Construction engineering quality data management method based on artificial intelligence

Abstract

The invention discloses a construction engineering quality data management method based on artificial intelligence, and relates to the technical field of data management, and the method comprises the steps: obtaining construction engineering quality data, extracting each modal feature vector in the construction engineering quality data, fusing each modal feature vector, and obtaining a fused feature vector; preliminarily grouping the fusion feature vectors to generate a quality data cluster, and taking the quality data cluster as a map node to generate an initial quality data map; expanding the initial quality data graph, calculating and judging a causal direction, generating a causal edge, and obtaining a dynamic causal quality data graph; carrying out distributed embedding optimization, and optimizing the dynamic causal knowledge graph; and based on the query vector and the optimized dynamic causal quality data graph, using a deep reinforcement learning network to optimize a query path, and by constructing the dynamic causal quality data graph, accurately identifying the causal relationship between the data, and enhancing the intelligent level of constructional engineering quality data management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data management, and particularly relates to a building engineering quality data management method based on artificial intelligence. BACKGROUND

[0002] Current building engineering quality data management mainly adopts two types of technical routes: one type is a structured storage scheme (such as MySQL) based on a relational database, which records data such as material detection and construction logs through a standardized table, but it is difficult to process unstructured data such as images and videos; the other type is to use traditional machine learning methods (such as SVM and random forest) to analyze the quality of single modal data, for example, to process crack images through a convolutional neural network.

[0003] However, these methods have significant limitations: first, multi-source heterogeneous data (such as detection values, construction texts, and field images) lack an effective fusion mechanism, resulting in a single dimension of quality evaluation. For example, a project has both abnormal concrete strength detection values and missing maintenance records in the construction log, and the existing system cannot establish a correlation analysis between the two. Secondly, the dynamic evolution relationship between quality data is not modeled, and traditional methods can only output static quality scores, which cannot predict the time sequence causal chain such as “rebar corrosion → concrete cracking → structure deformation”. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a building engineering quality data management method based on artificial intelligence to solve the problem that the quality evolution causal relationship is difficult to dynamically model.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a building engineering quality data management method based on artificial intelligence, which comprises, obtaining building engineering quality data and extracting modal feature vectors in the building engineering quality data, fusing the modal feature vectors to obtain a fused feature vector; preliminarily grouping the fused feature vector to generate an initial cluster set, clustering the initial cluster set to generate a quality data cluster, taking the quality data cluster as a graph node, and calculating the cosine similarity between two graph nodes as a semantic edge weight to generate an initial quality data graph; extending the initial quality data graph and performing time series preprocessing on the extended initial quality data graph, generating a causal edge by calculating and judging the causal direction, adding the causal edge to the extended initial quality data graph to obtain a dynamic causal quality data graph; The dynamic causal quality data graph is fragmented, distributed embedding optimization of the fragmented sub-graph is performed on a local computing node, and the dynamic causal knowledge graph is optimized; The multi-modal query input by the user is processed into a query vector, and based on the query vector and the optimized dynamic causal quality data graph, a deep reinforcement learning network is used to optimize the query path As a preferred scheme of the building engineering quality data management method based on artificial intelligence, the initial quality data graph is expanded, and the specific steps are as follows, The L2 norm of the difference between the fusion feature vector of the new building engineering quality data and each quality data cluster center is calculated, and the fusion feature vector of the new building engineering quality data is attributed by using the minimum value of all L2 norms; According to the sum of the average cluster distance mean and the standard deviation of all quality data clusters, the quality data cluster expansion threshold is set; When the minimum value of all L2 norms is less than or equal to the quality data cluster expansion threshold, the fusion feature vector of the new building engineering quality data is assigned to the corresponding quality data cluster, and the quality data cluster center is updated; Otherwise, the fusion feature vector of the new building engineering quality data is taken as a new quality data cluster center, a new graph node is generated, and the expansion of the initial quality data graph is completed.

