Rolling compaction rate dynamic evaluation method based on Beidou and machine learning

Abstract

The invention provides a rolling compaction rate dynamic evaluation method based on Beidou and machine learning, and overcomes the defects of'replacing surfaces with points' and poor adaptability of a fixed formula method in traditional test pit sampling. The compaction rate evaluation model based on machine learning is established, and the whole warehouse surface real-time acquisition of the rolling parameters is realized by using the Beidou system, so that the continuous, lossless and comprehensive evaluation of the compaction quality of the construction warehouse surface is realized. In addition, the method also introduces a model dynamic correction mechanism based on field measured data. By acquiring the real compaction rate of the verification point in the construction process and comparing the real compaction rate with the model prediction value, the evaluation model can be continuously self-optimized and adjusted under the on-site complex working condition, so that the overall precision and reliability of the whole warehouse surface compaction rate prediction are improved. A more scientific and accurate intelligent acceptance means is provided for dam rolling construction quality, and a construction monitoring technology based on a domestic Beidou system is promoted to develop to a self-adaptive and highly-credible intelligent stage.

Description

Technical Field

[0001] This invention relates to the field of process control and management technology for the compaction construction quality of rockfill dams, and in particular to a dynamic evaluation method for compaction rate based on BeiDou and machine learning. Background Technology

[0002] As a critical water conservancy infrastructure, the quality of the compaction construction of rockfill dams directly affects the overall stability and long-term durability of the project. During this construction process, the compaction quality of the filling materials is a core control indicator for ensuring dam safety; therefore, strict and effective quality control of the compaction process is crucial.

[0003] Currently, the quality control system for roller compaction construction in rockfill dams generally follows the "dual control" principle: firstly, "process control" of construction parameters such as compaction speed, number of compaction passes, and excitation force; and secondly, "final parameter control" through test pit sampling to detect the dry density or compaction degree of the fill. With technological advancements, real-time monitoring of compaction machinery using positioning technologies such as GNSS and surveying robots has enabled relatively good "process control." However, the acceptance of the final compaction quality currently relies mainly on two technical approaches: one is the traditional, discrete test pit sampling method; and the other is the indirect calculation method based on GPS elevation data.

[0004] Both of the aforementioned existing technical approaches have significant drawbacks. First, the traditional test pit sampling method uses a point-to-surface approach, failing to reflect the uniformity of compaction quality across the entire compaction surface, and is inefficient and destructive. Second, while the indirect calculation method based on GPS elevation data achieves full surface coverage, it relies on a fixed theoretical formula, and its evaluation accuracy is limited by the absolute accuracy of the elevation measurement, making it susceptible to interference from complex on-site conditions. Most importantly, this fixed formula cannot adaptively characterize the complex nonlinear relationship between rolling process parameters and final compaction quality under different material sources and mechanical conditions, resulting in insufficient reliability and universality of the evaluation results. Therefore, how to overcome the limitations of existing technologies and achieve a method that can accurately and adaptively evaluate the compaction quality of the entire compaction surface has become an urgent technical challenge to be solved in this field. Summary of the Invention

[0005] In view of the above problems, a dynamic evaluation method for compaction rate based on BeiDou and machine learning is proposed to overcome or at least partially solve these problems, including: S1. In the experimental area, the filling material source is compacted using a rolling machine, and several sets of compaction rate data under different rolling speeds and rolling passes are collected. Using a machine learning regression algorithm, the rolling speed and rolling passes are used as input features, and the compaction rate is used as the output label to train an initial compaction rate evaluation model. S2, on the actual construction site, the real-time compaction speed and number of compaction passes at all points within the entire site are obtained using a BeiDou-based compaction construction quality system, and the real-time compaction parameters are input into the initial compaction rate evaluation model to obtain the initial evaluated compaction rate of the entire site; wherein, the compaction machinery and fill material source of the actual construction site are the same as in step S1; the BeiDou-based compaction construction quality system includes: a BeiDou compaction construction quality monitoring data acquisition unit, a BeiDou satellite signal receiving antenna, and a 5G communication antenna; S3, Select some verification points on the actual construction site and measure the verification compaction rate data of the points using the additional mass method; Compare the verification compaction rate data with the predicted value of the initial evaluation model at the points; When the error exceeds a preset threshold, automatically use the verification compaction rate data and its corresponding real-time rolling speed and number of rolling passes as new samples to incrementally learn the initial compaction rate evaluation model and generate a corrected compaction rate evaluation model; S4, based on the revised model, updates and evaluates the compaction rate of the entire warehouse surface.

