A Data Governance and Decision-Making Method for Construction Enterprises Based on Lakewarehouse Integration Architecture

Abstract

This invention provides a data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture, relating to the field of construction enterprise data management technology. The method includes collecting data information from the construction phases of projects managed by the construction enterprise, storing it in a data lake, structuring all data in the data warehouse, calculating construction phase labels based on spatial and business characteristics, determining the current construction phase of the project managed by the construction enterprise based on the construction phase labels, constructing a spatiotemporal demand field based on the current construction phase, predicting the resource demand for all spatial locations on the construction site during construction, partitioning the construction site based on the peak point of the spatiotemporal demand field as seed points (including physical partitioning and digital twin partitioning), obtaining the S2-predicted resource demand and data from the data lake, and calculating anomaly fields for the construction phase based on the obtained data.

Description

Technical Field

[0001] This invention relates to the field of data management technology for construction enterprises, specifically a data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture. Background Technology

[0002] Construction company data encompasses various types of information generated during the operation of construction companies. This data runs through the entire life cycle of a project, from project planning, design, construction to operation and maintenance. In the construction phase, because the total project time is fixed, decisions on the overall construction progress need to be made based on changes in construction phase data.

[0003] To meet the data needs of construction companies, existing enterprises adopt a lake-warehouse integrated architecture for data governance and decision-making. The lake-warehouse integrated architecture combines the flexibility of a data lake with the efficiency of a data warehouse, enabling unified storage, management, and analysis of data. A data lake can store massive amounts of raw data in different formats, preserving the integrity of the data for the enterprise; while a data warehouse cleans, organizes, and models the data to provide high-quality, structured data that is easy to query and analyze quickly.

[0004] However, in actual use, the construction phase data is complex and involves multiple stages, with rapid changes in the data of each stage, and the decision-making timeliness of the construction phase is also required to meet the needs of construction companies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture.

[0006] A data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture includes the following steps:

[0007] S1. Collect data information from construction companies responsible for the project construction phase, store it in a data lake, and structure all data in the data warehouse;

[0008] Spatial feature information is extracted, specifically including Building Information Modeling (BIM) data and Ultra-Wideband (UWB) data from the data lake. Then, business feature information is extracted, specifically including the entropy value of the current spectrum of construction equipment and the gradient of construction material consumption.

[0009] The construction stage label is calculated based on spatial feature information and business feature information. Based on the construction stage label, the construction stage of the project currently under the responsibility of the construction company is determined.

[0010] S2. Construct a spatiotemporal demand field based on the current construction phase of the project undertaken by the construction company, and predict the resource demand of all spatial locations on the construction site during the construction process.

[0011] S3. The construction site is divided into zones based on the peak points of the spatiotemporal demand field as seed points, specifically including physical zoning and digital twin zoning.

[0012] When S1 determines that the construction phase of the project currently managed by the construction company has changed, the demand field is reconstructed through S2, and the partitioning is re-dated based on the reconstructed demand field to obtain the partitioning results.

[0013] S4. Receive the partitioning results from S3, obtain resource demand data for different partitions during the construction phase, and make corresponding decisions based on the comparison results of spatiotemporal demand field data.

[0014] S5. Obtain the resource demand predicted in S2 and the data lake data, and calculate the anomaly field in the construction phase based on the obtained data;

[0015] After obtaining the abnormal field, the abnormal state of the current construction stage is taken as the starting point, and the expected state of the construction stage is set as the ending point to obtain the adjustment path. The abnormal state of the construction stage is then repaired according to the adjustment path.

[0016] Preferably, the specific working steps of S1 are as follows:

[0017] Spatial feature information extraction specifically includes: encoding the Building Information Model (BIM) coordinates to obtain BIM coordinates. The trajectory vector G of the ultra-wideband device is collected, and after fusion, it is represented as... ;

[0018] The extraction of business feature information specifically includes: the construction equipment is a concrete mixer, and the entropy value of the current spectrum is calculated by collecting the current information of the construction equipment;

[0019] Construction materials specifically refer to prefabricated components during the construction phase. The average gradient within that time period is obtained by calculating the difference in the amount of construction materials consumed at two adjacent time points and dividing it by the time interval. This average gradient is then used as the gradient of construction material consumption.

[0020] Preferably, the specific steps for calculating the construction stage label in S1 are as follows:

[0021] First, construct the graph structure and determine the nodes of the graph. Divide the construction phases of the project currently under the responsibility of the construction company into stages, and take the beginning and end of each stage as the construction key points. The feature vector of each construction key point is composed of spatial feature information and business feature information.

[0022] Next, an adjacency matrix is ​​constructed to establish the connection relationships between nodes, forming the adjacency matrix;

[0023] Next, initialize the GCN model, define the number of layers, the number of units per layer, activation functions and other hyperparameters of the GCN, construct the graph convolutional layer and classification layer of the GCN model, construct the graph convolutional layer, the input is the node feature matrix and the adjacency matrix, and the output is the updated node feature matrix;

[0024] After the last graph convolutional layer, a classification layer is added to map node features to the probability distribution of stage labels.

[0025] The constructed node feature matrix and adjacency matrix are used as input data, and the real stage label corresponding to each node is prepared as training label.

[0026] Select the optimizer and loss function to compile the GCN model;

[0027] Prepare training data, input the training data into the GCN model, and perform iterative training. In each round of training, adjust the weights according to the value of the loss function.

[0028] When new data is input into the trained GCN model, for new construction key point data, it is also necessary to first calculate its spatial and business features and construct the corresponding graph structure.

[0029] The node feature matrix and adjacency matrix of the new data are input into the trained GCN model. The model will output the probability distribution of each node belonging to different stage labels. Based on the probability distribution, the stage label with the highest probability is selected as the predicted stage label output for the construction key point.

