Two-parameter identification method for moisture content and dry density of soil body

Abstract

The invention discloses a two-parameter identification method for soil moisture content and dry density, which comprises the following steps: S1, based on data of samples with different dry densities and moisture content, adopting a temperature tracing method of a cooling mode to obtain a cooling time history curve of a heat source in the samples with different dry densities and moisture content; s2, carrying out segmented averaging on the temperature time history data of a certain time period, namely carrying out a plurality of isochronous segments on the time period, and then carrying out averaging processing on the temperature of each time period to obtain segmented average temperature time history data of the time period; and S3, taking the segmented average temperature time history data in the S2 as a neural network input sample, taking the known soil dry density and moisture content as network output, and obtaining a soil dry density and moisture content two-parameter identification model. According to the method, the problem of how to identify the moisture content and dry density parameters of the soil body at the same time when the moisture content and dry density of the soil body are not determined is solved, and the problem of unstable heating of a heating mode is avoided.

Description

Technical Field

[0001] The present invention relates to the field of soil moisture content and dry density detection, and in particular to a dual-parameter identification method for soil moisture content and dry density. Background Art

[0002] The physical and mechanical properties of soil are related to its density and moisture content. Therefore, understanding the temporal and spatial evolution of soil physical properties is of great guiding significance for assessing the safety and stability of geotechnical engineering and geological bodies.

[0003] The heat transfer characteristics of soil are closely related to soil density and moisture content. Therefore, these physical quantities can be indirectly measured using reliable and inexpensive temperature measurement techniques, known as the temperature tracer method. This method monitors the transient temperature response of a heat source in the soil (heating or cooling patterns) and defines discriminant indicators based on the temperature response curve (commonly used indicators include the temperature rise amplitude, heating rate, cooling rate, and the integral of the heating time curve over time). The measured physical quantity is then inverted based on the statistical relationship between the discriminant indicators and the measured physical quantity (preliminarily calibrated through laboratory experiments). Compared to conventional measurement and geophysical methods, this method offers many advantages, including wide coverage, high efficiency, low power consumption, miniaturization, environmental friendliness, and long-term automated monitoring, and holds great promise for development and application.

[0004] Current methods for measuring soil moisture include temperature tracing and probe-based moisture content measurement. For example, in 2020, Junyi Guo et al. used a corundum tube encapsulated with an FBG sensor, which also served as a heating device, to study the impact of ambient temperature on moisture content measurements. The active heating method uses a linear heat source, requiring heating of the entire monitoring line during testing. This method requires high heating power when the optical cable is buried over long distances. This can affect the determination of temperature characteristic values, leading to measurement errors.

[0005] The temperature tracer method assumes that there is only one uncertain factor in soil properties and uses a single discriminant index to identify the inversion model for a soil property parameter. Common discriminant indexes include the temperature rise amplitude, temperature rise rate, temperature drop rate, and the time integral of the temperature rise time history curve. The heat transfer characteristics in soil are affected by multiple factors. Under non-permeable conditions, soil thermophysical properties are related not only to moisture content but also to soil density. However, when both soil density and moisture content are uncertain, this single discriminant index identification method cannot achieve the simultaneous inversion of multiple unknown physical quantities. Furthermore, soils are heterogeneous, and the statistical relationship between the discriminant index and the measured physical quantity is generally pre-calibrated through laboratory testing, using nearly homogeneous samples. Statistical models established in homogeneous samples can lead to errors when used to identify heterogeneous soil property parameters, and improving identification accuracy remains to be studied.

[0006] CN202211303390.0 discloses a soil moisture content and dry density detection device and detection method thereof. The detection method actually first uses an FDR sensor to directly measure the soil volume moisture content, and then uses the ring knife method to directly measure the soil wet density in the second step, and then converts it to obtain the soil mass moisture content and dry density. In essence, the detection operation is still divided into two steps. Summary of the Invention

[0007] To address the above-mentioned issues, the present invention aims to provide a dual-parameter identification method for soil moisture content and dry density. Unlike previous heating test methods, this method uses a cooling mode test method. By obtaining the cooling time history curve of the heat source in samples with different dry densities and moisture contents, the segmented average temperature time history data is obtained. This data is input into a neural network as a eigenvalue. The known soil dry density and moisture content are used as the network output to obtain a dual-parameter identification model for soil dry density and moisture content. This achieves simultaneous dual-parameter detection of soil moisture content and dry density, solving the problem of how to simultaneously identify soil moisture content and dry density parameters when both soil moisture content and dry density are uncertain. The cooling mode avoids the heating instability problem of the heating mode.