[0006] Optionally, before compacting the fill material source using compaction machinery in the experimental area, the following steps are also included: installing and fixing the Beidou compaction construction quality monitoring data acquisition unit in the compaction machinery cab, fixing the Beidou satellite signal receiving antenna and the 5G communication antenna on the top of the compaction machinery cab, and calibrating the geometric position parameters of the Beidou satellite signal receiving antenna and the compaction machinery rollers.

[0007] Optionally, step S1 specifically includes: Based on the dam's design requirements, determine the source of filling material, the type of compaction machinery, and the control requirements for compaction speed, number of compaction passes, and compaction rate; Select an experimental area for compaction experiments. Based on the control requirements, collect several sets of compaction rate data covering different combinations of compaction speed and number of compaction passes using the test pit sampling additional mass method. Then, associate the compaction rate data corresponding to each sampling point with the compaction speed and number of compaction passes to form a training dataset. An initial compaction rate evaluation model was trained based on the training dataset.

[0008] Optionally, the step of training an initial compaction rate evaluation model based on the training dataset includes: dividing the training dataset into a training set and a test set several times, training the model using the training set, and quantifying the prediction accuracy of the model using the test set; and selecting the model configuration with the highest prediction accuracy as the initial compaction rate evaluation model through a cross-validation process.

[0009] Optionally, the machine learning regression algorithm is any one of linear regression, elastic network regression, support vector regression, random forest regression, and gradient boosting regression.

[0010] Optionally, the control requirements for the rolling speed, number of rolling passes, and compaction rate shall meet the following: In the formula, N is the number of compaction passes, v is the compaction speed, and Q is the compaction rate.

[0011] Optionally, the step of using a BeiDou-based compaction construction quality system to obtain real-time compaction speed and number of compaction passes at all points within the entire work area includes: Satellite signals are received using a Beidou satellite signal receiving antenna fixed on the compaction machine, and the original spatiotemporal data of the compaction machine is collected by the Beidou compaction construction quality monitoring data acquisition unit. Based on the pre-calibrated geometric position parameters of the Beidou satellite signal receiving antenna and the rolling mechanical roller, the original spatiotemporal data is corrected to the actual rolling position of the roller to obtain spatiotemporal monitoring data; The spatiotemporal monitoring data is transmitted in real time to a remote monitoring center server via the 5G communication antenna. In the monitoring center server, based on the spatiotemporal monitoring data, the real-time compaction speed and number of compaction passes at each point within the entire warehouse surface are calculated using a spatial data mining algorithm.

[0012] Optionally, the spatiotemporal monitoring data includes at least timestamps, planar coordinates, and elevation information.

[0013] This invention effectively overcomes the shortcomings of traditional test pit sampling methods that rely on "points instead of areas" and the poor adaptability of fixed formula methods. By establishing a compaction rate evaluation model based on machine learning and utilizing the BeiDou system to achieve real-time acquisition of compaction parameters across the entire compaction surface, continuous, non-destructive, and comprehensive evaluation of the compaction quality of the construction surface is realized. Furthermore, this method introduces a dynamic model correction mechanism based on field measurement data. By obtaining the actual compaction rate at verification points during construction and comparing and learning from the model's predicted values, the evaluation model can continuously self-optimize and adjust under complex on-site conditions, thereby significantly improving the overall accuracy and reliability of the overall compaction rate prediction. This not only provides a more scientific and accurate intelligent acceptance method for dam compaction construction quality but also promotes the development of construction monitoring technology based on the domestic BeiDou system towards an adaptive, highly reliable, and intelligent stage. Attached Figure Description

[0014] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1This is a flowchart of a dynamic evaluation method for compaction rate based on BeiDou and machine learning provided in an embodiment of the present invention; Figure 2 This is a flowchart of another dynamic evaluation method for compaction rate based on BeiDou and machine learning provided in an embodiment of the present invention; Figure 3 This is a structural diagram of the dam filling and compaction construction quality monitoring system based on BeiDou provided in an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0017] Reference Figure 1 , Figure 2 The flowchart of a dynamic evaluation method for compaction rate based on BeiDou and machine learning, according to an embodiment of the present invention, is shown. Specifically, it may include steps S1-S4: Before compacting the fill material source using compaction machinery in the experimental area, the Beidou compaction construction quality monitoring data acquisition unit can be installed and fixed in the compaction machinery cab, the Beidou satellite signal receiving antenna and the 5G communication antenna can be fixed on the top of the compaction machinery cab, and the geometric position parameters of the Beidou satellite signal receiving antenna and the compaction machinery rollers can be calibrated.

