Rolling density prediction method and equipment based on additional mass method and machine learning
Through the combined method of additional mass method and machine learning, the problem of insufficient detection accuracy of rock pile compaction density in the existing technology is solved, and efficient and lossless compaction quality detection of rock pile materials is achieved, which improves detection accuracy and efficiency.
Patent Information
- Application Number
- CN202510680541.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-26
- Publication Date
- 2025-08-26
AI Technical Summary
In the prior art, the rock pile compaction density detection method based on a linear regression model is difficult to consider the influence of multiple factors at the same time, resulting in insufficient detection accuracy, especially after the data volume is expanded, the model calculation accuracy cannot be improved, and the traditional detection method has problems such as high cost, low efficiency, and high destructiveness.
The additional mass method combined with machine learning method is adopted, and the parameter prediction model is trained using the data set, and the influence of different factors is comprehensively considered, the best machine learning model is selected and fused to predict the moisture density and moisture content to determine the rolling density.
The accuracy and efficiency of the prediction of the crush density of the rock pile body are improved, the shortcomings of the linear regression model are avoided, and efficient and lossless compaction quality detection of the rock pile material is achieved.
Smart Images
Figure CN120541806A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of non-destructive testing of rockfill compaction quality, and in particular to a method and device for predicting compaction density based on an added mass method and machine learning. Background Art
[0002] Levees and dams serve as primary water-retaining structures in water conservancy project construction. The proper selection of fill materials and construction processes significantly impacts the safe operation of these projects. Rockfill, due to its readily available, simple construction, and adaptability, is widely used as a fill material in levee and dam construction. The quality of its compaction directly impacts the long-term deformation and stability of the levees and dams.
[0003] Traditional water injection and sand injection methods are limited by the pit detection method, and have shortcomings such as small measurement area, high cost, and low efficiency. They will also cause a certain degree of damage to the embankment and dam body, and have certain defects in the comprehensiveness and refined control of project quality.
[0004] Related technologies have proposed using the added mass method to measure the compaction density and quality of rockfill materials. This method, based on the energy-vibration-inversion theory, has been gradually applied to rockfill compaction quality testing due to its high efficiency, accuracy, and non-destructive nature. However, existing technologies, after collecting parameters such as the rockfill's vibration stiffness and parametric mass, still use linear regression models or multiple linear regression models to invert the rockfill's compaction density. This inversion method struggles to simultaneously consider the impact of multiple factors on the rockfill's compaction quality. In particular, as the amount of collected data increases, relying on the number of samples in the expanded linear model to calculate the rockfill's compaction quality will hardly change the model's calculation accuracy and may even result in reduced accuracy. Summary of the Invention
[0005] In order to help improve the accuracy of compaction density prediction, the present application provides a compaction density prediction method and device based on added mass method and machine learning.
[0006] In a first aspect, the present application provides a method for predicting compaction density based on the added mass method and machine learning, which adopts the following technical solutions: A compaction density prediction method based on an added mass method and machine learning, the method comprising: Acquiring test data corresponding to a test point, wherein the test data is obtained by testing the test point using an added mass method; Inputting the test data into a parameter prediction model to obtain a predicted wet density and a predicted moisture content; Determining the compaction density corresponding to the test point based on the predicted wet density and the predicted moisture content; In which, the parameter prediction model is obtained by training the initial model using a data set, and the data set includes at least one group of sample data, each group of sample data includes sample test data, sample wet density and sample moisture content, the sample test data is obtained by testing the sample points using the added mass method, and the sample wet density and the sample moisture content are obtained by testing the sample points using the water injection method.
[0007] By adopting the above technical solution, the test data of the added mass method can be processed through the parameter prediction model to realize the prediction of wet density and moisture content. Therefore, the influence of factors of different dimensions can be comprehensively considered through the parameter prediction model, avoiding the shortcomings of the linear regression model, and thus ensuring the accuracy of the final density prediction.
[0008] Optionally, the initial model includes at least two, and the method further includes: For each of the initial models, the data set is used to train and test the initial model to obtain a trained model corresponding to the initial model and a test result corresponding to the trained model; The parameter prediction model is determined based on the test results corresponding to each of the trained models.
[0009] By adopting the above technical solution, the best model can be selected from different types of machine learning models based on the actual training results as the parameter prediction model, which can help improve the prediction accuracy of the parameter prediction model finally used.