[0008] The present invention is achieved through the following technical solutions:

[0009] A dual-parameter identification method for soil moisture content and dry density includes the following steps: S1. Based on sample data of different dry densities and moisture contents, a temperature tracing method in a cooling mode is used to obtain cooling time-history curves of a heat source in samples of different dry densities and moisture contents; S2. Segmentally averaging the temperature-history data for a certain period of time, i.e., dividing the period into multiple equal-time segments and averaging the temperatures of each segment to obtain the segment-wise average temperature-history data for that segment; S3. Using the segment-wise average temperature-history data in S2 as neural network input samples and the known soil dry density and moisture content as network outputs, the neural network is trained and tested to obtain a dual-parameter identification model for soil dry density and moisture content. When testing a measuring point, the temperature-history data for a certain period of time during the cooling phase of the measuring point is segmentally averaged. The resulting segment-wise average temperature-history data is used as the neural network input to the dual-parameter identification model for soil dry density and moisture content, thereby simultaneously outputting the identification results for the measuring point's dry density and moisture content. In implementation, the cooling period can be selected within a time range of 0-1000 seconds. It can be flexibly selected according to actual test needs.

[0010] The cooling mode is to start heating and make the temperature reach the predetermined temperature, then stop heating and let the heat source cool naturally to perform temperature sampling until the sampling wavelength is close to the ambient temperature wavelength.

[0011] The test system used in the temperature tracing method using the cooling mode includes a heating system, a temperature measurement system and a monitoring tube. The heating system includes a voltage-stabilized power supply and a heating tube. The temperature measurement system includes multiple fiber Bragg grating temperature sensors arranged in series at equal intervals and connected by optical cables, and a fiber Bragg grating demodulator. The heating tube is arranged outside the fiber Bragg grating temperature sensor to form a heating-sensing unit. The power supply is electrically connected to the multiple fiber Bragg grating temperature sensors connected in series. The heating-sensing unit is arranged in the monitoring tube and can move up and down in the monitoring tube. During testing, the bottom of the monitoring tube needs to be blocked and a heat transfer fluid is placed in the monitoring tube. The heating-sensing unit is moved up and down to complete the test of the monitoring point.

[0012] The fiber Bragg grating temperature sensor is also covered with a first copper material layer, the heating tube is also covered with a second copper material layer, thermal conductive glue is provided between the first copper material layer and the heating tube, and thermal conductive glue is provided between the heating tube and the second copper material layer.

[0013] The monitoring tube is a carbon fiber tube and the thermal fluid is water.

[0014] Acquiring data for samples with varying dry densities and moisture contents: In the calibration test model, gravel soil with consistent particle size distribution is layered into the mold. The dry density of each gravel type is calculated by weighing. The moisture content of each gravel type is also measured using a drying method. Each layer must be compacted evenly, and the strength and frequency of compaction must be maintained consistently.

[0015] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0016] Unlike existing techniques that primarily measure soil moisture content and dry density separately, this method differs from previous heating testing methods by pioneering a cooling mode test method. By obtaining the cooling time history curve of a heat source in samples with different dry densities and moisture contents, segmented average temperature time history data is obtained. This data is input into a neural network as a eigenvalue. The known soil dry density and moisture content are used as network outputs to obtain a dual-parameter identification model for soil dry density and moisture content. This enables simultaneous dual-parameter detection of soil moisture content and dry density, solving the problem of how to simultaneously identify soil moisture content and dry density parameters when both are uncertain. The cooling mode avoids the heating instability problem of the heating mode. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0018] Figure 1 Schematic diagram of the monitoring structure of the present invention;

[0019] Figure 2 Schematic diagram of the sensor-heating unit structure;

[0020] Figure 3 This is the original curve with increased moisture content and dry density;

[0021] Figure 4 This is a comparison chart after extracting 40 average values;

[0022] Figure 5 This is a comparison chart of the calibration test training set error;

[0023] Figure 6 This is a comparison chart of the calibration test set error;

[0024] Figure 7 Schematic diagram of the verification test model structure.