[0030] Preferably, the specific steps for S2 to construct the spatiotemporal demand field are as follows:

[0031] S21. Divide the site of the project currently under the construction company's responsibility into a three-dimensional grid, and set the grid size to... Individual grid cells;

[0032] S22. Define the formula for calculating resource demand, construct a spatiotemporal demand field, and obtain the time-dependent demand by the combined changes in the construction phase rate and the spatial distribution of resources. At that time, the grid Resource demand at the location ;

[0033] The specific calculation formula is as follows:

[0034] ;

[0035] in Representation Grid exist Resource requirements at any given time, expressed in standard workdays equivalent. The rate of change during the construction phase;

[0036] For the first The basic demand weight of resource categories;

[0037] For the time of the first The center coordinates of the spatial distribution heatmap of resource types;

[0038] and These are regression coefficients used to balance the impact of the rate of change during the construction phase and the spatial distribution of resources on resource demand.

[0039] It is a very small positive constant used to prevent the denominator from being zero;

[0040] S23. Based on the predicted stage label, determine when the construction stage of the construction site changes, trigger the stage change instruction, repeat the demand field reconstruction process of S22, and update the resource demand field at fixed time intervals.

[0041] Preferably, step S2 further includes step S24, used to correct the resource requirement calculation formula based on the business feature information and spatial feature information extracted in real time in step S1:

[0042] S24. Adjust the basic demand weights of various resources based on the real-time extracted business characteristic information. ;

[0043] Based on the extracted spatial feature information, the coordinates of the resource thermal center are determined. Make corrections.

[0044] Preferably, the specific steps of S3 are as follows:

[0045] S31. Identify the peak points in the demand field, perform DBSCAN clustering on the peak points, and generate a set of k seed points;

[0046] S32. Using the seed point set as the center, divide the physical construction area and complete the physical partitioning;

[0047] Specifically, based on the seed point set, the construction site is divided into k physical construction areas using the Voronoi diagram algorithm. Each seed point corresponds to a Voronoi region, which contains all the grid cells closest to that seed point.

[0048] Periodically calculate the difference in resource requirements between adjacent partitions. For each adjacent partition, calculate the ratio of its resource requirement difference to the maximum resource requirement.

[0049] When this ratio exceeds the set value, adjust the partition boundaries;

[0050] The adjustment method is to reassign grid cells near the boundary to the less demanded partitions in the partitions with higher demand, until the ratio is lower than the set value or can no longer be adjusted.

[0051] S33. Map physical partitions in the BIM model, and add a data layer containing resource requirements D(x,y,z,t) to complete the digital twin partitioning. When the demand field is reconstructed, recalculate the seed points and update the partitions.

[0052] Preferably, the specific working steps of S31 are as follows:

[0053] Extract the resource demand D(x,y,z,t) of all grid cells from the demand field and calculate its mean. and standard deviation ;

[0054] Iterate through each grid cell in the demand field to find the cell that satisfies D(x,y,z,t)> +2 The grid points are the candidate peak points;

[0055] Steps: Use the DBSCAN clustering algorithm to cluster the candidate peak points;

[0056] Candidate peak points are divided into multiple clusters, each representing a high-demand region. The center point of each cluster is selected as a seed point, and the seed point set is represented as follows: .

[0057] Preferably, the specific steps of S4 are as follows:

[0058] Establish a demand feature vector for each partition k, and specifically calculate the partition's average demand, maximum demand, demand gradient, and material turnaround time.

[0059] Set threshold , and ;

[0060] When the maximum value of the demand gradient exceeds the threshold When the ratio of the maximum demand to the average demand in a partition exceeds a threshold, a device scheduling command is triggered. When the material turnaround time exceeds a threshold, a material replenishment command is triggered. At that time, a logistics optimization instruction is triggered.

[0061] Preferably, the specific working steps of S5 are as follows:

[0062] Obtain the predicted resource demand data from step S2, including the predicted resource demand for each grid point. ;

[0063] Obtain real-time observational data on resource demand from the data lake. ;

[0064] For each grid point, calculate the relative error between the observed resource demand and the predicted resource demand, and then calculate the average across all grid points to obtain the spatial anomaly degree. ;

[0065] The absolute error of the time derivative of resource demand is calculated and integrated over a set time period to obtain the time anomaly degree. ;

[0066] By combining spatial and temporal anomalies, a comprehensive anomaly value is calculated for each grid point, and a three-dimensional anomaly matrix is ​​constructed. ;

[0067] By associating the IDs of prefabricated components in the BIM model with the trajectory data of UWB devices, we can identify devices or personnel that enter or approach abnormal grids during abnormal time periods, thus identifying the sources of the anomalies.

[0068] Preferably, the specific steps of S5 in repairing the abnormal state during the construction phase according to the adjusted path are as follows:

[0069] Based on the three-dimensional anomaly matrix Set the starting state S0 and the target state ST;

[0070] Constructing decision-making processes and action sets Define a reward function to evaluate the immediate reward of performing action a in state s;

[0071] The algorithm learns the optimal action selection under different states to obtain the optimal repair path from the starting state to the target state.

[0072] This invention provides a data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture. It has the following beneficial effects:

[0073] 1. Multi-source data fusion can comprehensively reflect the construction situation, construct construction stage labels to accurately grasp construction progress, construct spatiotemporal demand fields to predict resource demand, reasonable zoning facilitates construction management, and the repair mechanism based on anomaly fields can improve the stability and controllability of the construction process. Overall, it can effectively manage construction enterprise data and assist decision-making, thereby improving construction efficiency and quality. Attached Figure Description

[0074] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0075] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0076] like Figure 1 This invention proposes a data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture, comprising the following steps:

[0077] Includes the following steps:

[0078] S1. Collect data information from construction companies responsible for the project construction phase, store it in a data lake, and structure all data in the data warehouse;

[0079] Spatial feature information is extracted, specifically including Building Information Modeling (BIM) data and Ultra-Wideband (UWB) data from the data lake. Then, business feature information is extracted, specifically including the entropy value of the current spectrum of construction equipment and the gradient of construction material consumption. It should be noted that relevant data from the construction phase of projects undertaken by construction companies are collected, covering BIM data (including spatial coordinates of the building structure), UWB data (trajectory vector information of equipment), current spectrum data of construction equipment, and construction material consumption data. The collected data is stored in the data lake for unified management and analysis in the future.

[0080] The data warehouse is used to structure all the data in the data lake, and different types of data are organized into structured data formats that are easy to analyze and extract. For example, BIM data and UWB data are stored in corresponding structured tables to prepare for subsequent extraction of spatial features and business features.

