Method and system for detecting sandstone moisture content of combined model under data fusion

Abstract

The invention provides a gravel water content detection method and system of a combined model under data fusion, and relates to the technical field of gravel water content detection. The method comprises the following steps: acquiring related detection data, preprocessing, combining MIC and GRA for correlation analysis, inputting the fused data to construct a BWO-RF detection model, obtaining first analysis data, obtaining second analysis data by using a DNN model, and superposing the first analysis data and the second analysis data to obtain the final moisture content. According to the method, the data quality and consistency are ensured, model optimization is carried out, and the random forest is optimized through the BWO algorithm, so that the model prediction capability is improved; dNN is used for error correction, the final prediction accuracy is improved, multiple methods and models are combined, and through multi-level data processing and optimization, the accuracy of the detection result of the water content of the sandstone is improved, and the detection speed of the water content of the sandstone is greatly increased.

Description

Technical Field

[0001] The present invention relates to the technical field of sand and gravel moisture content detection, and in particular to a sand and gravel moisture content detection method and system based on a combined model under data fusion. Background Art

[0002] As the most widely used building material in the world, concrete is widely used in practical engineering fields such as roads and bridges. Among the raw materials of concrete, the moisture content in sand and gravel is most affected by the geographical environment, weather, and season. In order to ensure the compressive strength and durability of concrete, the moisture content of sand and gravel is accurately and quickly measured during the production process, and corresponding compensation and correction are made, which is of great significance to improving the production quality of concrete.

[0003] At present, the main detection methods for sand and gravel moisture content at home and abroad include drying method, microwave method, infrared method and capacitance method, among which, the drying method has high detection accuracy, but it is time-consuming and labor-intensive. Although the detection time of microwave method, infrared method and capacitance method has been significantly improved, the accuracy needs to be improved. With the development of artificial intelligence and sensor technology, how to achieve faster and more accurate detection of sand and gravel moisture content has become one of the key research directions now. Summary of the invention

[0004] The main purpose of the present invention is to provide a sand and gravel moisture content detection method and system based on a combined model under data fusion, which solves the technical problems that the current sand and gravel moisture content detection is time-consuming and labor-intensive, the detection time is long, and the detection accuracy is not high enough.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a sand and gravel moisture content detection method of a combined model under data fusion, comprising the following steps: S1: Use multiple sensors to obtain relevant detection data of sand and gravel moisture content, merge them into a multidimensional array, perform data preprocessing, and save them into a data set. The relevant detection data include: microwave original value, sand and gravel sample temperature value, ambient humidity value, ambient temperature value and installation angle of microwave sensor; S2: Build a BWO-RF detection model, based on multi-data fusion, and use the detection data set to train the BWO-RF detection model, and retain the optimal BWO-RF detection model; S3: building a DNN model, where the DNN model is used to perform error correction on the first analysis data output by the optimal BWO-RF detection model to obtain second analysis data; S4: Input the data to be tested into the BWO-RF detection model and the DNN model, obtain the first analysis data and the second analysis data respectively, and superimpose the two to obtain the final moisture content detection value.

[0006] The preferred scheme also includes: before step S2, combining MIC and GRA, performing correlation analysis on the test data after each data preprocessing and the sand and gravel moisture content value, fully considering linear and nonlinear relationships, combining MIC and GRA, obtaining the correlation degree between the test data and the sand and gravel moisture content, and visualizing the correlation analysis results.

[0007] In the preferred embodiment, the visualization processing is specifically: calling the Seaborn library in Python to intuitively display the correlation analysis results through a heat map.

[0008] In a preferred embodiment, the data set further includes: calculating different sand and gravel mud content values ​​and fineness modulus values ​​according to the relevant detection data; The data preprocessing includes: data filtering, filtering out unqualified invalid data and abnormal values; Normalize or standardize the data to ensure consistency and comparability; Among them, data filtering is to scatter plot the sand and gravel moisture content detection values, determine the outliers as abnormal points, and remove the abnormal points and their corresponding characteristic values.

