A waterproof sand laying state determination method and system

Abstract

The present application relates to waterproof sand laying state measurement method technical field, especially in kind, a kind of waterproof sand laying state determination method and system, the method comprises: after waterproof sand laying is completed, uniform spatial grid division is carried out to laying area, the area is divided into multiple spatial detection units, and the waterproof sand state data of each spatial detection unit is collected;Preset time interval, collect waterproof sand state data on different time nodes after laying for each spatial detection unit, construct state evolution track, based on the state evolution track, the change rate of each state parameter in waterproof sand state data is calculated, and time sequence feature vector is formed.The present application realizes the change from single-point early warning to regional risk prediction, through the calculation of node risk coefficient and influence coefficient, high-risk unit can be accurately identified, targeted intervention basis is provided for construction site, repair and maintenance cost is reduced, and the long-term stability of waterproof sand laying layer is improved.

Description

Technical Field

[0001] This invention relates to the technical field of methods for measuring the laying status of waterproof sand, specifically a method and system for determining the laying status of waterproof sand. Background Technology

[0002] Waterproof sand paving layers are widely used in waterproofing systems for roofing, underground engineering, and municipal structures. They typically serve as a protective and leveling layer for the waterproofing layer, and their structural uniformity and compaction directly affect the integrity and durability of the waterproofing system. In actual construction, waterproof sand paving often relies on manual spreading combined with simple mechanical leveling. Quality control is mainly achieved through local thickness measurements, compaction checks, or empirical visual inspections. These methods have a point-like, discrete characteristic and are difficult to reflect the overall spatial distribution of the paving layer.

[0003] Furthermore, in the initial stage after the completion of construction, the waterproof sand layer will still be affected by subsequent construction loads, changes in environmental temperature and humidity, and micro-deformation of the base layer. Its internal particle structure, thickness distribution, and compaction state exhibit dynamic evolution characteristics. Existing detection methods are mostly based on a single time point for quality judgment, lacking the ability to continuously monitor and dynamically analyze the evolution process of the layer structure. This makes it difficult to identify areas that gradually evolve from a stable state to an abnormal state in a timely manner, and the potential waterproof performance defects have a delayed exposure characteristic.

[0004] Furthermore, traditional waterproof sand quality evaluation typically relies on thickness deviation or compaction indexes, using preset thresholds to determine compliance. However, changes in the structure of the paving layer usually manifest as a coupled evolution process involving multiple parameters such as thickness, surface morphology, and particle density. Single parameter thresholds or empirical rules are insufficient to cover complex structural changes, leading to misjudgments or omissions, and making it difficult to effectively distinguish between structural adjustment stages and potential instability stages. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a method and system for determining the state of waterproof sand paving.

[0006] This invention adopts the following technical solution: a method for determining the state of waterproof sand paving, comprising:

[0007] After the waterproof sand is laid, the laying area is divided into a unified spatial grid, and the area is divided into multiple spatial detection units. The waterproof sand status data of each spatial detection unit is collected.

[0008] At a preset time interval, waterproof sand state data is collected for each detection unit at different time points after laying, and a state evolution trajectory is constructed. Based on the state evolution trajectory, the change rate of each state parameter in the waterproof sand state data is calculated to form a time-series feature vector describing the dynamics of structural evolution.

[0009] The DBSCAN clustering algorithm is used to dynamically segment the time-series feature vectors at different time points, dividing the evolution process of the laying layer into multiple stages, including the stable stage, the structural adjustment stage, and the potential instability stage.

[0010] Feature parameters are extracted from the water and sand state data within each evolution cluster. Combined with preset stage determination rules, the evolution clusters are mapped to stable stages, structural adjustment stages, and potential instability stages. A temporal correspondence between stage labels and spatial detection units is established.

[0011] After obtaining the temporal correspondence and state data between stage labels and spatial detection units, an evolutionary cluster association graph is constructed.

[0012] Based on the evolution cluster association graph, a graph propagation algorithm is used to simulate the propagation path of potential anomalies. When any spatial detection unit enters the potential instability stage, it is marked as the main spatial detection unit. The neighboring units that may be affected are calculated according to the edge weights between nodes, and the spread range of potential instability is predicted, so as to carry out on-site construction intervention.

[0013] As a further description of the above technical solution: the waterproof sand condition data includes the thickness of the paving layer, surface roughness, and particle density;

[0014] The thickness of the paving layer was obtained using a laser rangefinder, the surface roughness was obtained using structured light scanning, and the particle density was obtained by inversion using a lightweight penetration compaction testing device.

[0015] As a further description of the above technical solution: the method of dividing the evolution process of the laying layer into multiple stages includes:

[0016] The time-series feature vector is input into the DBSCAN clustering algorithm. By setting the neighborhood radius and the minimum number of samples, density clustering is performed on the multidimensional state evolution trajectory to obtain multiple evolution clusters. Each evolution cluster represents a stage in the evolution process of the laying layer.