[0018] In the actual operation of this embodiment, the core data acquisition host, which integrates the Beidou positioning chip and data processing unit, can be securely installed in the cab of the compaction machine. The Beidou satellite signal receiving antenna and the 5G communication antenna are installed in an open and unobstructed area on the top of the cab to ensure stable reception of satellite and communication signals.

[0019] After physical installation is completed, geometric calibration is required to correct the spatial position information to the roller of the compaction machine: measure and record the three-dimensional spatial offset (ΔX, ΔY, ΔH) between the phase center of the Beidou antenna and the ground contact center of the vibrating roller of the compaction machine.

[0020] Specifically, in a static environment, the compaction machinery can be parked on a flat, hard surface. Using a high-precision total station, the three-dimensional coordinates of the BeiDou antenna phase center in the construction coordinate system are first measured. Then, the coordinates of the axes on both sides of the vibrating roller are measured, and the three-dimensional coordinates of the roller center are calculated. Based on the roller radius and the vertical direction relationship, the three-dimensional coordinates of the roller's ground contact center point are deduced. Finally, the coordinate differences between the antenna phase center and the roller's ground contact center in the X, Y, and H directions are calculated, which gives the required three-dimensional spatial offset.

[0021] This calibration parameter will be used to uniformly convert all spatial position coordinates received by the antenna to the actual coordinates of the actual compaction point (the roller), so as to ensure the accuracy of subsequent calculations of compaction trajectory, speed, number of passes and other parameters.

[0022] S1. In the experimental area, compaction machinery is used to compact the fill material source, and several sets of compaction rate data under different compaction speeds and compaction passes are collected. Using a machine learning regression algorithm, the compaction speed and compaction pass are used as input features, and the compaction rate is used as the output label to train an initial compaction rate evaluation model.

[0023] Specifically, step S1 can be achieved through the following steps: Based on the dam design requirements, the source of filling material, the type of compaction machinery, and the control requirements for compaction speed, number of compaction passes, and compaction rate were determined. Compaction experiments were conducted in an experimental area. Based on the control requirements, several sets of compaction rate data covering different combinations of compaction speed and number of compaction passes were collected using the test pit sampling with added mass method. The compaction rate data corresponding to each sampling point was correlated with the compaction speed and number of compaction passes to form a training dataset. An initial compaction rate evaluation model was trained based on this training dataset.

[0024] In practice, based on the design documents of the target dam project, the specific fill material source (such as the gradation of rockfill from a certain quarry) and the specific type of compaction machinery to be used in this experiment can be determined first. Under these constraints, several combinations of compaction speed (e.g., 1.0 km / h, 2.0 km / h, 3.0 km / h) and number of compaction passes (e.g., 4 passes, 6 passes, 8 passes, 10 passes) can be initially set according to the construction specifications.

[0025] In embodiments of the present invention, the control requirements for rolling speed, number of rolling passes, and compaction rate can be satisfied as follows: In the formula, N is the number of compaction passes, v is the compaction speed, and Q is the compaction rate.

[0026] Furthermore, compaction can be carried out on a flat, representative experimental surface according to the above combination scheme. The system automatically and continuously records the spatiotemporal trajectory data of the compaction machinery through calibrated Beidou equipment. For each area completed with each combination of compaction parameters, the "additional mass method" is used to test the on-site compaction rate. Specifically, representative points can be selected in the area, test pits can be excavated, and the wet density and moisture content of the fill material can be measured using the additional mass method instrument, thereby calculating the dry density and compaction rate (i.e., the measured compaction rate) at that point. At the same time, by using the spatial coordinates of the test pit point, the Beidou monitoring data can be retrieved back to accurately obtain the median compaction speed and the cumulative number of compaction passes used when compacting that point. This set of data, "(compaction speed, number of compaction passes) → measured compaction rate", is recorded. By changing the parameter combination and repeating the above process, dozens or even hundreds of sets of effective sample data covering a sufficient parameter space are collected to form the original dataset for model training.