[0010] Optionally, determining the parameter prediction model based on the test results corresponding to each of the trained models includes: determining a weight coefficient corresponding to each of the trained models based on the test results corresponding to each of the trained models; The trained models are fused based on the weight coefficients corresponding to the trained models to obtain the parameter prediction model.
[0011] By adopting the above technical solution, the prediction results of different types of machine learning models can be taken into consideration during the prediction process, and the degree of influence of the prediction results of different trained models on the final prediction results can be determined based on the test results corresponding to different trained models. This can help to comprehensively utilize the advantages of different types of machine learning models, thereby improving the accuracy of the prediction results of the parameter prediction model.
[0012] Optionally, determining the weight coefficient corresponding to each of the trained models based on the test results corresponding to each of the trained models includes: Initializing the weight coefficients corresponding to the trained models based on the test results corresponding to the trained models, and fusing the trained models based on the initialized weights corresponding to the trained models to obtain an initial fusion model; The weight coefficients corresponding to the trained models in the initial fusion model are optimized based on the data set to obtain an optimal combination of the weight coefficients corresponding to the trained models.
[0013] By adopting the above technical solution, after the trained models are fused based on the weight coefficients determined by the test results, the weights corresponding to different trained models in the fused model can be further adjusted by training the fused model to determine the weight coefficient combination with the highest prediction confidence, which can help to further improve the accuracy of the fused model.
[0014] Optionally, the using the data set to train the initial model includes: Grid search and cross-validation methods are used for parameter optimization during model training.
[0015] Optionally, the initial model includes an artificial neural network model, a support vector machine model and an extreme gradient boosting machine model.
[0016] Optionally, the test data includes first characteristic data and second characteristic data, and inputting the test data into a parameter prediction model to obtain predicted wet density and predicted moisture content includes: Performing feature interaction on the first feature data and the second feature data to obtain interaction data; The interactive data is input into a parameter prediction model to obtain the predicted wet density and the predicted moisture content.
[0017] By adopting the above technical solutions, the relationship between different test data can be captured based on the interaction features, thereby improving the accuracy of modeling and prediction.
[0018] Optionally, performing feature interaction on the first feature data and the second feature data to obtain interaction data includes: Performing feature conversion on the first feature data to obtain third feature data; Perform feature interaction on the third feature data and the second feature data to obtain interaction data.
[0019] By adopting the above technical solution, the nonlinear relationship between test data and density can be captured based on feature transformation, thereby improving the accuracy of modeling and prediction.
[0020] Optionally, the first characteristic data includes the circular frequency ω, the second characteristic data includes the dynamic stiffness K, the third characteristic data includes the product of the square of the circular frequency ω, and the interaction data includes the product of the dynamic stiffness K and the square of the circular frequency ω.
[0021] By adopting the above technical solution, the interactive data can enable the model not only to learn the correlation between dynamic stiffness and circular frequency, but also to capture the nonlinear changes of circular frequency, which can help improve the effect and accuracy of model training.
[0022] In a second aspect, the present application provides an electronic device, which adopts the following technical solution: An electronic device, comprising: at least one processor; Memory; At least one application, wherein the at least one application is stored in a memory and configured to be executed by at least one processor, and the at least one application is configured to: execute any one of the compaction density prediction methods based on the added mass method and machine learning provided in the first aspect.
[0023] In summary, this application includes at least one of the following beneficial technical effects: 1. The parameter prediction model can be used to process the test data of the added mass method to predict the wet density and moisture content. The parameter prediction model can comprehensively consider the influence of factors in different dimensions, avoid the shortcomings of the linear regression model, and thus ensure the accuracy of the final density prediction.
[0024] 2. Based on the actual training results, the best model can be selected from different types of machine learning models as the parameter prediction model. This can help improve the prediction accuracy of the parameter prediction model used in the end. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 1 is a flow chart of a method for predicting compaction density based on added mass method and machine learning provided in an embodiment of the present application; Figure 2 It is a schematic diagram of the added mass method; Figure 3 Schematic diagram of the parameter prediction modeling process provided in the embodiment of the present application; Figure 4 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0026] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-4It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.
[0027] Rockfill dam construction is characterized by a wide compaction construction area, large scale, and high safety requirements. It is urgent to effectively evaluate the compaction quality of rockfill dams and optimize the compaction plan in real time, and to seek a set of efficient and non-destructive rockfill material compaction quality detection technologies.