[0017] As a further description of the above technical solution: the method for mapping evolutionary clusters into stable stages, structural adjustment stages, and potential instability stages, based on preset stage determination rules, includes:

[0018] For each parameter within each evolutionary cluster, calculate the mean of that parameter at all time points within that evolutionary cluster;

[0019] For each parameter within each evolutionary cluster, calculate its variance and rate of change;

[0020] The mean, variance, and rate of change of each parameter within each evolutionary cluster are summarized to form the statistical feature vector of the cluster;

[0021] The obtained statistical feature vectors are input into the pre-constructed evolutionary stage prediction model, and the stage labels corresponding to the evolutionary clusters are output.

[0022] As a further description of the above technical solution: the method for establishing the temporal correspondence between stage labels and spatial detection units includes:

[0023] Based on the number of evolutionary clusters acquired within each spatial detection unit, and combined with the acquisition time interval covered by each evolutionary cluster and its corresponding stage label, the time range of the evolutionary cluster is associated with the stage label, thereby realizing the stage labeling of each spatial detection unit on the entire monitoring time series.

[0024] As a further description of the above technical solution: the training method for the evolutionary stage prediction model includes:

[0025] F sets of training data are collected in advance, where F is a positive integer greater than 0. The training data includes statistical feature vectors and corresponding stage labels. The stage labels include 0, 1, and 2, which correspond to the stable stage, structural adjustment stage, and potential instability stage, respectively.

[0026] Gradient boosting tree classifier was selected as the prediction model for the evolutionary stage, and initial hyperparameters were set.

[0027] During the model training phase, a multi-class cross-entropy loss function is used, with the statistical feature vector of the cluster as the model input feature and the encoding of the corresponding stage label as the prediction target. Model training is carried out based on the training data. The construction of each new tree aims to fit the classification residual of the training set loss function. The optimal splitting feature is selected through the Gini impurity criterion to divide the samples into different child nodes until the preset stopping condition is met.

[0028] The gradient descent method is used to optimize the weight of the leaf nodes of each new tree. The contribution weight of the new tree to the final stage judgment result is adjusted by the learning rate. In the hyperparameter optimization stage, the Bayesian optimization method is used to search for the optimal hyperparameter combination within the preset optimization range. The optimization objective is to maximize the F1 score on the validation set.

[0029] An early stopping mechanism is introduced during training. The collected training data is divided into training set, validation set and test set in a ratio of 7:2:1. Every 20 trees are trained, the macro average F1 score of the validation set is calculated. When the macro average F1 score of the validation set improves by less than 0.01 in 3 consecutive iterations, the model training is stopped. After training, the model parameters with the highest macro average F1 score of the validation set are saved.

[0030] As a further description of the above technical solution: the method for constructing the evolutionary cluster association graph includes:

[0031] Each spatial detection unit is used as a node in the evolution cluster association graph. Each node includes its current evolution stage, layer thickness, surface roughness, particle density, and evolution stage trajectory information.

[0032] The evolutionary stage trajectory information refers to the evolutionary stages experienced by the space detection unit.

[0033] The edge weights between nodes are defined, and the edge weights consist of spatial adjacency coefficients and evolution pattern similarity. The spatial adjacency coefficients represent the adjacent spatial positions of spatial detection units within the laying area, and the evolution pattern similarity represents the similarity of the changes in state parameters of spatial detection units during the evolution process. The degree of matching of evolution trends between nodes is calculated based on Euclidean distance.

[0034] The spatial adjacency coefficient and the evolutionary pattern similarity are weighted and combined to form network edges, where nodes represent spatial detection units and network edges represent spatial and evolutionary associations between nodes.

[0035] As a further description of the above technical solution: In the spatial adjacency coefficient, the weight of adjacent unit edges is set to 1, and the weight of non-adjacent unit edges is set to 0. In the evolution pattern similarity, the similarity of evolution trajectories is calculated using Euclidean distance, and then the spatial adjacency coefficient and the evolution similarity are weighted and summed to form network edges.

[0036] As a further description of the above technical solution: methods for predicting the spread range of potential instability include:

[0037] Obtain the node risk coefficient of the main space detection unit, which is obtained by weighted summation based on the rate of change of the waterproof sand state data;

[0038] Multiply the obtained node risk coefficient by the network edge to obtain the influence coefficient of the spatial detection unit through the network edge and the main spatial detection unit. Set an influence coefficient threshold. When the influence coefficient is greater than or equal to the influence coefficient threshold, mark the spatial detection unit as a high-risk unit.

[0039] The high-risk units and potential instability stages are then mapped onto the corresponding spatial detection units to guide intervention decisions at the construction site.

