Curing agent mix proportion design method for nomadic flow-state stabilized soil production line

Abstract

The invention provides a curing agent mix proportion design method for a nomadic type flow state solidified soil production line, and belongs to the technical field of computational chemistry, the method comprises the following steps: collecting an original soil sample of a to-be-borrowed area such as a beach culture sludge area, carrying out basic property determination, and establishing a multilayer neural network original soil evaluation system to carry out quantitative evaluation on the adaptability of original soil; constructing a five-layer neural network curing agent constraint system to determine a ratio range; designing a test scheme by adopting a Latin hypercube sampling method; establishing a multi-layer neural network performance evaluation system and a reaction mechanism system for comprehensive evaluation; preparing a standard test block to determine engineering performance; a matching scheme is verified through production line trial production, an engineering performance verification matrix is established, and a fuzzy comprehensive evaluation method is adopted to ensure that the performance meets engineering requirements. According to the scheme, the systematic design and optimization of the curing agent ratio are realized through the synergistic effect of a plurality of neural network models.

Description

Technical Field

[0001] The invention belongs to the technical field of computational chemistry, and in particular relates to a curing agent mix ratio design method for a nomadic fluidized solidified soil production line. Background Art

[0002] Soil solidification treatment in beach aquaculture silt areas is an important engineering technology to improve the value of land use. Among them, fluidized soil solidification technology is widely used because of its convenient construction and low cost. The traditional solidifier ratio design method mainly relies on empirical formulas and orthogonal experiments, and is evaluated and optimized through single performance indicators such as strength and permeability. Currently, commonly used solidifiers include inorganic materials such as quicklime, slaked lime, cement, fly ash, and various admixtures. These materials improve soil properties through mechanisms such as hydration reaction, ion exchange, and gelation.

[0003] However, the traditional curing agent ratio design method has obvious limitations. First, there is a lack of systematic evaluation of the original soil characteristics, and it is impossible to accurately reflect the correlation between the physical and chemical properties of the original soil and the curing effect. Secondly, the curing agent ratio optimization process relies too much on experimental data and fails to establish an effective theoretical guidance model. Thirdly, the performance evaluation system is single and it is difficult to fully reflect the engineering performance and durability of the cured soil. In addition, for special soils with high organic matter and high salt content such as beach aquaculture silt, traditional methods are difficult to accurately evaluate the impact of organic matter on the curing effect, and cannot effectively predict the chemical reaction mechanism during the curing process.

[0004] In actual engineering applications, due to the lack of systematic evaluation and optimization methods, the design of curing agent ratios often requires a lot of experimental exploration, which is not only time-consuming and labor-intensive, but also difficult to ensure the scientificity and reliability of the ratio scheme. Especially for engineering projects with complex components and high performance requirements, traditional methods are difficult to achieve accurate design and optimization of curing agent ratios, leading to problems such as unstable engineering quality and waste of resources. Therefore, it is urgent to establish a systematic evaluation and optimization method based on multi-layer neural networks. In summary, there is a technical problem in the prior art that the curing agent ratio of fluidized solidified soil is not optimized. Summary of the invention

[0005] In view of this, the present invention provides a curing agent mix ratio design method for a nomadic fluidized solidified soil production line, which can solve the technical problem in the prior art that the curing agent mix ratio of the fluidized solidified soil is not optimized to a high degree.

[0006] The present invention is implemented as follows: The present invention provides a curing agent mix ratio design method for a nomadic fluidized solidified soil production line, comprising the following steps: collecting original soil samples, and measuring the moisture content, organic matter content, particle size distribution, pH, cation exchange capacity, and salt content of the original soil samples; establishing a multi-layer neural network original soil evaluation system, constructing an L-th layer initial convolution kernel through a Gaussian kernel function, the L-th layer initial convolution kernel containing M characteristic parameters, and the M characteristic parameters containing soil quality indicators, chemical indicators, and physical indicators; establishing a five-layer neural network curing agent constraint system, constructing a first layer initial convolution kernel through a least squares method, and the characteristic parameters of the first layer initial convolution kernel containing quicklime mass fraction, slaked lime mass fraction, cement mass fraction, fly ash mass fraction, and admixture mass fraction; establishing a multi-layer neural network performance evaluation system, and using a local linear embedding algorithm to construct an L-th layer performance initial convolution kernel and an L-th layer performance equivalent convolution kernel; determining the optimal curing agent ratio scheme based on the multi-layer neural network performance evaluation system, and verifying it through production line trial production.

[0007] Among them, the first initial parameter of the Lth layer initial convolution kernel in the multi-layer neural network original soil evaluation system is the reference parameter K, and the reference parameter K is a non-zero number. The quotient of the M feature parameters and the reference parameter K is calculated by the radial basis function algorithm to obtain the Lth layer equivalent convolution kernel. When the quotient is greater than the first threshold, the M feature parameters are reduced by m times. When the quotient is less than the second threshold, the M feature parameters are expanded by n times. L represents the number of layers, L∈[3,8].

