operator to obtain a quality and safety deep learning model; and performing quality safety evaluation on the to-be-evaluated constructional engineering data by using a quality safety spectrum embedding operator, a quality safety multi-scale feature extraction operator and a quality safety deep learning model to obtain a quality safety evaluation result. The method is used for constructional engineering quality supervision and alarm.

Description

Technical Field

[0001] The invention relates to the technical field of building quality supervision, and in particular to a building engineering quality safety data identification system based on a quality and safety spectrum embedding operator. Background Art

[0002] The quality and safety of construction projects are directly related to the development of the national economy, the process of urbanization, and the quality of people's lives. Accurate and comprehensive analysis of construction project quality and safety data is of great significance to ensuring public safety, optimizing resource allocation, and improving the overall level of the construction industry. Therefore, obtaining scientific and objective identification and analysis results of construction project quality and safety data is of great value. Among the many data identification and analysis methods, computer-assisted quantitative analysis methods have become the key way to quickly obtain comprehensive analysis results due to their high efficiency, accuracy, and strong repeatability. In view of this, the development of advanced computer-assisted dimensionality reduction and feature characterization methods for construction project quality and safety data has important practical significance and application value.

[0003] At present, there are two main methods for identifying and analyzing the quality and safety data of construction projects: the first is the traditional expert analysis method, which uses an experienced team of experts to conduct on-site inspections and document reviews, and gives analysis results based on a preset indicator system. This method can comprehensively consider many factors, but it has the problems of strong subjectivity, low efficiency, and difficulty in handling large-scale projects. In addition, there may be large differences in the analysis results between different expert teams, which affects the consistency and comparability of the analysis. The second type is an analysis method based on statistical models, such as multivariate regression analysis and fuzzy comprehensive evaluation. This type of method collects a large amount of historical data and establishes a mathematical model to predict and analyze the quality and safety status of construction projects. This method can handle large-scale data, and the identification results obtained are more objective and efficient. In the quality and safety assessment of construction projects, accurately identifying and representing complex multidimensional data is a key challenge. Although the second type of method improves the objectivity and repeatability of the analysis, since the quality and safety data of construction projects usually involve multiple dimensions and there are complex relationships between indicators, this type of method often has difficulty in effectively processing high-dimensional, nonlinear complex indicator systems, and therefore cannot capture these relationships, resulting in a lack of comprehensiveness in the analysis results. In particular, in a high-dimensional data environment, the assumptions of the identification model are easily challenged, resulting in low identification accuracy. At the same time, traditional statistical models have high requirements for data quality and are difficult to adapt to the incomplete and uncertain data often encountered in construction projects. Incomplete or uncertain data may cause the identification model to ignore or mishandle certain key features, resulting in large deviations in the identification results, and ultimately leading to low identification accuracy of construction project quality and safety data. Summary of the invention

[0004] The purpose of the present invention is to solve the problem of low recognition accuracy in existing construction engineering safety data recognition methods, and propose a construction engineering quality safety data recognition system based on quality and safety spectrum embedding operator.

[0005] The construction engineering quality and safety data identification system based on the quality and safety spectrum embedding operator includes: a fusion tensor acquisition operator construction module, a quality and safety spectrum embedding operator construction module, a quality and safety multi-scale feature extraction operator construction module, a quality and safety deep learning model training module and a construction engineering quality and safety evaluation result acquisition module;

[0006] The fused tensor acquisition operator construction module is used to construct a fused tensor acquisition operator;

[0007] The fusion tensor acquisition operator input is the quality and safety tensor QualitySafetyTensor, and the fusion tensor acquisition operator output is the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor;

[0008] The quality and safety spectrum embedding operator construction module is used to establish the quality and safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding;

[0009] The QualitySafetyTensorSpectralEmbedding input is the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor, and the QualitySafetyTensorSpectralEmbedding output is the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix;

[0010] The quality safety multi-scale feature extraction operator construction module is used to establish the quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor;

[0011] The QualitySafetyMultiscaleFeatureExtractor input is the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix, and the QualitySafetyMultiscaleFeatureExtractor output is the quality and safety multi-scale feature vector QualitySafetyMultiscaleFeatureVector;

[0012] The quality safety deep learning model training module is used to input the quality safety training data set QualitySafetyTrainSet, call the fusion tensor acquisition operator, the quality safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, and the quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to process the QualitySafetyTrainSet, and use the processed QualitySafetyTrainSet to train the quality safety deep learning model QualitySafetyDeepModel to obtain the trained QualitySafetyDeepModel;

[0013] The construction project quality and safety evaluation result acquisition module uses a fusion tensor acquisition operator, QualitySafetyTensorSpectralEmbedding, QualitySafetyMultiscaleFeatureExtractor and a trained QualitySafetyDeepModel to perform quality and safety evaluation on the construction project data to be evaluated, and obtains the quality and safety evaluation result of the construction project to be evaluated.

[0014] Furthermore, the fused tensor acquisition operator construction module is used to construct a fused tensor acquisition operator, specifically:

[0015] S101, construct a fusion tensor acquisition operator, the input of which is the quality safety tensor QualitySafetyTensor;

[0016] The quality and safety tensor QualitySafetyTensor stores construction engineering quality evaluation and scoring data from different sources;

[0017] Among them, the sources of construction project quality evaluation and scoring data include: design drawings, on-site testing, sensor monitoring, and manual scoring;

[0018] The construction project quality evaluation scoring data includes: structural strength scoring data, material performance scoring data, and construction quality scoring data;

[0019] S102, obtaining the order of the quality and safety tensor QualitySafetyTensorOrder=the order of QualitySafetyTensor;

[0020] S103, establish a quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor = establish a tensor with the same dimension as QualitySafetyTensor, and initialize all elements to 0;

[0021] S104, initializing the quality and safety feature fusion counter QualitySafetyFusionCounter=1;

[0022] S105, establish a quality safety single feature temporary storage tensor QualitySafetySingleFeatureTensor = extract the QualitySafetyFusionCounterth feature of QualitySafetyTensor;

[0023] S106, obtaining the maximum value of the quality and safety single feature maximum value QualitySafetyFeatureMax=QualitySafetySingleFeatureTensor;

[0024] Get the minimum value of the quality and safety single feature QualitySafetyFeatureMin = the minimum value of QualitySafetySingleFeatureTensor;

[0025] S107, update QualitySafetySingleFeatureTensor using QualitySafetyFeatureMax and QualitySafetyFeatureMin;

[0026] S108, update QualitySafetyFeatureFusionTensor:

[0027] QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor+QualitySafetySingle FeatureTensor

[0028] S109, let QualitySafetyFusionCounter=QualitySafetyFusionCounter+1;

[0029] S110, if QualitySafetyFusionCounter>QualitySafetyTensorOrder, go to S111, otherwise go to S105;

[0030] S111, update QualitySafetyFeatureFusionTensor:

[0031] QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor / QualitySafetyTensor Order;

[0032] S112, obtaining the standard deviation of the current quality and safety feature standard deviation QualitySafetyFeatureStandardDev = QualitySafetyFeatureFusionTensor;

[0033] S113, using QualitySafetyFeatureStandardDev to obtain the quality and safety feature discrimination QualitySafetyFeatureDiscrimination;

[0034] S114, update QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor×QualitySafetyFeatureDiscrimination;

[0035] S115, taking the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor as the output of the fusion tensor acquisition operator.