[0007] As a preferred scheme of the building engineering quality data management method based on artificial intelligence, the causal direction is calculated and judged, and the specific steps are as follows, According to the sum of the average causal strength mean and the standard deviation of all time series quality data subsets, the causal strength threshold is set; The causal direction is judged by comparing the causal strength threshold and the causal strength; When the causal strength is greater than the causal strength threshold, it means that the causal direction is effective, otherwise it means that the causal direction is ineffective.

[0008] As a preferred scheme of the building engineering quality data management method based on artificial intelligence, the dynamic causal quality data graph is fragmented, and the specific steps are as follows, All graph semantic edge weight values in the dynamic causal quality data graph are traversed, graph nodes with weight values higher than the median of the graph semantic edge weight values are preferentially merged, and a rough quality data graph is obtained; The initial partition of the rough quality data graph is performed by using the Kuhn-Munkres algorithm, and the sub-graph of the rough quality data graph is obtained; For each boundary node of the sub-graph, the inter-sub-graph edge weight sum after the boundary node moves to other sub-graphs is calculated; Select the edge weight between sub-graphs and the most reduced boundary node moving operation, and update the edge weight and re-allocate the boundary node.

[0009] As a preferred scheme of the building engineering quality data management method based on artificial intelligence, the initial cluster set is generated, and the specific steps are as follows, Calculate the Euclidean distance between the fusion feature vectors; According to the mean and standard deviation of the Euclidean distance between the fusion feature vectors, the similarity threshold is set by adopting the three standard deviation principle; The fusion feature vectors with the Euclidean distance between the fusion feature vectors less than the similarity threshold are assigned to the same cluster, otherwise they are assigned to different clusters, and the initial cluster set is generated.

[0010] As a preferred scheme of the building engineering quality data management method based on artificial intelligence, the initial cluster set is generated, and the specific steps are as follows, Define the current sub-graph as the state space, select the next graph node, semantic edge and causal edge as the action space, and set the reward function based on the cosine similarity of the global embedding vector of the current sub-graph and the query vector; Input the current sub-graph into the reinforcement learning network, calculate the Q value of all possible actions, select the action with the maximum Q value, update the state space, and re-calculate the reward function for iterative optimization; When the maximum number of iterations is reached, stop optimization, and take the current sub-graph as the candidate query sub-graph; Calculate the cosine similarity of the global embedding vector of all graph nodes in the candidate query sub-graph and the query vector, and retain the graph node, semantic edge and causal edge with the highest cosine similarity to obtain the optimal query path.

[0011] As a preferred scheme of the building engineering quality data management method based on artificial intelligence, the initial cluster set is generated, and the specific steps are as follows, Set the minimum number of neighbors and the initial neighborhood radius; According to the initial cluster set, the initial neighborhood radius is optimized by calculating the contour coefficient; Based on the optimized neighborhood radius and the minimum number of neighbors, the initial cluster set is clustered to generate the quality data cluster.

[0012] As a preferred scheme of the building engineering quality data management method based on artificial intelligence, the building engineering quality data includes material detection records, construction logs, field photos, construction videos and voice records; The various modal feature vectors include text feature vectors, image feature vectors, video feature vectors and voice feature vectors.

[0013] In a second aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the artificial intelligence-based construction engineering quality data management method according to the first aspect of the present application.

[0014] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the artificial intelligence-based construction engineering quality data management method according to the first aspect of the present application.

[0015] The present application has the following beneficial effects: the present application can effectively capture the semantic association between different modal data by unified cleaning and feature extraction of multi-modal data such as text, image, video and voice, and generate a fusion feature vector, thereby improving the comprehensiveness and accuracy of data processing; by constructing a dynamic causal quality data graph, using a dynamic time warping algorithm and a mutual information calculation method, the causal relationship between data is accurately identified, the potential association between material defects and construction process errors is revealed, a reliable basis is provided for the tracing and prediction of quality problems, and the intelligent level of construction engineering quality data management is enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Fig. 1 Flowchart of the artificial intelligence-based construction engineering quality data management method.

