Whole-process rapid detection and evaluation method for embankment robweg soil material filling and compacting quality

Abstract

The invention discloses a whole-process rapid detection and evaluation method for filling and compacting quality of a dike soil material, which comprises the following steps of: simulating the compacting effect of the soil material under different water contents through rolling process parameters, and determining optimal rolling construction combination parameters under the condition of lowest cost; secondly, constructing a system model for rapidly detecting and evaluating the whole process of the filling and compacting quality of the embankment robba soil material, and inputting related parameters by establishing a long-short-time neural network structure, so that a compaction degree parameter result in the filling process of the on-site embankment robba soil material can be timely, effectively and accurately detected; compared with a traditional compaction degree detection method, the model method is used for evaluating the filling and compaction quality on the construction site, on one hand, operability is quicker, simpler and more convenient, quality control of the whole filling construction process can be covered, the construction period is better shortened, engineering construction is effectively guided, and the construction efficiency is improved. And on the other hand, the accuracy of the measured result is relatively accurate, and the cost of human resources and material resources is greatly reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field, and particularly relates to a method for rapidly detecting and evaluating the whole process of the compaction quality of embankment pond soil materials. Background Technique

[0002] A large amount of soil materials are required during the filling process of embankment projects. However, in actual work, soil material resources are relatively scarce, and most embankment filling design schemes adopt the method of excavating nearby pond soil for embankment construction. However, there are many soil layers excavated in the pond channels involved in the embankment. The water content, clay particle content, and plasticity index of some soil layers are high, which are higher than the requirements of the embankment specifications for soil materials. The compaction quality and detection method of embankment pond soil materials are directly related to the quality and progress of the project.

[0003] The compaction degree of pond soil materials is a key index for controlling the construction quality of embankment filling. The traditional detection method measures the compaction quality by detecting the compaction degree of the filler. During the operation, it is usually necessary to excavate the filled soil in a small area for destructive testing, which is likely to damage the surrounding soil mass; at the same time, in the traditional method, sampling is used for detection, the number of samples is small, and it cannot fully represent the compaction degree of the filled soil within the detection range; the detection period of the compaction degree is long, with a certain lag, which affects the project filling progress to a certain extent. Therefore, only by selecting a rapid and reliable detection technology can the quality and efficiency of embankment filling construction be ensured. At present, there are many detection methods for compaction degree at home and abroad, including direct detection and indirect detection. The commonly used methods are mainly the core cutter method, sand replacement method, and water replacement method, etc. However, the specified core cutter method, sand replacement method, and water replacement method all stay at the result detection and belong to destructive testing. The test operation has certain operations, is time-consuming and laborious, and also consumes a large amount of financial and material resources. Using the compaction degree detection result at the test point to replace the overall compaction effect within the area, the compaction quality of the entire construction area cannot be obtained. It can be seen that the commonly used methods such as the core cutter method, sand replacement method, and water replacement method can no longer meet the current detection requirements of large scale, non-destructive, fast speed, and high precision.

[0004] In the prior art "A Rapid Detection Method for the Compaction Degree of Red Clay CN202110668694.6", the PFWD device is used to obtain the load and displacement time - history data, and a convolutional neural network is established to obtain the compaction degree information of red clay. However, this method only targets the test area of the PFWD device, which only represents a local area. Taking a point to represent the whole, it is impossible to obtain the compaction degree information of the red clay in the entire construction area, and there are certain limitations for engineering quality control. In addition, in "A Rapid Evaluation Method for Subgrade Dynamic Elastic Modulus and Compaction Degree Based on Improved PFWD CN202311036899.8", based on the field measured data of dynamic elastic modulus and compaction degree, the corresponding required compaction degree and dynamic elastic modulus are calculated by back - calculation according to the optimal regression equation. The field measured data of compaction degree and the field measured data of dynamic elastic modulus are respectively compared with the required compaction degree and the required dynamic modulus to evaluate the construction quality of the road subgrade. This technical method still has certain limitations and is only limited to the scope of the tests conducted. How to quickly, accurately, and effectively obtain the compaction quality of the embankment filling project's Hong soil and control the quality of the entire construction process of the embankment filling project is a problem that must be solved in the current compaction of the Hong soil embankment. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for rapid detection and evaluation of the whole - process compaction quality of embankment Hong soil filling to solve the above problems.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A method for rapid detection and evaluation of the whole - process compaction quality of embankment Hong soil filling, the specific steps are as follows:

[0007] S1: Sampling and testing the soil before the project to obtain various parameters of the soil, and the specific parameters include particle size distribution, liquid - plastic limit, maximum dry density, and optimum moisture content;

[0008] S2: Carry out the rolling construction process, which is divided into several test small blocks with the same shape and size according to different paving thicknesses and rolling machines. The paving thickness of each test small block is the same, and there is a transition section between the test sections;

[0009] Perform different rolling passes on different test small blocks and roll with different filling thickness parameters to obtain the soil settlement and soil compaction degree during the rolling process;

[0010] S3: Determine the best filling construction combination parameters:

[0011] (1). According to the different test small blocks in S2, sample a data point p(x) = p(a, b, c, d, e), where the data point information corresponds to the rolling passes a, filling thickness b, in - zone soil moisture content c, compaction degree d, and cost e. The control index for compaction degree is not less than 95%;

[0012] (2). Set up the generation network G(x), and randomly generate a large number of sample data p g (x) according to different rolling passes and filling thickness parameters, with a uniform distribution;

[0013] (3). Set up the discriminant network D(x) to discriminate the probability of the sample data;

[0014] (4). Adopt the min-max loss function V(D, G) to make the sample data generated by the above generation network as qualified as possible, and make the discriminant network distinguish whether the data is qualified as much as possible, so as to obtain the best sample data under the condition of the lowest cost;

[0015] S4: After the project is completed, the detection and evaluation of the embankment are carried out, and the plane point coordinates are set according to the area during the actual rolling process; through the setting of the coordinates, the coordinate parameters are given to the soil settlement amount S n , the Rayleigh wave velocity V n , the dynamic elastic modulus E n , the soil compaction degree D n ;

[0016] Through the long short-term neural network structure, the above-collected soil settlement amount S n and the soil compaction degree D n are extracted, and then compared with the best sample data of the on-site soil obtained through S1, S2, and S3 calculations to judge whether it is qualified. It is realized that only the soil used needs to be detected in the early stage of a project, and there is no need for multiple sampling detections after the project is completed. The present invention has an overall reference, and only needs to make a comparison and judgment, without the need to detect each place.

[0017] Preferably, the judgment rule of the discriminant network D(x) in S3 is: the larger or smaller the value, the less real the result is, and the closer the value is to 2 / 3, the more real the result is, that is, the sample data is qualified;

[0018] The above formula of D(x) is:

[0019] p*(x) is the qualified sample data of historical projects.

[0020] Preferably, the formula of the min-max loss function V(D, G) in S3 is: V(D,G) = log(D(x)) + log(1 - D(G(x));

[0021] Specifically is a maximum-minimum problem, and its optimal solution is the best rolling passes, laying thickness, and water content of the embankment soil under the condition of the lowest cost.

[0022] Preferably, the calculation formula of the cost e in S3 is: e = ξ1e1 + ξ2e2 + ξ3e3 + ξ41e4 + ξ51e5;

[0023] Among them, ξ1 is the weight coefficient of the rolling machinery, e1 is the cost of the rolling machinery, ξ2 is the weight coefficient of the on-site labor force, e2 is the cost of the on-site labor force, ξ3 is the weight coefficient of the rolled soil material, e3 is the cost of the rolled soil material, ξ4 is the weight coefficient of the test and inspection, e4 is the cost of the test and inspection during the rolling process, ξ5 is the weight coefficient of the safety and environmental measures during the rolling process, and e5 is the cost of the safety and environmental measures during the rolling process.