[0027] The step of training an initial compaction rate evaluation model based on the training dataset includes: dividing the training dataset into a training set and a test set several times, training the model using the training set, and quantifying the prediction accuracy of the model using the test set; and selecting the model configuration with the highest prediction accuracy as the initial compaction rate evaluation model through a cross-validation process.

[0028] Specifically, after normalizing and preprocessing the collected raw dataset, a supervised learning sample set is constructed with "compaction speed" and "number of compaction passes" as input features (X) and "measured compaction rate" as the output label (Y). Machine learning algorithms such as random forest regression or gradient boosting regression, which can effectively handle nonlinear relationships and have strong anti-overfitting capabilities, are selected for model training. The technical principle is that these algorithms, by integrating a large number of decision trees, automatically learn and summarize the implicit, complex functional relationship Y=f(X) between compaction parameters and compaction rate from the provided sample data. This function f is the "initial compaction rate evaluation model" obtained through training.

[0029] To further evaluate the model's generalization ability, cross-validation can be used, dividing the dataset multiple times into training and validation sets to ensure the model maintains stable prediction accuracy even on unseen data. Finally, a reliable initial evaluation model optimized for the specific material source and compaction machinery is obtained and deployed to the server of the compaction construction quality monitoring system.

[0030] The machine learning regression algorithm can be any one of linear regression, elastic network regression, support vector regression, random forest regression, or gradient boosting regression.

[0031] S2, on the actual construction site, the real-time compaction speed and number of compaction passes of all points within the entire site are obtained using the BeiDou-based compaction construction quality system, and the real-time compaction parameters are input into the initial compaction rate evaluation model to obtain the initial evaluated compaction rate of the entire site; wherein, the compaction machinery and fill material source of the actual construction site are the same as in step S1; the BeiDou-based compaction construction quality system includes: a BeiDou compaction construction quality monitoring data acquisition unit, a BeiDou satellite signal receiving antenna, and a 5G communication antenna.

[0032] The method of using a BeiDou-based compaction construction quality system to obtain real-time compaction speed and number of compaction passes at all points within the entire surface area includes: Satellite signals are received using a Beidou satellite signal receiving antenna fixed on the compaction machine, and the original spatiotemporal data of the compaction machine is collected by the Beidou compaction construction quality monitoring data acquisition unit. Based on the pre-calibrated geometric position parameters of the Beidou satellite signal receiving antenna and the rolling mechanical roller, the original spatiotemporal data is corrected to the actual rolling position of the roller to obtain spatiotemporal monitoring data; The spatiotemporal monitoring data is transmitted in real time to a remote monitoring center server via the 5G communication antenna. In the monitoring center server, based on the spatiotemporal monitoring data, the real-time compaction speed and number of compaction passes at each point within the entire warehouse surface are calculated using a spatial data mining algorithm.

[0033] In a preferred embodiment of the present invention, the spatiotemporal monitoring data includes at least a timestamp, planar coordinates, and elevation information. For example: {time, X-coordinate of the construction coordinate system, Y-coordinate of the construction coordinate system, H-coordinate of the construction coordinate system}.

[0034] In practical applications, before starting construction operations, the operator of the compaction machinery needs to activate the data acquisition unit in the cab. Once activated, the system enters a fully automatic monitoring state without manual intervention. The BeiDou antenna continuously receives satellite signals, and the data acquisition unit synchronously calculates and records the raw spatiotemporal data corresponding to each observation moment at a high frequency (e.g., 1Hz or higher). This raw data is then corrected in real-time using the aforementioned calibration parameters by the built-in software, resulting in "spatiotemporal monitoring data" that accurately reflects the center position of the roller. The standard format of this data is: {Time, X-coordinate of construction coordinate system, Y-coordinate of construction coordinate system, H-coordinate of construction coordinate system}. Here, "Time" represents the observation moment; "X and Y coordinates" precisely characterize the planar position of the roller in the horizontal construction coordinate system; and "H coordinate" represents its elevation. Subsequently, this serialized spatiotemporal monitoring data is transmitted in real-time and stably to the remote "monitoring center data server" via a 5G communication antenna, leveraging the high bandwidth and low latency characteristics of the 5G network, achieving instantaneous synchronization and persistent storage of on-site sensor data with the cloud server.