[0018] Fig. 2 Flowchart of graph expansion and causal edge generation.

[0019] Fig. 3 Flowchart of distributed graph optimization.

[0020] Fig. 4 Flowchart of deep reinforcement learning query optimization. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0022] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the present application.

[0023] It is also noted that, as used herein, "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the application. The appearances of the phrase "in one embodiment" or "in an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.

[0024] Reference will now be made to Figs. 1-4 For one embodiment of the present application, the embodiment provides an artificial intelligence-based construction quality data management method, comprising the following steps: S1, obtaining construction quality data and extracting modal feature vectors in the construction quality data, fusing the modal feature vectors to obtain a fused feature vector.

[0025] Collecting construction quality data, including material test records (such as concrete strength and steel bar specifications), construction logs (such as construction time, location, and responsible person), on-site photos (such as cracks and defects), construction videos, and voice recordings; Cleaning the construction quality data, the specific steps are as follows: For numerical data in the material test records, check for missing values, fill in with the median of the same batch of numerical data, for example, for concrete strength values, if there are missing values, calculate the median of the known strength values of the same batch, and replace the missing values with the median, remove outliers from the filled numerical data, calculate the average and standard deviation of the numerical data, and if the numerical data is outside the range of the average plus or minus 3 times the standard deviation, delete it; For construction logs, encode and unify, convert all text data to UTF-8 encoding format, and remove redundant symbols in the text data, such as extra spaces, repeated punctuation, or invalid characters; For on-site photos and construction videos, use Gaussian filtering to smooth the image or video frame through convolution operation to remove background noise; For voice recordings, use spectral subtraction to calculate the noise spectrum in the voice recording and delete it from the voice recording; Extract semantic features from the cleaned construction quality data to obtain a multi-modal feature vector, the specific steps are as follows: The material detection record and the construction log are segmented into words by using a Chinese word segmentation tool, such as Jieba word segmentation, and stop words such as "de" and "shi" are removed, the TF-IDF value of each word is calculated by the TF-IDF method, and a text feature vector is generated by mapping the TF-IDF value of each word to a fixed-dimensional vector space, specifically, the total number of occurrences of each word is calculated, and the words are sorted in descending order, and the top several words are selected as the building quality dictionary, for example, the top 768, an empty text feature vector is initialized, each dimension corresponds to a word in the building quality dictionary, and the TF-IDF value of each word is filled into the empty text feature vector, if the word is not in the building quality dictionary, it is ignored, and the corresponding dimension is set to 0, to obtain the text feature vector; The gradient direction histogram is extracted from the field photos to generate an image feature vector, specifically, the field photos are converted into field grayscale images, the horizontal direction gradient and the vertical direction gradient of the field grayscale images are calculated using the Sobel operator, the gradient intensity and direction of each pixel point in the field photos are calculated, the field gradient intensity map and the field gradient direction map are generated, the field gradient direction map is divided into a plurality of cells, for example, 16x16 cells, the direction range (0~360°) is evenly divided into a plurality of intervals, for example, 9 direction intervals, each interval is 40°, the gradient intensity weighted sum of each direction interval is calculated, the gradient direction in each cell is quantized to obtain a 9-dimensional gradient direction histogram of the cell, 2x2 adjacent cells are combined into a block, the 9-dimensional gradient direction histograms of the four cells in each block are normalized by L2 norm, and the normalization process is repeated by sliding window until the entire field photo is covered, dimensionality reduction is performed by principal component analysis to obtain the image feature vector; The key frames are extracted from the construction video by using the uniform sampling method, specifically, the construction video is divided into equal-interval segments by time length, and one frame is taken from each segment as a key frame, for example, for a 60-second construction video, 30 frames are sampled at every 2 seconds to obtain a key frame set, the image feature vector is obtained by extracting the gradient direction histogram of each key frame in the key frame set, the key frame feature vector is obtained, and the video feature vector is obtained by averaging all key frame feature vectors; The mel-frequency cepstral coefficient method is used for speech recording to extract spectral coefficients and generate a speech feature vector. Specifically, the speech recording is pre-emphasized by a first-order high-pass filter to enhance the resolution of high-frequency signals, obtaining pre-emphasized speech. The pre-emphasized speech is windowed by a Hamming window function, and the pre-emphasized speech is segmented into short-time speech frames, obtaining a short-time speech frame sequence. The short-time speech frame sequence is subjected to fast Fourier transform (FFT) to convert the time-domain signal into a frequency-domain signal. The amplitude square of the frequency-domain signal is calculated to obtain a speech power spectrum. The speech power spectrum is passed through a mel filter bank to output log energy. The number of log energies is equal to the number of mel filters. The log energy is subjected to discrete cosine transform to obtain mel-frequency cepstral coefficients. The mel-frequency cepstral coefficients of each frame in the short-time speech frame sequence are arranged in time, and principal component analysis is performed to reduce the dimensionality, obtaining a speech feature vector. By constructing a quality data feature graph, the text feature vector, the image feature vector, the video feature vector and the speech feature vector are fused through a graph attention mechanism to obtain a fused semantic vector. The specific steps are as follows: Each piece of building quality data is defined as a node, and the node attributes are the text feature vector, the image feature vector, the video feature vector and the speech feature vector. For any two nodes, the Euclidean distance between the node attributes is calculated. The Euclidean distance set of all node pairs is calculated. The mean and standard deviation of the Euclidean distance are calculated. The Euclidean distance threshold is set as the difference between the mean and the single standard deviation. When the Euclidean distance between the attributes of two nodes is less than the Euclidean distance threshold, an edge is established between the two nodes. The initial edge weight can be set to 1. The quality data feature graph is obtained. The text feature vector, the image feature vector, the video feature vector and the speech feature vector of the building quality data are spliced to obtain an initial fusion feature vector. Based on the quality data feature graph, the attention coefficient of two adjacent nodes is calculated. Based on the graph attention mechanism, the feature vectors of adjacent nodes are aggregated by weighting with the attention coefficient to update the initial fusion feature vector and obtain the fusion feature vector.