[0025] Markings and corresponding parts names in the accompanying drawings:

[0026] 1-Fiber Bragg grating demodulator, 2-Power supply, 3-Monitoring tube, 4-Sensing-heating unit, 4-1-Fiber Bragg grating temperature sensor, 4-2-Heating tube, 5-First copper material layer, 6-Second copper material layer, 7-Thermal conductive adhesive, 8-Measuring point, 9-Verification model, 10-Water tank. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with examples and drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0028] Example 1

[0029] A dual-parameter identification method for soil moisture content and dry density comprises the following steps: S1, based on sample data of different dry densities and moisture contents, adopting a temperature tracing method in a cooling mode to obtain cooling time-history curves of a heat source in samples of different dry densities and moisture contents; S2, performing segmented averaging on the temperature-history data of a certain period, i.e., dividing the period into a plurality of equal-time segments, and then averaging the temperature of each period to obtain the segmented average temperature-history data of the period; S3, using the segmented average temperature-history data in S2 as neural network input samples and the known soil dry density and moisture content as network outputs, performing neural network training and testing to obtain a dual-parameter identification model for soil dry density and moisture content; when detecting a measuring point, performing segmented averaging on the temperature-history data of a certain period of the cooling stage of the measuring point to obtain the segmented average temperature-history data of the measuring point as the neural network input to the dual-parameter identification model for soil dry density and moisture content, thereby simultaneously outputting the identification results of the measuring point dry density and moisture content.

[0030] 1. Monitoring principle

[0031] Heat transfer in porous media occurs primarily through conduction and convection. However, in studies of common moisture content and dry density measurements, fluid flow within porous media is minimal, and convection can usually be ignored. Therefore, heat transfer in soil occurs primarily through conduction.

[0032] If the convective heat transfer of the fluid in the porous medium is ignored, according to the law of conservation of heat, the transient heat conduction control equation of the calculation model can be expressed as:

[0033]

[0034] Where: x ,λ y and λ z Represent the thermal conductivity in the x, y and z directions respectively; ρ eq is the equivalent density of the porous medium; c eq is the equivalent specific heat of the porous medium; Q T is the rate of heat generation per unit volume.

[0035] According to the heat conduction theory, the heat transfer between high temperature medium and low temperature medium can be expressed as:

[0036] q(t)=λ·ζ(t)·A (2)

[0037] Where: t is time; q(t) is the heat transfer per unit time at time t; λ is the heat transfer coefficient; ζ(t) is the temperature gradient at time t; A is the heat transfer area.

[0038] It can be seen from formula (2) that when the temperature gradient and heat transfer area are constant, the heat transfer rate is proportional to the thermal conductivity of the medium.

[0039] During the temperature rise or fall process, the heat input or output per unit volume of heat transfer medium from the outside can be

[0040] Expressed as: dQ = ρ·c·dT (3)

[0041] Where: Q is the heat; ρ is the density of the medium; c is the specific heat of the medium; T is the temperature.

[0042] It can be seen from formula (3) that the greater the product of the soil density and specific heat around the internal heat source, the more heat needs to be input or output from the outside to increase or decrease the unit volume temperature.

[0043] The thermal conductivity λ of the porous medium and the product of soil density and specific heat ρc can be calculated by weighted average of the relevant material parameters of the porous medium solid particles and fluid, as shown below:

[0044]

[0045] Where: ρ f and ρ s are the densities of fluid and solid particles respectively; c f and c s are the specific heats of the fluid and solid particles, respectively; λ f and λ s are the thermal conductivity coefficients of fluid and solid particles, respectively; ω and ρ d are the moisture content and dry density of the porous medium, respectively.

[0046] Equation (4) shows that for the same porous medium, the greater the moisture content and dry density, the greater the product of the equivalent average thermal conductivity, equivalent average density, and equivalent average specific heat capacity. Therefore, the moisture content and dry density of the soil surrounding the internal heat source significantly affect the speed of heat dissipation. Therefore, the moisture content and dry density can be measured by analyzing the cooling rate of the heat source.