[0008] Among them, in the five-layer neural network curing agent constraint system, the mass fraction of the admixture includes the mass fraction of the water reducer, the mass fraction of the early strength agent, the mass fraction of the swelling reducer, and the mass fraction of the organic stabilizer. The characteristic parameters of the first layer initial convolution kernel are processed by the regularization method, and the parameter quotient is calculated based on the radial basis function algorithm to obtain the first layer equivalent convolution kernel.

[0009] Among them, the curing agent component is mixed with the standardized soil sample according to a ratio, and the Latin hypercube sampling method is used to design test schemes for the characteristic parameters based on the five-layer neural network curing agent constraint system. Multiple groups of test samples are prepared according to the moisture content gradient, and the moisture content gradient interval is 5%. The slump, initial setting time, final setting time, pH, and organic stability index of the test samples are measured.

[0010] Among them, the L-th layer performance initial convolution kernel includes fluidity index, coagulation time index, chemical stability index, and organic matter decomposition index. The L-th layer performance equivalent convolution kernel is obtained by calculating the quotient of the L-th layer performance initial convolution kernel and the benchmark parameter K through the radial basis function algorithm.

[0011] Among them, a multi-layer neural network reaction mechanism system is established, and the principal component analysis method is used to construct the initial convolution kernel of the reaction. The initial convolution kernel of the reaction includes hydration reaction indicators, ion exchange indicators, mineral generation indicators, and organic matter transformation indicators. The quotient of the initial convolution kernel of the reaction and the benchmark parameter K is calculated by the radial basis function algorithm, and the reaction equivalent convolution kernel is determined in combination with the first threshold and the second threshold.

[0012] The test samples were made into standard cubic test blocks of 70 mm × 70 mm × 70 mm, and cured in a standard curing room at a temperature of 20 ± 2 degrees Celsius and a relative humidity of more than 95% for 3 days, 7 days, and 28 days, and the unconfined compressive strength, flexural strength, impermeability, frost resistance, and salt resistance of the standard cubic test blocks were measured.

[0013] Among them, according to the optimal ratio of the curing agent, a trial production of no less than 100 cubic meters is carried out on the production line, sampling is taken every 2 hours, engineering performance inspection data is obtained, and an engineering performance inspection matrix is ​​established. The engineering performance inspection matrix includes production stability indicators, product consistency indicators, construction adaptability indicators, and organic matter stability indicators.

[0014] The performance satisfaction is calculated by a fuzzy comprehensive evaluation method. When the performance satisfaction reaches above 0.85, the final curing agent ratio design scheme is output. When the performance satisfaction is lower than 0.85, the weight coefficient is adjusted and the optimization calculation is performed again.

[0015] Among them, a genetic algorithm is used to iteratively solve the multi-objective optimization function, and each generation of the population establishes an inter-layer mapping relationship through the equivalent convolution kernel. The number of iterations is not less than 1000 times, and the optimal ratio of the curing agent is calculated.

[0016] Compared with the prior art, the present invention provides a curing agent mix ratio design method for a nomadic fluidized solidified soil production line. The present invention proposes a curing agent mix ratio design method for a nomadic fluidized solidified soil production line based on a multi-layer neural network, and realizes the systematic design and optimization of the curing agent ratio by establishing an original soil evaluation system, a curing agent constraint system, a performance evaluation system, and a reaction mechanism system. This method overcomes the shortcomings of traditional technologies such as incomplete evaluation of original soil characteristics, lack of theoretical guidance for ratio optimization, and single performance evaluation.

[0017] Through the multi-layer neural network original soil evaluation system, the mapping relationship between the physical and chemical properties of the original soil and the curing effect was established, and the applicability of the original soil was accurately evaluated. The five-layer neural network curing agent constraint system provides a theoretical basis for ratio optimization, and ensures the scientific nature of the ratio scheme through nonlinear mapping and adaptive optimization. The introduction of the performance evaluation system and the reaction mechanism system makes the performance evaluation of the cured soil more comprehensive and the prediction of the chemical reaction process more accurate.

[0018] The present invention solves the technical problem that the curing agent ratio of fluidized solidified soil is not optimized in the prior art. Based on the feature extraction and nonlinear mapping capabilities of a multi-layer neural network, the method can accurately identify the original soil characteristics, predict the curing effect, and optimize the ratio scheme, significantly improving the scientificity and reliability of the curing agent ratio design. At the same time, by establishing a complete evaluation system, a comprehensive evaluation of the performance of solidified soil is achieved, providing strong technical support for engineering practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0021] like Figure 1 FIG. 1 is a flow chart of a method for designing a curing agent mix ratio for a nomadic fluidized solidified soil production line provided by the present invention. The method comprises the following steps:

[0022] S01. Collect original soil samples from the area to be sampled (such as the aquaculture sludge area on the beach). The original soil sample collection depth is 0 cm to 20 cm below the surface, 20 cm to 60 cm in the middle layer, and 60 cm to 100 cm in the bottom layer. The original soil sample collection interval is 100 m to 200 m. The sampling weight of each layer at each collection point is 2 kg to 5 kg. The moisture content, organic matter content, particle size distribution, pH, cation exchange capacity, and salt content of the original soil sample are measured;