[0040] A waterproof sand laying status determination system is provided for implementing the aforementioned waterproof sand laying status determination method. The system includes:

[0041] The grid division module divides the paving area into a unified spatial grid after the waterproof sand is laid, dividing the area into multiple spatial detection units and collecting waterproof sand status data for each spatial detection unit.

[0042] The trajectory construction module collects waterproof sand state data at different time points after each spatial detection unit is laid, based on a preset time interval, and constructs a state evolution trajectory. Based on the state evolution trajectory, the change rate of each state parameter in the waterproof sand state data is calculated to form a temporal feature vector describing the dynamics of structural evolution.

[0043] The dynamic segmentation module uses the DBSCAN clustering algorithm to dynamically segment the time-series feature vectors at different time points, dividing the evolution process of the laying layer into multiple stages, including the stable stage, the structural adjustment stage, and the potential instability stage.

[0044] The stage mapping module extracts feature parameters from the water and sand state data within each evolution cluster and, in conjunction with preset stage determination rules, maps the evolution clusters into stable stages, structural adjustment stages, and potential instability stages, establishing a temporal correspondence between stage labels and spatial detection units.

[0045] The association mapping module constructs an evolutionary cluster association graph after obtaining the temporal correspondence and state data between stage labels and spatial detection units.

[0046] The propagation prediction module, based on the evolution cluster association graph, uses a graph propagation algorithm to simulate the propagation path of potential anomalies. When any spatial detection unit enters the potential instability stage, it is marked as the main spatial detection unit. The module calculates the neighboring units that may be affected based on the edge weights between nodes, predicts the spread range of potential instability, and conducts on-site construction intervention.

[0047] The beneficial effects of this invention are as follows:

[0048] This invention continuously collects waterproof and sandy state data from each spatial detection unit and constructs a state evolution trajectory, enabling dynamic monitoring of the change process of the laying layer from stable to potentially unstable, and early identification of possible structural anomalies. The DBSCAN clustering algorithm is used to dynamically segment the time-series feature vector, accurately dividing the laying layer evolution process into a stable stage, a structural adjustment stage, and a potentially unstable stage. The stage determination is made by combining multi-parameter statistical features, ensuring the accuracy and robustness of anomaly identification.

[0049] Furthermore, by constructing an evolutionary cluster association graph and integrating spatial adjacency coefficients with evolutionary pattern similarity, this invention provides a quantitative model for abnormal propagation paths. It can simulate the impact of potential unstable units on surrounding areas, predict the spread of instability in advance, and realize the transformation from single-point early warning to regional risk prediction. Through the calculation of node risk coefficients and influence coefficients, this invention can accurately identify high-risk units, provide targeted intervention basis for construction sites, reduce rework and maintenance costs, and improve the long-term stability of waterproof sand paving layers. Attached Figure Description

[0050] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0051] Figure 1 This is a flowchart of a method for determining the state of waterproof sand laying according to Embodiment 1 of the present invention;

[0052] Figure 2 This is a flowchart of a method for mapping evolutionary clusters into stable stages, structural adjustment stages, and potential instability stages, as provided in Embodiment 1 of the present invention.

[0053] Figure 3 This is a flowchart of the method for constructing an evolutionary cluster association graph provided in Embodiment 1 of the present invention;

[0054] Figure 4 This is a module connection diagram of a waterproof sand laying status determination system provided in Embodiment 2 of the present invention. Detailed Implementation

[0055] To make the technical means, creative features, objectives, and effects of this invention readily understandable, the invention is further described below with reference to specific illustrations. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0056] Example 1

[0057] Please see Figures 1-3 This invention provides a technical solution: a method for determining the state of waterproof sand paving, comprising:

[0058] After the waterproof sand is laid, the laying area is divided into a unified spatial grid, and the area is divided into multiple spatial detection units. The status data of the waterproof sand in each spatial detection unit is collected.

[0059] It should be noted that the size of each detection unit can be manually set according to the uniformity requirements of the laying layer, the construction accuracy, and the resolution requirements of subsequent status judgment. Optionally, the grid size is 50cm-50cm.

[0060] The waterproof sand condition data includes the thickness of the paving layer, surface roughness, and particle density.

[0061] The thickness of the paving layer was obtained using a laser rangefinder, the surface roughness was obtained using structured light scanning, and the particle density was obtained by inversion using a lightweight penetration compaction testing device.

[0062] In this embodiment, the particle density of the waterproof sand paving layer is measured using a lightweight penetration compaction testing device and obtained by inversion calculation. The specific steps are as follows:

[0063] The lightweight penetration compaction testing device includes a penetration rod, a load sensor, and a depth measurement unit. The penetration rod has a diameter of 10 mm, the load sensor is used to measure the force applied during penetration, and the depth measurement unit is used to record the penetration depth.