[0047] 2. Monitoring system

[0048] like Figure 1 This monitoring system uses a mobile distributed point heat source for monitoring, and mainly includes three key components: a heating system, a temperature measurement system, and a monitoring tube. The heating system consists of a heating element, a voltage-stabilized power supply, and a wire connected in series, wherein the voltage-stabilized power supply supplies power to each heating element through a wire. The temperature measurement system consists of a multi-point serial fiber Bragg grating (FBG) temperature sensor 4-1, an FBG demodulator 1, and an optical cable. The optical cable is an armored optical cable. The FBG temperature sensor 4-1 senses the temperature information and transmits it to the FBG demodulator 1 through the optical cable to obtain the corresponding temperature wavelength. The sensing-heating unit 4 integrates the multi-point serial heating elements, namely the heating tube 4-2 and the FBG temperature sensor 4-1. Figure 2 The schematic diagram is shown below; the sensor-heating system integrates the corresponding heating and temperature measurement circuits. Mobile distributed monitoring involves dragging the sensor-heating unit 4 within the monitoring tube to achieve full-line measurement coverage. To achieve this, the inner diameter of the monitoring tube 3 should be larger than the outer diameter of the sensor-heating unit 4. The gap between the two is filled with an electrically insulating liquid with a higher thermal conductivity, effectively transferring heat released by the heat source. Furthermore, the insulating properties of the insulating liquid prevent leakage, improving the system's measurement stability to a certain extent.

[0049] The specific dimensions and configuration of the sensing and heating elements are as follows: FBG sensor 4-1 has an outer diameter of 4mm and a length of 60mm, and is covered with a first copper layer 5. This outer layer is an annular ceramic heating tube with an outer diameter of 10mm, a wall thickness of 2mm, and a length of 12mm. The outer layer of the annular heating tube 4-2 is covered with a second copper layer 6 (copper tube). The gaps between the FBG sensor 4-1 and the annular ceramic heating tube 4-2, as well as between the annular ceramic heating tube 4-2 and the outer second copper layer 6, are filled with thermally conductive adhesive 7, which provides isolation, fixation, and heat conduction. This design not only ensures the stability and performance of the sensor element, but also improves the durability and reliability of the entire system.

[0050] 3. Monitoring methods and steps

[0051] (1) Connect the FBG temperature sensor wire to the FBG demodulator, set the sampling frequency, and observe whether the sampling wavelength is normal. If there is a large fluctuation in the frequency band, it may be caused by unstable temperature. Wait until the temperature stabilizes before sampling.

[0052] (2) Connect the regulated power supply, wait for the sampling wavelength to stabilize, start recording the sample, and then turn on the regulated power supply for heating;

[0053] (3) After heating for a fixed time until the temperature reaches the predetermined temperature, the voltage regulator is turned off and the heat source is allowed to cool naturally until the sampling wavelength approaches the ambient temperature wavelength;

[0054] (4) Using the above steps, test samples with known dry density and moisture content to obtain the cooling time history curves of the heat source in samples with different dry density and moisture content;

[0055] (5) Perform segmented averaging on the temperature time history data of the period of 0 to 1000 s, for example, divide it into 40 segments, i.e., one segment every 25 s, and average the temperature of these 25 s; the time period can also be flexibly divided according to the actual test needs.

[0056] (6) The segmented average temperature time history data is used as the neural network input sample, and the known soil dry density and moisture content are used as the network output. The neural network is trained and tested to obtain the soil dry density and moisture content dual parameter identification model;

[0057] (7) Conducting tests in actual application scenarios using steps (1) to (3), and achieving full coverage of measurement points by dragging the sensing and heating units sequentially along the line;

[0058] (8) The temperature time history data of the cooling stage of each measuring point from 0 to 1000s are averaged in sections, and the section-wise average results are used as neural network inputs to the trained soil dry density and moisture content dual-parameter identification model to output the dry density and moisture content identification results of each measuring point.

[0059] Example 2 Calibration model experiment

[0060] Test model

[0061] The soil used in the calibration model test is gravel soil with consistent particle grading. The monitoring system adopts the point heat source mobile distributed monitoring system introduced in Example 1. The test model, that is, the mold material for loading the soil, is made of polyethylene, and its internal net volume is 200mm×200mm×200mm. It can be seen from the temperature cloud map of the cooling 1000s obtained in the previous numerical simulation that the diameter of the heat source heat dissipation range is less than 200mm, so in theory the size of the mold is sufficient. In order to ensure the feasibility of the experiment, a temperature sensor is placed on the inner wall of one side of the mold to measure whether the temperature at the inner wall changes during the test. Before the start of the test, the ambient temperature is turned on and recorded continuously until the cooling 1000s is completed and the record is turned off. After measurement, the temperature at the inner wall throughout the test is consistent with the ambient temperature at the time of initial heating, and no temperature rise or fall occurs. This shows that the heat transfer process of the heat source does not touch the edge of the mold, verifying that a mold of this size can provide an approximately adiabatic environment for the test.