[0081] The construction stage label is calculated based on spatial feature information and business feature information. Based on the construction stage label, the construction stage of the project currently under the responsibility of the construction company is determined.

[0082] S2. Construct a spatiotemporal demand field based on the current construction phase of the project undertaken by the construction company, and predict the resource demand of all spatial locations on the construction site during the construction process.

[0083] S3. The construction site is divided into zones based on the peak points of the spatiotemporal demand field as seed points, specifically including physical zoning and digital twin zoning.

[0084] When S1 determines that the construction phase of the project currently managed by the construction company has changed, the demand field is reconstructed through S2, and the partitioning is re-dated based on the reconstructed demand field to obtain the partitioning results.

[0085] S4. Receive the partitioning results from S3, obtain resource demand data for different partitions during the construction phase, and make corresponding decisions based on the comparison results of spatiotemporal demand field data.

[0086] S5. Obtain the resource demand predicted in S2 and the data lake data, and calculate the anomaly field in the construction phase based on the obtained data;

[0087] After obtaining the abnormal field, the abnormal state of the current construction stage is taken as the starting point, and the expected state of the construction stage is set as the ending point to obtain the adjustment path. The abnormal state of the construction stage is then repaired according to the adjustment path.

[0088] It should be noted that, based on the integrated lake-warehouse architecture, construction data from construction companies is collected into a data lake and structured using a data warehouse. By integrating multi-source data such as BIM data, UWB data, current spectrum entropy values ​​of construction equipment, and consumption gradients of construction materials, construction stage labels are constructed through feature extraction to determine the construction stage. Based on the construction stage, a spatiotemporal demand field is constructed to predict resource demand. Based on this, the construction site is partitioned, and decisions are made by comparing the partitioning results with the spatiotemporal demand field. Furthermore, the calculation of abnormal fields in the construction stage and the adjustment of paths to repair abnormal states are also introduced.

[0089] The integrated lake-warehouse architecture facilitates efficient data storage, management, and analysis, breaks down data silos, integrates multi-source data to comprehensively reflect construction status, constructs construction phase labels to accurately grasp construction progress, constructs spatiotemporal demand fields to predict resource demand, rationally partitions to facilitate construction management, and a repair mechanism based on anomalies can improve the stability and controllability of the construction process. Overall, it can effectively manage construction enterprise data and assist in decision-making, thereby improving construction efficiency and quality.

[0090] As an optional embodiment, the specific working steps of S1 are as follows:

[0091] Spatial feature information extraction specifically includes: encoding the Building Information Model (BIM) coordinates to obtain BIM coordinates. The trajectory vector G of the ultra-wideband device is collected, and after fusion, it is represented as... It should be noted that the Geohash algorithm is used to encode BIM coordinates by dividing the coordinate space into a grid and then encoding according to the grid in which the coordinates are located. The encoding length can control the precision, and the encoded result can represent the approximate location of the geographical location, converting continuous building space coordinates into discrete Geohash code.

[0092] By receiving signals from multiple UWB base stations, the real-time location coordinates of the device are calculated to form a trajectory vector, which can be represented as a series of location coordinate points corresponding to timestamps.

[0093] The specific fusion method is concatenation fusion. For example, the character sequence of Geohash code and the coordinate point sequence in UWB trajectory vector are concatenated in a certain order to form a longer feature vector. Similarly, the device current spectrum entropy value and material consumption gradient value are concatenated to fuse feature information from different sources and form a comprehensive feature representation for subsequent data analysis, machine learning modeling and other tasks.

[0094] The extraction of business feature information specifically includes: the construction equipment is a concrete mixer, and the entropy value of the current spectrum is obtained by collecting the current information of the construction equipment; it should be noted that in this embodiment, it is specifically calculated using the Shannon entropy formula.

[0095] Construction materials specifically refer to precast components used during the construction phase. The average gradient of material consumption over a given time period is obtained by calculating the difference in material consumption between two adjacent points in time and dividing by the time interval. This average gradient serves as the gradient for material consumption. Precast components specifically include various types of precast concrete parts.

[0096] It should be noted that the fusion of Geohash encoding and UWB trajectory vectors can convert continuous building spatial coordinates into discrete codes, while simultaneously grasping the real-time location of equipment. This provides accurate spatial information for subsequent analysis, and the refined calculation of business feature information can more accurately extract equipment operating status and material consumption trends, which helps to more accurately determine the construction stage and predict resource demand.

[0097] As an optional embodiment, the specific steps for calculating the construction stage label in S1 are as follows:

[0098] First, construct the graph structure and determine the nodes of the graph. Divide the construction phases of the project currently under the responsibility of the construction company into stages, and take the beginning and end of each stage as the construction key points. The feature vector of each construction key point is composed of spatial feature information and business feature information. For example, assuming that each node has 5-dimensional spatial features and 3-dimensional business features, the shape of the feature matrix X is (N,8), where N is the number of nodes.

[0099] Next, an adjacency matrix is ​​constructed to establish the connection relationship between nodes, forming an adjacency matrix. For example, if there is a sequential or mutually restrictive relationship between two construction key points, the corresponding position in the adjacency matrix is ​​marked as 1, otherwise it is marked as 0. For 10 construction key points, if key point 1 must be completed before key point 2, the position in the first row and second column of adjacency matrix A is marked as 1, otherwise it is marked as 0.

[0100] Next, initialize the GCN model, define the number of layers, the number of units per layer, activation functions and other hyperparameters of the GCN, construct the graph convolutional layer and classification layer of the GCN model, construct the graph convolutional layer, the input is the node feature matrix and the adjacency matrix, and the output is the updated node feature matrix;

[0101] After the last graph convolutional layer, a classification layer is added to map node features to the probability distribution of stage labels. It should be noted that the appropriate number of GCN layers should be selected based on the complexity of the problem and the amount of data.

[0102] Determine the number of units in each layer. Typically, as the number of layers increases, the number of units can be gradually reduced so that the model can progressively extract higher-level feature representations.

[0103] In this embodiment, a GCN model with two graph convolutional layers is initialized. The first layer has 16 units and the second layer has 8 units, both using the ReLU activation function. This is followed by a classification layer using the Softmax activation function.