[0064] At least three penetration measurements are performed at the center of each spatial detection unit, and the average value is taken. The penetration depth of each spatial detection unit is fixed at a preset value D, which is equal to the average value of the design thickness of the paving layer. When the penetrating rod is pressed vertically into the waterproof sand layer, the relationship curve between the penetration force F(d) and the penetration depth d is recorded in real time. Then, based on the pressure force-depth curve, combined with the particle friction characteristics of the waterproof sand and the empirical compaction model, the particle density is obtained by inversion.

[0065] An example of the inversion formula is as follows: In the formula, The particle density (g / cm³) is obtained from the inversion. The initial loose density (determined experimentally, approximately 1.5 g / cm³) is the initial density. The penetration force is N. Let be the cross-sectional area of ​​the penetration rod (m²), and k be an empirical coefficient, optionally k∈(0.08–0.12), obtained based on the properties of the waterproof sand particles and experimental calibration. The reference penetration stress, used for dimensional normalization, is preferably determined through calibration tests. Penetration tests are performed on standard specimens or reference paving layers with known density, and the mean or median penetration stress of the stable section is taken as the reference stress. .

[0066] In this embodiment, by dividing the laying area into multiple spatial detection units and using a unified gridded approach for status monitoring, the entire laying layer can be fully covered. This overcomes the limitations of traditional point-discrete and single-point detection methods, and realizes spatially distributed real-time monitoring of the quality status of the laying layer, effectively avoiding local omissions and false compliance.

[0067] At preset time intervals, waterproof sand state data are collected for each spatial detection unit at different time points after laying. By continuously collecting the waterproof sand state data of the laid layer, a state evolution trajectory is constructed. Based on the state evolution trajectory, the change rate of each state parameter in the waterproof sand state data is calculated to form a time-series feature vector describing the dynamics of structural evolution.

[0068] Here is an example of how to obtain the rate of change of the lay layer thickness:

[0069] The expression for the rate of change of the thickness of the laying layer is: In the formula, The rate of change of the layer thickness. For time intervals, For a moment The thickness of the underlying layer.

[0070] In this embodiment, by continuously collecting waterproof sand state data at different time points for each detection unit and constructing a state evolution trajectory, the evolution process of the paving layer can be dynamically captured, accurately describing the changing trend of the structure from a stable state to a potentially unstable state. This dynamic monitoring capability allows construction quality inspection to go beyond a single point in time, enabling in-depth analysis of the continuous changes in the paving layer and providing forward-looking data for subsequent maintenance.

[0071] The DBSCAN clustering algorithm is used to dynamically segment the time-series feature vectors at different time points, dividing the evolution process of the laying layer into multiple stages, including the stable stage, the structural adjustment stage, and the potential instability stage.

[0072] Methods that divide the evolution of the layup layer into multiple stages include:

[0073] The time-series feature vector is input into the DBSCAN clustering algorithm. By setting the neighborhood radius and the minimum number of samples, density clustering is performed on the multidimensional state evolution trajectory to obtain multiple evolution clusters. Each evolution cluster represents a stage in the evolution process of the laying layer.

[0074] It should be noted that the neighborhood radius is adaptively determined based on the K-distance curve, and the minimum number of samples is proportionally mapped according to the number of detection units and the trajectory length, thereby ensuring that the clustering results are sensitive to the evolution of local anomalies; optionally, the neighborhood radius is 0.18 and the minimum number of samples is 9.

[0075] Feature parameters are extracted from the water-proof sand state data within each evolution cluster. Combined with preset stage determination rules, the evolution clusters are mapped to stable stages, structural adjustment stages, and potential instability stages. A temporal correspondence between stage labels and spatial detection units is established.

[0076] Methods for mapping evolutionary clusters into stable stages, structural adjustment stages, and potential instability stages, based on pre-defined stage determination rules, include:

[0077] For each parameter within each evolutionary cluster, calculate the mean of that parameter at all time points within that cluster.

[0078] For each parameter within each evolutionary cluster, calculate its variance and rate of change.

[0079] The mean, variance, and rate of change of each parameter within each evolutionary cluster are summarized to form the statistical characteristic vector of the cluster.

[0080] The obtained statistical feature vectors are input into a pre-built evolutionary stage prediction model, and the stage labels corresponding to the evolutionary clusters are output.

[0081] Methods for establishing the temporal correspondence between stage labels and spatial detection units include:

[0082] Based on the number of evolutionary clusters acquired within each spatial detection unit, and combining the acquisition time interval covered by each evolutionary cluster with its corresponding stage label, the time range of the evolutionary cluster is associated with the stage label, thus achieving stage labeling for each spatial detection unit across the entire monitoring time series. This method establishes a one-to-one correspondence between spatial detection units and their evolutionary stages in the time dimension, providing an accurate spatiotemporal mapping basis for subsequent state analysis, risk prediction, and evolutionary trend identification.