[0062] The monitoring tube is a carbon fiber tube with an inner diameter of 16mm and a wall thickness of 1mm, and the thermal fluid is pure water. Before conducting the temperature test, the following test preparations are required: (1) Prepare a certain amount of pure water in advance and place it in the test site so that its temperature tends to be consistent with the ambient temperature; (2) Use a plug to plug the bottom of the monitoring tube, inject the thermal fluid, and observe for half an hour; if the liquid level in the tube does not drop, it means that the bottom of the monitoring tube is well sealed and will not cause the thermal fluid to lose; if the liquid level drops, the tightness of the plug needs to be readjusted. Because during the cooling process, the loss of thermal fluid will cause the loss of heat, resulting in temperature measurement data errors, so it is necessary to ensure the sealing of the bottom of the monitoring tube.

[0063] Test steps

[0064] A monitoring tube, filled with a plug, is placed vertically at the center of the mold bottom. Dry gravel soil is then filled in layers around the tube. Each layer is compacted evenly using a steel rod and hammer. The strength and frequency of compaction must be maintained consistently to ensure a uniform dry density distribution throughout the soil model. After filling, the sample is weighed and the dry density is recorded. The moisture content of the gravel soil used in the sample is measured and recorded using the drying method. The monitoring tube is then filled with thermal fluid and an FBG temperature sensor is placed inside the tube, centered in the mold. Temperature calibration tests are then performed at these moisture contents and dry densities.

[0065] After completing a set of tests, place the FBG temperature sensor in pure water and cool it to room temperature. Following the steps for filling the gravel soil described above, prepare soil samples with different dry densities at the same moisture content. It is important to note that the force and number of compactions used when compacting the gravel soil layer by layer must be varied to ensure that soil models with different dry densities are obtained. After completing all dry density tests at a specific moisture content, add a certain amount of water to all gravel soil samples and stir to prepare soil samples with a higher moisture content. After the samples are allowed to stand for one day to achieve a uniform moisture content, perform calibration tests for the next set of moisture content conditions. Follow this procedure to complete multiple sets of calibration tests at different moisture contents and dry densities.

[0066] In addition, the operating steps for temperature measurement of the monitoring system are the same under each set of working conditions. The detailed steps are as follows: (1) Connect the connector of the temperature sensing line to the FBG demodulator and set the wavelength acquisition frequency to 1Hz; (2) Connect the connector of the heating line to the regulated power supply and set the voltage to 11V. Too high a voltage may cause the heat diffusion range to exceed the model boundary, while too low a voltage may make the heating temperature insufficient to cause a significant thermal effect; (3) Start the FBG demodulator and observe the temperature wavelength. Only after it stabilizes can monitoring begin; if the temperature wavelength does not stabilize for a long time during the observation process, the connection channel should be changed and observed again; (4) After monitoring and recording the ambient temperature data for 60 seconds, turn on the regulated power supply, heat for 120 seconds, and then turn off the regulated power supply; (5) Wait for the sample to cool down naturally for 1000 seconds so that the heat source temperature approaches the ambient temperature, and stop data recording; (6) Replace the soil sample with a different moisture content or dry density and repeat the above steps for testing.

[0067] Get the cooling time curve

[0068] Newton's law of cooling is an empirical law that describes the rate of heat exchange between an object and its surroundings. Its core idea is that the rate of temperature change of an object is proportional to the temperature difference between the object and the surroundings. It can be expressed mathematically as:

[0069]

[0070] Where: T(t) is the temperature of the heat source at any time; t is time; T θ is the ambient temperature; k is the cooling constant, which is related to factors such as the object material, surface properties, and convection conditions. The larger k is, the faster the cooling rate.

[0071] The solution of the differential equation shown in formula (5) is:

[0072] T(t)=T θ +(T0-T θ ).e -kt (6)

[0073] Where: T0 is the initial temperature of the heat source.

[0074] Define the dimensionless parameter ζ:

[0075]

[0076] From formula (6) and formula (7), we can get:

[0077] ζ=-ktlg(e)=ξt (8)

[0078] In the formula: ξ = -klg(e), which is similar to the cooling constant k and is also a coefficient that reflects the speed of cooling.