[0024] Preferably, in S4, the soil compression settlement amount and the soil compaction degree at the coordinate position during the rolling process are obtained. The soil compaction degree is measured by the ring knife method required by the specification, and the Rayleigh wave velocity and the dynamic elastic modulus value are also measured.

[0025] Preferably, the long short-term neural network structure specifically includes: input parameters, forget gate, input gate, cell update, output gate and output parameters. The information m obtained in S4 n (x n , y n , S n , V n , E n , D n ) is input into the long short-term neural network and the final compaction quality is output.

[0026] In S4, the site is coordinateized. x n , y n are the point coordinates respectively, S n is the soil settlement amount, V n is the Rayleigh wave velocity of the multi-state transient surface wave Rayleigh wave, En is the dynamic elastic modulus, and Dn is the soil compaction degree.

[0027] Preferably, an activation function is used to determine which data in the input sample parameters need to be retained. The output result of the activation function is between 0 and 1. The closer it is to 1, the more it is retained. The closer it is to 0, the less it is retained, that is, forgotten, which determines how much information in the previous sample data is retained in the current calculation.

[0028] Among them, the activation formula of the above forget gate is: f n =σ(W f *[y n-1 , m n +b f );

[0029] σ is the activation function, Wf is the weight matrix, bf is the bias term, yn-1 is the hidden state at the previous moment, and the activation function is:

[0030] Preferably, the input gate operation is as follows: determine which value information in the current input mn and the previous hidden state yn-1 needs to be incorporated into the cell state of the neural network structure, avoid long-term memory, determine the modification and update of the cell state, and calculate the possible new information data that may be added after the cell state is updated according to the cell formula;

[0031] Among them, the activation formula of the above input gate is: i n = σ(W i *[y n-1 , m n +b i );

[0032] σ is the activation function, Wi is the weight matrix, bi is the bias term, and the activation function is:

[0033] Preferably, the cell formula of the input gate is:

[0034] tanh is the activation function, WC is the weight matrix, bC is the bias term, and the activation function is:

[0035] Cell update operation: The in obtained through the input gate operation and the updated cell state are used to update the cell state in cooperation with the forget gate. The forget gate controls the retention ratio of old memories and the input gate controls the addition ratio of new memories, enabling this model structure to simultaneously forget some old information and learn new information, maintaining the dynamic balance of memory, and also ensuring the collaborative update of data samples based on previous data samples.

[0036] The cell state update formula in the above neural network is:

[0037] Preferably, the output gate operation: Integrate the previous forget gate and input gate, and cooperate to update the cell output model parameters to evaluate the whole process rapid detection of the filling and compaction quality of the embankment Hong soil material;

[0038] The formula of the above output gate is: o n = σ(W o *[y n-1 , m n +b o );

[0039] σ is the activation function, Wo is the weight matrix, bo is the bias term, and the activation function is:

[0040] The output formula is: y n = o n *tanh(C n );

[0041] tanh activation function:

[0042] Output parameter: By outputting the value of the compactness yn, the compactness value calculated by the model is output, and then the compactness quality of the embankment Hong soil filling is evaluated by combining the best sample data obtained from S3.

[0043] Technical effects and advantages of the present invention: By simulating the compacting effect of soil materials under different moisture contents through rolling process parameters, the optimal rolling construction combination parameters under the lowest cost are determined; secondly, a whole-process rapid detection and evaluation system model for the compacting quality of embankment Hong soil filling is constructed. By establishing a long-short-term neural network structure and inputting relevant parameters, the compactness parameter results during the on-site embankment Hong soil filling process can be measured in a timely, effective and accurate manner; when using the model method of the present invention to evaluate the compacting quality of the construction site, compared with the traditional compactness detection method, on the one hand, the operation is faster, simpler and more convenient, and it can cover the whole-process quality control of the filling construction, better saving the construction period and effectively guiding the engineering construction. On the other hand, the measured results are also more accurate in terms of accuracy, and it also greatly reduces the human and material costs. Compared with the traditional core cutter method, water injection method and sand replacement method, the influence of human operation error is ignored, the operation is simple, and the human and material costs are better saved. By simulating the compacting effect of soil materials under different moisture contents through rolling process parameters based on deep learning algorithms, the optimal rolling construction combination parameters under the lowest cost are determined, saving the construction input cost, having a certain scientific nature, controlling the quality of the whole process of embankment filling construction, the method is simple to operate, more effectively ensuring the construction quality of the embankment, not only ensuring the speed of the detection work but also ensuring the accuracy of the detection results, and at the same time better ensuring the construction progress and saving the construction period. Brief Description of the Drawings