[0026] S2, the fusion feature vector is preliminarily grouped to generate an initial cluster set. The initial cluster set is clustered to generate a quality data cluster. The quality data cluster is taken as a graph node, and the cosine similarity between two graph nodes is calculated as a semantic edge weight to generate an initial quality data graph.

[0027] The fusion feature vector is preliminarily grouped to generate an initial cluster set. Specifically, the Euclidean distance between the fusion feature vectors is calculated. According to the mean and standard deviation of the Euclidean distance between the fusion feature vectors, a similarity threshold is set using the three-sigma principle, for example, 0.3. The fusion feature vectors with an Euclidean distance between them less than the similarity threshold are assigned to the same cluster, and vice versa. The initial cluster set is generated. The DBSCAN algorithm (density clustering algorithm) is used to cluster the initial cluster set to generate quality data clusters. Specifically, the DBSCAN algorithm parameters are initialized, including setting the minimum neighbor number and the initial neighborhood radius, wherein the minimum neighbor number represents the minimum number of fused feature vectors in a cluster, and the initial neighborhood radius is equal to the similarity threshold. According to the initial cluster set, the initial neighborhood radius is optimized by calculating the contour coefficient. Specifically, within the value range of the neighborhood radius, for example fixed step length, such as 0.01, iterative testing is performed, the average contour coefficient of all fused feature vectors is calculated, and the neighborhood radius corresponding to the maximum average contour coefficient is selected, wherein the contour coefficient calculation formula is as follows: ; In the formula, the contour coefficient of the i-th fused feature vector is denoted as the average Euclidean distance between the i-th fused feature vector and other fused feature vectors in the cluster is denoted as the average Euclidean distance between the i-th fused feature vector and all fused feature vectors in the nearest neighbor cluster is denoted as Based on the optimized neighborhood radius and the minimum neighbor number, the initial cluster set is clustered to generate quality data clusters, each quality data cluster represents an entity, and serves as a graph node, such as "concrete member" and "construction unit", etc. The weight of the graph semantic edge is obtained by calculating the cosine similarity between two graph nodes. Each graph semantic edge records the start graph node ID, the end graph node ID and the graph semantic edge weight value, and the construction of the initial quality data graph is completed. The initial quality data graph is converted into the storage format of the distributed graph database Neo4j. Specifically, the graph node ID and the graph edge ID are written into the node table and the relationship table of the distributed graph database Neo4j as unique identifiers through the batch import tool of Neo4j. Based on the graph node ID and the graph edge ID, a primary key index is created to support fast retrieval. Specifically, the CREATE INDEX statement is executed in the distributed graph database Neo4j to create a B+ tree index for the graph node ID and the graph edge ID. At the same time, a secondary index is created for the timestamp attribute of the graph node to support time-based queries. The index is stored in the index partition of the distributed graph database Neo4j to ensure query efficiency. S3, the initial quality data graph is extended, and the extended initial quality data graph is time series preprocessed. By calculating and judging the causal direction, a causal edge is generated. The causal edge is added to the extended initial quality data graph to obtain a dynamic causal quality data graph.