[0104] The constructed node feature matrix and adjacency matrix are used as input data, and the real stage label corresponding to each node is prepared as training label. It should be noted that the real stage label corresponding to each node is the historical construction stage label of the construction company.

[0105] Select the optimizer and loss function to compile the GCN model;

[0106] Choose an appropriate optimizer; in this embodiment, the Adam optimizer is used.

[0107] The loss function is selected based on the problem type. In this embodiment, the cross-entropy loss function is used.

[0108] The optimizer and loss function are associated with the GCN model to complete the model compilation process;

[0109] Prepare training data and input it into the GCN model for iterative training. In each round of training, adjust the weights according to the value of the loss function. It should be noted that the training data consists of a node feature matrix and an adjacency matrix. Combine the spatial features and business features of each extracted key point into a feature vector to form the feature matrix X.

[0110] Ensure that the construction of the adjacency matrix A is consistent with the actual construction logic dependencies, and that the dimension of the matrix corresponds to the number of nodes in the feature matrix X;

[0111] It should also be noted that in this embodiment, a larger batch size is set, such as the same as the number of nodes, because graph data is usually processed as a whole.

[0112] In each training round, the model automatically adjusts the weights based on the value of the loss function to minimize the difference between the predicted results and the true labels;

[0113] The model was trained for 100 rounds, with all nodes used in each round.

[0114] When new data is input into the trained GCN model, for new construction key point data, it is also necessary to first calculate its spatial and business features and construct the corresponding graph structure.

[0115] The node feature matrix and adjacency matrix of the new data are input into the trained GCN model. The model will output the probability distribution of each node belonging to different stage labels. Based on the probability distribution, the stage label with the highest probability is selected as the predicted stage label output for the construction key point.

[0116] It should be noted that the specific steps for calculating construction stage labels by constructing a graph structure and using a GCN model are clearly defined, including determining the nodes of the graph, constructing the adjacency matrix, initializing the GCN model, and training the GCN model.

[0117] Graph structures can intuitively represent the logical relationships between key points in the construction phase. The GCN model can comprehensively consider spatial and business feature information and learn the deep feature representation of key points, thereby more accurately classifying the construction phase, improving the accuracy of construction phase judgment, and providing a reliable basis for subsequent resource demand forecasting and decision-making.

[0118] As an optional embodiment, the specific steps for S2 to construct the spatiotemporal demand field are as follows:

[0119] S21. Divide the site of the project currently under the construction company's responsibility into a three-dimensional grid, and set the grid size to... Each unit grid has a size of no more than [number] cells; it should be noted that, in this embodiment, the size of each unit grid does not exceed [number]. ;

[0120] S22. Define the formula for calculating resource demand, construct a spatiotemporal demand field, and obtain the time-dependent demand by the combined changes in the construction phase rate and the spatial distribution of resources. At that time, the grid Resource demand at the location ;

[0121] The specific calculation formula is as follows:

[0122] ;

[0123] in Representation Grid exist Resource requirements at any given time, expressed in standard workdays equivalent. The rate of change during the construction phase reflects the speed of progress in each phase and is obtained by differencing the time series of phase labels.

[0124] For the The basic demand weight of a resource category; representing the contribution weight of different prefabricated component types to the resource demand.

[0125] For the time of the first The center coordinates of the spatial distribution heatmap of similar resources; it should be noted that the specific method of obtaining this information is to add construction status markers to each prefabricated component in the BIM model, including statuses such as "not under construction", "under construction", and "completed";

[0126] Based on the construction schedule and actual construction conditions, prefabricated components that are in the "under construction" stage are selected.

[0127] For precast components that are under construction, calculating their geometric center coordinates can serve as the coordinates of the construction hotspots of the resource.

[0128] and The regression coefficients are used to balance the impact of the rate of change in the construction phase and the spatial distribution of resources on resource demand. By collecting actual resource demand data and corresponding data on the rate of change in the construction phase and the spatial distribution of resources from multiple historical projects, a regression model is established and solved using regression analysis methods such as the least squares method.

[0129] It is a very small positive constant used to prevent the denominator from being zero; in this embodiment, it is taken as 1×10^-6;

[0130] S23. Based on the predicted stage label, determine when the construction stage of the construction site changes, trigger the stage change instruction, repeat the demand field reconstruction process of S22, and update the resource demand field at fixed time intervals.

[0131] In this embodiment, the fixed time interval is 5 minutes. Within 5 minutes of receiving the stage change instruction, the resource requirement D(x,y,z,t) for each grid cell is recalculated based on the latest construction stage information and resource requirements, including:

[0132] Update the rate of change during the construction phase ;

[0133] Adjust the basic demand weights of various resources according to the new construction phase. ;

[0134] Redetermine the center coordinates of the resource spatial distribution heat map ;

[0135] By substituting the calibrated regression coefficients into the resource demand forecasting model, a new resource demand field distribution can be calculated.

[0136] The reconstructed resource demand data is pushed to construction management personnel, the resource scheduling system, and on-site workers in real time. Construction management personnel can intuitively view the changes in the resource demand field through mobile terminals or management platforms; the resource scheduling system optimizes resource allocation plans based on the updated resource demand field; and on-site workers can rationally arrange construction tasks and resource usage based on the latest resource demand information.

[0137] It should be noted that the paper describes how to divide the building site into a three-dimensional grid when constructing the spatiotemporal demand field, define the formula for calculating resource demand, and determine the resource demand based on the rate of change of the construction stage and the spatial distribution of resources. It also proposes a process for updating the rate of change of the construction stage, adjusting the weight of the basic resource demand, redetermining the center coordinates of the resource spatial distribution heat map, and reconstructing the resource demand field after triggering the stage change instruction.

[0138] Three-dimensional meshing can discretize complex construction sites, making it easier to accurately calculate resource requirements in each area. The defined resource requirement calculation formula takes into account construction progress and spatial factors, which can more accurately predict resource requirements. When changes occur in the construction phase, relevant parameters can be updated and the demand field can be reconstructed in a timely manner, which can ensure the real-time performance and accuracy of resource requirement prediction and help optimize resource allocation.

[0139] As an optional embodiment: S2 further includes S24, used to correct the resource demand calculation formula based on the business feature information and spatial feature information extracted in real time by S1.

