Roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid

Abstract

The invention discloses an intelligent heating geogrid-based roadbed settlement and frozen soil plasticity prediction system, which comprises an intelligent heating geogrid, a data acquisition unit, an intelligent heating control unit, a moisture content calculation unit and an analysis prediction unit, the intelligent heating control unit performs starting control on the heating resistance wire according to a soil body temperature value collected by the optical fiber temperature sensor so as to heat the self-heating steel pipe, the self-heating steel pipe heats the soil body, and the moisture content calculation unit calculates the moisture content of the soil body according to a temperature rise value of the soil body in unit time. And the analysis and prediction unit predicts roadbed settlement and frozen soil plasticity according to the soil moisture content and the grating dependent variable. The self-heating steel pipe is installed on the grating body, the water content of the soil body is monitored in real time while the soil body is heated, the subgrade settlement state and the frozen soil plastic characteristics are accurately predicted according to the collected data, and therefore scientific support is provided for safety evaluation and maintenance of subgrade engineering.

Description

Technical Field

[0001] The invention relates to the technical field of railway roadbed monitoring, in particular to a roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrids. Background Art

[0002] The construction of high-speed railways is of great strategic significance to supporting my country's transportation power. my country's railway operating mileage has increased from 98,000 kilometers in 2012 to 159,000 kilometers in 2023, of which high-speed railways have increased from 9,000 kilometers to 45,000 kilometers, ranking first in the world. However, my country is also a country with widespread permafrost distribution. Seasonally frozen areas account for 53.5% of the total land area, and these areas cover more than 7,000 kilometers of high-speed railway lines.

[0003] In seasonally frozen soil areas, the subgrade structure of high-speed railways faces unprecedented challenges. These challenges come not only from the seasonal freeze-thaw cycle, but also from the strong dynamic effect generated by the high-speed passage of trains, making the filling and structural state of the subgrade extremely complex and changeable. This complex and changeable environment can easily lead to the gradual degradation of subgrade performance, and then cause a series of frost damage problems such as uneven settlement, damage to the sealing layer, and separation of the base, posing a severe challenge to the safe operation of high-speed railways. These problems may not only force trains to reduce their operating speeds, but may also trigger a large amount of maintenance work and resource investment. Therefore, how to scientifically evaluate the evolution trend of the service performance of high-speed railway subgrades in seasonal freeze-thaw environments, and effectively predict and control potential disaster risks to ensure that the subgrade maintains stable performance during its service life of up to a hundred years is a major issue that needs to be urgently addressed in the current field of high-speed railway construction, operation and maintenance in seasonally frozen soil areas.

[0004] As a high-performance geosynthetics, geogrid plays a vital role in the construction of cold-region railways. In cold regions, due to the extremely low winter temperatures, soil and roadbed materials are susceptible to freeze-thaw cycles, resulting in unstable roadbed structures, which may lead to safety hazards such as deformation and cracking of railway lines. In order to meet these challenges, geogrids have become an ideal choice for cold-region railway construction with their excellent physical and mechanical properties. Geogrids are usually made of high-strength, corrosion-resistant materials, such as basalt fibers or high-molecular polymers such as polypropylene and polyvinyl chloride. These materials have excellent frost resistance and durability, and can maintain stable performance under extreme low temperature conditions. In the construction of cold-region railways, geogrids are widely used in key links such as roadbed reinforcement, slope stabilization, and retaining wall construction. By laying geogrids, the bearing capacity and stability of the roadbed can be effectively enhanced, and the damage to the roadbed structure caused by freeze-thaw cycles can be reduced. At the same time, geogrids can also effectively disperse and transfer loads, prevent uneven settlement and cracking of the roadbed, and thus ensure the safety and stability of the railway line.

[0005] In cold-region railway construction, preventing soil from freezing is one of the key issues to ensure stable railway operation. In the existing technology, passive insulation measures are usually adopted, such as laying insulation materials and increasing soil thickness, to slow down the heat loss of the soil, but these methods have limited effects, especially under extremely low temperature conditions, it is still difficult to completely avoid soil freezing.