[0028] ​​​Based on the construction progress of the construction project, new construction project quality data is obtained, and the initial quality data graph is expanded based on the fusion feature vector of the new construction project quality data. Specifically, the L2 norm of the difference between the fusion feature vector of the new construction project quality data and each quality data cluster center is calculated, and the minimum value of all L2 norms is used to determine the attribution of the fusion feature vector of the new construction project quality data. Further, the average cluster distance mean and standard deviation sum of all quality data clusters are used to set the quality data cluster expansion threshold. When the minimum value of all L2 norms is less than or equal to the quality data cluster expansion threshold, the fusion feature vector of the new construction project quality data is assigned to the corresponding quality data cluster, and the quality data cluster center is updated. Otherwise, the fusion feature vector of the new construction project quality data is taken as a new quality data cluster center, and a new graph node is generated. The initial quality data graph is expanded by calculating the initial weight of the graph edge between the new graph node and the existing graph node, and stored in the distributed graph database Neo4j. The expanded initial quality data graph includes the timestamp and value sequence of each graph node, wherein the timestamp records the construction project quality data collection time, and the value sequence represents the quantitative indicators related to the graph node, such as concrete strength value, environmental humidity value, or stress data, etc. The value sequence is sorted in ascending order according to the timestamp to obtain time series quality data. The time series quality data is divided into multiple subsets according to the graph nodes, and the dynamic time warping (DTW) algorithm is used to calculate the optimal alignment path between two time series quality data subsets, such as A and B. A is time-shifted according to a preset lag time step range, such as , to generate a lag sequence A, and the optimal alignment path between the lag sequence A and B is calculated. The lag time step number with the smallest optimal alignment path is selected. If the lag time step number with the smallest optimal alignment path is greater than 0, it indicates that A is the cause and B is the effect, and the causal direction is represented as . If the lag time step number with the smallest optimal alignment path is less than 0, it is the opposite, and the causal direction is represented as . Further, for each causal direction, the mutual information of the time series quality data subsets A and B is calculated to represent the causal strength, and the causal strength threshold is set according to the sum of the average and standard deviation of all time series quality data subset pairs. By comparing the causal strength threshold and the causal strength, it is determined whether the causal direction is valid. When the causal strength is greater than the causal strength threshold, the causal edge of A and B is generated, and the causal edge attribute is represented as mutual information and lag step number. Otherwise, the causal direction is invalid, and no processing is performed. The generated causal edge is added to the expanded initial quality data graph, and the distributed graph database Neo4j is updated to obtain a dynamic causal quality data graph.