[0140] S24. Adjust the basic demand weights of various resources based on the real-time extracted business characteristic information. It should be noted that the specific formula is as follows: For the initial first The basic demand weight of a resource category represents the basic demand for that resource under normal construction conditions;

[0141] ;

[0142] For the initial first The basic demand weight of a resource category represents the basic demand for that resource under normal construction conditions;

[0143] For the first after dynamic calibration Weights of basic resource requirements;

[0144] The current spectrum entropy value of the device at the current moment. When the device malfunctions, the current spectrum entropy value will increase, indicating that the device's operating state becomes unstable.

[0145] Specifically, a high-precision current sensor, such as a Hall current sensor, is connected to the power supply line of the device, and the voltage signal output by the current sensor is connected to the data acquisition card.

[0146] The current signal is collected over a period of time, the length of which needs to be determined based on the equipment's operating cycle and possible abnormal situations.

[0147] The acquired time-domain current signal is converted to the frequency domain using a fast Fourier transform.

[0148] For example, given a current signal sequence x(n) of length N, after FFT, a frequency domain sequence X(k) is obtained, where k = 0, 1, ..., N-1. Each point in the frequency domain sequence corresponds to the amplitude and phase information of a frequency component.

[0149] Power spectral density (PSD) reflects the power distribution of a signal at different frequencies. It can be calculated from the square of the magnitude of the FFT result, i.e., PSD(k) = |X(k)|². For example, if the complex amplitude of a frequency point k after FFT is 3 + 4i, then its power spectral density is (3)² + (4)² = 25.

[0150] The power spectral density is normalized so that its sum is 1, and it is used as the probability density function of the frequency component, i.e., p(k)=PSD(k) / ΣPSD(k), where ΣPSD(k) is the sum of the power spectral densities at all frequency points;

[0151] The spectral entropy is calculated according to the formula of information entropy: device current spectral entropy Hcurrent=-Σp(k)×log2p(k). If there are m frequency points with probability densities greater than 0 in the normalized probability density sequence, these m points are summed.

[0152] Specifically, at the construction site or production site, designated personnel should regularly record the consumption of materials. The recording frequency can be determined based on the project progress and the rate of material consumption. The collected material consumption data should be arranged in chronological order to construct a time series, and regression analysis should be used to fit the material consumption time series to obtain the desired results. ;

[0153] This is the feature influence coefficient, used to adjust the degree of influence of business features on resource weights;

[0154] In this embodiment The values ​​are 0.15 and 0.2, respectively, used to balance changes in equipment status and material consumption. In actual use, they should be adjusted as needed.

[0155] Based on the extracted spatial feature information, the coordinates of the resource thermal center are determined. Make corrections. It should be noted that the specific formula is as follows:

[0156] ;

[0157] : The center coordinates of the spatial distribution heatmap of the j-th type of resource at time t after correction.

[0158] Based on the center coordinates obtained by clustering UWB device trajectory points, UWB devices can track the location of personnel and equipment at the construction site in real time. By performing cluster analysis on UWB trajectory points using the DBSCAN clustering algorithm, the concentrated distribution area of ​​resources in actual construction can be obtained.

[0159] Construction hotspot coordinates identified from the construction status markers of BIM components. The BIM model contains detailed design information and construction schedule of the building. By analyzing the construction status of components, the location of resource demand hotspots can be predicted.

[0160] : Dynamic adjustment factor, used to control the weight of UWB data and BIM data in the calculation of thermal center coordinates. In this embodiment, the specific value is: when the construction stage label belongs to the main construction or equipment installation stage, Taking 0.7, at this point, more emphasis is placed on the actual location data provided by the UWB equipment, because the construction activities in these two stages are relatively complex and the location changes frequently; other stages Taking 0.3 means relying more on the predicted location of the BIM model, because construction activities are relatively regular at this point, and the prediction accuracy of the BIM model is relatively high.

[0161] By combining real-time location data provided by UWB devices with predicted location information from BIM models, the coordinates of the resource thermal center are corrected. This fusion approach fully leverages the advantages of both data sources, improving the accuracy and robustness of resource spatial distribution prediction. During complex phases such as main construction and equipment installation, real-time data from UWB devices better reflects the actual construction situation; while during other relatively stable construction phases, the predicted location from the BIM model is more reliable.

[0162] It should be noted that adjusting resource weights by combining real-time equipment current spectrum entropy values ​​and material consumption gradients can make resource demand forecasts more closely reflect actual construction conditions, such as addressing equipment failures or abnormal material consumption. Correcting the resource thermal center coordinates integrates real-time UWB equipment location data and BIM model predicted location information, improving the accuracy and robustness of resource spatial distribution forecasts and enhancing the reliability of resource demand forecasts.

[0163] As an optional embodiment, the specific steps of S3 are as follows:

[0164] S31. Identify the peak points in the demand field, perform DBSCAN clustering on the peak points, and generate a set of k seed points;

[0165] S32. Using the seed point set as the center, divide the physical construction area and complete the physical partitioning;

[0166] Specifically, based on the seed point set, the construction site is divided into k physical construction areas using the Voronoi diagram algorithm. Each seed point corresponds to a Voronoi region, which contains all the grid cells closest to that seed point.

[0167] Periodically calculate the difference in resource requirements between adjacent partitions. For each adjacent partition, calculate the ratio of its resource requirement difference to the maximum resource requirement.

[0168] When the ratio exceeds the set value, the partition boundary is adjusted; it should be noted that the set value in this embodiment is 0.3, which is set by the construction personnel, and the specific value range is between 0.2 and 0.35.

[0169] The adjustment method is to reassign grid cells near the boundary to the less demanded partitions in the partitions with higher demand, until the ratio is lower than the set value or can no longer be adjusted.