[0009] In a preferred embodiment, in S3, the BWO-RF detection model is trained using the detection data set, including the following steps: Dividing the data set into a training set and a test set; Set the number of iterations for training; Input the training set into the model for iterative training until the number of iterative training times is reached, and save the optimal model; The iterative training includes optimizing the key hyperparameters of the model using BWO, and continuously iterating so that the loss function is continuously reduced and tends to be stable, thereby obtaining the optimal hyperparameters and optimizing the model; The key hyperparameters are the number of decision trees and the maximum number of features.

[0010] In the preferred embodiment, the optimization of key hyper parameters by using BWO includes the following steps: S31: Initialize BWO related parameters; S32: Initialize the beluga population and calculate the corresponding fitness function value; S33: Calculate the falling probability and balance factor; S34: Position update calculations are performed in the three phases of exploration, development, and falling; S35: Calculate the fitness function value and calculate the current optimal value; S36: Determine whether the maximum number of iterations has been reached, if so, proceed to S37; if not, proceed to S33; S37: Output the best position of the white whale, that is, obtain the optimal parameter combination of the random forest.

[0011] A sand and gravel moisture content detection system of a combined model under data fusion, applied to a sand and gravel moisture content detection method of a combined model under data fusion, comprising: The data acquisition module is used to obtain relevant detection data of sand and gravel moisture content using multiple sensors, and save the data into a data set after data preprocessing. The relevant detection data includes: microwave original value, sand and gravel sample temperature value, ambient humidity value, ambient temperature value and installation angle of microwave sensor; BWO-RF detection model module, used to build BWO-RF detection model, based on multi-data fusion, and use the detection data set to train the BWO-RF detection model, and retain the optimal BWO-RF detection model; A DNN model module, used for building a DNN model, wherein the DNN model is used for performing error correction on the first analysis data output by the optimal BWO-RF detection model to obtain second analysis data; The moisture content calculation module is used to input the data to be tested into the BWO-RF detection model and the DNN model, obtain the first analysis data and the second analysis data respectively, and superimpose the two to obtain the final moisture content detection value.

[0012] The invention provides a sand and gravel moisture content detection method and system of a combined model under data fusion. The method obtains relevant detection data of sand and gravel moisture content through multiple sensors and performs preprocessing to obtain a data set. Then, MIC and GRA are combined to perform correlation analysis. The fused data are input into a BWO-RF detection model to obtain first analysis data. Then, the output of the BWO-RF detection model is input into a DNN model to obtain second analysis data. The second analysis data and the first analysis data are superimposed to obtain a final moisture content detection value. Through data preprocessing, data quality and consistency are ensured to avoid affecting model performance due to data problems. Through correlation analysis, MIC and GRA are selected to perform multi-level data association analysis, and model optimization is performed. Random forest is optimized through the BWO algorithm to improve the model prediction ability. Then, DNN is used to perform error correction to improve the final prediction accuracy. Combining multiple methods and models, through multi-level data processing and optimization, the accuracy of sand and gravel moisture content detection results is improved, and the detection speed of sand and gravel moisture content is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a flow chart of the sand and gravel moisture content detection method of the present invention; Figure 2It is a visualization diagram of the correlation between different types of sensor data and the water content of sand and gravel in the present invention; Figure 3 It is a schematic diagram of the algorithm flow of the BWO-RF model of the present invention; Figure 4 It is a curve diagram of the detection of the moisture content of sand and gravel under the combined model of the present invention; Figure 5 Schematic diagram of the error between the detected value and the actual value of the water content of sand and gravel in the present invention; Figure 6 This is a schematic diagram of data filtering outliers in the present invention; Figure 7 It is a schematic diagram of iterative training of hyperparameter optimization of the present invention. DETAILED DESCRIPTION

[0014] Example 1 like Figure 1-7 As shown, a sand and gravel moisture content detection method based on a combined model under data fusion includes the following steps: S1: Use multiple sensors to obtain relevant detection data on the moisture content of sand and gravel, merge them into a multidimensional array, perform data preprocessing, and save them into a data set. The relevant detection data include: microwave original value, sand and gravel sample temperature value, ambient humidity value, ambient temperature value and the installation angle of the microwave sensor.

[0015] S2: Build a BWO-RF detection model based on multi-data fusion and use the detection data set to train the BWO-RF detection model and retain the optimal BWO-RF detection model.

[0016] S3: Building a DNN model, where the DNN model is used to perform error correction on the first analysis data output by the optimal BWO-RF detection model to obtain second analysis data.