[0023] S02, drying the original soil sample at 105 degrees Celsius for 24 hours, grinding it through a 60-mesh standard sieve, measuring the calcium ion content, magnesium ion content, sodium ion content, and degree of organic matter decomposition in the original soil sample, and recording the original soil sample as a standardized soil sample;

[0024] S03, establish a multi-layer neural network original soil evaluation system, the multi-layer neural network original soil evaluation system constructs an L-th layer initial convolution kernel through a Gaussian kernel function, the L-th layer initial convolution kernel includes M characteristic parameters, the M characteristic parameters include soil quality indicators, chemical indicators, and physical indicators, the first initial parameter of the L-th layer initial convolution kernel is a reference parameter K, the reference parameter K is a non-zero number, and the quotient of the M characteristic parameters and the reference parameter K is calculated by a radial basis function algorithm to obtain an L-th layer equivalent convolution kernel, when the quotient value is greater than a first threshold, the M characteristic parameters are reduced by m times, and when the quotient value is less than a second threshold, the M characteristic parameters are reduced by m times. M characteristic parameters are expanded n times, and the L+1th layer initial convolution kernel is obtained through the back propagation algorithm. The L+1th layer initial convolution kernel and the Lth layer initial convolution kernel are mapped to each other through batch normalization. The parameters of the L+1th layer initial convolution kernel are expanded K times, and then the L+1th layer equivalent convolution kernel is determined through the radial basis function algorithm. The Lth layer equivalent convolution kernel and the L+1th layer equivalent convolution kernel are optimized by the cross-validation method. The feature transfer is realized through the activation function, and the original soil adaptability index is calculated. The original soil adaptability index includes soil adaptability coefficient, ion activity coefficient, organic matter stability coefficient, and physical property coefficient.

[0025] S04, establish a five-layer neural network curing agent constraint system, construct the first layer of initial convolution kernel by least square method, the characteristic parameters of the first layer of initial convolution kernel include quicklime mass fraction, slaked lime mass fraction, cement mass fraction, fly ash mass fraction, admixture mass fraction, the admixture mass fraction includes water reducer mass fraction, early strength agent mass fraction, swelling reducer mass fraction, organic stabilizer mass fraction, process the characteristic parameters of the first layer of initial convolution kernel by regularization method, calculate the parameter quotient based on radial basis function algorithm to obtain the first layer of equivalent convolution kernel, use optimization algorithm to construct the second to fifth layers of initial convolution kernel, establish inter-layer feature mapping relationship by batch normalization, optimize the first to fifth layers of equivalent convolution kernel parameters based on cross entropy loss function, and determine the upper and lower limit thresholds of the characteristic parameters in combination with the original soil adaptability index;

[0026] S05. Based on the five-layer neural network curing agent constraint system, the original soil adaptability index is used as the network input layer, the characteristic parameters are used as the hidden layer mapping parameters, the initial range of the curing agent component ratio is determined by the orthogonal test method, the network parameters are optimized by the stochastic gradient descent method, and the optimization model of the inter-layer characteristic mapping relationship is established;

[0027] S06, mixing the curing agent component with the standardized soil sample according to a ratio, designing test schemes for the characteristic parameters respectively based on the five-layer neural network curing agent constraint system, and preparing multiple groups of test samples according to a moisture content gradient, wherein the moisture content gradient interval is 5%, and measuring the slump, initial setting time, final setting time, pH value, and organic stability index of the test samples;

[0028] S07. Establish a multi-layer neural network performance evaluation system, use a local linear embedding algorithm to construct an L-th layer performance initial convolution kernel, the L-th layer performance initial convolution kernel includes a fluidity index, a coagulation time index, a chemical stability index, and an organic matter decomposition index, calculate the quotient of the L-th layer performance initial convolution kernel and the benchmark parameter K through a radial basis function algorithm to obtain an L-th layer performance equivalent convolution kernel, use a residual network structure to establish an inter-layer feature transfer mechanism, perform feature weighting through an attention mechanism, and calculate the comprehensive performance score of the test sample;

[0029] S08, establishing a multi-layer neural network reaction mechanism system, using principal component analysis to construct an initial reaction convolution kernel, the initial reaction convolution kernel includes a hydration reaction index, an ion exchange index, a mineral generation index, and an organic matter conversion index, calculating the quotient of the initial reaction convolution kernel and the reference parameter K by a radial basis function algorithm, and combining the first threshold and the second threshold to determine the reaction equivalent convolution kernel, using a gated recurrent unit to establish an inter-layer feature mapping relationship, and calculating the reaction activity index of the test sample by an adaptive learning rate optimization algorithm;

[0030] S09, the test sample is made into a standard cubic test block of 70 mm×70 mm×70 mm, and cured in a standard curing room at a temperature of 20±2 degrees Celsius and a relative humidity of more than 95% for 3 days, 7 days, and 28 days;