[0036] Furthermore, the use of QualitySafetyFeatureMax and QualitySafetyFeatureMin in S107 to update QualitySafetySingleFeatureTensor is specifically as follows:

[0037]

[0038] Furthermore, the method of using QualitySafetyFeatureStandardDev in S113 to obtain the quality and safety feature discrimination QualitySafetyFeatureDiscrimination is specifically as follows:

[0039]

[0040] Among them, round() is the rounding function.

[0041] Furthermore, the quality and safety spectrum embedding operator construction module is used to establish the quality and safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, specifically:

[0042] S201, establish a quality safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, the input of QualitySafetyTensorSpectralEmbedding is the quality safety feature fusion tensor QualitySafetyFeatureFusionTensor;

[0043] S202, calculate the high-order singular value decomposition of QualitySafetyFeatureFusionTensor:

[0044] HOSVD(QualitySafetyFeatureFusionTensor)=[QualitySafetyU1,QualitySafetyU2,...,QualitySafetyUn,QualitySafetyS]

[0045] Among them, HOSVD is the high-order singular value decomposition, QualitySafetyU1, QualitySafetyU2,..., QualitySafetyUn are orthogonal factor matrices, QualitySafetyS is the quality and safety Tucker core tensor, and n is the total number of orthogonal factor matrices;

[0046] S203, calculate the spectral norm QualitySafetyλ of the quality and safety Tucker core tensor QualitySafetyS:

[0047] QualitySafetyλ=||QualitySafetyS|| σ =max(QualitySafetyσ1,QualitySafetyσ2,...,QualitySafetyσr')

[0048] where QualitySafetyλ is the spectral norm of QualitySafetyS, ||QualitySafetyS|| σ is the spectral norm of QualitySafetyS, QualitySafetyσk is the singular value of QualitySafetyS, r' is the rank, and k ranges from 1 to r';

[0049] S204, construct the quality and safety low rank tensor QualitySafetyLowRankApproximation:

[0050] QualitySafetyLowRankApproximation=QualitySafetyS×1QualitySafetyU1(:,1:r1)×2QualitySafetyU2(:,1:r2)...×n QualitySafetyUn(:,1:rn)

[0051]

[0052] Where ε is the preset threshold, ×i represents the tensor-matrix multiplication along the i-th dimension, i ranges from 1 to n, QualitySafetyUi(:,1:ri) takes the first ri columns of the QualitySafetyUi matrix, and QualitySafetyσk_i is the k-th singular value in the i-th dimension;

[0053] S205, calculate the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix:

[0054] QualitySafetyEmbeddedFeatureMatrix=Matrix expansion (QualitySafetyLowRankApproximation)

[0055] S206, apply nonlinear activation function to update the quality and safety embedding feature matrix:

[0056] QualitySafetyEmbeddedFeatureMatrix=tanh(QualitySafetyW*QualitySafetyEmbeddedFeatureMatrix+QualitySafetyb)

[0057] Among them, QualitySafetyW and QualitySafetyb are learnable parameters, and tanh is the hyperbolic tangent function;

[0058] S207, use the QualitySafetyEmbeddedFeatureMatrix as the output of QualitySafetyTensorSpectralEmbedding.

[0059] Furthermore, the quality safety multi-scale feature extraction operator construction module is used to establish the quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor, specifically:

[0060] S301, establish a quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor, and input the quality safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix;

[0061] S302, initializing the quality safety multi-scale feature vector QualitySafetyMultiscaleFeatureVector; expanding the QualitySafetyEmbeddedFeatureMatrix matrix into a one-dimensional vector;

[0062] S303, looping through different scales, setting the quality and safety scale index QualitySafetyScaleIndex = 1;

[0063] S304, extract features from the current scale to obtain the quality and safety scale feature QualitySafetyScaleFeature:

[0064] QualitySafetyScaleFeature=[QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD]=Discrete wavelet transform (QualitySafetyEmbeddedFeatureMatrix,'db4',QualitySafetyScaleIndex)

[0065] Where db4 represents Daubechies wavelet, QualitySafetycA is the approximation coefficient, QualitySafetycH, QualitySafetycV, and QualitySafetycD are the horizontal detail coefficient, vertical detail coefficient, and diagonal detail coefficient, respectively;

[0066] S305, calculate the statistical feature QualitySafetyScaleFeature of the current scale:

[0067] QualitySafetyScaleFeature=[QualitySafetyμ,QualitySafetyσ',QualitySafetyE]

[0068] QualitySafetyμ=mean([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD])

[0069] QualitySafetyσ'=std([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD])

[0070] QualitySafetyE=entropy([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD])

[0071] Among them, QualitySafetyμ is the mean of the quality and safety data at the current scale, QualitySafetyσ' is the standard deviation of the quality and safety data at the current scale, QualitySafetyE is the entropy of the quality and safety data at the current scale, mean() is the mean function, std() is the standard deviation function, and entropy() is the entropy function;

[0072] S306, apply the attention mechanism to update QualitySafetyScaleFeature:

[0073] QualitySafetyScaleFeature=QualitySafetyAttentionWeight*QualitySafetyScaleFeature

[0074] Among them, QualitySafetyAttentionWeight is the quality and safety scale feature attention weight;

[0075] S307, adding QualitySafetyScaleFeature to QualitySafetyMultiscaleFeatureVector;

[0076] S308, let QualitySafetyScaleIndex=QualitySafetyScaleIndex+1;

[0077] S309, if QualitySafetyScaleIndex is greater than the preset maximum quality and safety scale QualitySafetyMaxScale, go to S310, otherwise go to S304;

[0078] S310, update QualitySafetyMultiscaleFeatureVector:

[0079] QualitySafetyMultiscaleFeatureVector=CNN1D(QualitySafetyMultiscaleFeatureVector)

[0080] Among them, CNN1D is a one-dimensional convolutional neural network;

[0081] S311, output QualitySafetyMultiscaleFeatureVector as the result of QualitySafetyMultiscaleFeatureExtractor.