[0170] S33. Map physical partitions in the BIM model. An additional data layer contains resource requirements D(x,y,z,t), completing the digital twin partitioning. When the demand field is reconstructed, seed points are recalculated and partitions are updated. It should be noted that the following additional data layers are integrated into the digital twin partitioning:

[0171] Resource demand matrix D(x,y,z,t): Stores the resource demand of each grid cell, used to visually display the distribution of resource demand;

[0172] Based on UWB device trajectories and device types, a heat map is generated to display device density and distribution, in order to optimize device scheduling;

[0173] Based on safety monitoring data (such as working at heights, temporary power supply, etc.), calculate and display the safety risk density of each area to assist in safety management;

[0174] It should be noted that dividing the construction site into zones and dynamically adjusting the zone boundaries according to resource demand can make the zones more in line with actual construction needs. Mapping physical zones in the BIM model and adding a data layer realizes the combination of physical construction and digital twin, which facilitates intuitive display and analysis of the construction situation and provides more comprehensive information support for construction management.

[0175] As an optional embodiment, the specific working steps of S31 are as follows:

[0176] Extract the resource demand D(x,y,z,t) of all grid cells from the demand field and calculate its mean. and standard deviation ;

[0177] Iterate through each grid cell in the demand field to find the cell that satisfies D(x,y,z,t)> +2 The grid points are the candidate peak points;

[0178] Steps: Use the DBSCAN clustering algorithm to cluster the candidate peak points;

[0179] Parameter settings: Set the neighborhood radius to 2-3 meters (adjust the specific value according to the resource density of the construction site), and set the minimum number of samples (MinPts) to 5-10 (depending on the clustering scale).

[0180] Candidate peak points are divided into multiple clusters, each representing a high-demand region. The center point of each cluster is selected as a seed point, and the seed point set is represented as follows: .

[0181] It should be noted that the ability to accurately identify high-demand areas in the demand field provides a key basis for the subsequent rational division of construction zones. Reasonable parameter settings ensure the accuracy and reliability of the clustering results, thereby improving the rationality of the zoning and helping to achieve effective allocation and management of construction resources.

[0182] As an optional embodiment, the specific steps of S4 are as follows:

[0183] Establish a demand feature vector for each partition k, and specifically calculate the partition's average demand, maximum demand, demand gradient, and material turnaround time.

[0184] Set threshold , and ;

[0185] When the maximum value of the demand gradient exceeds the threshold When the ratio of the maximum demand to the average demand in a partition exceeds a threshold, a device scheduling command is triggered. When the material turnaround time exceeds a threshold, a material replenishment command is triggered. At that time, a logistics optimization instruction is triggered. It should be noted that a statistical analysis of the demand gradient values ​​corresponding to the point where construction efficiency begins to decline significantly across all projects is performed, and the average value is taken as the... The initial value is set when the maximum value of the demand gradient exceeds the threshold. This indicates that demand is changing rapidly in space, and it may be necessary to reschedule equipment to meet the construction needs of different areas;

[0186] When the ratio of the maximum demand to the average demand in a region exceeds a threshold In such cases, materials need to be replenished promptly to prevent construction delays due to material shortages. Specifically, a statistical analysis should be conducted on the demand ratios corresponding to the initial risk of material shortages in all projects, and the average value should be taken as the benchmark. Initial value;

[0187] When material turnover time exceeds the threshold This indicates that material inventory is piling up or being consumed too slowly, requiring optimization of the logistics and distribution plan to reduce inventory costs and capital tied up. Specifically, statistical analysis should be conducted on the material turnover times corresponding to when costs and risks begin to increase significantly across all projects, and the average value should be taken as the benchmark. Initial value;

[0188] It should be noted that by establishing demand feature vectors for each zone, the construction demand of each zone can be fully understood. Setting thresholds and triggering corresponding instructions can promptly detect abnormalities in the construction process and take measures such as equipment scheduling, material replenishment, and logistics optimization, thereby ensuring the smooth progress of the construction process and improving construction efficiency and quality.

[0189] As an optional embodiment, the specific working steps of S5 are as follows:

[0190] Obtain the predicted resource demand data from step S2, including the predicted resource demand for each grid point. ;

[0191] Obtain real-time observational data on resource demand from the data lake. ;

[0192] For each grid point, calculate the relative error between the observed resource demand and the predicted resource demand, and then calculate the average across all grid points to obtain the spatial anomaly degree. ;

[0193] It should be noted that the specific calculation formula is as follows;

[0194] ;

[0195] Where N is the number of grid points. This represents the resource requirement data for the i-th grid point. Represents the predicted resource demand of the i-th grid point. ;

[0196] The absolute error of the time derivative of resource demand is calculated and integrated over a set time period to obtain the time anomaly degree. It should be noted that, for example, setting a time period as... The specific formula for integration is: ;

[0197] By combining spatial and temporal anomalies, a comprehensive anomaly value is calculated for each grid point, and a three-dimensional anomaly matrix is ​​constructed. It should be noted that the comprehensive outlier value is calculated by combining spatial and temporal outliers at time t, specifically by assigning equal weights to spatial and temporal outliers.

[0198] Set a threshold for the overall outlier value, and mark grid points whose overall outlier value is greater than the threshold as outlier grids;

[0199] The threshold for comprehensive outliers is obtained by calculating the distribution of historical outliers and selecting a high percentile (e.g., the 95th percentile).

[0200] By associating the IDs of prefabricated components in the BIM model with the trajectory data of UWB devices, we can identify devices or personnel that enter or approach abnormal grids during abnormal time periods, thus identifying the sources of the anomalies.

[0201] It should be noted that reflecting abnormal situations during the construction process and promptly identifying potential problems, by comparing with historical data and combining BIM model and UWB equipment data for anomaly tracing, can quickly locate the cause of the anomaly and provide strong support for taking targeted measures to repair the abnormal state.

[0202] As an optional embodiment, the specific steps of S5 in repairing the abnormal state during the construction phase according to the adjusted path are as follows:

[0203] Based on the three-dimensional anomaly matrix Set the starting state S0 and the target state ST; it should be noted that...

[0204] A(x,y,z,t): The three-dimensional anomaly matrix at the current moment, representing the comprehensive anomaly degree of each grid point in the construction site, obtained by fusing spatial anomaly degree and temporal anomaly degree;

[0205] The observed gradient of resource demand represents the rate of change of resource demand in space, reflecting the uneven spatial distribution of resource demand.

[0206] The delay time of the current construction phase relative to the planned schedule is used to measure the degree of lag in the construction progress.