[0035] Construction management and quality control personnel can access the external web service page of the "Intelligent System for Quality Monitoring of Dam Filling and Compaction Construction" deployed on the server via various network terminals such as computers and mobile phones, for reference. Figure 3 At the monitoring center server, the massive spatiotemporal monitoring data streams are simultaneously processed in real-time by a "spatial data mining algorithm," the core purpose of which is to extract the compaction process parameters required for evaluation. The algorithm first analyzes continuous time-location sequences. For any small area covered by compaction within the compaction zone, the system accurately calculates the "real-time compaction pass count" at that point by statistically analyzing the number of times the compaction machinery trajectory point falls into that area over a period of time, combined with the identification of the compaction machinery's operating mode. Simultaneously, based on the planar distance and time interval between two adjacent spatiotemporal monitoring points on the same compaction trajectory, the algorithm calculates the instantaneous "real-time compaction speed" at that moment, and assigns these speed values ​​to the corresponding compaction zone location through spatial interpolation and other methods. The key parameters such as the total compaction pass count and compaction speed obtained from the above calculations are dynamically updated and displayed on the user-accessible web system interface, enabling managers to intuitively and in real-time view the overall progress and compliance of process parameters at the construction site, and remotely grasp first-hand construction conditions.

[0036] Through the above processing, the entire construction surface is digitized into a dynamic matrix composed of numerous grids. Each grid has a "compaction pass count" and "compaction speed" value bound to it, calculated in real time based on BeiDou measured data. Finally, the real-time compaction parameters (speed and pass count) of all points on the entire surface obtained above are used as input variables and batch-input into the "initial compaction rate evaluation model" trained in step S1. It is important to emphasize that the model of compaction machinery and the source of fill material used in the actual construction must be consistent with the conditions set in the model training stage (S1) to ensure the consistency between the model input features and the training feature distribution, and to guarantee the effectiveness of the prediction. The model will calculate for each pair of input parameters and output the corresponding compaction rate prediction value. Finally, the system generates an "initial evaluation compaction rate" distribution map covering the entire construction surface, thereby completing the first comprehensive evaluation of the compaction quality of the surface in near real-time and non-destructively during construction.

[0037] S3. Select some verification points on the actual construction site and measure the verification compaction rate data of the points using the additional mass method. Compare the verification compaction rate data with the predicted value of the initial evaluation model at the points. When the error exceeds a preset threshold, automatically use the verification compaction rate data and its corresponding real-time rolling speed and number of rolling passes as new samples to incrementally learn the initial compaction rate evaluation model and generate a corrected compaction rate evaluation model.

[0038] After the initial evaluation model completes the preliminary compaction rate calculation for the entire surface, a certain number of verification points need to be selected on the actual construction surface. The selection of these points should follow the principle of uniform spatial distribution while also considering key areas. For example, points can be evenly distributed according to the construction grid, with additional points added at the edges of the compaction track, overlaps, and other areas with weak quality. At each verification point, an additional mass method is used for on-site measurement: First, a test pit meeting the required dimensions is excavated at the selected point using excavation equipment, and all fill material is carefully collected from the pit; then, the wet density and moisture content of the fill material are immediately measured, and the dry density of that point is calculated; finally, this dry density value is compared with the maximum dry density value determined experimentally to calculate the measured compaction rate of that verification point. The compaction rate data obtained in this process serves as the "true ground value" for evaluating the accuracy of the model's predictions.

[0039] While acquiring the verification compaction rate data, the system automatically extracts the historical compaction data corresponding to the coordinates of the verification point from the BeiDou-based construction quality monitoring database. This includes the real-time compaction speed and cumulative number of compaction passes when the compaction machinery passes through this point. Subsequently, the system calls the initial evaluation model, inputs the compaction speed and number of passes for this point, and obtains the model's predicted compaction rate for that point. The system automatically calculates the absolute or relative error between the measured and predicted values ​​and compares it with a pre-set tolerance threshold. This threshold is a pre-set value, such as ±3%, based on engineering acceptance standards and model accuracy requirements. The core principle is that while the initial model, based on historical experimental data, has universality, it cannot completely cover all micro-variations in actual construction (such as slight fluctuations in material moisture content and the influence of on-site temperature). By introducing a direct comparison between on-site measured data and model predictions, local deviations of the model in the current specific construction environment can be quantitatively diagnosed.