[0082] Furthermore, the quality safety scale feature attention weight QualitySafetyAttentionWeight in S306 is obtained by:

[0083] QualitySafetyAttentionWeight=softmax(QualitySafetyV*tanh(QualitySafetyW*QualitySafetyScaleFeature+QualitySafetyb))

[0084] Among them, tanh is the hyperbolic tangent function, QualitySafetyV, QualitySafetyW, and QualitySafetyb are learnable parameters.

[0085] Furthermore, the quality and safety deep learning model training module is used to input the quality and safety training data set QualitySafetyTrainSet, call the fusion tensor acquisition operator, the quality and safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, and the quality and safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to process the QualitySafetyTrainSet, and use the processed QualitySafetyTrainSet to train the quality and safety deep learning model QualitySafetyDeepModel to obtain the trained QualitySafetyDeepModel, specifically:

[0086] S401, inputting a quality safety training data set QualitySafetyTrainSet, wherein QualitySafetyTrainSet includes quality safety indicators and corresponding quality safety evaluation results;

[0087] The quality safety index values ​​in the quality safety training data set QualitySafetyTrainSet include: structural strength value, material performance value and construction quality value;

[0088] The quality and safety evaluation result is the safety risk level of the construction project;

[0089] S402, calling the fusion tensor acquisition operator, the quality and safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, and the quality and safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to process each sample in QualitySafetyTrainSet to obtain the quality and safety training feature matrix QualitySafetyTrainFeatureMatrix;

[0090] S403, constructing a quality safety deep neural network model QualitySafetyDeepModel;

[0091] The quality safety deep neural network model QualitySafetyDeepModel includes: an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer;

[0092] The number of nodes in the input layer is equal to the dimension of QualitySafetyMultiscaleFeatureVector;

[0093] The first hidden layer contains 256 nodes and uses the LeakyReLU activation function;

[0094] The second hidden layer contains 128 nodes and uses the LeakyReLU activation function;

[0095] The third hidden layer contains 64 nodes and uses the LeakyReLU activation function;

[0096] The output layer includes 1 node, uses Sigmoid activation function to output the safety risk level value of the construction project;

[0097] Among them, when the safety risk level of construction projects is within the range of [0.0-0.3], it means there is no safety risk; when the safety risk level of construction projects is within the range of [0.3-0.6], it means a first-level safety risk and an early warning report is required; when the safety risk level of construction projects is within the range of [0.6-1.0], it means there is a second-level safety risk and an early warning report and processing are required immediately;

[0098] S404, define a loss function QualitySafetyLoss;

[0099] S405, using the Adam optimizer, setting the learning rate to 0.001, and setting the learning rate decay strategy:

[0100] QualitySafetylr=QualitySafetylr0 / (1+QualitySafetydecay_rate*epoch)

[0101] Among them, QualitySafetylr is the current learning rate, QualitySafetylr0 is the initial learning rate, QualitySafetydecay_rate is the learning rate decay rate, and epoch is the number of training iterations;

[0102] S406, applying batch normalization and Dropout;

[0103] S407, based on the settings of S404, S405, and S406, the QualitySafetyDeepModel model is trained using k-fold cross validation, and an early stopping strategy is applied to obtain a trained QualitySafetyDeepModel.

[0104] Furthermore, the loss function QualitySafetyLoss defined in S404 is specifically:

[0105] QualitySafetyLoss=α*MSE+(1-α)*MAE+β*L2 regularization

[0106] Among them, α and β are hyperparameters, MSE is the mean square error between the predicted value and the actual value, and MAE is the mean absolute error between the predicted value and the actual value.

[0107] Furthermore, the construction project quality and safety evaluation result acquisition module uses the fusion tensor acquisition operator, QualitySafetyTensorSpectralEmbedding, QualitySafetyMultiscaleFeatureExtractor and the trained QualitySafetyDeepModel to perform quality and safety evaluation on the construction project data to be evaluated, and obtains the quality and safety evaluation result of the construction project to be evaluated, specifically:

[0108] S501, inputting new quality and safety data QualitySafetyNewData of the construction project quality and safety data to be tested, calling the fusion tensor acquisition operator to process QualitySafetyNewData, and obtaining a new fusion tensor of the quality and safety features of the construction project to be evaluated;

[0109] S502, calling QualitySafetyTensorSpectralEmbedding to process the new tensor of quality and safety feature fusion to obtain new quality and safety embedded data QualitySafetyEmbeddedNewData;

[0110] S503, calling QualitySafetyMultiscaleFeatureExtractor to process the quality safety embedded new data QualitySafetyEmbeddedNewData to obtain the quality safety multi-scale new data QualitySafetyMultiscaleNewData;

[0111] S504, inputting QualitySafetyMultiscaleNewData into the trained QualitySafetyDeepModel to obtain the construction engineering safety risk level value QualitySafetyEvaluation of the quality and safety data to be tested;

[0112] S505, determine the quality and safety risk level according to the value of QualitySafetyEvaluation:

[0113] if QualitySafetyEvaluation<=0.3:

[0114] Quality and safety risk level = "no safety risk";

[0115] elif 0.3 <QualitySafetyEvaluation<=0.6:

[0116] Quality and safety risk level = "Level 1 safety risk exists, and early warning report is required";

[0117] else:

[0118] Quality and safety risk level = "Second-level safety risk exists, and immediate warning report and processing are required";

[0119] S506, output QualitySafetyEvaluation and the corresponding quality and safety risk level as the quality and safety evaluation result of the construction project to be evaluated.

[0120] The beneficial effects of the present invention are:

[0121] The present invention proposes a construction project quality and safety data identification system based on a quality and safety spectrum embedding operator, establishes a heterogeneous tensor spectrum embedding dimensionality reduction operator, can extract and fuse key features from a multi-dimensional, multi-scale complex indicator system, and effectively process high-dimensional nonlinear information in construction project quality and safety data. The present invention uses a deep learning model to learn the features of construction project quality and safety data to obtain a trained quality and safety deep learning model. By using a heterogeneous tensor spectrum embedding dimensionality reduction operator and a trained quality and safety deep learning model, an objective and comprehensive representation of construction project quality and safety data can be obtained, the efficiency and accuracy of feature extraction are improved, thereby improving the accuracy and efficiency of data identification of construction project quality and safety data, and more accurately supervising the construction quality of construction projects. The present invention can effectively process high-dimensional, nonlinear complex indicator systems, while overcoming the influence of incomplete and uncertain data, and improving the accuracy and reliability of construction project quality and safety data identification analysis. The present invention can effectively reduce the dimension and extract features of complex multidimensional data, can better capture key information and potential patterns in the data, and provide more reliable data identification support for subsequent quality and safety analysis and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0122] Figure 1 It is a radar chart of evaluation index scores. DETAILED DESCRIPTION