[0044] Figure 1 It is a schematic diagram of the construction site division of the rolling process of the present invention;

[0045] Figure 2 It is a schematic diagram of the method model structure of the present invention;

[0046] Figure 3 It is a schematic diagram of the process flow of the whole-process rapid detection and evaluation system and method for the compacting quality of embankment Hong soil filling of the present invention;

[0047] Figure 4 It is a schematic diagram of the on-site test detection steps of the rolling test of the present invention. Detailed Embodiments

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] The present invention provides a Figures 1-4 rapid full-process detection and evaluation method for the filling and compaction quality of embankment Hong soil materials as shown in

[0050] Embodiment 1: Step 1: Take samples of Hong soil materials at the construction site and carry out the rolling construction process;

[0051] First, analyze the rolling test construction process. The site area is 32m×22m; it is divided into 4 test small blocks of 10m×15m according to different paving thicknesses and rolling machines. The paving thickness of each test small block is 30cm, and a 2m transition section is set between the test sections; conduct tests on the physical parameters of the soil properties of the Hong soil materials to obtain physical parameters such as the soil particle gradation, liquid-plastic limit, maximum dry density, and optimum moisture content. According to the requirement in the specification that the allowable deviation of the moisture content of the filling soil materials from the optimum moisture content is ±3%, combined with the results of the compaction test, the moisture content of the soil materials in Area A-1 and Area A-2 is the optimum moisture content of compaction, the moisture content of the soil materials in Area B-1 is the optimum moisture content of compaction - 2%, and the moisture content of the soil materials in Area B-2 is the optimum moisture content of compaction + 2%, as shown in Figure 1 .

[0052] Step 2: Intelligently optimize the filling construction ratio parameters of the Hong soil materials to determine the best filling construction combination parameters; combine the rolling test in Step 1 above, take different rolling passes and filling thickness parameters for rolling, obtain the soil settlement and soil compaction degree during the rolling process, and construct the best rolling construction combination parameter model.

[0053] The relationships among different rolling passes, paving thicknesses, and moisture contents will definitely result in different inputs of human resources, material resources, and costs. The larger the number of rolling passes and the thinner the thickness, it will definitely be qualified, but the resulting input of human resources, material resources, and costs will also be greater; based on the deep learning algorithm, simulate the compaction effects of soil materials under different costs, rolling passes, paving thicknesses, and moisture contents, analyze parameters such as soil particle gradation, liquid-plastic limit, maximum dry density, and optimum moisture content, combine historical engineering data and specification requirements, combine the results of rolling with different rolling passes and filling thickness parameters in the actual process, and comprehensively consider the costs of machinery, manpower, soil materials, test detection, and safety environment measures to determine the best rolling construction combination parameters under the lowest cost.

[0054] The specific steps are as follows: For the above different regions A-1, A-2, B-1, and B-2, sample a data point p(x) = p(a, b, c, d, e). The data point information corresponds to the number of rolling passes a, the filling thickness b, the moisture content c of the soil property within the region, the compaction degree d, and the cost e. The control index for the compaction degree is not less than 95%.

[0055] The calculation formula for the cost e is: e = ξ1e1 + ξ2e2 + ξ3e3 + ξ41e4 + ξ51e5;

[0056] ξ1 is the weight coefficient of the rolling machinery, e1 is the cost of the rolling machinery, ξ2 is the weight coefficient of the on-site labor force, e2 is the cost of the on-site labor force, ξ3 is the weight coefficient of the rolled soil material, e3 is the cost of the rolled soil material, ξ4 is the weight coefficient of the test and inspection, e4 is the cost of the test and inspection during the rolling process, ξ5 is the weight coefficient of the safety and environmental measures during the rolling process, and e5 is the cost of the safety and environmental measures during the rolling process.