[0028] Currently, coarse aggregate compaction quality testing methods in current Chinese standards include water injection, sand injection, Rayleigh wave, and γ-γ methods. While water injection and sand injection methods are simple, they are expensive, inefficient, and require limited coverage. They also cause some damage to the dam body, severely hindering on-site engineering progress and refined control of compaction quality. Researchers at home and abroad have attempted to use techniques such as Rayleigh wave velocity measurement and wave-electric field coupling imaging to test the compaction quality of fill materials, achieving some success. However, these methods are limited by the selection of structural control parameters, statistical interference, and the influence of on-site testing equipment and signal acquisition. When measuring foundations with rock content exceeding 50%, these methods often produce numerous gaps in signal acquisition. Their imaging accuracy and resolution are unreliable and unrepresentative, making them unsuitable for controlling the compaction quality of rockfill dams with high rock content. γ-γ scattering testing is only suitable for thin rockfill layers, and gamma rays are harmful to humans, making it unsuitable for long-term monitoring of rockfill dam compaction quality.
[0029] The added mass method uses a graded loading method to treat the rockfill as an undamped single-degree-of-freedom linear elastic system under vibration conditions, thereby obtaining the stiffness and parametric mass of the rockfill material and inverting the density of the rockfill material, thereby achieving compaction evaluation of the rockfill material. However, existing models are mainly based on linear models or multi-linear models, and the calculation method mainly uses the collected dynamic stiffness K, parametric mass m and / or circular frequency ω. 2 The wet density ρ of the rockfill obtained by the water injection method is t The inversion model for the compaction density of the rockfill is obtained by linearly fitting parameters such as the water content w. This method makes it difficult to simultaneously consider the impact of multiple test parameters on the test results. Although simple linear fitting can achieve test accuracy within the specification requirements, it does not improve as the amount of test data increases.
[0030] Machine Learning (ML) is a multi-disciplinary interdisciplinary subject that involves multiple disciplines such as statistics, probability theory, advanced mathematics, and linear algebra. Machine learning specializes in how computers simulate and implement human learning behavior, and is the basis and fundamental way to achieve artificial intelligence. Specifically, in this application, the parameters obtained through field tests are used to construct a data set for machine learning model training, and the trained machine learning model is used to predict the wet density and moisture content, and finally the compaction density is determined based on the predicted wet density and predicted moisture content. Experiments have shown that this solution has the advantages of high prediction accuracy, strong portability, and high iteration efficiency.
[0031] The present application embodiment discloses a method for predicting compaction density based on the added mass method and machine learning. Figure 1 , the compaction density prediction method based on the added mass method and machine learning includes the following steps: Step 101: Acquire test data corresponding to the test points.
[0032] The test data is obtained by testing the test points using the added mass method. In one example, the test data includes dynamic stiffness K, parametric mass m, and circular frequency ω.
[0033] Specifically, refer to Figure 2 The added mass method considers the rock pile as an equivalent linear spring system with a single degree of freedom. According to the properties of the mass-elastic model, when the added mass is free from external forces and vibrates freely, the following formula can be obtained: K=ω 2 m (2) Where Z is the displacement of the signal receiving point, m is the sum of the rockfill and the additional mass, i.e., the parametric mass, K is the dynamic stiffness of the rockfill system, ω is the circular frequency of the signal receiving point, and Z0 is the vibration displacement amplitude of the signal receiving point.
[0034] In the added mass method, the total mass participating in the vibration is the sum of the vibration-parameter mass of the rockfill and the mass of the added mass block, so the following formula is obtained: K=ω 2 (m0+Δm) (4) Right now: From formula (5), we can see that 1 / ω 2 The inverse slope of the curve of Δm is the stiffness K. During the measurement process, by changing the number of additional mass blocks Δm, the change of the main frequency of the curve when the velocity is extracted at the top center of the additional mass blocks is studied. The dynamic stiffness K and circular frequency ω of the rockfill are obtained by calculation.
[0035] Step 102: Input the test data into the parameter prediction model to obtain the predicted wet density and the predicted moisture content.