[0017] S4: Input the data to be tested into the BWO-RF detection model and the DNN model, obtain the first analysis data and the second analysis data respectively, and superimpose the two to obtain the final moisture content detection value.

[0018] In this embodiment, BWO (Beluga Whale Optimization) performs well in solving one-dimensional and multi-dimensional problems, and demonstrates extremely strong benchmark function scalability analysis capabilities. BWO has good global convergence. RF (Random Forest) is an integrated method composed of a decision tree and a bagging framework. RF can effectively solve nonlinear problems between multiple parameters. BWO iteratively optimizes RF key parameters. The combination of the two is the BWO-RF detection model used in this embodiment.

[0019] DNN (Deep Neural Networks) is a complex form of machine learning, belonging to the category of Artificial Neural Networks (ANNs). MIC (Maximal Information Coefficient) is a statistic used to quantify the strength of the dependency relationship between two variables; GRA (Grey Relation Analysis) is a multi-factor statistical analysis method used to analyze and evaluate the correlation and influence between multiple indicators.

[0020] Relevant detection data of sand and gravel moisture content is obtained through multi-sensor and preprocessed to obtain a data set, and then MIC and GRA are combined for correlation analysis, and the fused data is input into a BWO-RF detection model to obtain first analysis data, and then the output of the BWO-RF detection model is input into a DNN model to obtain second analysis data, and the second analysis data and the first analysis data are superimposed to obtain a final moisture content detection value; this embodiment ensures data quality and consistency through data preprocessing to avoid affecting model performance due to data problems; performs model optimization, optimizes random forests through the BWO algorithm, and improves model prediction ability; and then uses DNN to perform error correction to improve the final prediction accuracy. Combining multiple methods and models, through multi-level data processing and optimization, the accuracy of sand and gravel moisture content detection results is improved, and the detection speed of sand and gravel moisture content is greatly improved.

[0021] The preferred solution also includes: before step S2, combining MIC and GRA, performing correlation analysis on the test data after each data preprocessing and the sand and gravel moisture content value, fully considering the linear and nonlinear relationship, combining MIC and GRA, obtaining the correlation between the test data and the sand and gravel moisture content, and visualizing the correlation analysis results. In this embodiment, the correlation analysis selects MIC and GRA to perform multi-level data association analysis, and the correlation degree is obtained, which is between (0, 1). The closer to 1, the better the correlation.

[0022] In the preferred solution, the visualization processing is specifically: calling the Seaborn library in Python to intuitively display the correlation analysis results through a heat map.

[0023] like Figure 2As shown, it is a visualization diagram of the correlation between different types of sensor data and sand and gravel moisture content of the present invention. The visualization of this embodiment uses the Seaborn library in Python based on the calculation results of the correlation analysis to intuitively visualize the correlation between each input feature, including fineness modulus, mud content, sample temperature, contact angle, ambient temperature, and microwave original value, and the sand and gravel moisture content detection value.

[0024] In the preferred embodiment, the data set also includes: calculating different sand and gravel mud content values ​​and fineness modulus values ​​based on relevant test data.

[0025] Data preprocessing includes: data filtering, filtering out unqualified invalid data and outliers.

[0026] Normalize or standardize the data to ensure its consistency and comparability.

[0027] Among them, data filtering is to scatter plot the sand and gravel moisture content detection values, determine the outliers as abnormal points, and remove the abnormal points and their corresponding characteristic values.

[0028] like Figure 6 As shown, data filtering is to filter out abnormal data. By performing a chemical analysis on the original data, as shown in the figure, outliers in the sand and gravel moisture content test values ​​(red dots in the figure) are found and their corresponding data are filtered out.

[0029] In the preferred solution, in step S3, the BWO-RF detection model is trained using the detection data set, including the following steps: Divide the dataset into training and testing sets.

[0030] Set the number of training iterations.

[0031] The training set is input into the model for iterative training until the number of iterative training times is reached and the optimal model is saved.

[0032] Among them, iterative training includes the model using BWO to optimize its key hyperparameters. Through continuous iteration, the loss function is continuously reduced and tends to be stable, the optimal hyperparameters are obtained, and the model is optimized.

[0033] Key hyperparameters are the number of decision trees and the maximum number of features.