[0057] Set up the generation network G(x), and randomly generate a large number of sample data pg(x) according to different numbers of rolling passes and filling thicknesses, which are uniformly distributed.

[0058] Set up the discriminant network D(x), which outputs a probability D(X) between 0 and 1 to discriminate the probability of the sample data. The larger or smaller the value, the less real the result is, and the closer the value is to 2 / 3, the more real the result is (that is, p(x) and P * (x) are approximately equal in the case of the same soil, and at this time P g (X) is closer to p(x) or P * (x), it means the result is more real), that is, the sample data is qualified.

[0059] The above formula for D(x) is:

[0060] p*(x) is the qualified sample data of historical projects.

[0061] Adopt the min-max loss function V(D, G) to make the sample data generated by the above generation network as qualified as possible, and make the discriminant network distinguish whether the data is qualified as much as possible, so as to obtain the best sample data under the lowest cost.

[0062] The above formula for the min-max loss function V(D, G) is: V(D,G) = log(D(x)) + log(1 - D(G(x));

[0063] Specifically It is a max-min problem. Find its optimal solution, which is the best number of rolling passes, laying thickness, and moisture content of the embankment soil under the lowest cost.

[0064] Step 3: Determine the optimal number of rolling passes, laying thickness, and moisture content of Hong soil for the levee filling construction, and then start the levee filling construction. During the rolling process, the plane coordinates are determined according to the area, and the soil compression settlement and soil compaction degree at the coordinate positions during the rolling process are obtained. The soil compaction degree is tested by the core cutter method as required by the specifications, and the Rayleigh wave velocity and dynamic elastic modulus value are also tested. Through the coordinate setting, the soil settlement Sn, Rayleigh wave velocity Vn, dynamic elastic modulus En, and soil compaction degree Dn are assigned to the coordinate parameters.

[0065] Construct a rapid detection and evaluation system model for the whole process of the compaction quality of Hong soil for levee filling, and establish a long short-term neural network structure, which includes: input parameters, forget gate, input gate, cell update, output gate, and output parameters. The process is shown in the appendix Figure 2 , and the specific operation of the long short-term neural network structure is as follows:

[0066] Input parameters: Input the above information mn(xn, yn, Sn, Vn, En, Dn) into the model;

[0067] The site is coordinated, where xn and yn are the point coordinates, Sn is the soil settlement, Vn is the Rayleigh wave velocity of the multi-state transient surface wave, En is the dynamic elastic modulus, and Dn is the soil compaction degree.

[0068] Forget gate operation: Determine which data in the input sample parameter state needs to be retained through an activation function. The output result of the activation function is between 0 and 1. The closer it is to 1, the more it is retained; the closer it is to 0, the less it is retained, that is, forgotten, which determines how much information in the previous sample data is retained in the current calculation.

[0069] Among them, the activation formula of the above forget gate is: f n =σ(W f *[y n-1 ,m n +b f );

[0070] σ is the activation function, Wf is the weight matrix, bf is the bias term, yn-1 is the hidden state at the previous moment, and the activation function is: The parameter X in the activation function here is the following in the above formula: W i *[y n-1 ,m n +b i , and the principle of the substitution items of the activation function parameter X described below is the same

[0071] Input gate operation: Determine which valuable information in the current input mn and the hidden state yn-1 at the previous moment needs to be incorporated into the cell state of the neural network structure, avoid long-term memory, determine the modification and update of the cell state, and calculate the possible new information data that may be added after the cell state is updated according to the cell formula.