[0083] In this embodiment, the DBSCAN clustering algorithm is used for dynamic segmentation to accurately divide the evolution stages of the laying layer, including the stabilization stage, the structural adjustment stage, and the potential instability stage. By comprehensively analyzing the parameter characteristics of each stage, key nodes of structural changes can be identified more accurately, providing clear early warnings for potential risks during construction.

[0084] The training methods for the evolution stage prediction model include:

[0085] F sets of training data are collected in advance, where F is a positive integer greater than 0. The training data includes statistical feature vectors and corresponding stage labels. The stage labels include 0, 1, and 2, which correspond to the stable stage, the structural adjustment stage, and the potential instability stage, respectively.

[0086] Gradient boosting tree classifier was selected as the prediction model for the evolution stage. Initial hyperparameters were set, including: number of decision trees 100-150, maximum depth of a single tree 4-6, minimum number of samples for node splitting 8-12, maximum number of features considered during splitting 3, splitting criterion Gini impurity, learning rate 0.05-0.1, and regularization coefficient (L2) 0.1-0.2.

[0087] During the model training phase, a multi-class cross-entropy loss function is used to measure the deviation between the model's predicted probability and the true stage label of the evolutionary cluster. The statistical feature vector of the cluster is used as the model input feature, and the encoding of the corresponding stage label is used as the prediction target. Model training is carried out based on the training data. The construction of each new tree is aimed at fitting the classification residual of the training set loss function. The optimal splitting feature is selected through the Gini impurity criterion, such as prioritizing the parameter change rate as the splitting feature, and the samples are divided into different child nodes until the preset stopping conditions are met, such as reaching the preset maximum depth, the number of child node samples being less than the minimum number of samples, or the splitting gain being lower than the threshold.

[0088] The gradient descent method is used to optimize the weights of the leaf nodes of each new tree. The contribution weight of the new tree to the final judgment result is adjusted by the learning rate, so as to avoid a single tree dominating the prediction process and ensure the stability of the model judgment.

[0089] In the hyperparameter optimization stage, the Bayesian optimization method is adopted to search for the optimal combination of hyperparameters within the preset optimization range. The optimization objective is to maximize the F1 score (macro average) on the validation set. The specific hyperparameter optimization range is as follows: number of decision trees 80-180, maximum depth of a single tree 3-7, learning rate 0.03-0.12, and regularization coefficient (L2) 0.05-0.25.

[0090] An early stopping mechanism is introduced during training. The collected training data is divided into training, validation, and test sets in a 7:2:1 ratio. Every 20 trees are trained, and the macro-average F1 score on the validation set is calculated. Training is stopped when the macro-average F1 score on the validation set improves by less than 0.01 over three consecutive iterations to prevent overfitting. After training, the model parameters with the highest macro-average F1 score on the validation set are saved to ensure the model has stable stage-determining capabilities for evolutionary clusters with different statistical characteristics.

[0091] The trained model is validated using a test set. Precision, recall, and macro average F1 score are calculated for each stage. If the precision of each stage on the test set is ≥90% and the macro average F1 score is ≥89%, the model performance evaluation is satisfactory and it can be deployed in practice for predicting stage labels of evolutionary clusters and establishing correspondence with spatial detection units.

[0092] After obtaining the temporal correspondence and state data between stage labels and spatial detection units, an evolutionary cluster association graph is constructed.

[0093] The method for constructing the evolutionary cluster association graph includes:

[0094] Each spatial detection unit is used as a node in the evolution cluster association graph. Each node includes its current evolution stage, layer thickness, surface roughness, particle density, and evolution stage trajectory information.

[0095] The evolutionary stage trajectory information refers to the evolutionary stages experienced by the space detection unit.

[0096] The edge weights between nodes are defined, consisting of spatial adjacency coefficients and evolution pattern similarity. The spatial adjacency coefficients represent the adjacent spatial positions of the detection units within the laying area, and the evolution pattern similarity represents the similarity of the changes in state parameters of the detection units during the evolution process. The degree of matching of the evolution trends between nodes is calculated based on Euclidean distance or cosine similarity.

[0097] The spatial adjacency coefficient and the similarity of evolution patterns are weighted and combined to form network edges, where nodes represent each detection unit and network edges represent the spatial and evolutionary relationships between nodes, reflecting the possibility of propagation of potential instability information.

[0098] Specifically, in the spatial adjacency coefficient, the weight of adjacent unit edges is set to 1, and the weight of non-adjacent unit edges is set to 0. In the evolution pattern similarity, the similarity of evolution trajectories is calculated using Euclidean distance. Then, the spatial adjacency coefficient and the evolution similarity are weighted and summed to form network edges.