[0031] S10, measuring the unconfined compressive strength, flexural strength, impermeability, frost resistance, and salt resistance of the standard cubic test block, and establishing an engineering performance matrix, wherein the engineering performance matrix includes a strength growth coefficient, a compressive strength coefficient, a flexural strength coefficient, a permeability coefficient, and a durability coefficient;

[0032] S11, using the analytic hierarchy process to determine the comprehensive performance score, the reaction activity index, and the weight coefficient of the engineering performance matrix, constructing a multi-objective optimization function, and introducing an organic matter content correction coefficient;

[0033] S12, using a genetic algorithm to iteratively solve the multi-objective optimization function, each generation of populations establishes an inter-layer mapping relationship through the equivalent convolution kernel, the number of iterations is not less than 1000 times, and the optimal ratio of the curing agent is calculated;

[0034] S13, according to the optimal ratio of the curing agent, conduct a trial production of no less than 100 cubic meters on the production line, take samples every 2 hours, and obtain engineering performance test data;

[0035] S14, establishing an engineering performance inspection matrix, wherein the engineering performance inspection matrix includes a production stability index, a product consistency index, a construction adaptability index, and an organic matter stability index, and calculating performance satisfaction by a fuzzy comprehensive evaluation method;

[0036] S15, when the performance satisfaction reaches above 0.85, output the final curing agent ratio design scheme; when the performance satisfaction is lower than 0.85, return to step S11, adjust the weight coefficient and re-calculate the optimization.

[0037] The specific implementation of the above steps is described in detail below. The specific implementation of step S01 is to carry out stratified sampling and determination of basic property indicators in the area to be sampled (such as the beach aquaculture sludge area). This step is intended to obtain the basic characteristic parameters of the original soil and provide data support for the subsequent curing agent ratio design. The stratified sampling method is adopted when collecting the original soil sample, and the sampling depth is divided into three layers: 0 cm to 20 cm in the surface layer, 20 cm to 60 cm in the middle layer, and 60 cm to 100 cm in the bottom layer. The sampling point spacing is set to 100 meters to 200 meters to ensure that the sample is representative. The sampling weight of each layer of each sampling point is 2 kg to 5 kg, and the sample is reduced by the quartering method. The moisture content is determined by the drying method, and the drying temperature is 105 degrees Celsius. The organic matter content is determined by the potassium dichromate volumetric method, and the organic matter content is obtained by measuring the organic carbon content in the sample. The particle size distribution is determined by a laser particle size analyzer, and the measurement range is 0.1 microns to 1000 microns. The pH is determined by the potentiometric method, and the soil-water ratio is 1 to 2.5. The cation exchange capacity was determined by the ammonium acetate exchange method, and the exchangeable cation content was determined by a flame photometer. The salt content was determined by the conductivity method, with a soil-water ratio of 1:5.

[0038] The specific implementation method of step S02 is to pre-treat the original soil sample and determine the ion content and organic properties. This step is intended to obtain the chemical composition characteristics of the original soil and provide basic data for establishing an original soil evaluation system. The original soil sample was dried in a constant temperature drying oven at 105 degrees Celsius for 24 hours, ground with an agate mortar, and passed through a 60-mesh standard sieve. The calcium ion content was determined by ethylenediaminetetraacetic acid titration. The magnesium ion content was determined by atomic absorption spectrophotometry. The sodium ion content was determined by flame photometry. The degree of decomposition of organic matter was determined by alkaline nitrogen method, and the stability of organic matter was evaluated by measuring the content of easily decomposable nitrogen in the soil.

[0039] The specific implementation method of step S03 is to establish a multi-layer neural network original soil evaluation system, which is intended to quantitatively evaluate the adaptability of the original soil. The initial convolution kernel is constructed using the Gaussian kernel function, and the characteristic parameters include soil quality indicators, chemical indicators, physical indicators, etc. The reference parameter K is 1. The quotient of the characteristic parameter and the reference parameter is calculated by the radial basis function algorithm. When the quotient is greater than the first threshold, the parameter is reduced. The first threshold is 1.5, and the reduction factor m is 0.8. When the quotient is less than the second threshold, the parameter is expanded. The second threshold is 0.5, and the expansion factor n is 1.2. The back propagation algorithm is used to construct the next layer of convolution kernels, and the inter-layer mapping relationship is established by batch normalization. The cross-validation method is used to optimize the equivalent convolution kernel, and the feature transfer is realized by the activation function, and the original soil adaptability index is calculated. The soil adaptability coefficient reflects the degree of matching between the physical and chemical properties of the original soil and the curing requirements. The ion activity coefficient characterizes the influence of ions in the original soil on the curing process. The organic stability coefficient reflects the influence of organic matter on the curing effect. The physical property coefficient represents the influence of the engineering properties of the original soil on the curing effect.