[0036] The parameter prediction model is obtained by training the initial model using the data set. Specifically, the data set includes at least one set of sample data, each set of sample data includes sample test data, sample wet density and sample moisture content. The sample test data is obtained by testing the sample points using the added mass method, and the sample wet density and sample moisture content are obtained by testing the sample points using the water filling method. In an example, reference Figure 3 In the data preprocessing part, the test data includes dynamic stiffness K, parametric mass m, and circular frequency ω.
[0037] In one example, the initial model is any one of a machine learning model, such as an artificial neural network (ANN) model, a support vector machine (SVM) model, or an eXtreme Gradient Boosting (XGBoost) model. Furthermore, different types of initial models can be trained using training data, and the model with the best training effect can be determined as the parameter prediction model.
[0038] Optionally, use the dataset to train the initial model, including optimizing parameters using grid search and cross-validation (e.g., 5-fold cross-validation) during model training. The grid search range varies depending on the type of initial model. See Tables 1-3 for details. Table 1 shows the grid search range for ANN model parameters; Table 2 shows the grid search range for SVM model parameters; and Table 3 shows the grid search range for XGBoost model parameters.
[0039] Table 1 parameter scope hidden_layer_sizes (5),(5,5),(10),(10,10),(50),(50,50) alpha 0.0001,0.001,0.01 learning_rate_init 0.0001,0.001,0.01 Tol 1e-5, 1e-4, 1e-3 Table 2 parameter scope C 0.001,0.01,0.1,10,100,1000 Table 3 In one example, during model training, mean square error (MSE) can be used as an evaluation metric. Specifically, when hyperparameters are selected so that the mean square error achieved by the machine learning model on the test set is minimized or less than a preset threshold, the hyperparameter combination is retained as the optimal hyperparameter, resulting in a usable parameter prediction model. Specifically, the formula for calculating mean square error is expressed as follows: Where n represents the number of samples, Y i Indicates the actual value, Represents the predicted value.
[0040] In actual implementation, after using the parameter prediction model to predict the wet density and moisture content, the sample points can be randomly inspected. For example, the water irrigation method can be used to obtain the wet density and moisture content corresponding to some sample points to verify the accuracy of the corresponding prediction results of the parameter prediction model, and the sample data can be supplemented to supplement the training of the parameter prediction model. This can help to continuously improve the accuracy of the prediction results of the parameter prediction model during use.
[0041] Step 103: Determine the compaction density based on the predicted wet density and the predicted moisture content.
[0042] Specifically, the compacted density is determined based on the predicted wet density and the predicted moisture content, and is expressed by the following formula: Where ρ is the predicted wet density; w is the predicted moisture content; ρ dt is the compaction density.
[0043] To verify the practical effectiveness of the technical solution provided in this embodiment, during the research process, the density detection data of the Wuyue Pumped Storage Power Station and the Liyuan Hydropower Station, as well as the pit measurement and additional method detection data of the Nuozadu Hydropower Station, were collated and statistically analyzed. A database was constructed for training and testing the parameter prediction model. The database contains a total of 131 sets of data. In this example, 90 sets of data were randomly selected to form the training set for training the parameter prediction model, and the remaining data constituted the test set for testing the training effect of the parameter prediction model. Table 4 below is a statistical table of the test results of the Nuozadu Hydropower Station.
[0044] Table 4 Relative error range <1％ 1％～2％ 2％～3％ 3％～4％ 4％～5％ >5％ total Number of measuring points 20 6 8 1 5 1 41 Proportion 48.78％ 14.63％ 19.51％ 2.44％ 12.20％ 2.44％ 100％ Table 4 shows that the technical solution provided by this embodiment can accurately deduce the compaction density of rockfill bodies and can accurately detect a variety of materials. This model overcomes the poor adaptability and generalizability of multilinear models. Furthermore, the technical solution provided by this embodiment can be embedded in computing software. Through post-processing, it can quickly display multidimensional on-site inspection data, providing real-time feedback on compaction operations to construction workers and management personnel, thereby improving the accuracy of compaction operations.