[0006] In addition, in order to monitor the moisture content of the soil, the existing technology usually uses an independent moisture content monitoring system, which is generally installed separately from the geogrid. This not only increases the complexity of construction, but also makes it difficult to directly link the monitoring data with the reinforcement effect of the geogrid, which is not conducive to the comprehensive control of the soil state. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrids. Self-heating steel pipes are installed on the grid body to heat the soil while monitoring the moisture content of the soil in real time. The roadbed settlement state and frozen soil plasticity characteristics are accurately predicted based on the collected data, thereby providing scientific support for the safety assessment and maintenance of roadbed projects.

[0008] The technical solution of the present invention is:

[0009] A roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrids, including intelligent heating geogrids, data acquisition units, intelligent heating control units, moisture content calculation units and analysis and prediction units;

[0010] The intelligent heating geogrid includes a grid body and a self-heating steel pipe connected to the grid body, the self-heating steel pipe includes a steel pipe body, a heating resistance wire, an optical fiber temperature sensor and an optical fiber strain sensor, the heating resistance wire and the optical fiber temperature sensor are both arranged in the steel pipe body, the optical fiber strain sensor is connected to the outer tube wall of the steel pipe body, and the optical fiber temperature sensor and the optical fiber strain sensor are both connected to the data acquisition unit;

[0011] The data acquisition unit and the heating resistance wire are respectively connected to the intelligent heating control unit, and the intelligent heating control unit starts and controls the heating resistance wire according to the soil temperature value collected by the optical fiber temperature sensor, thereby heating the self-heating steel pipe, and the self-heating steel pipe heats the soil;

[0012] The moisture content calculation unit is connected to the data acquisition unit, and the moisture content calculation unit calculates the temperature rise value of the soil per unit time, and then calculates the moisture content of the soil according to the following formula (1);

[0013] w=a+bΔT+cΔT 2 +dΔT 3 +eΔT 4(1);

[0014] In formula (1), w represents the soil moisture content, ΔT represents the temperature rise value of the soil per unit time collected by the optical fiber temperature sensor, and a, b, c, d, and e are all calibration coefficients of the third-order polynomial shown in formula (1);

[0015] The moisture content calculation unit and the data acquisition unit are both connected to the analysis and prediction unit, and the analysis and prediction unit predicts the roadbed settlement and frozen soil plasticity according to the calculated soil moisture content and the grid strain collected by the optical fiber strain sensor;

[0016] The analysis and prediction unit includes multiple deep learning algorithm models, and the multiple deep learning algorithm models all use roadbed settlement and soil plasticity index as sample labels. The collected roadbed settlement, soil plasticity index, soil moisture content, and grid strain are input into the multiple deep learning algorithm models as real sample data. The multiple deep learning algorithm models are trained separately to obtain trained algorithm models after training. The multiple trained algorithm models perform predictions separately and output intermediate results. Then, the contribution weights of each algorithm model are dynamically adjusted through the weighted fusion method to improve the prediction accuracy of the analysis and prediction unit.

[0017] The self-heating steel pipe is fixedly connected to the top of the grid body by bolts. The steel pipe body is a rectangular pipe. The steel pipe body is filled with high thermal conductivity epoxy resin. The high thermal conductivity epoxy resin is provided with a resistance wire hole and an optical fiber hole. The resistance wire hole is a U-shaped hole. Both ends of the resistance wire hole are located at one end of the steel pipe body. The U-shaped bent end of the resistance wire hole is adjacent to the other end of the steel pipe body. The axis of the optical fiber hole overlaps with the axis of the steel pipe body. The heating resistance wire is arranged in the resistance wire hole. Both ends of the heating resistance wire are led out from one end of the steel pipe body. The optical fiber temperature sensor is arranged in the optical fiber hole, and both ends are led out from the two ends of the steel pipe body respectively. An optical fiber positioning groove is arranged on the outer tube wall of the steel pipe body. Both ends of the optical fiber positioning groove extend along the axial direction of the steel pipe body and pass through both ends of the steel pipe body. The optical fiber strain sensor is a glass fiber encapsulated optical fiber strain sensor, and the optical fiber strain sensor is fixed in the optical fiber positioning groove.