[0123] Specific implementation method 1: The construction project quality and safety data identification system based on the quality and safety spectrum embedding operator in this implementation method includes: a fusion tensor acquisition operator construction module, a quality and safety spectrum embedding operator construction module, a quality and safety multi-scale feature extraction operator construction module, a quality and safety deep learning model training module and a construction project quality and safety evaluation result acquisition module;

[0124] The fused tensor acquisition operator construction module is used to construct a fused tensor acquisition operator;

[0125] The fusion tensor acquisition operator input is the quality and safety tensor QualitySafetyTensor, and the fusion tensor acquisition operator output is the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor, specifically:

[0126] S101, construct a fusion tensor acquisition operator, the operator input is the quality and safety tensor QualitySafetyTensor;

[0127] The quality and safety tensor is a multi-source heterogeneous data tensor of construction project quality and safety;

[0128] The QualitySafetyTensor is a multidimensional array that stores construction quality evaluation and scoring data from different sources, including design drawings, on-site inspections, sensor monitoring, manual scoring, etc. The data type is a floating point number, including but not limited to the values ​​of indicators such as structural strength, material properties, and construction quality.

[0129] S102, obtaining the order of the quality and safety tensor QualitySafetyTensorOrder=the order of QualitySafetyTensor;

[0130] S103, establish a quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor = establish a tensor with the same dimension as QualitySafetyTensor, and initialize all elements to 0;

[0131] S104, initializing the quality and safety feature fusion counter QualitySafetyFusionCounter=1;

[0132] S105, establish a quality safety single feature temporary storage tensor QualitySafetySingleFeatureTensor = extract the QualitySafetyFusionCounterth feature of QualitySafetyTensor;

[0133] S106, obtaining the maximum value of the quality and safety single feature QualitySafetyFeatureMax=QualitySafetySingleFeatureTensor, and obtaining the minimum value of the quality and safety single feature QualitySafetyFeatureMin=QualitySafetySingleFeatureTensor;

[0134] S107, update QualitySafetySingleFeatureTensor:

[0135]

[0136] S108, update QualitySafetyFeatureFusionTensor:

[0137] QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor+QualitySafetySingle FeatureTensor;

[0138] S109, let QualitySafetyFusionCounter=QualitySafetyFusionCounter+1;

[0139] S110, if QualitySafetyFusionCounter>QualitySafetyTensorOrder, go to S111, otherwise go to S105;

[0140] S111, update QualitySafetyFeatureFusionTensor:

[0141] QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor / QualitySafetyTensor Order;

[0142] S112, obtaining the standard deviation of the current quality and safety feature standard deviation QualitySafetyFeatureStandardDev = QualitySafetyFeatureFusionTensor;

[0143] S113, obtain the quality and safety feature discrimination QualitySafetyFeatureDiscrimination:

[0144]

[0145] Among them, round() is the rounding function;

[0146] S114, update QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor×QualitySafetyFeatureDiscrimination;

[0147] S115, taking the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor as the output of the fusion tensor acquisition operator.

[0148] The quality and safety spectrum embedding operator construction module is used to establish the quality and safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding;

[0149] The input of QualitySafetyTensorSpectralEmbedding is the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor, and the output of QualitySafetyTensorSpectralEmbedding is the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix, specifically:

[0150] S201, establish a quality safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, the input of QualitySafetyTensorSpectralEmbedding is the quality safety feature fusion tensor QualitySafetyFeatureFusionTensor;

[0151] S202, calculate the high-order singular value decomposition of QualitySafetyFeatureFusionTensor:

[0152] HOSVD(QualitySafetyFeatureFusionTensor)=[QualitySafetyU1,QualitySafetyU2,...,QualitySafetyUn,QualitySafetyS]

[0153] Among them, HOSVD is a high-order singular value decomposition, which is used to decompose the high-dimensional tensor into the quality and safety Tucker core tensor and a series of factor matrices. QualitySafetyU1, QualitySafetyU2,..., QualitySafetyUn are orthogonal factor matrices, QualitySafetyS is the quality and safety Tucker core tensor, and n is the total number of orthogonal factor matrices;

[0154] S203, calculate the spectral norm of the quality and safety Tucker core tensor QualitySafetyS:

[0155] QualitySafetyλ=||QualitySafetyS|| σ =max(QualitySafetyσ1,QualitySafetyσ2,...,QualitySafetyσr')

[0156] where QualitySafetyλ is the spectral norm of QualitySafetyS, ||QualitySafetyS|| σ is the spectral norm of QualitySafetyS, QualitySafetyσk is the singular value of QualitySafetyS, r' is the rank, and k ranges from 1 to r'.

[0157] The singular value is the degree to which a matrix or tensor is stretched or compressed in different directions. The spectral norm represents the largest singular value and is used to measure the "size" of a tensor.

[0158] S204, construct the quality and safety low rank tensor QualitySafetyLowRankApproximation:

[0159] QualitySafetyLowRankApproximation=QualitySafetyS×1QualitySafetyU1(:,1:r1)×2QualitySafetyU2(:,1:r2)...×n QualitySafetyUn(:,1:rn)

[0160]

[0161] Among them, ×i represents the tensor-matrix multiplication along the i-th dimension, QualitySafetyUi(:,1:ri) means taking the first ri columns of the QualitySafetyUi matrix, i ranges from 1 to n, ε is the preset threshold, and QualitySafetyσk_i is the k-th singular value in the i-th dimension;

[0162] S205, calculate the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix:

[0163] QualitySafetyEmbeddedFeatureMatrix=Matrix expansion (QualitySafetyLowRankApproximation)

[0164] S206, apply nonlinear activation function to update the quality and safety embedding feature matrix:

[0165] QualitySafetyEmbeddedFeatureMatrix=tanh(QualitySafetyW*QualitySafetyEmbeddedFeatureMatrix+QualitySafetyb)

[0166] Among them, QualitySafetyW and QualitySafetyb are learnable parameters, and tanh is the hyperbolic tangent function;

[0167] S207,

[0168] Use QualitySafetyEmbeddedFeatureMatrix as the output of QualitySafetyTensorSpectralEmbedding.