[0045] The implementation principle of a compaction density prediction method based on the added mass method and machine learning in an embodiment of the present application is as follows: obtain test data corresponding to the test point, and the test data is obtained by testing the test point using the added mass method; input the test data into a parameter prediction model to obtain predicted wet density and predicted moisture content; determine the compaction density corresponding to the test point based on the predicted wet density and predicted moisture content; the parameter prediction model is obtained by training the initial model using a data set, and the data set includes at least one group of sample data, each group of sample data includes sample test data, sample wet density and sample moisture content, the sample test data is obtained by testing the sample point using the added mass method, and the sample wet density and sample moisture content are obtained by testing the sample point using the water injection method. By adopting the above scheme, the test data of the added mass method can be processed by the parameter prediction model to realize the prediction of wet density and moisture content, so that the influence of factors of different dimensions can be comprehensively considered by the parameter prediction model, avoiding the shortcomings of the linear regression model, and thus ensuring the accuracy of the final density prediction.
[0046] In some embodiments, the initial model includes at least two. In one example, the initial model includes: an artificial neural network model, a support vector machine model, and an extreme gradient boosting machine model.
[0047] Accordingly, the compaction density prediction method based on the added mass method and machine learning provided in this embodiment also includes: for each initial model, using a data set to train and test the initial model to obtain a trained model corresponding to the initial model and a test result corresponding to the trained model; determining a parameter prediction model based on the test results corresponding to each trained model.
[0048] The test results can be expressed as the confidence of the prediction of the trained model. Figure 3 In the model training and testing part, the dataset can be divided into a training set (for example, about 70% of the total data in the dataset) and a test set (for example, about 30% of the total data in the dataset). The training set is used to train the initial model, and the test set is used to test the trained model to obtain the test results corresponding to the trained model.
[0049] In one example, reference Figure 3 In the model selection section, a parameter prediction model is determined based on the test results corresponding to each trained model. This includes determining the trained model with the best prediction performance indicated by the test results as the parameter prediction model. For example, if the test results are expressed as confidence levels, the trained model with the highest prediction confidence level is determined as the parameter prediction model. In this way, the model with the best prediction performance can be selected from the total number of trained models as the parameter prediction model.
[0050] In another example, the parameter prediction model is a fusion model. In this case, determining the parameter prediction model based on the test results corresponding to each trained model includes: determining weight coefficients corresponding to each trained model based on the test results corresponding to each trained model; and fusing the trained models based on the weight coefficients corresponding to each trained model to obtain the parameter prediction model. For example, when the test results are expressed as confidence levels, the weight coefficients are positively correlated with the confidence levels, i.e., the higher the confidence level, the larger the weight coefficient corresponding to the fused model.
[0051] Among them, the sum of the weight coefficients corresponding to each trained model is 1.
[0052] Correspondingly, the parameter prediction model integrates various fused models. The input layer of the parameter prediction model is used to receive data and input the received input into each fused model respectively. The output layer of the parameter prediction model is used to receive the output data of each fused model, and perform weighted fusion on the output data of each fused model according to the weights corresponding to the fused model, and output the fused result.
[0053] By adopting the above technical solution, the prediction results of different types of machine learning models can be taken into consideration during the prediction process, and the degree of influence of the prediction results of different trained models on the final prediction results can be determined based on the test results corresponding to different trained models. This can help to comprehensively utilize the advantages of different types of machine learning models, thereby improving the accuracy of the prediction results of the parameter prediction model.
[0054] Furthermore, the weight coefficients corresponding to the respective trained models are determined based on the test results corresponding to the respective trained models, including: initializing the weight coefficients corresponding to the respective trained models based on the test results corresponding to the respective trained models, and fusing the respective trained models based on the initialized weights corresponding to the respective trained models to obtain an initial fusion model; and optimizing the weight coefficients corresponding to the respective trained models in the initial fusion model based on the data set to obtain the optimal combination of the weight coefficients corresponding to the respective trained models.
[0055] Among them, the method of fusing the trained models based on the initialization weights corresponding to the trained models to obtain the initial fusion model is the same as the method of fusing the trained models based on the weight coefficients corresponding to the trained models to obtain the parameter prediction model, which will not be repeated here.
[0056] Specifically, in order to help further improve the accuracy of the fused model prediction, after the trained models are fused based on the weight coefficients determined by the test results, the weights corresponding to different trained models in the fused model are further adjusted by training the fused model to determine the weight coefficient combination with the highest prediction confidence.
[0057] In addition, since the weight coefficients corresponding to each trained model are initialized based on the test results corresponding to the trained model, the accuracy of the weight coefficient initialization can be improved compared to random initialization or initialization with fixed weights, thereby shortening the training time of the model.