[0040] In embodiments of the present invention, the threshold can be determined based on a comprehensive analysis of multi-source information: First, referring to the requirements for the compaction rate qualification standard in engineering design and construction specifications, the threshold is set to a reasonable error range that allows deviation from the standard; second, it can be statistically determined based on the average prediction error (such as root mean square error) exhibited by the initial evaluation model in historical verification data or cross-validation stages; in addition, the measurement accuracy of the on-site additional quality method testing means itself must also be considered. The threshold can be in the form of a relative threshold (such as set to ±2% to ±5% of the predicted value) to adapt to the evaluation of different compaction level areas; or it can be in the form of an absolute threshold (such as set to a difference of 1.5 percentage points in the absolute value of the compaction rate). The specific value of this threshold can be configured and adjusted during system deployment according to the characteristics of the material source, the importance level of the project, and the strictness of quality control.

[0041] When the system determines that the error at a certain point exceeds a preset threshold, it automatically triggers the model correction procedure. At this time, the complete data triplet for that verification point—"compaction speed, number of compaction passes, and measured compaction rate"—is added as a new high-quality sample to the existing model training sample library. Subsequently, the system starts an incremental learning algorithm to update the initial evaluation model. The principle of incremental learning differs from retraining; it does not discard the old model and all old data to start from scratch. Instead, it retains the knowledge structure already learned by the initial model (i.e., the basic mapping relationship between compaction parameters and compaction rate) and, with a small learning rate, mainly uses the newly added sample data to fine-tune and optimize the model's parameters. The aim is to enable the model to absorb information from the new environment, correct its local prediction biases, and avoid catastrophic forgetting. Commonly used incremental learning strategies include online sequence learning or sliding window-based sample update learning. Finally, the system generates a corrected compaction rate evaluation model. This model integrates the initial experimental patterns and the latest field verification information, and its predictive ability for the current construction site is specifically enhanced. The revised model will be immediately used to replace the initial model, serving subsequent compaction rate calculations and quality assessments, thus forming a closed-loop quality control process of "assessment-verification-correction-reassessment" to continuously improve the reliability and scientific nature of the overall warehouse quality evaluation.

[0042] In embodiments of the present invention, when a random forest regression model is used, the incremental learning can be achieved in the following ways: adding new samples to the training set, and while retaining most of the original decision tree structure, reconstructing or adjusting the weights of local subtrees that are highly correlated with the new samples; alternatively, a gradient boosting framework that supports online learning can be used to fine-tune the model by using new samples as input for new training rounds with an extremely low learning rate.

[0043] S4, based on the revised model, updates and evaluates the compaction rate of the entire warehouse surface.

[0044] Specifically, the system divides the actual construction surface into spatial grids and reads the rolling speed and number of rolling passes parameters corresponding to each grid, which are obtained in real time by the Beidou-based monitoring system, forming a complete input dataset. Then, it calls the corrected compaction rate evaluation model generated through incremental learning in step S3 to perform batch calculations on the dataset, outputting the predicted compaction rate value for each grid, thereby generating a new compaction rate digital matrix covering the entire surface. Next, the system compares the calculated compaction rate matrix with the compaction rate quality control standards specified in the dam design, automatically identifying and locating all substandard areas where the compaction rate is below the design threshold. After the analysis is complete, the system immediately highlights these substandard areas in a prominent color through the monitoring platform's graphical interface and generates a quality warning report containing information such as specific location, area, and compaction rate deviation. The report is promptly fed back to the handheld terminals of on-site construction management and supervision personnel or the large screen of the command center via the 5G network, guiding them to immediately carry out targeted compaction or process adjustments in substandard areas. This enables precise, dynamic, and closed-loop control of construction quality, ensuring that the overall compaction quality of the dam fill fully meets the design and safety requirements.

[0045] The above provides a detailed description of a dynamic evaluation method for compaction rate based on BeiDou and machine learning. Specific examples are used to illustrate the principle and implementation of the invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the invention. Therefore, the content of this specification should not be construed as a limitation of the invention.