[0040] The specific implementation of step S04 is to establish a five-layer neural network curing agent constraint system, which is intended to determine the reasonable ratio range of each component of the curing agent. The first layer of initial convolution kernel is constructed by least squares method, and the characteristic parameters include quicklime mass fraction, slaked lime mass fraction, cement mass fraction, fly ash mass fraction, admixture mass fraction, etc. The characteristic parameters are processed by regularization method to avoid overfitting. The parameter quotient is calculated by radial basis function algorithm. The initial convolution kernels of the second to fifth layers are constructed by optimization algorithm, and the inter-layer feature mapping relationship is established by batch normalization. The equivalent convolution kernel parameters of each layer are optimized based on the cross entropy loss function, and the upper and lower limit thresholds of the characteristic parameters are determined in combination with the original soil adaptability index. The upper limit of quicklime mass fraction is 30% and the lower limit is 10%. The upper limit of slaked lime mass fraction is 25% and the lower limit is 8%. The upper limit of cement mass fraction is 20% and the lower limit is 5%. The upper limit of fly ash mass fraction is 15% and the lower limit is 3%. The upper limit of admixture mass fraction is 5% and the lower limit is 0.5%.

[0041] The specific implementation method of step S05 is to determine the initial range of the curing agent ratio based on the neural network curing agent constraint system. This step aims to establish a curing agent ratio optimization model. The original soil adaptability index is used as the network input layer, and the characteristic parameters are used as the hidden layer mapping parameters. The orthogonal test method is used to determine the initial range of the curing agent group ratio, and the level combination of each factor is determined by designing an orthogonal test table. The random gradient descent method is used to optimize the network parameters, the learning rate is set to 0.01, and the maximum number of iterations is 1000 times. An optimization model of the inter-layer feature mapping relationship is established to guide the curing agent ratio design.

[0042] The specific implementation method of step S06 is to conduct a curing agent ratio test, which is intended to obtain the performance indicators of the test samples under different ratio conditions. The Latin hypercube sampling method is used to design the test plan for the characteristic parameters to ensure that the test points are evenly distributed in the parameter space. The test samples are prepared according to the moisture content gradient, and the moisture content gradient interval is 5%, ranging from 15% to 40%. The slump is determined by the slump cone method. The initial setting time and final setting time are determined by the Vicat instrument method. The pH is determined by the potentiometric method. The organic stability index is determined by the alkaline nitrogen method.

[0043] The specific implementation method of step S07 is to establish a multi-layer neural network performance evaluation system, which is intended to conduct a comprehensive performance evaluation of the test samples. The local linear embedding algorithm is used to construct the initial convolution kernel of the performance, including fluidity index, coagulation time index, chemical stability index, organic matter decomposition index, etc. The equivalent convolution kernel is obtained by calculating the quotient through the radial basis function algorithm. The inter-layer feature transfer mechanism is established using the residual network structure to avoid the gradient disappearance problem. Feature weighting is performed through the attention mechanism, and the importance weight ranges from 0 to 1. The comprehensive performance score of the test sample is calculated, and the score ranges from 0 to 100 points.

[0044] The specific implementation method of step S08 is to establish a multi-layer neural network reaction mechanism system, which is intended to evaluate the activity of the curing reaction. The principal component analysis method is used to construct the initial convolution kernel of the reaction, including hydration reaction indicators, ion exchange indicators, mineral generation indicators, organic matter conversion indicators, etc. The quotient is calculated by the radial basis function algorithm and the reaction equivalent convolution kernel is determined in combination with the threshold. The inter-layer feature mapping relationship is established by the gated recurrent unit, and the reaction activity index is calculated by the adaptive learning rate optimization algorithm, and the index range is 0 to 1.

[0045] The specific implementation of step S09 is to prepare a standard test block and perform curing. This step is intended to obtain engineering performance test samples. A 70 mm × 70 mm × 70 mm cubic test block is prepared using a standard test mold, and the test block is formed using a vibration molding method. The test block is placed in a standard curing room for curing, the temperature is controlled at 20 ± 2 degrees Celsius, the relative humidity is maintained at above 95%, and the curing period is 3 days, 7 days, and 28 days.

[0046] The specific implementation method of step S10 is to carry out engineering performance testing, which is intended to obtain the mechanical properties and durability indicators of the solidified soil. The unconfined compressive strength is determined using a universal testing machine with a loading rate of 1 mm per minute. The flexural strength is determined using an electronic universal testing machine with a span of 200 mm. The permeameter is used to measure the impermeability, and the water pressure is 1 MPa. The quick freezing method is used to measure the anti-freeze performance, and the number of freeze-thaw cycles is 50 times. The dry-wet cycle method is used to measure the salt resistance, and the number of cycles is 15 times. Establish an engineering performance matrix, including indicators such as strength growth coefficient, compressive strength coefficient, flexural strength coefficient, permeability coefficient, and durability coefficient.

[0047] The specific implementation method of step S11 is to construct a multi-objective optimization function, which is intended to establish a curing agent ratio optimization model. The weight coefficients of the comprehensive performance score, the reaction activity index, and the engineering performance matrix are determined by the hierarchical analysis method, and the weight coefficient ranges from 0 to 1, and the sum of each weight is 1. The organic matter content correction coefficient is introduced, and the correction coefficient is negatively correlated with the organic matter content.