[0018] In the third-order polynomial shown in formula (1), the calibration process of calibration coefficients a, b, c, d, and e is as follows: first, soil is filled in each insulation model box, a self-heating steel pipe and a soil moisture sensor are arranged in the soil, a water pipe is connected to the insulation model box, the heating resistance wire of the self-heating steel pipe is connected to the heating control module outside the insulation model box, the optical fiber temperature sensor of the self-heating steel pipe is connected to the optical fiber demodulator, and the optical fiber demodulator and the soil moisture sensor are connected to the data acquisition unit; before calibration, water is added to the soil in multiple insulation model boxes through the water pipe according to the size of the soil volume in the insulation model box, and the soil moisture sensor collects the moisture content of the soil in the insulation model box in real time, thereby obtaining soil with different moisture contents, and then the heating control module controls the self-heating steel pipe to perform a fixed power and fixed time heating operation on the soil with different moisture contents. After the heating is completed, the soil temperature rise corresponding to the different moisture contents is recorded, and a linear fitting is performed according to the soil temperature rise corresponding to the different moisture contents to obtain the calibration coefficients a, b, c, d, and e.

[0019] b, c, d and e.

[0020] The subgrade settlement is divided into five intervals, specifically:

[0021] Section 1: The roadbed is stable, and the settlement is ≤5mm;

[0022] Interval 2: Slight settlement, 5mm<settlement≤20mm;

[0023] Interval 3: medium settlement, 20mm<settlement≤50mm;

[0024] Interval 4: severe settlement, 50mm<settlement≤100mm;

[0025] Range 5: Extreme settlement, settlement > 100 mm.

[0026] The soil plasticity index is divided into five intervals, specifically:

[0027] Interval 1: non-plastic, soil plasticity index PI = 0;

[0028] Interval 2: Low plasticity, 0 <PI≤10;

[0029] Interval 3: Medium plasticity, 10 <PI≤20;

[0030] Interval 4: High plasticity, 20 <PI≤40;

[0031] Range 5: Extremely high plasticity, PI>40.

[0032] The specific steps of dynamically adjusting the contribution weights of each algorithm model by weighted fusion method are as follows:

[0033] S1. Interval classification accuracy assessment: The target variable, i.e., roadbed settlement or soil plasticity index, is divided into five intervals {Z1, Z2, Z3, Z4, Z5};

[0034] For each algorithm model M i , in the interval Z j Internal calculation classification accuracy A i,j , see the following formula (2) for details:

[0035]

[0036] S2, weighted adjustment: calculate each algorithm model in interval Z j The weight ω i,j , see the following formula (3) for details:

[0037]

[0038] In formula (3), I represents the number of algorithm models;

[0039] The algorithm model is in interval Z j The comprehensive prediction result is P(Z j ), see the following formula (4) for details:

[0040]

[0041] In formula (4), P i (Z j ) represents each algorithm model M i In the interval Z j The prediction results;

[0042] S3. Combine the importance of sample distribution in each interval to obtain the global weight, as shown in the following formula (5):

[0043]

[0044] In formula (5), N j Represents interval Z j The number of samples, N is the total number of samples;

[0045] There are three deep learning algorithm models, namely XGBoost regression model, LightGBM algorithm model and DNN deep neural network model; prediction probability P = ω1P1(x)+ω2P2(x)+ω3P3(x);

[0046] P is the predicted probability of soil settlement or plasticity in a certain interval in the area; p1(x), p2(x), and p3(x) represent the predicted probability of settlement or plasticity in a certain interval predicted by the XGBoost regression model, LightGBM algorithm model, and DNN deep neural network model, respectively, and x represents settlement or soil plasticity.

[0047] The algorithm model dynamically generates interactive features by calculating the nonlinear correlation between soil moisture content, grid strain and target variable. Specifically, the nonlinear correlation between soil moisture content w and grid strain ε and target variable-roadbed settlement S or soil plasticity index PI is calculated first, and interactive features are constructed and generated;

[0048] Use polynomial fitting to calculate the interaction term θ(ω,ε)=α×w×ε+β×w 2 +γ×ε 2 , where α, β, and γ are fitting parameters used to enhance nonlinear effects;

[0049] The interaction term of soil moisture content w-grid strain variable ε is f1 = θ(ω, ε);

[0050] The local grid strain gradient is Used to describe the impact of the local grid strain change rate on the target variable;

[0051] Use the attention mechanism to generate feature weights and define the dynamic weight function for each feature See the following formula (6) for details:

[0052]

[0053] In formula (6), It is the attention score of feature i at time t. The attention mechanism is used to calculate the contribution of each feature to the target variable at a certain moment or time period. Through the attention score, the algorithm model adaptively assigns different weights according to the change pattern of the feature over time, thereby improving the prediction performance.