[0072] Among them, the activation formula of the above input gate is: i n = σ(W i *[y n-1 , m n +b i );

[0073] σ is the activation function, Wi is the weight matrix, bi is the bias term, and the activation function is:

[0074] The cell formula of the above input gate is:

[0075] tanh is the activation function, WC is the weight matrix, bC is the bias term, and the activation function is:

[0076] Cell update operation: in obtained through the input gate operation and the updated cell state cooperate with the forget gate to update the cell state. The forget gate controls the retention ratio of old memories and the input gate controls the addition ratio of new memories, enabling this model structure to simultaneously forget some old information and learn new information, maintaining the dynamic balance of memory, and also ensuring the collaborative update of data samples based on previous data samples.

[0077] The cell state update formula in the above neural network is:

[0078] Output gate operation: Integrate the previous forget gate and input gate, and cooperate to update the cell output model parameters to evaluate the whole process rapid detection of the filling and compaction quality of Hong soil for the dike.

[0079] The formula of the above output gate is: o n = σ(W o *[y n-1 , m n +b o );

[0080] σ is the activation function, Wo is the weight matrix, bo is the bias term, and the activation function is:

[0081] The output formula is: y n = o n *tanh(C n );

[0082] tanh activation function:

[0083] Output parameter: By outputting the value of the compaction degree yn, the compaction degree value calculated by the model can be output, and the filling and compaction quality of Hong soil for the dike can be evaluated.

[0084] Specific implementation method: In the construction of filling and compaction of Hong soil for a newly built dike, all the filled Hong soil is taken from the river beach surface, and the water content of the soil is strictly controlled at the material yard. The backfill soil meets the specifications and design requirements, and the water content is controlled at the optimal water content. Specific measures such as loosening, watering, and sunning are adopted. When the natural water content of the soil at the material yard is greater than or less than the water content for construction filling, the water content of the soil at the material yard is adjusted according to the soil excavation method, the loading, unloading, and transportation process, and meteorological conditions. The adjustment method mainly focuses on sunning or adding water. Compaction tests are carried out on the Hong soil. According to the results of indoor compaction tests, the optimal water content of the soil sample for this rolling test is determined to be 17.4%, and the maximum dry density is 1.64 g / cm 3 .

[0085] First, the intelligent optimization of the filling construction ratio parameters of the Hong soil is carried out to determine the best filling construction combination parameters. Control parameters are set, and different rolling passes and filling thickness parameters are adopted for rolling. During the rolling process, the plane is coordinateized by area to obtain the soil compression settlement and soil compaction degree at the coordinate positions during the rolling process. The soil compaction degree is tested by the core cutter method required by the specifications, and the Rayleigh wave velocity and dynamic elastic modulus value are also tested. Through the coordinate setting, the coordinate parameters of the soil settlement Sn, Rayleigh wave velocity Vn, dynamic elastic modulus En, and soil compaction degree Dn are given, and the soil compaction degree Dn is mn(xn, yn, Sn, Vn, En, Dn) (n = 1, 2, 3 ···, n). Among them: Multiple transient surface wave method tests for Rayleigh wave velocity are carried out in the rolling test area, with an area of 15 * 10 m. 4 survey lines are arranged along the advancing direction of the rolling machine, with a spacing of 2 m, and 7 survey lines are arranged in the cross-sectional direction, with a spacing of 2 m; Core cutter method detection: 4 survey lines are arranged along the advancing direction of the rolling machine, and 7 survey lines are arranged in the cross-sectional direction, with a spacing of 2 m. Core cutter method detection is carried out at equal intervals of 2 m on the survey lines; The dynamic elastic modulus test is arranged with 6 points around the core cutter detection points, 2 times for each point, a total of 12 times, as shown in the appendix Figure 4 . Through the rolling test detection, a large amount of mn data is obtained as training samples, and the test results are given to them for training. Through model training and using the error discovery propagation algorithm to fine-tune the weights, the optimized model is obtained. At any point on the site where no test detection has been carried out, test detections are carried out to obtain mn+1 to mn+20 points as test samples to test and debug the model.