[0169] The quality safety multi-scale feature extraction operator construction module is used to establish a quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor;

[0170] The QualitySafetyMultiscaleFeatureExtractor input is the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix, and the QualitySafetyMultiscaleFeatureExtractor output is the quality and safety multi-scale feature vector QualitySafetyMultiscaleFeatureVector, specifically:

[0171] S301, establish a quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor, and input the quality safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix;

[0172] S302, initializing the quality safety multi-scale feature vector QualitySafetyMultiscaleFeatureVector; expanding the QualitySafetyEmbeddedFeatureMatrix matrix into a one-dimensional vector;

[0173] S303, looping through different scales, setting the quality and safety scale index QualitySafetyScaleIndex = 1;

[0174] S304, extract features from the current scale to obtain the quality and safety scale feature QualitySafetyScaleFeature:

[0175] QualitySafetyScaleFeature=[QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD]=Discrete wavelet transform (QualitySafetyEmbeddedFeatureMatrix,'db4',QualitySafetyScaleIndex)

[0176] Where db4 represents Daubechies wavelet, QualitySafetycA is the approximation coefficient, QualitySafetycH, QualitySafetycV, and QualitySafetycD are the horizontal detail coefficient, vertical detail coefficient, and diagonal detail coefficient, respectively;

[0177] In this step, the QualitySafetyEmbeddedFeatureMatrix is ​​subjected to discrete wavelet transform to obtain QualitySafetycA, QualitySafetycH, QualitySafetycV and QualitySafetycD;

[0178] S305, calculate the statistical feature QualitySafetyScaleFeature of the current scale:

[0179] QualitySafetyμ=mean([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD])

[0180] QualitySafetyσ'=std([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD])

[0181] QualitySafetyE=entropy([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD])

[0182] QualitySafetyScaleFeature=[QualitySafetyμ,QualitySafetyσ',QualitySafetyE]

[0183] Among them, QualitySafetyμ is the mean of the quality and safety data at the current scale, QualitySafetyσ' is the standard deviation of the quality and safety data at the current scale, QualitySafetyE is the entropy of the quality and safety data at the current scale, mean() is the mean function, std() is the standard deviation function, and entropy() is the entropy function;

[0184] S306, apply the attention mechanism to update QualitySafetyScaleFeature:

[0185] QualitySafetyScaleFeature=QualitySafetyAttentionWeight*QualitySafetyScaleFeature

[0186] QualitySafetyAttentionWeight=softmax(QualitySafetyV*tanh(QualitySafetyW*QualitySafetyScaleFeature+QualitySafetyb))

[0187] Among them, QualitySafetyAttentionWeight is the quality and safety scale feature attention weight, tanh is the hyperbolic tangent function, QualitySafetyV, QualitySafetyW, and QualitySafetyb are learnable parameters;

[0188] S307, add QualitySafetyScaleFeature to QualitySafetyMultiscaleFeatureVector;

[0189] S308, let QualitySafetyScaleIndex=QualitySafetyScaleIndex+1;

[0190] S309, if QualitySafetyScaleIndex is greater than the preset maximum quality and safety scale QualitySafetyMaxScale, go to S310, otherwise go to S304;

[0191] S310, apply one-dimensional convolutional neural network for feature fusion and update QualitySafetyMultiscaleFeatureVector:

[0192] QualitySafetyMultiscaleFeatureVector=CNN1D(QualitySafetyMultiscaleFeatureVector)

[0193] Among them, CNN1D is a one-dimensional convolutional neural network;

[0194] S311, output QualitySafetyMultiscaleFeatureVector as the result of QualitySafetyMultiscaleFeatureExtractor.

[0195] The quality safety deep learning model training module is used to input the quality safety training data set QualitySafetyTrainSet, call the fusion tensor acquisition operator, the quality safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, and the quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to process the QualitySafetyTrainSet, and use the processed QualitySafetyTrainSet to train the quality safety deep learning model QualitySafetyDeepModel to obtain the trained QualitySafetyDeepModel, specifically:

[0196] S401, inputting a quality safety training data set QualitySafetyTrainSet, wherein QualitySafetyTrainSet includes quality safety index values ​​and corresponding quality safety evaluation results;

[0197] The quality safety index values ​​in the quality safety training data set QualitySafetyTrainSet include: structural strength value, material performance value and construction quality value;

[0198] The quality and safety evaluation result is the safety risk level of the construction project;

[0199] S402, calling the fusion tensor acquisition operator, the quality and safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, and the quality and safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to process each sample in QualitySafetyTrainSet to obtain the quality and safety training feature matrix QualitySafetyTrainFeatureMatrix;

[0200] S403, constructing a quality safety deep neural network model QualitySafetyDeepModel;

[0201] The quality safety deep neural network model QualitySafetyDeepModel includes: an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer;

[0202] The number of nodes in the input layer is equal to the dimension of QualitySafetyMultiscaleFeatureVector;

[0203] The first hidden layer contains 256 nodes and uses the LeakyReLU activation function;

[0204] The second hidden layer contains 128 nodes and uses the LeakyReLU activation function;

[0205] The third hidden layer contains 64 nodes and uses the LeakyReLU activation function;

[0206] The output layer includes 1 node, uses Sigmoid activation function to output the safety risk level value of the construction project;

[0207] The output value range of a single node in the output layer is [0,1], which indicates the degree of quality and safety risk of the construction project. The larger the value of the safety risk degree of the construction project, the higher the safety risk of the construction project:

[0208] When the safety risk level of a construction project is within the range of [0.0-0.3], it indicates that there is no safety risk. When the safety risk level of a construction project is within the range of [0.3-0.6], it indicates that there is a safety risk (level 1 safety risk) and an early warning report is required. When the safety risk level of a construction project is within the range of [0.6-1.0], it indicates that there is a greater safety risk (level 2 safety risk) and an early warning report and processing are required immediately.

[0209] S404, define the loss function QualitySafetyLoss:

[0210] QualitySafetyLoss=α*MSE+(1-α)*MAE+β*L2 regularization

[0211] Among them, α and β are hyperparameters, MSE is the mean square error between the predicted value and the actual value, and MAE is the mean absolute error between the predicted value and the actual value;

[0212] S405, using the Adam optimizer, setting the learning rate to 0.001, and setting the learning rate decay strategy:

[0213] QualitySafetylr=QualitySafetylr0 / (1+QualitySafetydecay_rate*epoch)

[0214] Among them, QualitySafetylr is the current learning rate, QualitySafetylr0 is the initial learning rate set to 0.001, QualitySafetydecay_rate is the learning rate decay rate, and epoch is the number of training iterations;

[0215] S406, apply batch normalization and Dropout (rate = 0.5) to prevent overfitting;

[0216] S407, using a k-fold cross validation method (k=5) to train the QualitySafetyDeepModel, and applying an early stopping strategy to obtain a trained QualitySafetyDeepModel;

[0217] The loss function for training QualitySafetyDeepModel is QualitySafetyLoss, the optimizer is Adam optimizer, the optimization strategy adopts learning rate decay strategy, and batch normalization and Dropout (rate = 0.5) are applied to prevent overfitting.