[0054] Advantages of the present invention:

[0055] (1) The present invention installs a self-heating steel pipe on the grid body. The steel pipe body of the self-heating steel pipe is equipped with a heating resistor and an optical fiber temperature sensor. The optical fiber temperature sensor can monitor the temperature of the soil in real time. When the temperature of the soil is lower than a set threshold, the intelligent heating control unit can control the heating resistor to start heating the soil, which can continuously and evenly transfer heat to the soil, effectively preventing the freezing of the soil and ensuring the stability and safety of the railway structure.

[0056] (2) The moisture content calculation unit of the present invention can calculate the moisture content according to the temperature rise value of the soil per unit time, without the need to set up a separate moisture content monitoring system, which greatly reduces the construction complexity.

[0057] (3) The self-heating steel pipe of the present invention is equipped with an optical fiber strain sensor, and the optical fiber strain sensor is a glass fiber encapsulated optical fiber strain sensor, which is resistant to high temperature and high pressure. The analysis and prediction unit can predict the roadbed settlement and soil plasticity index based on the real-time collected soil moisture content and the grid strain collected by the optical fiber strain sensor. According to the prediction results, it can effectively guide the roadbed reinforcement and maintenance measures to ensure the stability of the roadbed. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a control principle block diagram of the present invention.

[0059] Figure 2 It is a structural schematic diagram of the self-heating steel pipe of the present invention connected to the grid body.

[0060] Figure 3 It is an axial longitudinal sectional view of the self-heating steel pipe of the present invention.

[0061] Figure 4 It is a cross-sectional view of the self-heating steel pipe of the present invention.

[0062] Figure numerals: 1-intelligent heating geogrid, 11-grid body, 12-self-heating steel pipe, 121-steel pipe body, 122-heating resistance wire, 123-optical fiber temperature sensor, 124-optical fiber strain sensor, 125-high thermal conductivity epoxy resin, 126-optical fiber positioning groove, 127-bolt hole, 2-data acquisition unit, 3-intelligent heating control unit, 4-moisture content calculation unit, 5-analysis and prediction unit. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] See Figure 1-Figure 4 , a roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid, including an intelligent heating geogrid 1, a data acquisition unit 2, an intelligent heating control unit 3, a moisture content calculation unit 4 and an analysis and prediction unit 5;

[0065] The intelligent heating geogrid includes a grid body 11 and a self-heating steel pipe 12 connected to the grid body 11, the self-heating steel pipe includes a steel pipe body 121, a heating resistor 122, an optical fiber temperature sensor 123 and an optical fiber strain sensor 124, the steel pipe body 121 is a rectangular tube, the steel pipe body 121 is filled with a high thermal conductivity epoxy resin 125, and the high thermal conductivity epoxy resin 125 is provided with a resistor wire hole and an optical fiber hole, the resistor wire hole is a U-shaped hole, both ends of the resistor wire hole are located at one end of the steel pipe body 121, the U-shaped bending end of the resistor wire hole is adjacent to the other end of the steel pipe body 121, the axis of the optical fiber hole overlaps with the axis of the steel pipe body 121, the heating resistor wire 122 is arranged in the resistor wire hole, and both ends of the heating resistor wire 122 are Leading out from one end of the steel pipe body 122, the optical fiber temperature sensor 123 is arranged in the optical fiber hole, and the two ends are respectively led out from the two ends of the steel pipe body 121, and the outer tube wall of the steel pipe body 121 is activated and etched with an optical fiber positioning groove 126, and the two ends of the optical fiber positioning groove extend along the axial direction of the steel pipe body 121 and penetrate the two ends of the steel pipe body 121, and the optical fiber strain sensor 124 is a glass fiber encapsulated optical fiber strain sensor, and the optical fiber strain sensor 124 is fixed in the optical fiber positioning groove 126, and a bolt hole 127 is provided on the self-heating steel pipe 12, and the self-heating steel pipe 12 is fixedly connected to the top of the grid body 11 by a bolt passing through the bolt hole 127, and the optical fiber temperature sensor 123 and the optical fiber strain sensor 124 are both connected to the data acquisition unit 2;