[0099] Optionally, network edge = W1 × spatial adjacency coefficient + W2 × evolution trajectory similarity; W1 and W2 are weight coefficients, optionally W1 = 0.6, W2 = 0.4.

[0100] Based on the evolution cluster association graph, a graph propagation algorithm is used to simulate the propagation path of potential anomalies. When any spatial detection unit enters the potential instability stage, it is marked as the main spatial detection unit. The neighboring units that may be affected are calculated according to the edge weights between nodes, and the spread range of potential instability is predicted, so as to carry out on-site construction intervention.

[0101] It should be noted that the graph propagation prediction adopts an iterative propagation method: the main space detection unit is used as the initial source node, and the source node carries the initial risk amount; in each round of propagation, the risk amount is distributed to adjacent nodes according to the edge weight, and distance decay or round decay is applied to the propagation risk amount; when the cumulative risk amount of a node reaches the preset judgment condition, the node is added to the diffusion range set; when the propagation round reaches the maximum round or the newly added node is empty, the propagation stops, and the diffusion range set and the corresponding propagation path record are output.

[0102] Specifically, methods for predicting the spread of potential instability include:

[0103] The node risk coefficient of the main space detection unit is obtained. Optionally, the node risk coefficient is obtained by weighted summation of the change rate of the waterproof sand state data.

[0104] Example: Node risk coefficient = α × thickness change rate + β × density change rate + γ × roughness change rate; It should be noted that α, β and γ are weighting coefficients. The formulas mentioned above are all numerical calculations after removing dimensions. They are formulas that are closest to the real situation obtained by software simulation based on a large amount of data. The weighting coefficients in the formulas and the preset thresholds in the analysis process are set by those skilled in the art according to the actual situation or obtained by simulation based on a large amount of data.

[0105] The obtained node risk coefficient is multiplied by the network edge to obtain the influence coefficient of the spatial detection unit through that network edge and the main spatial detection unit. A preset influence coefficient threshold is set. When the influence coefficient is greater than or equal to the threshold, the spatial detection unit is marked as a high-risk unit; when the influence coefficient is less than the threshold, the spatial detection unit is not marked. Specifically, the influence coefficient threshold can be obtained statistically based on historical detection data or simulation results. For example, the mean of the influence coefficient plus a certain number of standard deviations can be taken as the threshold to ensure that the marked spatial detection units are indeed high-risk units.

[0106] The high-risk units and potential instability stages are then mapped onto the corresponding spatial detection units to guide intervention decisions at the construction site.

[0107] In this embodiment, by constructing an evolutionary cluster association graph and employing a graph propagation algorithm to simulate the propagation path of potential anomalies, the propagation range of anomaly regions can be predicted. When a spatial detection unit enters a potential instability stage, this invention can calculate the neighboring units it may affect and conduct timely risk assessments and construction interventions to prevent the instability phenomenon from spreading during construction, thus ensuring the long-term stability of construction quality and waterproofing function.

[0108] In this embodiment, by dividing the paving area into spatial grids and collecting multi-parameter data, the entire paving layer is fully covered, overcoming the shortcomings of traditional point sampling inspections that cannot reflect the overall quality status. By continuously collecting the waterproof and sandy state data of each spatial detection unit and constructing the state evolution trajectory, this invention can dynamically monitor the change process of the paving layer from stable to potentially unstable, identify possible structural anomalies in advance, and use the DBSCAN clustering algorithm to dynamically segment the time-series feature vector, accurately dividing the paving layer evolution process into a stable stage, a structural adjustment stage, and a potentially unstable stage. Combined with multi-parameter statistical features for stage determination, the accuracy of anomaly identification is ensured.

[0109] Furthermore, this invention constructs an evolutionary cluster association graph, integrating spatial adjacency coefficients and evolutionary pattern similarity to provide a quantitative model for anomaly propagation paths. This model can simulate the impact of potentially unstable units on surrounding areas, predict the spread of instability in advance, and achieve a shift from single-point early warning to regional risk prediction. By calculating node risk coefficients and influence coefficients, this invention can accurately identify high-risk units, providing targeted intervention basis for construction sites, reducing rework and maintenance costs, and improving the long-term stability of the waterproof sand paving layer. This invention enhances risk early warning and intervention capabilities while improving the accuracy of paving quality assessment, and has significant engineering application value.

[0110] Example 2

[0111] Please see Figure 4This invention provides a technical solution: a waterproof sand laying status determination system, which is used to implement the aforementioned waterproof sand laying status determination method. The system includes:

[0112] The grid division module divides the paving area into a unified spatial grid after the waterproof sand is laid, dividing the area into multiple spatial detection units and collecting waterproof sand status data for each spatial detection unit.

[0113] The trajectory construction module collects waterproof sand state data at different time points after each spatial detection unit is laid, based on a preset time interval, and constructs a state evolution trajectory. Based on the state evolution trajectory, the change rate of each state parameter in the waterproof sand state data is calculated to form a temporal feature vector describing the dynamics of structural evolution.