[0218] The construction engineering quality and safety evaluation result acquisition module uses the fusion tensor acquisition operator, QualitySafetyTensorSpectralEmbedding, QualitySafetyMultiscaleFeatureExtractor and the trained QualitySafetyDeepModel to perform quality and safety evaluation on the construction engineering data to be evaluated, and obtains the quality and safety evaluation result of the construction engineering to be evaluated, specifically:

[0219] S501, input new quality and safety data QualitySafetyNewData of the construction project quality and safety data to be tested; call the fusion tensor acquisition operator to process QualitySafetyNewData to obtain a new tensor of quality and safety feature fusion;

[0220] S502, calling QualitySafetyTensorSpectralEmbedding to process the new tensor of quality and safety feature fusion to obtain new quality and safety embedded data QualitySafetyEmbeddedNewData;

[0221] S503, calling QualitySafetyMultiscaleFeatureExtractor to process QualitySafetyEmbeddedNewData to obtain new quality safety multi-scale data QualitySafetyMultiscaleNewData;

[0222] S504, inputting QualitySafetyMultiscaleNewData into the trained QualitySafetyDeepModel to obtain the construction engineering safety risk level value QualitySafetyEvaluation of the quality and safety data to be tested;

[0223] S505, determine the quality and safety risk level according to the value of QualitySafetyEvaluation:

[0224] if QualitySafetyEvaluation<=0.3:

[0225] Quality and safety risk level = "no safety risk";

[0226] elif 0.3 <QualitySafetyEvaluation<=0.6:

[0227] Quality and safety risk level = "Level 1 safety risk exists, and early warning report is required";

[0228] else:

[0229] Quality and safety risk level = "There is a large safety risk (secondary safety risk), and immediate warning report and processing are required";

[0230] S506, output QualitySafetyEvaluation and the corresponding quality and safety risk level as the quality and safety evaluation result of the construction project to be evaluated.

[0231] Example:

[0232] In order to verify the beneficial effects of the present invention, this embodiment carried out experiments according to the specific implementation methods, specifically:

[0233] First, obtain the quality safety tensor QualitySafetyTensor. Table 1 shows some data examples of QualitySafetyTensor:

[0234] Table 1. Some data examples of QualitySafetyTensor

[0235] Structural safety Material quality Construction technology Environmental factors 0.85 0.92 0.78 0.88 0.79 0.88 0.82 0.91 0.91 0.86 0.75 0.84 0.83 0.90 0.80 0.87 0.88 0.89 0.77 0.86

[0236] Then, the quality and safety multi-scale feature vector is obtained, as shown in Table 2:

[0237] Table 2 Example of multi-scale feature extraction results

[0238]

[0239]

[0240] Then, the QualitySafetyTrainSet is processed by using the quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to obtain the quality safety deep learning model QualitySafetyDeepModel. The performance of QualitySafetyDeepModel in this embodiment is evaluated on the test set, and the results are shown in Table 3;

[0241] Table 3 Performance evaluation of the model on the test set

[0242] index value Accuracy 0.92 Accuracy 0.9 Recall 0.89 F1 score 0.895

[0243] Finally, the trained QualitySafetyDeepModel is used to evaluate the construction engineering data to be tested, and the results are shown in Table 4;

[0244] Table 4 Evaluation results of new construction project data

[0245] Evaluation indicators Score Structural safety 0.88 Material quality 0.92 Construction technology 0.82 Environmental factors 0.86 Overall Rating 0.85

[0246] According to Table 4 and Figure 1 According to the evaluation results, the comprehensive quality and safety evaluation of the current construction project is 0.85, which belongs to the high-risk range of quality and safety. It can be seen intuitively from the radar chart that the current project performs well in terms of material quality and structural safety, but there is still room for improvement in construction technology and environmental factors. It is recommended to take the following measures:

[0247] 1. Conduct a comprehensive safety inspection immediately, focusing on potential risks in construction technology and environmental factors.

[0248] 2. Develop and implement a detailed risk management plan, giving priority to improving construction techniques and improving construction quality.

[0249] 3. Strengthen the monitoring and management of the construction environment to reduce the safety risks caused by environmental factors.

[0250] 4. Organize safety training for relevant personnel to improve safety awareness and operational standards.

[0251] 5. Conduct quality and safety assessments regularly and continuously monitor changes in various indicators.

[0252] By using this invention, we can comprehensively and objectively evaluate the quality and safety status of construction projects, providing strong support for subsequent operations. This method not only takes into account multiple evaluation dimensions, but also extracts deep features through advanced data recognition technology, making the evaluation results more accurate and reliable. This comprehensive quantitative evaluation method for construction project quality and safety based on heterogeneous tensor spectrum embedded in dimensionality reduction operators provides an innovative solution for quality supervision in the construction industry, which helps to improve the quality and safety level of the entire industry.

Claims

1. Construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator, characterized by The system includes: a fusion tensor acquisition operator construction module, a quality and safety spectrum embedding operator construction module, a quality and safety multi-scale feature extraction operator construction module, a quality and safety deep learning model training module and a construction project quality and safety evaluation result acquisition module; The fused tensor acquisition operator construction module is used to construct a fused tensor acquisition operator; The fusion tensor acquisition operator input is the quality and safety tensor QualitySafetyTensor, and the fusion tensor acquisition operator output is the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor; The quality and safety spectrum embedding operator construction module is used to establish the quality and safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding; The QualitySafetyTensorSpectralEmbedding input is the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor, and the QualitySafetyTensorSpectralEmbedding output is the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix; The quality safety multi-scale feature extraction operator construction module is used to establish a quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor; The QualitySafetyMultiscaleFeatureExtractor input is the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix, and the QualitySafetyMultiscaleFeatureExtractor output is the quality and safety multi-scale feature vector QualitySafetyMultiscaleFeatureVector; The quality safety deep learning model training module is used to input the quality safety training data set QualitySafetyTrainSet, call the fusion tensor acquisition operator, the quality safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, and the quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to process the QualitySafetyTrainSet, and use the processed QualitySafetyTrainSet to train the quality safety deep learning model QualitySafetyDeepModel to obtain the trained QualitySafetyDeepModel; The construction project quality and safety evaluation result acquisition module uses a fusion tensor acquisition operator, QualitySafetyTensorSpectralEmbedding, QualitySafetyMultiscaleFeatureExtractor and a trained QualitySafetyDeepModel to perform quality and safety evaluation on the construction project data to be evaluated, and obtains the quality and safety evaluation result of the construction project to be evaluated.