[0048] The specific implementation method of step S12 is to use a genetic algorithm to solve the optimal ratio scheme, which is intended to obtain the optimal ratio of the curing agent. The population size is set to 100, the crossover probability is 0.8, the mutation probability is 0.1, and the number of iterations is not less than 1000 times. Each generation of populations establishes an inter-layer mapping relationship through an equivalent convolution kernel, calculates the fitness function value, and generates a new generation of populations through roulette selection, crossover, and mutation operations, and finally obtains the optimal ratio of the curing agent.

[0049] The specific implementation of step S13 is to conduct a trial production on the production line, which is intended to verify the engineering applicability of the optimal ratio scheme. According to the optimal ratio scheme, a trial production of no less than 100 cubic meters is conducted on the production line, and samples are taken every 2 hours for inspection to measure the engineering performance indicators.

[0050] The specific implementation method of step S14 is to establish an engineering performance test matrix, which is intended to evaluate the comprehensive performance of the solidified soil. The engineering performance test matrix includes production stability indicators, product consistency indicators, construction adaptability indicators, organic matter stability indicators, etc. The fuzzy comprehensive evaluation method is used to calculate the performance satisfaction, establish an evaluation factor set and an evaluation level set, construct a fuzzy relationship matrix, determine the weight vector, and calculate the comprehensive evaluation result.

[0051] The specific implementation of step S15 is to output the final design scheme, which is intended to ensure that the curing agent ratio meets the engineering requirements. When the performance satisfaction reaches above 0.85, the final curing agent ratio design scheme is output. When the performance satisfaction is lower than 0.85, return to step S11 and re-optimize the calculation by adjusting the weight coefficient until the performance requirements are met.

[0052] The specific implementation method of the multi-layer neural network original soil evaluation system is to evaluate the suitability of the original soil in a hierarchical and progressive manner. First, the input layer is constructed to contain three categories of characteristic parameters: soil quality indicators, chemical indicators, and physical indicators. Soil quality indicators include physical characteristic parameters such as particle size distribution, moisture content, organic matter content, and density, which are obtained through laboratory tests. Chemical indicators include chemical characteristic parameters such as pH, cation exchange capacity, calcium ion content, magnesium ion content, and sodium ion content, which are obtained through chemical analysis methods. Physical indicators include engineering characteristic parameters such as compressibility, permeability, and shear strength, which are determined by standard test methods. Next, the Gaussian kernel function is used to construct the L-th layer initial convolution kernel to achieve nonlinear mapping of the feature space, and the bandwidth parameter of the Gaussian kernel function is determined by the cross-validation method to determine the optimal value. For each characteristic parameter, the quotient value with the benchmark parameter is calculated, and adaptive adjustment is performed according to the threshold to ensure that the characteristic parameter is within a reasonable range. Then, the next layer of convolution kernel is constructed by the back propagation algorithm, the error gradient between each layer is calculated, and the network parameters are updated by the stochastic gradient descent method. The batch normalization method is used to process the output of each layer to reduce the internal covariate shift and speed up the training. The deep features of the original soil are extracted through multi-layer convolution operations, and the activation function is used to introduce nonlinear characteristics. Finally, the original soil adaptability index is calculated, which reflects the comprehensive applicability of the original soil. The soil adaptability coefficient reflects the degree of matching between the physical and chemical properties of the original soil and the curing requirements, and is calculated by comparing the original soil indicators with the standard requirements. The ion activity coefficient characterizes the influence of ions in the original soil on the curing process, taking into account the influence of cation exchange capacity and ion concentration. The organic stability coefficient reflects the influence of organic matter on the curing effect, which is calculated by the decomposition degree and content of organic matter. The physical property coefficient characterizes the influence of the engineering properties of the original soil on the curing effect, and comprehensively considers the strength, deformation and other characteristics. The entire evaluation system adopts a hierarchical structure, and through multi-layer feature extraction and nonlinear mapping, it realizes the accurate evaluation of the applicability of the original soil.

[0053] The specific implementation method of the five-layer neural network curing agent constraint system is to establish a deep neural network with five hidden layers to determine the reasonable range of curing agent ratio. The first layer uses the least squares method to construct the initial convolution kernel, and the input features include the mass fraction of quicklime, the mass fraction of slaked lime, the mass fraction of cement, the mass fraction of fly ash and the mass fraction of admixtures. The mass fraction of quicklime ranges from 10% to 30% of the total mass, which is determined by the material ratio test. The mass fraction of slaked lime ranges from 8% to 25% of the total mass, which is determined according to the hydration reaction requirements. The mass fraction of cement ranges from 5% to 20% of the total mass, which is determined by considering the strength development requirements. The mass fraction of fly ash ranges from 3% to 15% of the total mass, which is determined based on the working performance requirements. The mass fraction of admixtures includes the mass fractions of water reducers, early strength agents, swelling agents and organic stabilizers, and the total amount is controlled between 0.5% and 5%. The regularization method is used to process the characteristic parameters, and L1 and L2 regularization terms are added to prevent the network from overfitting. The parameter quotient is calculated by the radial basis function algorithm to obtain the first layer of equivalent convolution kernel. The optimization algorithm is used to construct the initial convolution kernel from the second to the fifth layer, and the number of neurons in each layer decreases layer by layer. Batch normalization is used between layers to establish feature mapping relationships to ensure stable data distribution. The equivalent convolution kernel parameters of each layer are optimized based on the cross entropy loss function, and the stochastic gradient descent method is used to update the parameters. The upper and lower limit thresholds of the characteristic parameters are determined in combination with the original soil adaptability index, and the ratio constraint conditions are established. The entire constraint system realizes the precise constraint of the curing agent ratio range through deep learning methods, and takes into account the interaction relationship between the components.