[0086] The dike is filled according to the best construction rolling combination parameters, and 3 site areas are tested. Each site is filled with 3 layers, and the filling thickness of each layer of Hong soil is 30 cm. We take samples for comparison, and the comparison results are shown in Table 1 below:

[0087] Table 1 Summary table of actual parameters and model calculation results (the maximum dry density is 1.64 g / cm 3 )

[0088]

[0089] It can be seen from the table that the relative errors are all less than 0.05 (5%), realizing the rapid detection of the filling quality of the embankment Hong soil materials, effectively controlling the compaction quality of the excavated soil materials in the Hong channel, being able to represent the compaction degree of the fill within the construction filling range, and realizing the rapid detection and evaluation of the whole construction process; it shows that the method of the present invention can be applied to the actual application of this project, and the results obtained by the method have high accuracy, whole-process control, and simple operation.

[0090] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A rapid detection and evaluation method for the whole process of the filling and compaction quality of embankment pond soil materials, characterized in that: The specific steps are as follows: S1: Sampling and testing the soil materials before the project to obtain various parameters of the soil materials. The specific parameters include particle size distribution, liquid-plastic limit, maximum dry density, and optimum moisture content; S2: Carry out the rolling construction process. According to different paving thicknesses and rolling machinery, it is divided into several test small blocks with the same shape and size. The paving thickness of each test small block is the same, and there is a transition section between the test sections; Perform different rolling passes on different test small blocks and roll with different filling thickness parameters to obtain the soil settlement and soil compaction degree during the rolling process; S3: Determine the best filling construction combination parameters: (1). According to the different test small blocks in S2, sample a data point p(x) = p(a, b, c, d, e). The data point information corresponds to the rolling pass number a, filling thickness b, in-zone soil moisture content c, compaction degree d, and cost e. The control index of the compaction degree is not less than 95%; (2) Set up the generation network G(x), and randomly generate a large number of sample data p g (x) according to different rolling passes and filling thickness parameters, with a uniform distribution; (3). Set up a discriminant network D(x) to discriminate the probability of the sample data; (4). Adopt the minimax loss function V(D, G) to make the sample data generated by the above generation network as qualified as possible, and make the discriminant network distinguish whether the data is qualified as much as possible, and obtain the best sample data under the lowest cost; S4: Detection and evaluation of the embankment after the project is completed, and plane point coordinate conversion according to the area during the actual rolling process; through the coordinate conversion setting, the coordinate parameters are given for the soil settlement amount S n , Rayleigh wave velocity V n , dynamic elastic modulus E n , soil compaction degree D n ; Through the long short-term neural network structure, the collected soil settlement S n and soil compaction degree D n are extracted, and then compared with the optimal sample data of the on-site soil calculated through S1, S2, and S3 to determine whether it is qualified.

2. A method for rapid detection and evaluation of the whole process of the filling and compaction quality of embankment hydraulic fill materials according to claim 1, characterized in that: The judgment rule of the discriminant network D(x) in S3 is: The larger or smaller the value, the less real the result is. The closer the value is to 2 / 3, the more real the result is, that is, the sample data is qualified; The above D(x) formula is as follows: p * (x) is the qualified sample data of historical projects.

3. A method for rapid detection and evaluation of the whole process of the filling and compaction quality of embankment fine soil materials according to claim 1, characterized in that: The formula of the minimax loss function V(D, G) in S3 is: V(D,G) = log(D(x)) + log(1 - D(G(x)); Specifically It is a maximum-minimum problem. Finding its optimal solution means finding the best rolling passes, laying thickness, and moisture content of the embankment soil under the condition of the lowest cost.

4. A method for rapid detection and evaluation of the whole process of the filling and compaction quality of embankment soil materials according to claim 1, characterized in that: The calculation formula of the cost e in S3 is: e = ξ1e1 + ξ2e2 + ξ3e3 + ξ41e4 + ξ51e5; The above ξ1 is the weight coefficient of the rolling machinery, e1 is the cost of the rolling machinery, ξ2 is the weight coefficient of the on-site labor force, e2 is the cost of the on-site labor force, ξ3 is the weight coefficient of the rolled soil materials, e3 is the cost of the rolled soil materials, ξ4 is the weight coefficient of the test and detection, e4 is the cost of the test and detection during the rolling process, ξ5 is the weight coefficient of the safety and environmental measures during the rolling process, and e5 is the cost of the safety and environmental measures during the rolling process.