2. The construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator according to claim 1 is characterized by: The fused tensor acquisition operator construction module is used to construct a fused tensor acquisition operator, specifically: S101, construct a fusion tensor acquisition operator, the input of which is the quality safety tensor QualitySafetyTensor; The quality and safety tensor QualitySafetyTensor stores construction engineering quality evaluation and scoring data from different sources; Among them, the sources of construction project quality evaluation and scoring data include: design drawings, on-site testing, sensor monitoring, and manual scoring; The construction project quality evaluation scoring data includes: structural strength scoring data, material performance scoring data, and construction quality scoring data; S102, obtaining the order of the quality and safety tensor QualitySafetyTensorOrder=the order of QualitySafetyTensor; S103, establish a quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor = establish a tensor with the same dimension as QualitySafetyTensor, and initialize all elements to 0; S104, initializing the quality and safety feature fusion counter QualitySafetyFusionCounter=1; S105, establish a quality safety single feature temporary storage tensor QualitySafetySingleFeatureTensor = extract the QualitySafetyFusionCounterth feature of QualitySafetyTensor; S106, obtaining the maximum value of the quality and safety single feature maximum value QualitySafetyFeatureMax=QualitySafetySingleFeatureTensor; Get the minimum value of the quality and safety single feature QualitySafetyFeatureMin = the minimum value of QualitySafetySingleFeatureTensor; S107, update QualitySafetySingleFeatureTensor using QualitySafetyFeatureMax and QualitySafetyFeatureMin; S108, update QualitySafetyFeatureFusionTensor: QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor+QualitySafetySingle FeatureTensor S109, let QualitySafetyFusionCounter=QualitySafetyFusionCounter+1; S110, if QualitySafetyFusionCounter>QualitySafetyTensorOrder, go to S111, otherwise go to S105; S111, update QualitySafetyFeatureFusionTensor: QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor / QualitySafetyTensor Order; S112, obtaining the standard deviation of the current quality and safety feature standard deviation QualitySafetyFeatureStandardDev = QualitySafetyFeatureFusionTensor; S113, using QualitySafetyFeatureStandardDev to obtain the quality and safety feature discrimination QualitySafetyFeatureDiscrimination; S114, update QualitySafetyFeatureFusionTensor=QualitySafetyFeatureFusionTensor×QualitySafetyFeatureDiscrimination; S115, taking the quality and safety feature fusion tensor QualitySafetyFeatureFusionTensor as the output of the fusion tensor acquisition operator.

3. The construction engineering quality and safety data identification system based on the quality and safety spectrum embedding operator according to claim 2 is characterized by: The use of QualitySafetyFeatureMax and QualitySafetyFeatureMin to update QualitySafetySingleFeatureTensor in S107 is specifically as follows:

4. The construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator according to claim 3 is characterized by: The method of using QualitySafetyFeatureStandardDev in S113 to obtain the quality and safety feature discrimination QualitySafetyFeatureDiscrimination is specifically as follows: Among them, round() is the rounding function.

5. The construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator according to claim 4 is characterized by: The quality safety spectrum embedding operator construction module is used to establish the quality safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, specifically: S201, establish a quality safety spectral embedding operator QualitySafetyTensorSpectralEmbedding, the input of QualitySafetyTensorSpectralEmbedding is the quality safety feature fusion tensor QualitySafetyFeatureFusionTensor; S202, calculate the high-order singular value decomposition of QualitySafetyFeatureFusionTensor: HOSVD(QualitySafetyFeatureFusionTensor)=[QualitySafetyU1,QualitySafetyU2,...,QualitySafetyUn,QualitySafetyS] Among them, HOSVD is the high-order singular value decomposition, QualitySafetyU1, QualitySafetyU2,..., QualitySafetyUn are orthogonal factor matrices, QualitySafetyS is the quality and safety Tucker core tensor, and n is the total number of orthogonal factor matrices; S203, calculate the spectral norm QualitySafetyλ of the quality and safety Tucker core tensor QualitySafetyS: QualitySafetyλ=||QualitySafetyS|| σ =max(QualitySafetyσ1,QualitySafetyσ2,...,QualitySafetyσr') where QualitySafetyλ is the spectral norm of QualitySafetyS, ||QualitySafetyS|| σ is the spectral norm of QualitySafetyS, QualitySafetyσk is the singular value of QualitySafetyS, r' is the rank, and k ranges from 1 to r'; S204, construct the quality and safety low rank tensor QualitySafetyLowRankApproximation: QualitySafetyLowRankApproximation=QualitySafetyS×1QualitySafetyU1(:,1:r1)×2QualitySafetyU2(:,1:r2)...×n QualitySafetyUn(:,1:rn) Where ε is the preset threshold, ×i represents the tensor-matrix multiplication along the i-th dimension, i ranges from 1 to n, QualitySafetyUi(:,1:ri) takes the first ri columns of the QualitySafetyUi matrix, and QualitySafetyσk_i is the k-th singular value in the i-th dimension; S205, calculate the quality and safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix: QualitySafetyEmbeddedFeatureMatrix=Matrix expansion (QualitySafetyLowRankApproximation) S206, apply nonlinear activation function to update the quality and safety embedding feature matrix: QualitySafetyEmbeddedFeatureMatrix=tanh(QualitySafetyW*QualitySafetyEmbeddedFeatureMatrix+QualitySafetyb) Among them, QualitySafetyW and QualitySafetyb are learnable parameters, and tanh is the hyperbolic tangent function; S207， Use QualitySafetyEmbeddedFeatureMatrix as the output of QualitySafetyTensorSpectralEmbedding.

6. The construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator according to claim 5 is characterized by: The quality safety multi-scale feature extraction operator construction module is used to establish the quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor, specifically: S301, establish a quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor, and input the quality safety embedded feature matrix QualitySafetyEmbeddedFeatureMatrix; S302, initializing the quality safety multi-scale feature vector QualitySafetyMultiscaleFeatureVector; expanding the QualitySafetyEmbeddedFeatureMatrix matrix into a one-dimensional vector; S303, looping through different scales, setting the quality and safety scale index QualitySafetyScaleIndex = 1; S304, extract features from the current scale to obtain the quality and safety scale feature QualitySafetyScaleFeature: QualitySafetyScaleFeature=[QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD]=Discrete wavelet transform (QualitySafetyEmbeddedFeatureMatrix,'db4',QualitySafetyScaleIndex) Where db4 represents Daubechies wavelet, QualitySafetycA is the approximation coefficient, QualitySafetycH, QualitySafetycV, and QualitySafetycD are the horizontal detail coefficient, vertical detail coefficient, and diagonal detail coefficient, respectively; S305, calculate the statistical feature QualitySafetyScaleFeature of the current scale: QualitySafetyScaleFeature=[QualitySafetyμ,QualitySafetyσ',QualitySafetyE] QualitySafetyμ=mean([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD]) QualitySafetyσ'=std([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD]) QualitySafetyE=entropy([QualitySafetycA,QualitySafetycH,QualitySafetycV,QualitySafetycD]) Among them, QualitySafetyμ is the mean of the quality and safety data at the current scale, QualitySafetyσ' is the standard deviation of the quality and safety data at the current scale, QualitySafetyE is the entropy of the quality and safety data at the current scale, mean() is the mean function, std() is the standard deviation function, and entropy() is the entropy function; S306, apply the attention mechanism to update QualitySafetyScaleFeature: QualitySafetyScaleFeature=QualitySafetyAttentionWeight*QualitySafetyScaleFeature Among them, QualitySafetyAttentionWeight is the quality and safety scale feature attention weight; S307, adding QualitySafetyScaleFeature to QualitySafetyMultiscaleFeatureVector; S308, let QualitySafetyScaleIndex=QualitySafetyScaleIndex+1; S309, if QualitySafetyScaleIndex is greater than the preset maximum quality and safety scale QualitySafetyMaxScale, go to S310, otherwise go to S304; S310, update QualitySafetyMultiscaleFeatureVector: QualitySafetyMultiscaleFeatureVector=CNN1D(QualitySafetyMultiscaleFeatureVector) Among them, CNN1D is a one-dimensional convolutional neural network; S311, output QualitySafetyMultiscaleFeatureVector as the result of QualitySafetyMultiscaleFeatureExtractor.