[0054] The specific implementation method of the multi-layer neural network performance evaluation system is to establish a deep learning model for evaluating the performance of solidified soil. First, the local linear embedding algorithm is used to construct the initial convolution kernel of the performance, which can maintain the local geometric characteristics of the data. The initial convolution kernel contains fluidity index, setting time index, chemical stability index and organic matter decomposition index. The fluidity index is measured by the slump test to reflect the working performance of the solidified soil. The setting time index includes the initial setting time and the final setting time, which are measured by the Vicat instrument to reflect the solidification reaction rate. The chemical stability index is measured by pH value and ion concentration to reflect the chemical stability of the solidified soil. The organic matter decomposition index is measured by the change of organic matter content to reflect the stability of organic matter. The quotient of the initial convolution kernel and the benchmark parameter is calculated by the radial basis function algorithm to obtain the equivalent convolution kernel. The residual network structure is used to establish the inter-layer feature transfer mechanism, and the jump connection is introduced to avoid the gradient vanishing problem of the deep network. Different features are weighted by the attention mechanism, and important features obtain higher weight coefficients. Finally, the comprehensive performance score of the test sample is calculated, and the score ranges from 0 to 100 points. The entire evaluation system takes into account various performance indicators of solidified soil and achieves accurate performance evaluation through deep learning methods.

[0055] The specific implementation method of the multi-layer neural network reaction mechanism system is to establish a deep learning model for evaluating the activity of the curing reaction. First, the principal component analysis method is used to construct the initial convolution kernel of the reaction to extract the main features of the reaction process. The initial convolution kernel contains hydration reaction index, ion exchange index, mineral formation index and organic matter transformation index. The hydration reaction index is determined by the hydration heat and the content of hydration products to reflect the progress of the curing reaction. The ion exchange index is determined by the change of ion concentration to reflect the ion exchange capacity. The mineral formation index is determined by X-ray diffraction analysis to reflect the type and content of the new mineral. The organic matter transformation index is determined by the change of organic matter morphology to reflect the degree of stabilization of organic matter. The quotient of the initial convolution kernel of the reaction and the benchmark parameter is calculated by the radial basis function algorithm, and the equivalent convolution kernel of the reaction is determined by combining the threshold. The inter-layer feature mapping relationship is established by the gated recurrent unit. The structure can process time series data and is suitable for reaction kinetics analysis. The network parameters are adjusted by the adaptive learning rate optimization algorithm to improve the convergence speed and stability of the model. Finally, the reaction activity index is calculated, which reflects the comprehensive activity of the curing reaction, and the index range is 0 to 1. The entire system achieves quantitative characterization of the curing reaction mechanism through deep learning methods.

[0056] The Gaussian kernel function in the multi-layer neural network original soil evaluation system is expressed as follows:

[0057]

[0058] Where x is the input feature vector; x i is the i-th training sample; σ is the bandwidth parameter of the Gaussian kernel, ranging from 0.1 to 1.0; || xx i || 2 is the Euclidean distance. This function is used for feature space mapping and can handle nonlinear data.

[0059] The characteristic parameter calculation of the initial convolution kernel of the Lth layer is expressed as follows:

[0060]

[0061] In the formula, F L is the initial convolution kernel output of the Lth layer; w i is the weight coefficient of the i-th characteristic parameter; P i is the ith feature parameter; b is the bias term; M is the total number of feature parameters. The feature parameter weights are optimized through the back propagation algorithm.

[0062] The quotient calculation of the characteristic parameter and the reference parameter K is expressed as follows:

[0063]

[0064] In the formula, R is the quotient; FL is the initial convolution kernel output of the Lth layer; K is the reference parameter. When R>1.5, it is reduced by 0.8 times; when R<0.5, it is expanded by 1.2 times.

[0065] The basis function of the radial basis function algorithm is expressed as follows:

[0066]

[0067] Where r is the distance and σ is the expansion constant, ranging from 0.1 to 1.0. This function is used to construct a nonlinear mapping relationship.