[0034] In this embodiment, BWO (Beluga Optimization): the Beluga Optimization algorithm is used to optimize the parameters of the random forest to improve the model performance and enhance the model prediction capability.

[0035] RF (Random Forest): An integrated learning model suitable for handling high-dimensional data and multi-feature problems.

[0036] Model training: Use the dataset to train the BWO-RF model and select appropriate hyperparameters (such as the number of trees, the maximum depth of each tree, etc.).

[0037] Keep the best model: Select the best performing model through cross-validation or other evaluation methods.

[0038] In the preferred solution, BWO is used to optimize its key hyper parameters, including the following steps: S31: Initialize BWO related parameters.

[0039] S32: Initialize the beluga population and calculate the corresponding fitness function value.

[0040] S33: Calculate the falling probability and balance factor.

[0041] S34: Position update calculations are performed during the exploration, development, and falling phases.

[0042] S35: Calculate the fitness function value and calculate the current optimal value.

[0043] S36: Determine whether the maximum number of iterations has been reached. If so, proceed to S37; if not, proceed to S33.

[0044] S37: Output the best position of the white whale, that is, obtain the optimal parameter combination of the random forest.

[0045] In this embodiment, Figure 7 As shown in the figure, specifically, in the optimization process, first determine the input and output of the model, and preprocess the sand and gravel moisture content detection data set, and divide it into a training set and a test set; secondly, set the relevant parameters of the BWO algorithm (maximum number of iterations 20, population size 50, number of decision trees in the range of (300,500), maximum number of features in the range of (1,6)), and determine MSE (mean square error) as the loss function in the training process. The change of the loss function during the training process is shown in the figure.

[0046] The optimal parameters are: the number of decision trees is 467 and the maximum number of features is 2.

[0047] In this embodiment, the BWO related parameters are initialized, including defining the population size, the maximum number of iterations, the initial step size, the falling probability and the balance factor, and initializing the positions and velocities of all beluga whales.

[0048] Through the above steps, the White Whale Optimization Algorithm uses the White Whale behavior model in nature and gradually finds the optimal hyperparameter combination through the search strategies of exploration, development and fall phases. The calculation and update of each step ensures that the algorithm can find the best solution in the complex search space.

[0049] After obtaining the optimal BWO-RF model and DNN model from S1-S4 above, actual detection is performed.

[0050] In step S4, the first analysis data is obtained from the BWO-RF model, and the second analysis data is obtained from the DNN model. The first analysis data and the second analysis data are superimposed or weighted fused to obtain a final moisture content detection value; the final moisture content detection value is output, and necessary post-processing or formatting is performed.

[0051] In this embodiment, it is assumed that: the input feature in the sand and gravel moisture content detection process is X, the first analysis data is Y1, the second analysis data is Y2, the data set is divided into a training set and a test set, and after the model training is completed, the parameters of the training model are retained for detection. The test set input of the BWO-RF detection model is X1, the output is Y1, the actual value is Y_true, and the prediction error is E (E=Y_true-Y1). The test set input of the DNN detection model is X2, the output is Y2, and the final moisture content value is Y (Y=Y1+Y2).

[0052] Example 2 In combination with Example 1, a sand and gravel moisture content detection system of a combined model under data fusion is provided, which is applicable to a sand and gravel moisture content detection method of a combined model under data fusion, including: The data acquisition module is used to obtain relevant test data of sand and gravel moisture content using multiple sensors, and save the data into a data set after data preprocessing. The relevant test data include: microwave original value, sand and gravel sample temperature value, ambient humidity value, ambient temperature value and the installation angle of the microwave sensor.

[0053] The BWO-RF detection model module is used to build the BWO-RF detection model. Based on multi-data fusion, the BWO-RF detection model is trained using the detection data set to retain the optimal BWO-RF detection model.

[0054] The DNN model module is used to build a DNN model, and the DNN model is used to perform error correction on the first analysis data Y1 output by the optimal BWO-RF detection model to obtain the second analysis data Y2.

[0055] The moisture content calculation module is used to input the data to be tested into the BWO-RF detection model and the DNN model, obtain the first analysis data Y1 and the second analysis data Y2 respectively, and superimpose the two to obtain the final moisture content detection value Y, Y=Y1+Y2.