[0079] When the temperature variation law of the heat source during the cooling stage is measured, ξ can be obtained using equations (7) and (8).

[0080] The calibration test initially obtained 24 sets of 0-1000s lg[(T(t)-T θ ) / (T0-T θ )] cooling time curve, but after extracting eigenvalues and conducting neural network training, it was found that the effect was poor. Despite continuous attempts to adjust network hyperparameters, sample division, and data eigenvalues, a satisfactory network model has not been obtained. Therefore, the moisture content and dry density of 24 groups of data were classified and compared, and it was found that some of the curves showed obvious large fluctuations. Compared with other moisture content and dry density curves, their trends were inconsistent with the corresponding theoretical curves. Since the BP neural network has high requirements for the accuracy of training samples, the presence of erroneous samples may lead to a significant decline in the performance of the neural network model. Therefore, some samples with erroneous trends in the 24 groups of samples were eliminated, and finally 18 groups of experimental samples were retained.

[0081] The main reasons for the error in the test data are as follows: (1) The dry density of the test soil is unevenly distributed, resulting in the average dry density of the specimen not being representative of the dry density of the soil around the heat source. Improvement methods include increasing the number of layers of compaction and keeping the compaction strength, number of times and method of each layer fixed; (2) The test interval between each specimen is too short, and the FBG sensor is not completely cooled to room temperature or the FBG demodulator wavelength is not completely stable before the test is started, resulting in differences in temperature data. The improvement method is to increase the test interval to ensure that the sensor temperature and demodulator wavelength are completely stable before starting the next set of tests; (3) During the process of filling gravel soil and moving the specimen, large disturbances are generated, causing the soil around the monitoring tube to loosen, resulting in a decrease in the dry density of the soil around the monitoring tube, thereby affecting the heat transfer efficiency.

[0082] After optimizing the experimental operation, in order to increase the training sample size, 9 additional groups of experiments were conducted, and a total of 27 groups of experimental data were obtained. Figure 3The original curves showing the increase in moisture content and dry density are shown (in order to facilitate the observation of the curve trend, the data with a shorter overlap section are selected for display, and the trends of the remaining data are similar); Figure 4 The 40 average values extracted from the corresponding original curves are shown. It can be observed that as the moisture content or dry density increases, the cooling curve decreases more rapidly, consistent with the cooling behavior of porous media heat sources. Temperature fluctuations caused by the test itself can be mitigated by taking an average value, resulting in smoother and more stable data.

[0083] Building a neural network model

[0084] BP neural network hyperparameters

[0085] According to the steps of establishing BP neural network in numerical simulation, a neural network corresponding to the calibration test was constructed. The network structure still uses a single hidden layer with 8 nodes. The activation function uses tansig function and purelin function, the training function is trainlm, and the network performance function uses mean square error function mse. The maximum number of training iterations is set to 1000, the learning rate is 0.01, and the training accuracy is 10 -7 Due to the relatively small number of samples, the automatic partitioning of the dataset into training, validation, and test sets in the Neural Network Toolbox was disabled. Of the 27 groups of samples, four were randomly selected as test set samples, while the remaining 23 were used as training set samples.

[0086] The trained neural network model's predictive performance on the test set failed to meet the specified error limits, so a trial-and-error algorithm was used to adjust hyperparameters. Attempts to modify hyperparameters, such as the number of hidden layers, the number of hidden nodes, and the activation function type, did not significantly improve network performance. However, changing the training function did significantly improve network performance. Preliminary trials using various training functions, including the trainlm function (Levenberg-Marquardt algorithm), the traingda function (adaptive learning rate gradient descent algorithm), the traingdm function (gradient descent algorithm with momentum), and the traingdx function (adaptive learning rate gradient descent algorithm with momentum), revealed that the trainlm function performed poorly. The neural network model trained with this function performed well on the training set but poorly on the test set, likely due to overfitting. Simplifying the network complexity (reducing the number of nodes) also failed to significantly improve final performance.

[0087] In contrast, the neural network model trained using the traingdx function performed better on both the training and test sets. The traingdx function introduces a momentum term, which uses information from historical weight updates to adjust the direction and magnitude of current weight updates, thereby accelerating convergence, avoiding local minima, and improving the search for the global minimum. Furthermore, it uses an adaptive learning rate, meaning that the size of the network weight update is dynamically adjusted based on the gradient of each weight. This facilitates the use of larger learning rates in the early stages of training to accelerate convergence, and smaller learning rates in the later stages of training to more precisely adjust weights, thereby improving training stability and effectiveness.