[0207] The target state is a desired construction state, defined as follows:

[0208] ;

[0209] The expected degree of abnormality is no more than 10% of the abnormality threshold, of which It is a threshold used to judge abnormalities. It can be determined through statistical analysis of historical data and is the average value of historical data. This condition ensures that abnormal situations during construction are effectively controlled and that the construction status is close to the normal level.

[0210] |∇D_obs-∇D_pred|< The difference between the expected gradient of resource demand and the predicted gradient is less than a set threshold. , The value can be determined based on the fluctuation range of the resource demand gradient in historical data;

[0211] Constructing decision-making processes and action sets Define a reward function to evaluate the immediate reward of performing action a in state s;

[0212] It should be noted that the decision-making process specifically defines three main actions that can be taken during the anomaly repair process:

[0213] Equipment adjustment: Adjusting construction equipment, including equipment repair, rescheduling, or changing equipment types, to improve equipment efficiency or adapt to new construction needs;

[0214] Material reallocation: The reallocation of construction materials, including the emergency transfer of materials from the material inventory to areas with high demand, or the optimization of material distribution plans, to ensure the timeliness and rationality of material supply;

[0215] Work process optimization: Optimize the construction process, including adjusting the construction sequence, merging or breaking down construction tasks, increasing or decreasing construction teams, etc., to improve construction efficiency and reduce construction delays.

[0216] It should also be noted that, in this embodiment, a reward function is defined to evaluate the immediate reward brought about by performing action a in state s, and its expression is:

[0217]

[0218] : Indicates the negative impact of the degree of abnormality on the reward, where It is a penalty coefficient used to measure the impact weight of the degree of abnormality on the construction process. Its value can be determined based on the historical impact data of abnormality on construction costs, schedule and quality. The higher the degree of abnormality, the lower the reward.

[0219] : Indicates the impact of the time required to perform action a on the reward, where t(a) is the time cost coefficient, which measures the cost weight of time in the abnormal repair process. Its value can be set according to the tightness of the construction project schedule. t(a) is the estimated time required to perform action a. The longer the time, the lower the reward.

[0220] The algorithm learns the optimal action selection under different states to obtain the optimal repair path from the starting state to the target state.

[0221] It should be noted that the specific steps are to create a Q table to store the Q values ​​corresponding to different combinations of states (S) and actions (A). Initially, all Q values ​​are set to 0. The rows of the Q table represent states, and the columns represent actions.

[0222] Set the abnormal state of the current construction phase as the current state Scurrent=S0;

[0223] In the state Scurrent, select an action 'a' from the action set, randomly select an action with a fixed probability to explore the environment, and select the action with the largest current Q value with a probability of 1-fixed change to utilize existing knowledge.

[0224] Perform the action and observe the new state and reward. Perform the selected action a, observe the new state Snew and the immediate reward R(Scurrent,a) from the environmental feedback.

[0225] Update the value of Q(Scurrent,a) in the Q table according to the update formula of Q-learning:

[0226]

[0227]

[0228] in:

[0229] The learning rate controls the degree to which new information replaces old information. It is usually a value between 0 and 1 and can be adjusted according to experimental results. In this embodiment, it is set to 0.5.

[0230] This is a discount factor used to balance the weights of immediate rewards and future rewards. It is usually between 0 and 1 and can be adjusted according to the importance attached to future rewards. In this example, it is set to 0.5.

[0231] Update the current state to the new state: Scurrent = Snew;

[0232] If the current state satisfies the target state ST, or the maximum number of iterations is reached, the learning process terminates; otherwise, it returns to continue learning.

[0233] Extract the optimal strategy from the updated Q table, that is, for each state, select the action with the largest Q value as the optimal action, and form the optimal repair path from the starting state to the target state.

[0234] It should be noted that, based on the current construction status and abnormal situations, the optimal repair action can be intelligently selected to achieve an effective transformation from the abnormal state to the target state, thereby improving the efficiency and accuracy of abnormal repair and ensuring the smooth implementation of the construction project.

[0235] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture, characterized in that, Includes the following steps: S1. Collect data information from construction companies responsible for the project construction phase, store it in a data lake, and structure all data in the data warehouse; Spatial feature information is extracted, specifically including Building Information Modeling (BIM) data and Ultra-Wideband (UWB) data from the data lake. Then, business feature information is extracted, specifically including the entropy value of the current spectrum of construction equipment and the gradient of construction material consumption. The construction stage label is calculated based on spatial feature information and business feature information. Based on the construction stage label, the construction stage of the project currently under the responsibility of the construction company is determined. S2. Construct a spatiotemporal demand field based on the current construction phase of the project undertaken by the construction company, and predict the resource demand of all spatial locations on the construction site during the construction process. S3. The construction site is divided into zones based on the peak points of the spatiotemporal demand field as seed points, specifically including physical zoning and digital twin zoning. When S1 determines that the construction phase of the project currently managed by the construction company has changed, the demand field is reconstructed through S2, and the partitioning is re-dated based on the reconstructed demand field to obtain the partitioning results. S4. Receive the partitioning results from S3, obtain resource demand data for different partitions during the construction phase, and make corresponding decisions based on the comparison results of spatiotemporal demand field data. S5. Obtain the resource demand predicted in S2 and the data lake data, and calculate the anomaly field in the construction phase based on the obtained data; After obtaining the abnormal field, the abnormal state of the current construction stage is taken as the starting point, and the expected state of the construction stage is set as the ending point to obtain the adjustment path. The abnormal state of the construction stage is then repaired according to the adjustment path.

2. The data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 1, characterized in that: The specific working steps of S1 are as follows: Spatial feature information extraction specifically includes: encoding the Building Information Model (BIM) coordinates to obtain BIM coordinates. The trajectory vector G of the ultra-wideband device is collected, and after fusion, it is represented as... ; The extraction of business feature information specifically includes: the construction equipment is a concrete mixer, and the entropy value of the current spectrum is calculated by collecting the current information of the construction equipment; Construction materials specifically refer to prefabricated components during the construction phase. The average gradient within that time period is obtained by calculating the difference in the amount of construction materials consumed at two adjacent time points and dividing it by the time interval. This average gradient is then used as the gradient of construction material consumption.