[0114] The dynamic segmentation module uses the DBSCAN clustering algorithm to dynamically segment the time-series feature vectors at different time points, dividing the evolution process of the laying layer into multiple stages, including the stable stage, the structural adjustment stage, and the potential instability stage.

[0115] The stage mapping module extracts feature parameters from the water and sand state data within each evolution cluster and, in conjunction with preset stage determination rules, maps the evolution clusters into stable stages, structural adjustment stages, and potential instability stages, establishing a temporal correspondence between stage labels and spatial detection units.

[0116] The association mapping module constructs an evolutionary cluster association graph after obtaining the temporal correspondence and state data between stage labels and spatial detection units.

[0117] The propagation prediction module, based on the evolution cluster association graph, uses a graph propagation algorithm to simulate the propagation path of potential anomalies. When any spatial detection unit enters the potential instability stage, it is marked as the main spatial detection unit. The module calculates the neighboring units that may be affected based on the edge weights between nodes, predicts the spread range of potential instability, and conducts on-site construction intervention.

[0118] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for determining the laying status of waterproof sand, characterized in that, include: After the waterproof sand is laid, the laying area is divided into a unified spatial grid, and the area is divided into multiple spatial detection units. The waterproof sand status data of each spatial detection unit is collected. At a preset time interval, waterproof sand state data is collected for each spatial detection unit at different time points after laying, and a state evolution trajectory is constructed. Based on the state evolution trajectory, the change rate of each state parameter in the waterproof sand state data is calculated to form a time-series feature vector describing the dynamics of structural evolution. The DBSCAN clustering algorithm is used to dynamically segment the time-series feature vectors at different time points, dividing the evolution process of the laying layer into multiple stages, including the stable stage, the structural adjustment stage, and the potential instability stage. Feature parameters are extracted from the water and sand state data within each evolution cluster. Combined with preset stage determination rules, the evolution clusters are mapped to stable stages, structural adjustment stages, and potential instability stages. A temporal correspondence between stage labels and spatial detection units is established. After obtaining the temporal correspondence and state data between stage labels and spatial detection units, an evolutionary cluster association graph is constructed. Based on the evolution cluster association graph, a graph propagation algorithm is used to simulate the propagation path of potential anomalies. When any spatial detection unit enters the potential instability stage, it is marked as the main spatial detection unit. The neighboring units that may be affected are calculated according to the edge weights between nodes, and the spread range of potential instability is predicted, so as to carry out on-site construction intervention.

2. The method for determining the state of waterproof sand laying according to claim 1, characterized in that, The waterproof sand condition data includes the thickness of the paving layer, surface roughness, and particle density; The thickness of the paving layer was obtained using a laser rangefinder, the surface roughness was obtained using structured light scanning, and the particle density was obtained by inversion using a lightweight penetration compaction testing device.

3. The method for determining the state of waterproof sand laying according to claim 2, characterized in that, The method for dividing the evolution process of the layup layer into multiple stages includes: The time-series feature vector is input into the DBSCAN clustering algorithm. By setting the neighborhood radius and the minimum number of samples, density clustering is performed on the multidimensional state evolution trajectory to obtain multiple evolution clusters. Each evolution cluster represents a stage in the evolution process of the laying layer.

4. The method for determining the state of waterproof sand laying according to claim 3, characterized in that, The method for mapping evolutionary clusters into stable stages, structural adjustment stages, and potential instability stages by combining preset stage determination rules includes: For each parameter within each evolutionary cluster, calculate the mean of that parameter at all time points within that evolutionary cluster; For each parameter within each evolutionary cluster, calculate its variance and rate of change; The mean, variance, and rate of change of each parameter within each evolutionary cluster are summarized to form the statistical feature vector of the cluster; The obtained statistical feature vectors are input into the pre-constructed evolutionary stage prediction model, and the stage labels corresponding to the evolutionary clusters are output.

5. The method for determining the state of waterproof sand laying according to claim 3, characterized in that, Methods for establishing the temporal correspondence between stage labels and spatial detection units include: Based on the number of evolutionary clusters acquired within each spatial detection unit, and combined with the acquisition time interval covered by each evolutionary cluster and its corresponding stage label, the time range of the evolutionary cluster is associated with the stage label, thereby realizing the stage labeling of each spatial detection unit on the entire monitoring time series.