[0088] Therefore, traingdx is chosen as the training function, which corresponds to the adaptive learning rate gradient descent algorithm with additional momentum.

[0089] BP neural network training results

[0090] The performance of the BP neural network on the training set is as follows: the root mean square error (RMSE) of moisture content and dry density is 0.0045 and 17.97 kg·m -3 , the corresponding determination coefficients are 0.9910 and 0.9755 respectively. Figure 5 On the test set, the RMSE of moisture content and dry density are 0.0100 and 30.73 kg·m -3 , which is significantly lower than the set error limits (0.02 and 50 kg·m -3 ). At the same time, the coefficients of determination of moisture content and dry density are 0.9915 and 0.9500 respectively, both exceeding the standard of 0.9. Figure 6 The error comparison is shown.

[0091] Model Validation

[0092] 1. Test model

[0093] In order to verify the generalization performance of the BP neural network established in the calibration test, a verification test with non-uniform distribution of moisture content and dry density was carried out. Figure 7The verification model 9 shown is tubular, 1000 mm high, with an inner diameter of 100 mm and a wall thickness of 5 mm. The outer shell is made of plexiglass. The bottom is sealed with a 5 mm thick plexiglass plate, and the bottom joints are sealed with hot-dry adhesive. A circular hole with a diameter of 3 mm is located in the center of the bottom plate to allow water to penetrate the soil through the circular hole. The center of the circular hole is aligned with the central axis of the tubular model. The entire tubular model is placed vertically in a water tank 10. A thin gasket is used to raise the bottom of the model to ensure a gap between the circular hole bottom plate and the bottom of the water tank 10 to ensure free water penetration. The outer wall of the verification model 9 is marked with a 1000 mm height scale to indicate the location of the measuring points 8. Measuring point 1 is located 100 mm from the bottom. Measuring points 8 are set every 100 mm thereafter, up to 1000 mm, for a total of nine measuring points distributed across the entire height of the model.

[0094] A metal plug was used to seal one end of the carbon fiber monitoring tube. The tube was pre-filled with water to test its tightness and ensure proper function. The monitoring tube was then placed vertically on the center axis of the model, with the bottom plug aligned with the circular hole and the opening of the monitoring tube facing upward. The same gravel soil used in the calibration test was layered around the monitoring tube and evenly compacted using a steel rod. The force and frequency of compaction were reduced with each additional layer to ensure that the dry density of the specimen gradually decreased from bottom to top. Finally, the filled plexiglass tube was placed in a water tank 10, which was filled with pure water until the water level reached approximately 200 mm above the scale on the tube. Water seeped in through the circular hole in the base of the plexiglass tube, saturating the soil sample below the water level. Due to the capillary effect, water in the soil moved upward through the pores under the action of surface tension, causing the moisture content of the sample to vary with height. The entire model was left in the water tank for 14 days. Once the water penetration level stabilized, the validation test was ready. At this time, the entire soil sample shows a distribution in which the dry density and moisture content gradually decrease from bottom to top.

[0095] After the dry-wet boundary of the soil sample stabilizes (i.e., the soil moisture content approaches stability), a validation test is conducted. As demonstrated in the validation numerical simulation, simultaneous heating and temperature measurement at all nine measuring points will result in overlapping temperature diffusion ranges, leading to errors in the temperature measurement data. Therefore, to ensure that adjacent measuring points do not interfere with each other, a single temperature sensor is connected in series to the temperature measurement circuit. Temperature measurements are taken at each of the nine measuring points by dragging the circuit. The operation steps of the temperature measurement system are the same as those for the calibration test. It is important to note that after the test at one measuring point is completed, the temperature of the thermal fluid at the adjacent measuring point may increase. In this case, the adjacent measuring point can be skipped and the temperature sensor can be dragged to the next measuring point. The test can then be resumed only after the temperature at that measuring point reaches the ambient temperature. Finally, all nine measuring points are measured sequentially, and soil samples are removed layer by layer at 5 cm intervals. During sampling, the sample must be kept flush with the scale mark and squeezed as little as possible to avoid disturbing the underlying soil and thus changing its dry density. After the soil samples were taken out and weighed, the moisture content and dry density of each layer of soil were determined using the drying method, and the average moisture content and dry density of the 5 cm layer were obtained.