7. The construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator according to claim 6 is characterized by: The quality safety scale feature attention weight QualitySafetyAttentionWeight in S306 is obtained by: QualitySafetyAttentionWeight=softmax(QualitySafetyV*tanh(QualitySafetyW*QualitySafetyScaleFeature+QualitySafetyb)) Among them, tanh is the hyperbolic tangent function, QualitySafetyV, QualitySafetyW, and QualitySafetyb are learnable parameters.

8. The construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator according to claim 7 is characterized by: The quality safety deep learning model training module is used to input the quality safety training data set QualitySafetyTrainSet, call the fusion tensor acquisition operator, the quality safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, and the quality safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to process the QualitySafetyTrainSet, and use the processed QualitySafetyTrainSet to train the quality safety deep learning model QualitySafetyDeepModel to obtain the trained QualitySafetyDeepModel, specifically: S401, inputting a quality safety training data set QualitySafetyTrainSet, wherein QualitySafetyTrainSet includes quality safety indicators and corresponding quality safety evaluation results; The quality safety index values ​​in the quality safety training data set QualitySafetyTrainSet include: structural strength value, material performance value and construction quality value; The quality and safety evaluation result is the safety risk level of the construction project; S402, calling the fusion tensor acquisition operator, the quality and safety spectrum embedding operator QualitySafetyTensorSpectralEmbedding, and the quality and safety multi-scale feature extraction operator QualitySafetyMultiscaleFeatureExtractor to process each sample in QualitySafetyTrainSet to obtain the quality and safety training feature matrix QualitySafetyTrainFeatureMatrix; S403, constructing a quality safety deep neural network model QualitySafetyDeepModel; The quality safety deep neural network model QualitySafetyDeepModel includes: an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer; The number of nodes in the input layer is equal to the dimension of QualitySafetyMultiscaleFeatureVector; The first hidden layer contains 256 nodes and uses the LeakyReLU activation function; The second hidden layer contains 128 nodes and uses the LeakyReLU activation function; The third hidden layer contains 64 nodes and uses the LeakyReLU activation function; The output layer includes 1 node, uses Sigmoid activation function to output the safety risk level of the construction project; Among them, when the safety risk level of construction projects is within the range of [0.0-0.3], it means there is no safety risk; when the safety risk level of construction projects is within the range of [0.3-0.6], it means a first-level safety risk and an early warning report is required; when the safety risk level of construction projects is within the range of [0.6-1.0], it means there is a second-level safety risk and an early warning report and processing are required immediately; S404, define a loss function QualitySafetyLoss; S405, using the Adam optimizer, setting the learning rate to 0.001, and setting the learning rate decay strategy: QualitySafetylr=QualitySafetylr0 / (1+QualitySafetydecay_rate*epoch) Among them, QualitySafetylr is the current learning rate, QualitySafetylr0 is the initial learning rate, QualitySafetydecay_rate is the learning rate decay rate, and epoch is the number of training iterations; S406, applying batch normalization and Dropout; S407, based on the settings of S404, S405, and S406, the QualitySafetyDeepModel model is trained using k-fold cross validation, and an early stopping strategy is applied to obtain a trained QualitySafetyDeepModel.

9. The construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator according to claim 8 is characterized by: The loss function QualitySafetyLoss defined in S404 is specifically: QualitySafetyLoss=α*MSE+(1-α)*MAE+β*L2 regularization Among them, α and β are hyperparameters, MSE is the mean square error between the predicted value and the actual value, and MAE is the mean absolute error between the predicted value and the actual value.

10. The construction engineering quality and safety data identification system based on quality and safety spectrum embedding operator according to claim 9 is characterized by: The construction engineering quality and safety evaluation result acquisition module uses the fusion tensor acquisition operator, QualitySafetyTensorSpectralEmbedding, QualitySafetyMultiscaleFeatureExtractor and the trained QualitySafetyDeepModel to perform quality and safety evaluation on the construction engineering data to be evaluated, and obtains the quality and safety evaluation result of the construction engineering to be evaluated, specifically: S501, inputting new quality and safety data QualitySafetyNewData of the construction project quality and safety data to be tested, calling the fusion tensor acquisition operator to process QualitySafetyNewData, and obtaining a new fusion tensor of the quality and safety features of the construction project to be evaluated; S502, calling QualitySafetyTensorSpectralEmbedding to process the new tensor of quality and safety feature fusion to obtain new quality and safety embedded data QualitySafetyEmbeddedNewData; S503, calling QualitySafetyMultiscaleFeatureExtractor to process the quality safety embedded new data QualitySafetyEmbeddedNewData to obtain the quality safety multi-scale new data QualitySafetyMultiscaleNewData; S504, inputting QualitySafetyMultiscaleNewData into the trained QualitySafetyDeepModel to obtain the construction engineering safety risk level value QualitySafetyEvaluation of the quality and safety data to be tested; S505, determine the quality and safety risk level according to the value of QualitySafetyEvaluation: if QualitySafetyEvaluation<=0.3: Quality and safety risk level = "no safety risk"; elif 0.3 <QualitySafetyEvaluation<=0.6: Quality and safety risk level = "Level 1 safety risk exists, and early warning report is required"; else: Quality and safety risk level = "Second-level safety risk exists, and immediate warning report and processing are required"; S506, output QualitySafetyEvaluation and the corresponding quality and safety risk level as the quality and safety evaluation result of the construction project to be evaluated.

Images