[0058] It should be noted that during the training of the initial fusion model, the model parameters of each trained model in the initial fusion model will not be adjusted, but only the weight coefficients corresponding to different trained models will be adjusted. This can help simplify the model training process and improve the efficiency of model training.
[0059] In one example, during the training of the initial fusion model, the direction of weight adjustment can be determined based on the relationship between the confidence of each trained model and the average confidence. For example, the weight coefficient of the model with a confidence greater than the average confidence is increased, and the weight coefficient of the model with a confidence less than the average confidence is decreased. The confidence of the initial fusion model is monitored during the training process. When the overall confidence of the initial fusion model no longer changes, the optimal combination of weight coefficients corresponding to the trained model is obtained. Furthermore, during the training of the initial fusion model, the maximum amplitude of a single adjustment of the weight system is limited. Within the range of the maximum amplitude, the specific adjustment amplitude needs to be determined based on the adjustment direction of the weight coefficients of different trained models. It is necessary to ensure that the sum of the weight coefficients of each trained model before and after the adjustment is 1.
[0060] In some embodiments, the test data includes first characteristic data and second characteristic data. The above step 102, inputting the test data into the parameter prediction model to obtain the predicted wet density and the predicted moisture content, includes: performing feature interaction on the first characteristic data and the second characteristic data to obtain interaction data; inputting the interaction data into the parameter prediction model to obtain the predicted wet density and the predicted moisture content.
[0061] Specifically, given that different test data often interact in complex ways in reality, dynamic stiffness (K) and circular frequency (ω) may interact together, and their impact on density is not simply the sum of the effects of either K or ω alone. By using interactive features (such as the interaction of K and ω), we can capture the relationships between different test data, leading to more accurate modeling and prediction.
[0062] In actual implementation, during the prediction process, the first feature data, the second feature data and the interaction data can be input into the parameter prediction model together. When the test data includes third feature data in addition to the first feature data and the second feature data, the third feature data and the interaction data also need to be output into the parameter prediction model together to enrich the data type of the input parameter prediction model.
[0063] Optionally, performing feature interaction on the first feature data and the second feature data to obtain interaction data includes: multiplying or dividing the first feature data by the second feature data to obtain the interaction data.
[0064] Optionally, the first feature data and the second feature data may include at least one of the following combinations: Combination 1: The first characteristic data is the dynamic stiffness K, and the second characteristic data is the circular frequency ω. The interactive data obtained in this case may be able to reflect a certain comprehensive stiffness characteristic of the soil under specific vibration conditions.
[0065] Combination 2: The first characteristic data is the parametric mass m, and the second characteristic data is the circular frequency ω or the dynamic stiffness K.
[0066] Further, performing feature interaction on the first feature data and the second feature data to obtain interaction data includes: performing feature conversion on the first feature data to obtain third feature data; performing feature interaction on the third feature data and the second feature data to obtain interaction data.
[0067] Feature conversion methods can include taking logarithms, squares, or higher powers, performing Fourier transforms on time features (e.g., acceleration response time, velocity response time, and displacement response time), and extracting corresponding frequency domain features (e.g., dominant frequency and frequency bandwidth). The specific feature conversion method can be determined based on the properties of different data types.
[0068] Specifically, considering that the relationship between test data and density may be nonlinear, feature transformation can capture the nonlinear relationship between test data and density, thereby achieving more accurate modeling and prediction.
[0069] In one example, the first characteristic data includes the circular frequency ω, the second characteristic data includes the dynamic stiffness K, and the third characteristic data includes the product of the square of the circular frequency ω. The interaction data includes the product of the dynamic stiffness K and the square of the circular frequency ω. This interaction data can not only reflect the interaction between the dynamic stiffness K and the circular frequency ω, but also reflect the nonlinear changes in the circular frequency ω, which can help improve the effectiveness and accuracy of model training.
[0070] In one example, the first feature data includes dynamic stiffness K, the second feature data includes the square of the circular frequency ω, and the third feature data also includes dynamic stiffness K. The interaction data includes the product of dynamic stiffness K and the square of the circular frequency ω. This interaction data allows the model to not only learn the relationship between dynamic stiffness and circular frequency, but also capture the nonlinear changes in circular frequency, thereby helping to improve the effectiveness and accuracy of model training.