3. The data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 2, characterized in that: The specific steps for calculating the construction stage label using S1 are as follows: First, construct the graph structure and determine the nodes of the graph. Divide the construction phases of the project currently under the responsibility of the construction company into stages, and take the beginning and end of each stage as the construction key points. The feature vector of each construction key point is composed of spatial feature information and business feature information. Next, an adjacency matrix is ​​constructed to establish the connection relationships between nodes, forming the adjacency matrix; Next, initialize the GCN model, define the number of layers, the number of units per layer, and the activation function hyperparameters, construct the graph convolutional layer and classification layer of the GCN model, construct the graph convolutional layer, the input is the node feature matrix and the adjacency matrix, and the output is the updated node feature matrix; After the last graph convolutional layer, a classification layer is added to map node features to the probability distribution of stage labels. The constructed node feature matrix and adjacency matrix are used as input data, and the real stage label corresponding to each node is prepared as training label. Select the optimizer and loss function to compile the GCN model; Prepare training data, input the training data into the GCN model, and perform iterative training. In each round of training, adjust the weights according to the value of the loss function. When new data is input into the trained GCN model, for new construction key point data, it is also necessary to first calculate its spatial and business features and construct the corresponding graph structure. The node feature matrix and adjacency matrix of the new data are input into the trained GCN model. The model will output the probability distribution of each node belonging to different stage labels. Based on the probability distribution, the stage label with the highest probability is selected as the predicted stage label output for the construction key point.

4. The data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 2, characterized in that: The specific steps for constructing the spatiotemporal demand field using S2 are as follows: S21. Divide the site of the project currently under the construction company's responsibility into a three-dimensional grid, and set the grid size to... Individual grid cells; S22. Define the formula for calculating resource demand, construct a spatiotemporal demand field, and obtain the time-dependent demand by the combined changes in the construction phase rate and the spatial distribution of resources. At that time, the grid Resource demand at the location ; The specific calculation formula is as follows: ； in Representation Grid exist Resource requirements at any given time, expressed in standard workdays equivalent. The rate of change during the construction phase; For the first The basic demand weight of resource categories; For the The center coordinates of the spatial distribution heatmap of resource types; and These are regression coefficients used to balance the impact of the rate of change during the construction phase and the spatial distribution of resources on resource demand. It is a very small positive constant used to prevent the denominator from being zero; S23. Based on the predicted stage label, determine when the construction stage of the construction site changes, trigger the stage change instruction, repeat the demand field reconstruction process of S22, and update the resource demand field at fixed time intervals.

5. A data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 4, characterized in that: S2 further includes S24, which is used to correct the resource requirement calculation formula based on the business feature information and spatial feature information extracted in real time by S1: S24. Adjust the basic demand weights of various resources based on the real-time extracted business characteristic information. ; Based on the extracted spatial feature information, the coordinates of the resource thermal center are determined. Make corrections.

6. A data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 4, characterized in that: The specific steps of S3 are as follows: S31. Identify the peak points in the demand field, perform DBSCAN clustering on the peak points, and generate a set of k seed points; S32. Using the seed point set as the center, divide the physical construction area and complete the physical partitioning; Specifically, based on the seed point set, the construction site is divided into k physical construction areas using the Voronoi diagram algorithm. Each seed point corresponds to a Voronoi region, which contains all the grid cells closest to that seed point. Periodically calculate the difference in resource requirements between adjacent partitions. For each adjacent partition, calculate the ratio of its resource requirement difference to the maximum resource requirement. When this ratio exceeds the set value, adjust the partition boundaries; The adjustment method is to reassign grid cells near the boundary to the less demanded partitions in the partitions with higher demand, until the ratio is lower than the set value or can no longer be adjusted. S33. Map physical partitions in the BIM model, with an additional data layer containing resource requirements. The digital twin partitioning is completed, and when the demand field is reconstructed, the seed points are recalculated and the partitioning is updated.

7. A data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 6, characterized in that: The specific working steps of S31 are as follows: Extract the resource requirements of all grid cells from the demand field. Calculate its mean and standard deviation ; Traverse each grid cell in the demand field to find those that meet the requirements. The grid points are the candidate peak points; Steps: Use the DBSCAN clustering algorithm to cluster the candidate peak points; Candidate peak points are divided into multiple clusters, each representing a high-demand region. The center point of each cluster is selected as a seed point, and the seed point set is represented as follows: .

8. A data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 6, characterized in that: The specific steps of S4 are as follows: Establish a demand feature vector for each partition k, and specifically calculate the partition's average demand, maximum demand, demand gradient, and material turnaround time. Set threshold , , ; When the maximum value of the demand gradient exceeds the threshold When the ratio of the maximum demand to the average demand in a partition exceeds a threshold, a device scheduling command is triggered. When the material turnaround time exceeds a threshold, a material replenishment command is triggered. At that time, a logistics optimization instruction is triggered.

9. A data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 3, characterized in that: The specific working steps of S5 are as follows: Obtain the predicted resource demand data from step S2, including the predicted resource demand for each grid point. ; Obtain real-time observational data on resource demand from the data lake. ; For each grid point, calculate the relative error between the observed resource demand and the predicted resource demand, and then calculate the average across all grid points to obtain the spatial anomaly degree. ; The absolute error of the time derivative of resource demand is calculated and integrated over a set time period to obtain the time anomaly degree. ; By combining spatial and temporal anomalies, a comprehensive anomaly value is calculated for each grid point, and a three-dimensional anomaly matrix is ​​constructed. ; By associating the IDs of prefabricated components in the BIM model with the trajectory data of UWB devices, we can identify devices or personnel that enter or approach abnormal grids during abnormal time periods, thus identifying the sources of the anomalies.

10. A data governance and decision-making method for construction enterprises based on a lake-warehouse integrated architecture as described in claim 9, characterized in that: The specific steps of S5 in repairing the abnormal state during the construction phase according to the adjusted path are as follows: Based on the three-dimensional anomaly matrix Set the starting state S0 and the target state ST; Constructing decision-making processes and action sets Define a reward function to evaluate the immediate reward of performing action a in state s; The algorithm learns the optimal action selection under different states to obtain the optimal repair path from the starting state to the target state.

Images