[0066] The data acquisition unit 2 and the heating resistance wire 122 are respectively connected to the intelligent heating control unit 3. The intelligent heating control unit 3 starts and controls the heating resistance wire 122 according to the soil temperature value collected by the optical fiber temperature sensor 123, thereby heating the self-heating steel pipe 12, and the self-heating steel pipe 12 heats the soil;

[0067] The control principle of the intelligent heating control unit 3 is:

[0068] First, the temperature feedback value F collected by the optical fiber temperature sensor 123 and the set temperature threshold S are subjected to difference calculation to obtain the temperature difference E, that is, E=SF;

[0069] Then, the PID control algorithm is used to control the output to h. The calculation is shown in the following formula (7):

[0070]

[0071] In formula (7), K p , K i and K d are proportional, integral and differential coefficients respectively'

[0072] According to the control output h, the required heating power P=h×η is calculated, where η is the efficiency coefficient of the intelligent heating control unit 3, so as to adjust the heating power of the heating resistor 122 so that the temperature reaches the set temperature threshold S;

[0073] The moisture content calculation unit 4 is connected to the data acquisition unit 2. The moisture content calculation unit 4 calculates the temperature rise value of the soil per unit time, and then calculates the moisture content of the soil according to the following formula (1);

[0074] w=a+bΔT+cΔT 2 +dΔT 3 +eΔT 4 (1);

[0075] In formula (1), w represents the soil moisture content, ΔT represents the temperature rise value of the soil per unit time collected by the optical fiber temperature sensor, and a, b, c, d, and e are all calibration coefficients of the third-order polynomial shown in formula (1);

[0076] In the third-order polynomial shown in formula (1), the calibration process of the calibration coefficients a, b, c, d, and e is as follows: first, soil is filled in each insulation model box, a self-heating steel pipe 12 and a soil moisture sensor are arranged in the soil, a water pipe is connected to the insulation model box, a heating resistor 122 of the self-heating steel pipe 12 is connected to a heating control module outside the insulation model box, an optical fiber temperature sensor 123 of the self-heating steel pipe is connected to an optical fiber demodulator, and the optical fiber demodulator and the soil moisture sensor are connected to a data acquisition unit, which collects soil temperature values ​​and moisture content. rate data; before calibration, according to the size of the soil volume in the insulation model box, water is added to the soil in multiple insulation model boxes through the water pipe, and the soil moisture sensor collects the moisture content of the soil in the insulation model box in real time, so as to obtain soil with different moisture contents, and then the heating control module controls the self-heating steel pipe to perform a fixed power and fixed time heating operation on the soil with different moisture contents. After the heating is completed, the soil temperature rise corresponding to the different moisture contents is recorded, and the linear fitting is performed according to the soil temperature rise corresponding to the different moisture contents to obtain the calibration coefficients a, b, c, d and e;

[0077] The moisture content calculation unit 4 and the data acquisition unit 2 are both connected to the analysis and prediction unit 5. The analysis and prediction unit 5 predicts the roadbed settlement and frozen soil plasticity according to the soil moisture content calculated by the moisture content calculation unit 4 and the grid strain collected by the optical fiber strain sensor 124.

[0078] The analysis and prediction unit 5 includes three deep learning algorithm models, namely, the XGBoost regression model, the LightGBM algorithm model and the DNN deep neural network model. The three deep learning algorithm models all use the roadbed settlement and the soil plasticity index as sample labels, and input the collected roadbed settlement (mm), soil plasticity index PI, soil moisture content w, and grid strain ε as real sample data into the three deep learning algorithm models. The three deep learning algorithm models are trained separately. In the real sample collection stage, 100 learning points are evenly selected for every 100 meters of geogrid, and the corresponding roadbed settlement, soil plasticity index, soil moisture content, and grid strain are recorded. The collected data needs to be cleaned and preprocessed before being input into the three deep learning algorithm models. The cleaning and preprocessing specifically include operations such as outlier removal, missing value filling, and feature normalization to ensure the quality of the input data. After training, the trained algorithm models are obtained. The three trained algorithm models are predicted separately and the intermediate results are output. Then, the contribution weights of each algorithm model are dynamically adjusted through the weighted fusion method to improve the accuracy of the prediction of the analysis and prediction unit.