[0029] S4. The dynamic causal quality data graph is segmented, and the segmented subgraphs are distributed and embedded on local computing nodes to optimize the dynamic causal knowledge graph.

[0030] A multi-level partitioning method based on METIS (Graph Partitioning Toolkit) is used to partition the dynamic causal quality data graph. This partitioning strategy divides the dynamic causal quality data graph into several subgraphs, including coarsening, initial partitioning, and refinement. Specifically, the weight values ​​of all semantic edges in the dynamic causal quality data graph are traversed. Graph node pairs with weight values ​​higher than the median of the semantic edge weight values ​​are merged first. During merging, the neighboring graph node with the largest semantic edge weight of the current graph node is selected. This merging process is repeated multiple times, for example, until the number of graph nodes is reduced to approximately 1 / 10 of the original size, completing the coarsening of the dynamic causal quality data graph and obtaining a coarsened quality data graph. The Kronenstein-based ... The Lin algorithm performs initial partitioning of the coarsened quality data graph. Specifically, it randomly divides the coarsened quality data graph into several subgraphs, such as four, and ensures that the number of graph nodes in each subgraph is approximately equal. By summing all cross-subgraph edge pairs, the sum of edge weights between subgraphs after the initial partitioning is calculated. By swapping graph node pairs in subgraphs, the initial partitioning is completed until the sum of edge weights between subgraphs no longer decreases after multiple consecutive swaps. For the boundary nodes of each subgraph, i.e., the graph nodes connected to other subgraphs, the sum of edge weights between subgraphs is calculated after the boundary node is moved to another subgraph. The boundary node with the largest reduction in the sum of edge weights between subgraphs is selected for the move operation, and the edge weights are updated and the boundary nodes are redistributed to complete the refinement of the subgraphs. Subgraphs are assigned to local computing nodes for embedding optimization, and the Adam optimizer is used to process the embedding vectors of the graph nodes of each subgraph. Specifically, the loss function for the subgraph is set based on the L2 norm squared difference between the embedding vectors, and the calculation formula is as follows: ; In the formula, This represents the loss value of the subgraph. Indicates the first subgraph Embedding vectors of graph nodes Indicates the first The first graph node The embedding vectors of the neighboring nodes should be noted. It should be pointed out that the starting point for optimizing the embedding vectors is obtained by reducing the dimensionality of the fused feature vectors. For the first The graph node and the first The semantic edge weights of each neighboring node. This is the index of the graph node. For graph nodes Indexes of neighboring nodes; The embedding vectors of each graph node in the subgraph are iteratively updated by minimizing the loss function of the subgraph, the embedding optimization of the subgraph is realized, and the locally optimized embedding vectors are obtained. The global embedding vectors are generated by averaging the embedding vectors of each graph node in all subgraphs on the master computing node using the distributed average aggregation algorithm, and the calculation formula is as follows: ; In the formula, represents the global embedding vector of the i-th graph node, is the number of subgraphs, is the index variable of the subgraph, represents the embedding vector of the i-th graph node in the j-th subgraph, The global embedding vectors are distributed back to each local computing node, and the optimization of the dynamic causal quality data graph is completed, and the distributed graph database Neo4j is updated.

[0031] S5, the user input multi-modal query is processed into a query vector, based on the query vector and the optimized dynamic causal quality data graph, a deep reinforcement learning network is used to optimize the query path.