[0071] The present application also provides an electronic device, such as Figure 4 As shown, Figure 4 The electronic device 200 shown includes a processor 201 and a memory 203. The processor 201 and the memory 203 are connected, for example, via a bus 202. Optionally, the electronic device 200 may further include a transceiver 204. It should be noted that in actual applications, the number of transceivers 204 is not limited to one, and the structure of the electronic device 200 does not constitute a limitation on the embodiments of the present application.
[0072] The processor 201 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor 201 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.
[0073] The bus 202 may include a path for transmitting information between the above components. The bus 202 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 202 may be divided into an address bus, a data bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.
[0074] The memory 203 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.
[0075] The memory 203 is used to store application code for executing the solution of the present application, and the execution is controlled by the processor 201. The processor 201 is used to execute the application code stored in the memory 203 to implement the content shown in the above method embodiment.
[0076] The electronic devices include, but are not limited to, mobile terminals such as mobile phones, laptop computers, PDAs (personal digital assistants), and PADs (tablet computers), and fixed terminals such as digital TVs and desktop computers, etc. They may also be server terminals. Figure 4 The electronic device shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.
[0077] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps and they may be performed in other orders.
[0078] The above are only some of the implementation methods of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. A compaction density prediction method based on added mass method and machine learning, characterized in that: The method comprises: Acquiring test data corresponding to a test point, wherein the test data is obtained by testing the test point using an added mass method; Inputting the test data into a parameter prediction model to obtain a predicted wet density and a predicted moisture content; Determining the compaction density corresponding to the test point based on the predicted wet density and the predicted moisture content; In which, the parameter prediction model is obtained by training the initial model using a data set, and the data set includes at least one group of sample data, each group of sample data includes sample test data, sample wet density and sample moisture content, the sample test data is obtained by testing the sample points using the added mass method, and the sample wet density and the sample moisture content are obtained by testing the sample points using the water injection method.
2. The method according to claim 1, characterized in that The initial model includes at least two, and the method further includes: For each of the initial models, the data set is used to train and test the initial model to obtain a trained model corresponding to the initial model and a test result corresponding to the trained model; The parameter prediction model is determined based on the test results corresponding to each of the trained models.
3. The method according to claim 2, characterized in that Determining a parameter prediction model based on the test results corresponding to each of the trained models includes: Determine the weight coefficient corresponding to each of the trained models based on the test results corresponding to each of the trained models; The trained models are fused based on the weight coefficients corresponding to the trained models to obtain the parameter prediction model.
4. The method according to claim 3, characterized in that Determining the weight coefficient corresponding to each of the trained models based on the test results corresponding to each of the trained models includes: Initializing the weight coefficients corresponding to the trained models based on the test results corresponding to the trained models, and fusing the trained models based on the initialized weights corresponding to the trained models to obtain an initial fusion model; The weight coefficients corresponding to the trained models in the initial fusion model are optimized based on the data set to obtain an optimal combination of the weight coefficients corresponding to the trained models.
5. The method according to claim 1, characterized in that The using the data set to train the initial model includes: Grid search and cross-validation methods are used for parameter optimization during model training.
6. The method according to claim 1, characterized in that The initial models include an artificial neural network model, a support vector machine model and an extreme gradient boosting machine model.
7. The method according to claim 1, characterized in that The test data includes first characteristic data and second characteristic data, and inputting the test data into a parameter prediction model to obtain a predicted wet density and a predicted moisture content includes: Performing feature interaction on the first feature data and the second feature data to obtain interaction data; The interactive data is input into a parameter prediction model to obtain the predicted wet density and the predicted moisture content.
8. The method according to claim 7, characterized in that The performing feature interaction on the first feature data and the second feature data to obtain interaction data includes: Performing feature conversion on the first feature data to obtain third feature data; Perform feature interaction on the third feature data and the second feature data to obtain interaction data.
9. The method according to claim 8, characterized in that The first characteristic data includes the circular frequency ω, the second characteristic data includes the dynamic stiffness K, the third characteristic data includes the product of the square of the circular frequency ω, and the interaction data includes the product of the dynamic stiffness K and the square of the circular frequency ω.
10. An electronic device, characterized in that: The electronic device comprises: at least one processor; Memory; At least one application, wherein the at least one application is stored in a memory and configured to be executed by at least one processor, and the at least one application is configured to: execute the compaction density prediction method based on the added mass method and machine learning as described in any one of claims 1 to 9.