[0056] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A sand and gravel moisture content detection method based on a combined model under data fusion, characterized in that: The following steps are involved: S1: Use multiple sensors to obtain relevant detection data of sand and gravel moisture content, merge them into a multidimensional array, perform data preprocessing, and save them into a data set. The relevant detection data include: microwave original value, sand and gravel sample temperature value, ambient humidity value, ambient temperature value and installation angle of microwave sensor; S2: Build a BWO-RF detection model, based on multi-data fusion, and use the detection data set to train the BWO-RF detection model, and retain the optimal BWO-RF detection model; S3: building a DNN model, where the DNN model is used to perform error correction on the first analysis data output by the optimal BWO-RF detection model to obtain second analysis data; S4: Input the data to be tested into the BWO-RF detection model and the DNN model, obtain the first analysis data and the second analysis data respectively, and superimpose the two to obtain the final moisture content detection value.

2. The sand and gravel moisture content detection method of the combined model under data fusion according to claim 1 is characterized in that: Also includes: Before step S2, MIC and GRA are combined to perform correlation analysis on the test data after each data preprocessing and the sand and gravel moisture content value, fully considering the linear and nonlinear relationships, combining MIC and GRA to obtain the correlation between the test data and the sand and gravel moisture content, and visualizing the correlation analysis results.

3. The sand and gravel moisture content detection method of the combined model under data fusion according to claim 1 is characterized in that: The visualization processing specifically includes: calling the Seaborn library in Python to intuitively display the correlation analysis results through a heat map.

4. The sand and gravel moisture content detection method of the combined model under data fusion according to claim 1 is characterized in that: The data set also includes: calculating different sand and gravel mud content values ​​and fineness modulus values ​​according to the relevant detection data; The data preprocessing includes: data filtering, filtering out unqualified invalid data and abnormal values; Normalize or standardize the data to ensure consistency and comparability; Among them, data filtering is to scatter plot the sand and gravel moisture content detection values, determine the outliers as abnormal points, and remove the abnormal points and their corresponding characteristic values.

5. The sand and gravel moisture content detection method of the combined model under data fusion according to claim 1 is characterized in that: In S3, the BWO-RF detection model is trained using the detection data set, including the following steps: Dividing the data set into a training set and a test set; Set the number of iterations for training; Input the training set into the model for iterative training until the number of iterative training times is reached, and save the optimal model; The iterative training includes optimizing the key hyperparameters of the model using BWO, and continuously iterating so that the loss function is continuously reduced and tends to be stable, thereby obtaining the optimal hyperparameters and optimizing the model; The key hyperparameters are the number of decision trees and the maximum number of features.

6. The sand and gravel moisture content detection method of the combined model under data fusion according to claim 5 is characterized in that: The optimization of key hyper parameters by using BWO includes the following steps: S31: Initialize BWO related parameters; S32: Initialize the beluga population and calculate the corresponding fitness function value; S33: Calculate the falling probability and balance factor; S34: Position update calculations are performed in the three phases of exploration, development, and falling; S35: Calculate the fitness function value and calculate the current optimal value; S36: Determine whether the maximum number of iterations has been reached, if so, proceed to S37; if not, proceed to S33; S37: Output the best position of the white whale, that is, obtain the optimal parameter combination of the random forest.

7. A sand and gravel moisture content detection system based on a combined model under data fusion, characterized in that: It is characterized in that The sand and gravel moisture content detection method applicable to the combined model under data fusion includes: The data acquisition module is used to obtain relevant detection data of sand and gravel moisture content using multiple sensors, and save the data into a data set after data preprocessing. The relevant detection data includes: microwave original value, sand and gravel sample temperature value, ambient humidity value, ambient temperature value and installation angle of microwave sensor; BWO-RF detection model module, used to build BWO-RF detection model, based on multi-data fusion, and use the detection data set to train the BWO-RF detection model, and retain the optimal BWO-RF detection model; A DNN model module, used for building a DNN model, wherein the DNN model is used for performing error correction on the first analysis data output by the optimal BWO-RF detection model to obtain second analysis data; The moisture content calculation module is used to input the data to be tested into the BWO-RF detection model and the DNN model, obtain the first analysis data and the second analysis data respectively, and superimpose the two to obtain the final moisture content detection value.

Images