[0032] Receive the multi-modal query data input by the user, including text, voice, image or video, generate a query vector by generating a fusion feature vector method, it should be pointed out that if the user input query data only contains part of the mode, for example, only text query, the corresponding dimension of voice, image or video is zero vector; Based on the query vector and the optimized dynamic causal quality data graph, a deep reinforcement learning network is used to optimize the query path, specifically, the current subgraph is defined as the state space, the next graph node, semantic edge and causal edge are selected as the action space, based on the cosine similarity of the global embedding vector of the current subgraph and the query vector, the reward function is set, and the calculation formula is as follows: ; In the formula, is the reward value, is the query vector, is the global embedding vector of the current subgraph, represents the number of query steps; Initialize the network parameters of the reinforcement learning network, for example, the number of fully connected layers, the maximum number of iterations, the learning rate, the reinforcement learning network loss function, etc. ​​​The current subgraph is input into the reinforcement learning network, the Q values of all possible actions are calculated, the action with the maximum Q value is selected, the state space is updated, and the reward function is recalculated. In the iteration process, the reinforcement learning network loss function is minimized by the Adam optimizer. When the maximum number of iterations is reached, the optimization is stopped, and the current subgraph is taken as the candidate query subgraph; The cosine similarity of the global embedding vector of all graph nodes in the candidate query subgraph and the query vector is calculated, the graph nodes, semantic edges and causal edges with the highest cosine similarity are retained, and the optimal query path is obtained.

[0033] It should be noted that the iteration of the deep reinforcement learning network may increase the computational overhead, which seems to conflict with the fast retrieval of the distributed graph database Neo4j. However, the goal of the deep reinforcement learning network is to refine the query range and reduce the retrieval burden of the distributed graph database Neo4j. The combination of the two improves the overall efficiency of building engineering quality data management and query.

[0034] The embodiment also provides a computer device suitable for the building engineering quality data management method based on artificial intelligence, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the building engineering quality data management method based on artificial intelligence proposed in the above embodiment.

[0035] The computer device can be a terminal, which comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, touchpad or mouse, etc.

[0036] The embodiment also provides a storage medium on which a computer program is stored, the program being executed by a processor to implement the method for managing construction engineering quality data based on artificial intelligence proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0037] To sum up, the application can effectively capture the semantic association between different modal data by unified cleaning and feature extraction of multi-modal data such as text, image, video and voice, and generate a fusion feature vector, thereby improving the comprehensiveness and accuracy of data processing. By constructing a dynamic causal quality data graph, using a dynamic time warping algorithm and a mutual information calculation method, the causal relationship between data is accurately identified, the potential association between material defects and construction process errors is revealed, a reliable basis is provided for the tracing and prediction of quality problems, and the intelligent level of construction engineering quality data management is enhanced.

[0038] It should be noted that the above embodiments are only used to illustrate the technical solutions of the application rather than limit the application. Although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the application, and all should be covered in the scope of the claims of the application.

Claims

1. An artificial intelligence-based construction engineering quality data management method, characterized by: The application relates to a method for constructing a dynamic causal quality data graph. The method comprises the following steps:

1. acquiring construction engineering quality data, extracting modal feature vectors in the construction engineering quality data, fusing the modal feature vectors, and obtaining fused feature vectors; 2. preliminarily grouping the fused feature vectors, generating an initial cluster set, clustering the initial cluster set, generating quality data clusters, taking the quality data clusters as graph nodes, calculating the cosine similarity of two graph nodes as a semantic edge weight, and generating an initial quality data graph; 3. extending the initial quality data graph, performing time sequence preprocessing on the extended initial quality data graph, generating a causal edge by calculating and judging a causal direction, adding the causal edge to the extended initial quality data graph, and obtaining a dynamic causal quality data graph; 4. fragmenting the dynamic causal quality data graph, performing distributed embedding optimization on the fragmented subgraphs in local computing nodes, and optimizing the dynamic causal knowledge graph; 2. The artificial intelligence-based construction engineering quality data management method of claim 1, wherein:

5. processing a multi-modal query input by a user into a query vector, using a deep reinforcement learning network to optimize a query path based on the query vector and the optimized dynamic causal quality data graph. The method for extending the initial quality data graph comprises the following steps:

1. calculating the L2 norm of the difference between the fused feature vector of the new construction engineering quality data and each quality data cluster center, and judging the attribution of the fused feature vector of the new construction engineering quality data by using the minimum value in all L2 norms; 2. setting a quality data cluster expansion threshold according to the sum of the average intra-cluster distance mean and the standard deviation of all quality data clusters; 3. when the minimum value in all L2 norms is less than or equal to the quality data cluster expansion threshold, the fused feature vector of the new construction engineering quality data is assigned to the corresponding quality data cluster, and the quality data cluster center is updated; 3.The artificial intelligence-based construction engineering quality data management method of claim 2, wherein:

4. otherwise, the fused feature vector of the new construction engineering quality data is taken as a new quality data cluster center, a new graph node is generated, and the extension of the initial quality data graph is completed. The method for calculating and judging the causal direction comprises the following steps:

1. setting a causal strength threshold according to the sum of the average causal strength mean and the standard deviation of all time sequence quality data subsets; 2. judging whether the causal direction is valid by comparing the causal strength threshold and the causal strength; 4. The artificial intelligence-based construction engineering quality data management method of claim 3, characterized by:

3. when the causal strength is greater than the causal strength threshold, the causal direction is valid, otherwise the causal direction is invalid. The method for fragmenting the dynamic causal quality data graph comprises the following steps:

1. traversing the weight values of all graph semantic edges in the dynamic causal quality data graph, preferentially merging the graph node pairs higher than the median of the graph semantic edge weight values, and obtaining a coarsened quality data graph; 2. using the Kuhn-Munkres algorithm to perform initial partitioning on the coarsened quality data graph, and obtaining subgraphs of the coarsened quality data graph; 3. for the boundary nodes of each subgraph, calculating the inter-subgraph edge weight sum after the boundary nodes are moved to other subgraphs; 5.The artificial intelligence-based construction engineering quality data management method of claim 4, wherein:

4. selecting the boundary node moving operation with the most reduced inter-subgraph edge weight sum, and updating the edge weight and reallocating the boundary nodes. The method for generating the initial cluster set comprises the following steps:

1. calculating the Euclidean distance between the fused feature vectors; 2. setting a similarity threshold according to the mean and the standard deviation of the Euclidean distance between the fused feature vectors, and using the three-sigma principle. The fusion feature vectors with the Euclidean distance between them less than the similarity threshold are assigned to the same cluster, and otherwise, they are assigned to different clusters, generating an initial cluster set. 6.The artificial intelligence-based construction engineering quality data management method of claim 5, wherein: The deep reinforcement learning network is used to optimize the query path, and the specific steps are as follows, The current subgraph is defined as the state space, the next graph node, semantic edge and causal edge are selected as the action space, and the reward function is set based on the cosine similarity between the global embedding vector of the current subgraph and the query vector; The current subgraph is input into the reinforcement learning network, the Q values of all possible actions are calculated, the action with the maximum Q value is selected, the state space is updated, and the reward function is recalculated, and iterative optimization is performed; When the maximum number of iterations is reached, the optimization is stopped, and the current subgraph is taken as a candidate query subgraph; The cosine similarity between the global embedding vector of all graph nodes in the candidate query subgraph and the query vector is calculated, the graph node, semantic edge and causal edge with the highest cosine similarity are retained, and the optimal query path is obtained. 7.The artificial intelligence-based construction engineering quality data management method of claim 6, wherein: The quality data cluster is generated, and the specific steps are as follows, The minimum number of neighbors and the initial neighborhood radius are set; According to the initial cluster set, the initial neighborhood radius is optimized by calculating the contour coefficient; Based on the optimized neighborhood radius and the minimum number of neighbors, the initial cluster set is clustered to generate the quality data cluster. 8.The artificial intelligence-based construction engineering quality data management method of claim 7, wherein: The construction engineering quality data includes material detection records, construction logs, on-site photos, construction videos and voice records; The modal feature vectors include text feature vectors, image feature vectors, video feature vectors and voice feature vectors. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the construction engineering quality data management method based on artificial intelligence according to any one of claims 1-8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the construction engineering quality data management method based on artificial intelligence according to any one of claims 1-8.

Images