[0079] The subgrade settlement in the real sample is collected by a settlement observation instrument or a settlement ruler. The soil plasticity index PI in the real sample is obtained by sampling the soil on site, and calculating the PI by testing the liquid limit (LL) and plastic limit (PL) of the soil samples.

[0080] Among them, the roadbed settlement is divided into five intervals, specifically:

[0081] Zone 1: The roadbed is stable, the settlement is ≤5mm, the roadbed has almost no settlement, and the structure is stable;

[0082] Interval 2: Slight settlement, 5mm<settlement≤20mm, slight settlement, little impact on the roadbed;

[0083] Interval 3: Moderate settlement, 20mm<settlement≤50mm, significant settlement, maintenance required;

[0084] Interval 4: Severe settlement, 50mm<settlement≤100mm, which has a significant impact on the roadbed structure and requires reinforcement measures;

[0085] Range 5: Extreme settlement, settlement > 100mm, the roadbed may fail and immediate protective treatment is required.

[0086] The soil plasticity index is divided into five intervals, specifically:

[0087] Interval 1: non-plastic, soil plasticity index PI = 0, frozen soil has no plasticity, mainly sandy soil;

[0088] Interval 2: Low plasticity, 0 < PI ≤ 10, the plasticity of frozen soil is weak and the deformation ability is limited;

[0089] Interval 3: Medium plasticity, 10 < PI ≤ 20, with obvious plasticity and significant deformation;

[0090] Interval 4: High plasticity, 20 < PI ≤ 40, the frozen soil has high plasticity and is prone to large-scale deformation;

[0091] Interval 5: Extremely high plasticity, PI > 40, the frozen soil shows extremely high plasticity, has a very strong deformation ability and is prone to instability.

[0092] The contribution weights of each algorithm model are dynamically adjusted by the method of weighted fusion, and the specific steps are as follows:

[0093] S1. Interval classification accuracy evaluation: The target variable, that is, the subgrade settlement or the soil plasticity index, is divided into 5 intervals {Z1, Z2, Z3, Z4, Z5};

[0094] For each algorithm model M i , the classification accuracy A j is calculated within the interval Z i,j , as shown in the following formula (2):

[0095]

[0096] S2. Weighted adjustment: Calculate the weight ω j of each algorithm model in the interval Z i,j , as shown in the following formula (3):

[0097]

[0098] In formula (3), I represents the number of algorithm models;

[0099] The comprehensive prediction result of the algorithm model in the interval Z j is P(Z j ), as shown in the following formula (4):

[0100]

[0101] In formula (4), P i (Z j represents the prediction result of each algorithm model M i in the interval Z j ;

[0102] S3. Combine the importance of the sample distribution in each interval to obtain the global weight, as shown in the following formula (5):

[0103]

[0104] In formula (5), N j Represents interval Z j The number of samples, N is the total number of samples;

[0105] Prediction probability P = ω1P1(x)+ω2P2(x)+ω3P3(x);

[0106] P is the predicted probability of soil settlement or plasticity in a certain interval in the area; P1(x), P2(x), and P3(x) represent the predicted probability of settlement or plasticity in a certain interval predicted by the XGBoost regression model, the LightGBM algorithm model, and the DNN deep neural network model, respectively, and x represents settlement or soil plasticity;

[0107] Each algorithm model dynamically generates interactive features by calculating the nonlinear correlation between soil moisture content, grid strain and target variable. Specifically, the nonlinear correlation between soil moisture content w and grid strain ε and target variable - roadbed settlement S or soil plasticity index PI is first calculated to construct and generate interactive features.

[0108] Use polynomial fitting to calculate the interaction term θ(ω,ε)=α×w×ε+β×w 2 +γ×ε 2 , where α, β, and γ are fitting parameters used to enhance nonlinear effects;

[0109] The interaction term of soil moisture content w-grid strain variable ε is f1 = θ(ω, ε);

[0110] The local grid strain gradient is Used to describe the impact of the local grid strain change rate on the target variable;

[0111] Use the attention mechanism to generate feature weights and define the dynamic weight function for each feature See the following formula (6) for details:

[0112]

[0113] In formula (6), It is the attention score of feature i at time t. The attention mechanism is used to calculate the contribution of each feature to the target variable at a certain moment or time period. Through the attention score, the algorithm model adaptively assigns different weights according to the change pattern of the feature over time, thereby improving the prediction performance.