Claims

1. A dynamic evaluation method for compaction rate based on BeiDou and machine learning, characterized in that, The method includes: S1. In the experimental area, the filling material source is compacted using a rolling machine, and several sets of compaction rate data under different rolling speeds and rolling passes are collected. Using a machine learning regression algorithm, the rolling speed and rolling passes are used as input features, and the compaction rate is used as the output label to train an initial compaction rate evaluation model. S2, on the actual construction site, the real-time compaction speed and number of compaction passes at all points within the entire site are obtained using a BeiDou-based compaction construction quality system, and the real-time compaction parameters are input into the initial compaction rate evaluation model to obtain the initial evaluated compaction rate of the entire site; wherein, the compaction machinery and fill material source of the actual construction site are the same as in step S1; the BeiDou-based compaction construction quality system includes: a BeiDou compaction construction quality monitoring data acquisition unit, a BeiDou satellite signal receiving antenna, and a 5G communication antenna; S3, Select some verification points on the actual construction site and measure the verification compaction rate data of the points using the additional mass method; Compare the verification compaction rate data with the predicted value of the initial evaluation model at the points; When the error exceeds a preset threshold, automatically use the verification compaction rate data and its corresponding real-time rolling speed and number of rolling passes as new samples to incrementally learn the initial compaction rate evaluation model and generate a corrected compaction rate evaluation model; S4, based on the revised model, updates and evaluates the compaction rate of the entire warehouse surface.

2. The method according to claim 1, characterized in that, Before compacting the fill material source using compaction machinery in the experimental area, the following steps are also included: installing and fixing the Beidou compaction construction quality monitoring data acquisition unit in the compaction machinery cab, fixing the Beidou satellite signal receiving antenna and the 5G communication antenna on the top of the compaction machinery cab, and calibrating the geometric position parameters of the Beidou satellite signal receiving antenna and the compaction machinery rollers.

3. The method according to claim 2, characterized in that, in, Step S1 specifically includes: Based on the dam's design requirements, determine the source of filling material, the type of compaction machinery, and the control requirements for compaction speed, number of compaction passes, and compaction rate; Select an experimental area for compaction experiments. Based on the control requirements, collect several sets of compaction rate data covering different combinations of compaction speed and number of compaction passes using the test pit sampling additional mass method. Then, associate the compaction rate data corresponding to each sampling point with the compaction speed and number of compaction passes to form a training dataset. An initial compaction rate evaluation model was trained based on the training dataset.

4. The method according to claim 3, characterized in that, The process of training an initial compaction rate evaluation model based on the training dataset includes: dividing the training dataset into a training set and a test set several times; training the model using the training set; and quantifying the prediction accuracy of the model using the test set; and selecting the model configuration with the highest prediction accuracy as the initial compaction rate evaluation model through a cross-validation process.

5. The method according to claim 4, characterized in that, The machine learning regression algorithm is any one of linear regression, elastic network regression, support vector regression, random forest regression, or gradient boosting regression.

6. The method according to claim 5, characterized in that, The control requirements for compaction speed, number of compaction passes, and compaction rate must meet the following: In the formula, N is the number of compaction passes, v is the compaction speed, and Q is the compaction rate.

7. The method according to claim 6, characterized in that, The method of using a BeiDou-based compaction construction quality system to obtain real-time compaction speed and number of compaction passes at all points within the entire work area includes: Satellite signals are received using a Beidou satellite signal receiving antenna fixed on the compaction machine, and the original spatiotemporal data of the compaction machine is collected by the Beidou compaction construction quality monitoring data acquisition unit. Based on the pre-calibrated geometric position parameters of the Beidou satellite signal receiving antenna and the rolling mechanical roller, the original spatiotemporal data is corrected to the actual rolling position of the roller to obtain spatiotemporal monitoring data; The spatiotemporal monitoring data is transmitted in real time to a remote monitoring center server via the 5G communication antenna. In the monitoring center server, based on the spatiotemporal monitoring data, the real-time compaction speed and number of compaction passes at each point within the entire warehouse surface are calculated using a spatial data mining algorithm.

8. The method according to claim 7, characterized in that, The spatiotemporal monitoring data includes at least timestamps, planar coordinates, and elevation information.

Images