[0096] After measuring the temperature wavelength information at the nine measuring points, 40 average values were extracted from each cooling time curve through data processing. Next, a trained BP calibration neural network model was used for prediction, thereby inverting the predicted moisture content and dry density at each measuring point. The moisture content and dry density of each layer of soil sample were measured using the drying method. The values of the two layers of soil above and below the measuring point (i.e., the soil within 5 cm above and below the measuring point) were then averaged to obtain the average moisture content and average dry density, which are the relatively true moisture content and dry density at that measuring point. The predicted values at the measuring points were compared with the corresponding true values. The comparison results are shown in Table 5.2.

[0097] Table 5.2 Comparison of moisture content and dry density errors at each measuring point (all linear function changes)

[0098]

[0099] Table 5.2 shows that the RMSEs for the predicted moisture content and dry density at the nine measurement points are 0.0165 and 34.09 kg·m⁻³, respectively, and their coefficients of determination are 0.9482 and 0.9359, respectively. Both the RMSEs and coefficients of determination are within the specified error limits. The predicted and true value curves also show a consistent trend.

[0100] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dual-parameter identification method for soil moisture content and dry density, characterized in that: The following steps are involved: S1. Based on the data of samples with different dry densities and moisture contents, the temperature tracing method in the cooling mode is used to obtain the cooling time history curve of the heat source in the samples with different dry densities and moisture contents. S2. The temperature time history data of a certain period is averaged in sections, that is, the period is divided into multiple equal time sections, and the temperature of each section is averaged to obtain the section-wise average temperature time history data of the period. S3, use the segmented average temperature time history data in S2 as the neural network input sample, and the known soil dry density and moisture content as the network output, and perform neural network training and testing. A dual-parameter identification model of soil dry density and moisture content is obtained. When detecting the measuring point, the temperature time history data of a certain period of the cooling stage of the measuring point is processed by segmented averaging. The segmented average temperature time history data of the measuring point is obtained as the neural network input to the dual-parameter identification model of soil dry density and moisture content, and the identification results of the dry density and moisture content of the measuring point can be output simultaneously.

2. The dual-parameter identification method according to claim 1, characterized in that: The cooling mode is to start heating and make the temperature reach the predetermined temperature, then stop heating and let the heat source cool naturally to perform temperature sampling until the sampling wavelength is close to the ambient temperature wavelength.

3. The dual parameter identification method according to claim 1, characterized in that: The test system used in the temperature tracer method using the cooling mode includes a heating system, a temperature measurement system, and a monitoring tube. The heating system includes a voltage-stabilized power supply and a heating tube. The temperature measurement system includes multiple fiber Bragg grating temperature sensors connected in series at equal intervals via optical cables and a fiber Bragg grating demodulator. The heating tube is sheathed over the fiber Bragg grating temperature sensors to form a heating-sensing unit. The power supply is electrically connected to the multiple fiber Bragg grating temperature sensors connected in series. The heating-sensing unit is arranged in the monitoring tube and can move up and down within the monitoring tube. During testing, the bottom of the monitoring tube needs to be blocked and a heat transfer fluid is placed in the monitoring tube. Move the heating-sensing unit up and down to complete the test of the monitoring point.

4. The dual parameter identification method according to claim 3, characterized in that: The fiber Bragg grating temperature sensor is also covered with a first copper material layer, the heating tube is also covered with a second copper material layer, thermal conductive glue is provided between the first copper material layer and the heating tube, and thermal conductive glue is provided between the heating tube and the second copper material layer.

5. The dual-parameter identification method according to claim 4, characterized in that: The monitoring tube is a carbon fiber tube and the thermal fluid is water.

6. The dual-parameter identification method according to claim 4, characterized in that: Acquisition of sample data with different dry densities and moisture contents: In the calibration test model, gravel soil with consistent particle grading is filled into the mold layer by layer, and the dry density of different types of gravel soil is obtained by weighing and calculation; at the same time, the drying method is used to measure and obtain the moisture content of different types of gravel soil.

7. The dual-parameter identification method according to claim 6, characterized in that: Each time a layer of calibration sample is filled in, it must be compacted evenly, and the compaction strength and number of times for each layer must remain stable.

Images