[0114] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid is characterized by: It includes intelligent heating geogrid, data acquisition unit, intelligent heating control unit, moisture content calculation unit and analysis and prediction unit; The intelligent heating geogrid includes a grid body and a self-heating steel pipe connected to the grid body. The self-heating steel pipe includes a steel pipe body, a heating resistance wire, an optical fiber temperature sensor and an optical fiber strain sensor. The heating resistance wire and the optical fiber temperature sensor are both arranged in the steel pipe body. The optical fiber strain sensor is connected to the outer wall of the steel pipe body. The optical fiber temperature sensor and the optical fiber strain sensor are both connected to the data acquisition unit. The data acquisition unit and the heating resistance wire are respectively connected to the intelligent heating control unit, and the intelligent heating control unit starts and controls the heating resistance wire according to the soil temperature value collected by the optical fiber temperature sensor, thereby heating the self-heating steel pipe, and the self-heating steel pipe heats the soil; The moisture content calculation unit is connected to the data acquisition unit, and the moisture content calculation unit calculates the temperature rise value of the soil per unit time, and then calculates the moisture content of the soil according to the following formula (1); w=a+bΔT+cΔT 2 +dΔT 3 +eΔT 4 (1); In formula (1), w represents the soil moisture content, ΔT represents the temperature rise value of the soil per unit time collected by the optical fiber temperature sensor, and a, b, c, d, and e are all calibration coefficients of the third-order polynomial shown in formula (1); The moisture content calculation unit and the data acquisition unit are both connected to the analysis and prediction unit, and the analysis and prediction unit predicts the roadbed settlement and frozen soil plasticity according to the calculated soil moisture content and the grid strain collected by the optical fiber strain sensor; The analysis and prediction unit includes multiple deep learning algorithm models, and the multiple deep learning algorithm models all use roadbed settlement and soil plasticity index as sample labels. The collected roadbed settlement, soil plasticity index, soil moisture content, and grid strain are input into the multiple deep learning algorithm models as real sample data. The multiple deep learning algorithm models are trained separately to obtain trained algorithm models after training. The multiple trained algorithm models perform predictions separately and output intermediate results. Then, the contribution weights of each algorithm model are dynamically adjusted through the weighted fusion method to improve the prediction accuracy of the analysis and prediction unit.

2. The roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid according to claim 1 is characterized by: The self-heating steel pipe is fixedly connected to the top of the grid body by bolts. The steel pipe body is a rectangular pipe. The steel pipe body is filled with high thermal conductivity epoxy resin. The high thermal conductivity epoxy resin is provided with a resistance wire hole and an optical fiber hole. The resistance wire hole is a U-shaped hole. Both ends of the resistance wire hole are located at one end of the steel pipe body. The U-shaped bent end of the resistance wire hole is adjacent to the other end of the steel pipe body. The axis of the optical fiber hole overlaps with the axis of the steel pipe body. The heating resistance wire is arranged in the resistance wire hole. Both ends of the heating resistance wire are led out from one end of the steel pipe body. The optical fiber temperature sensor is arranged in the optical fiber hole, and both ends are led out from the two ends of the steel pipe body respectively. An optical fiber positioning groove is arranged on the outer tube wall of the steel pipe body. Both ends of the optical fiber positioning groove extend along the axial direction of the steel pipe body and pass through both ends of the steel pipe body. The optical fiber strain sensor is a glass fiber encapsulated optical fiber strain sensor, and the optical fiber strain sensor is fixed in the optical fiber positioning groove.