6. The method for determining the state of waterproof sand laying according to claim 4, characterized in that, The training methods for the evolution stage prediction model include: F sets of training data are collected in advance, where F is a positive integer greater than 0. The training data includes statistical feature vectors and corresponding stage labels. The stage labels include 0, 1, and 2, which correspond to the stable stage, structural adjustment stage, and potential instability stage, respectively. Gradient boosting tree classifier was selected as the prediction model for the evolutionary stage, and initial hyperparameters were set. During the model training phase, a multi-class cross-entropy loss function is used, with the statistical feature vector of the cluster as the model input feature and the encoding of the corresponding stage label as the prediction target. Model training is carried out based on the training data. The construction of each new tree aims to fit the classification residual of the training set loss function. The optimal splitting feature is selected through the Gini impurity criterion to divide the samples into different child nodes until the preset stopping condition is met. The gradient descent method is used to optimize the weight of the leaf nodes of each new tree. The contribution weight of the new tree to the final stage judgment result is adjusted by the learning rate. In the hyperparameter optimization stage, the Bayesian optimization method is used to search for the optimal hyperparameter combination within the preset optimization range. The optimization objective is to maximize the F1 score on the validation set. An early stopping mechanism is introduced during training. The collected training data is divided into training set, validation set and test set in a ratio of 7:2:

1. Every 20 trees are trained, the macro average F1 score of the validation set is calculated. When the macro average F1 score of the validation set improves by less than 0.01 in 3 consecutive iterations, the model training is stopped. After training, the model parameters with the highest macro average F1 score of the validation set are saved.

7. The method for determining the state of waterproof sand laying according to claim 1, characterized in that, The method for constructing the evolutionary cluster association graph includes: Each spatial detection unit is used as a node in the evolution cluster association graph. Each node includes its current evolution stage, layer thickness, surface roughness, particle density, and evolution stage trajectory information. The evolutionary stage trajectory information refers to the evolutionary stages experienced by the space detection unit. The edge weights between nodes are defined, and the edge weights consist of spatial adjacency coefficients and evolution pattern similarity. The spatial adjacency coefficients represent the adjacent spatial positions of spatial detection units within the laying area, and the evolution pattern similarity represents the similarity of the changes in state parameters of spatial detection units during the evolution process. The degree of matching of evolution trends between nodes is calculated based on Euclidean distance. The spatial adjacency coefficient and the evolutionary pattern similarity are weighted and combined to form network edges, where nodes represent spatial detection units and network edges represent spatial and evolutionary associations between nodes.

8. The method for determining the state of waterproof sand laying according to claim 7, characterized in that, In the spatial adjacency coefficient, the weight of adjacent unit edges is set to 1, and the weight of non-adjacent unit edges is set to 0. In the evolution pattern similarity, the similarity of evolution trajectories is calculated using Euclidean distance. Then, the spatial adjacency coefficient and the evolution similarity are weighted and summed to form network edges.

9. The method for determining the state of waterproof sand laying according to claim 1, characterized in that, The method for predicting the spread of potential instability includes: Obtain the node risk coefficient of the main space detection unit, which is obtained by weighted summation based on the rate of change of the waterproof sand state data; Multiply the obtained node risk coefficient by the network edge to obtain the influence coefficient of the spatial detection unit connected to the main spatial detection unit through the network edge. Set an influence coefficient threshold. When the influence coefficient is greater than or equal to the influence coefficient threshold, mark the spatial detection unit as a high-risk unit. The high-risk units and potential instability stages are then mapped onto the corresponding spatial detection units to guide intervention decisions at the construction site.

10. A system for determining the state of waterproof sand paving, used to implement the method for determining the state of waterproof sand paving according to any one of claims 1-9, characterized in that, The system includes: The grid division module divides the paving area into a unified spatial grid after the waterproof sand is laid, dividing the area into multiple spatial detection units and collecting waterproof sand status data for each spatial detection unit. The trajectory construction module collects waterproof sand state data at different time points after each spatial detection unit is laid, based on a preset time interval, and constructs a state evolution trajectory. Based on the state evolution trajectory, the change rate of each state parameter in the waterproof sand state data is calculated to form a temporal feature vector describing the dynamics of structural evolution. The dynamic segmentation module uses the DBSCAN clustering algorithm to dynamically segment the time-series feature vectors at different time points, dividing the evolution process of the laying layer into multiple stages, including the stable stage, the structural adjustment stage, and the potential instability stage. The stage mapping module extracts feature parameters from the water and sand state data within each evolution cluster and, in conjunction with preset stage determination rules, maps the evolution clusters into stable stages, structural adjustment stages, and potential instability stages, establishing a temporal correspondence between stage labels and spatial detection units. The association mapping module constructs an evolutionary cluster association graph after obtaining the temporal correspondence and state data between the stage labels and spatial detection units; The propagation prediction module, based on the evolution cluster association graph, uses a graph propagation algorithm to simulate the propagation path of potential anomalies. When any spatial detection unit enters the potential instability stage, it is marked as the main spatial detection unit. The module calculates the neighboring units that may be affected based on the edge weights between nodes, predicts the spread range of potential instability, and conducts on-site construction intervention.

Images