5. A method for rapid detection and evaluation of the whole process of the filling and compaction quality of embankment fine soil materials according to claim 1, characterized in that: In S4, obtain the soil compression settlement and soil compaction degree at the coordinate position during the rolling process. The soil compaction degree is tested by the ring knife method required by the specification, and the Rayleigh wave velocity and the dynamic elastic modulus value are also tested.

6. A method for rapid detection and evaluation of the whole process of the filling and compaction quality of embankment fine soil materials according to claim 5, characterized in that: The long short-term neural network structure specifically includes: input parameters, forget gate, input gate, cell update, output gate, and output parameters. The information m obtained in S4 n (x n ,y n ,S n ,V n ,E n ,D n ) is input into the long short-term neural network and finally outputs the compaction quality; In S4, the site is coordinateized, where x n , y n are the point coordinates respectively, S n is the soil settlement, V n is the Rayleigh wave velocity of the multi-state transient surface wave, En is the dynamic elastic modulus, and Dn is the soil compaction degree.

7. A method for rapid detection and evaluation of the whole process of the filling and compaction quality of embankment fine soil materials according to claim 6, characterized in that: The forgetting gate operation is as follows: Determine which data in the input sample parameter state needs to be retained through an activation function. The output result of the activation function is between 0 and 1. The closer it is to 1, the more it is retained. The closer it is to 0, the less it is retained, that is, forgotten, and determine how much information in the previous sample data is retained in this calculation; Among them, the above forgetting gate activation formula is: f n =σ(W f *[y n-1 ,m n +b f ); σ is the activation function, Wf is the weight matrix, bf is the bias term, and yn-1 is the hidden state at the previous moment. The activation function is:

8. A method for rapid detection and evaluation of the whole process of the filling and compaction quality of embankment fine soil materials according to claim 7, characterized in that: The input gate operation is as follows: Determine which value information in the current input mn and the previous hidden state yn-1 needs to be incorporated into the cell state of the neural network structure, avoid long-term memory, determine the modification and update of the cell state, and calculate the possible new information data that may be added after the cell state is updated according to the cell formula; Among them, the activation formula of the above input gate is: i n = σ(W i * [y n-1 , m n + b i ) σ is the activation function, Wi is the weight matrix, and bi is the bias term. The activation function is as follows:

9. A method for rapid detection and evaluation of the whole process of the filling and compaction quality of embankment backfill soil according to claim 8, characterized in that: The cell formula of the input gate is as follows: The activation function is tanh, WC is the weight matrix, bC is the bias term, and the activation function is as follows: Cell update operation: The in obtained through the input gate operation and the updated cell state cooperate with the forget gate to update the cell state. The forget gate controls the retention ratio of old memories and the input gate controls the addition ratio of new memories, enabling this model structure to simultaneously forget some old information and learn new information, maintaining the dynamic balance of memories, and also ensuring the collaborative update of data samples based on previous data samples. The cell state update formula in the above neural network is as follows:

10. A method for rapid detection and evaluation of the whole process of filling and compaction quality of embankment hydraulic fill materials according to claim 9, characterized in that: Output gate operation: Integrate the previous forget gate and input gate, and the cell output model parameters updated collaboratively to evaluate the rapid detection of the whole process of the filling and compaction quality of the embankment Hong soil material; The above output gate formula is: o n = σ(W o * [y n-1 , m n + b o ); σ is the activation function, Wo is the weight matrix, and bo is the bias term. The activation function is: The output formula is: y n = o n *tanh(C n ); tanh activation function: Output parameters: By outputting the value of the compaction degree yn, the compaction degree value calculated by the model is output, and then the filling and compaction quality of the embankment Hong soil material is evaluated by combining the optimal sample data obtained in S3.

Images