3. The roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid according to claim 1 is characterized by: In the third-order polynomial shown in formula (1), the calibration process of the calibration coefficients a, b, c, d, and e is as follows: first, soil is filled in each thermal insulation model box, a self-heating steel pipe and a soil moisture sensor are arranged in the soil, a water pipe is connected to the thermal insulation model box, the heating resistance wire of the self-heating steel pipe is connected to the heating control module outside the thermal insulation model box, the optical fiber temperature sensor of the self-heating steel pipe is connected to the optical fiber demodulator, and the optical fiber demodulator and the soil moisture sensor are connected to the data acquisition unit; before calibration, according to the thermal insulation model, The volume of the soil in the mold box is determined by adding water to the soil in multiple insulation model boxes through a water pipe. The soil moisture sensor collects the moisture content of the soil in the insulation model box in real time, thereby obtaining soil with different moisture contents. Then the heating control module controls the self-heating steel pipe to perform fixed power and fixed time heating operations on the soil with different moisture contents. After the heating is completed, the soil temperature rise values ​​corresponding to different moisture contents are recorded, and linear fitting is performed according to the soil temperature rise values ​​corresponding to different moisture contents to obtain calibration coefficients a, b, c, d and e.

4. The roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid according to claim 1 is characterized by: The roadbed settlement is divided into five intervals, specifically: Section 1: The roadbed is stable, and the settlement is ≤5mm; Interval 2: Slight settlement, 5mm<settlement≤20mm; Interval 3: medium settlement, 20mm<settlement≤50mm; Interval 4: severe settlement, 50mm<settlement≤100mm; Range 5: Extreme settlement, settlement > 100 mm.

5. The roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid according to claim 1 is characterized by: The soil plasticity index is divided into five intervals, specifically: Interval 1: non-plastic, soil plasticity index PI = 0; Interval 2: Low plasticity, 0 <PI≤10; Interval 3: Medium plasticity, 10 <PI≤20; Interval 4: High plasticity, 20 <PI≤40; Range 5: Extremely high plasticity, PI>40.

6. The roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid according to claim 4 or 5, characterized in that: The specific steps of dynamically adjusting the contribution weights of each algorithm model by weighted fusion method are as follows: S1. Interval classification accuracy assessment: The target variable, i.e., roadbed settlement or soil plasticity index, is divided into five intervals {Z1, Z2, Z3, Z4, Z5}; For each algorithm model M i , in the interval Z j Internal calculation classification accuracy A i,j , see the following formula (2) for details: S2, weighted adjustment: calculate each algorithm model in interval Z j The weight ω i,j , see the following formula (3) for details: In formula (3), I represents the number of algorithm models; The algorithm model is in interval Z j The comprehensive prediction result is P(Z j ), see the following formula (4) for details: In formula (4), P i (Z j ) represents each algorithm model M i In the interval Z j The prediction results; S3. Combine the importance of sample distribution in each interval to obtain the global weight, as shown in the following formula (5): In formula (5), N j Represents interval Z j The number of samples, N is the total number of samples; There are three deep learning algorithm models, namely XGBoost regression model, LightGBM algorithm model and DNN deep neural network model; prediction probability P = ω1P1(x)+ω2P2(x)+ω3P3(x); P is the predicted probability of soil settlement or plasticity in a certain interval in the area; P1(x), P2(x), and P3(x) represent the predicted probability of settlement or plasticity in a certain interval predicted by the XGBoost regression model, LightGBM algorithm model, and DNN deep neural network model, respectively, and x represents settlement or soil plasticity.

7. The roadbed settlement and frozen soil plasticity prediction system based on intelligent heating geogrid according to claim 6 is characterized by: The algorithm model dynamically generates interactive features by calculating the nonlinear correlation between soil moisture content, grid strain and target variable. Specifically, the nonlinear correlation between soil moisture content w and grid strain ε and target variable-roadbed settlement S or soil plasticity index PI is calculated first, and interactive features are constructed and generated; Use polynomial fitting to calculate the interaction term θ(ω,ε)=α×w×ε+β×w 2 +γ×ε 2 , where α, β, and γ are fitting parameters used to enhance nonlinear effects; The interaction term of soil moisture content w-grid strain variable ε is f1 = θ(ω, ε); The local grid strain gradient is Used to describe the impact of the local grid strain change rate on the target variable; Use the attention mechanism to generate feature weights and define the dynamic weight function for each feature See the following formula (6) for details: In formula (6), It is the attention score of feature i at time t. The attention mechanism is used to calculate the contribution of each feature to the target variable at a certain moment or time period. Through the attention score, the algorithm model adaptively assigns different weights according to the change pattern of the feature over time, thereby improving the prediction performance.

Images