[0068] The original soil adaptability index is calculated as follows:

[0069] I soil =α1C s +α2C i +α3C o +α4C p ;

[0070] In the formula, I soil is the original soil adaptability index; C s is the soil adaptability coefficient; C i is the ion activity coefficient; C o is the stability coefficient of organic matter; C p is the physical property coefficient; α1, α2, α3, α4 are weight coefficients, and satisfy Each coefficient is obtained through experimental measurement.

[0071] The loss function of the five-layer neural network curing agent constraint system is expressed as follows:

[0072]

[0073] Where L is the cross entropy loss; y i is the true label; is the predicted value; n is the number of samples. This function is used to optimize network parameters.

[0074] The optimization model of curing agent component ratio is expressed as follows:

[0075]

[0076] Where f(x) is the objective function; w i is the weight of the i-th performance indicator; y i is the target value; f i (x) is the predicted value of the i-th performance indicator; x is the matching parameter vector; m is the number of performance indicators.

[0077] The engineering performance matrix is ​​expressed as follows:

[0078]

[0079] In the formula, S ij is the jth performance index of the ith curing period, i=1, 2, 3 correspond to 3 days, 7 days, and 28 days of curing period respectively, and j=1, 2, 3, 4, 5 correspond to the strength growth coefficient, compressive strength coefficient, flexural strength coefficient, permeability coefficient, and durability coefficient respectively.

[0080] The comprehensive performance satisfaction calculation is expressed as follows:

[0081]

[0082] In the formula, D is the performance satisfaction; β i is the weight of the i-th evaluation factor; r ij is the fuzzy relation matrix element; j is the membership degree of the jth evaluation level; k is the number of evaluation factors; l is the number of evaluation levels. This function is used to evaluate the comprehensive performance of solidified soil.

[0083] The reactivity index calculation is expressed as follows:

[0084] I r =λ1H+λ2E+λ3M+λ4O;

[0085] In the formula, I r is the reaction activity index; H is the hydration reaction index; E is the ion exchange index; M is the mineral formation index; O is the organic matter transformation index; λ1, λ2, λ3, λ4 are weight coefficients, and they satisfy Each index is obtained through experimental measurement.

[0086] The construction of the above equations takes into account the following principles: the Gaussian kernel function uses an exponential form because it has good nonlinear mapping capabilities and has infinite-dimensional properties in the feature space; the radial basis function uses an exponential form to ensure the continuity and smoothness of the function; the original soil adaptability index uses linear weighting to reflect the comprehensive influence of various factors; the cross entropy loss function is suitable for classification problems and can accelerate the training convergence speed; the engineering performance matrix uses a three-dimensional form to reflect the performance evolution law of different maintenance periods; the fuzzy comprehensive evaluation uses double-layer weighting to comprehensively consider the influence of qualitative and quantitative indicators.

[0087] The derivation process and parameters of the Gaussian kernel function are as follows:

[0088] It is derived from Taylor expansion, where σ is determined by the following steps: first, the training data set is standardized; then, 10 equally spaced σ values ​​are selected in the range of 0.1 to 1.0 using the grid search method; then, the model performance corresponding to each σ value is evaluated using the cross-validation method; finally, the σ value with the smallest validation error is selected as the optimal parameter. This function can map nonlinear problems in low-dimensional space to high-dimensional space for linear processing, and has good feature extraction capabilities.

[0089] The calculation formula of the characteristic parameters of the initial convolution kernel of the Lth layer is:

[0090] The derivation of is based on the following principle: First, for each characteristic parameter P i Normalize it to eliminate the dimension effect; then introduce the weight coefficient w i Characterize the importance of different features; finally, add the bias term b to increase the flexibility of the model. Weight coefficient w i It is obtained through back propagation algorithm optimization. The specific steps are: calculate the gradient of the loss function with respect to the weight; update the weight using the stochastic gradient descent method; iterate and optimize until convergence. This formula realizes the weighted combination of features and can reflect the influence of different features on the evaluation results.

[0091] The calculation formula of the quotient of characteristic parameter and reference parameter K is:

[0092] The design is based on the following considerations: the benchmark parameter K is set to 1 as the standard reference value; when the quotient R is greater than 1.5, it indicates that the characteristic parameter is too large and needs to be reduced by 0.8 times; when the quotient R is less than 0.5, it indicates that the characteristic parameter is too small and needs to be expanded by 1.2 times. This adaptive adjustment mechanism can keep the characteristic parameters within a reasonable range and improve the stability of the model.

[0093] The basis function expression of radial basis function:

[0094] The selection of is based on the following reasons: the exponential function has good continuity and differentiability; the distance r uses the Euclidean distance, which can reflect the similarity between samples; the expansion constant σ is determined by the cross-validation method. This function implements nonlinear mapping and can handle complex feature relationships.

[0095] Original soil adaptability index calculation formula:

[0096] I soil =α1C s +α2C i +α3C o +α4C pThe construction process is as follows: First, various coefficients are obtained through experimental determination; then, the weight coefficients α1, α2, α3, and α4 are determined by the hierarchical analysis method. The specific steps include: constructing a judgment matrix; calculating eigenvalues ​​and eigenvectors; and performing consistency tests. This formula comprehensively considers the physical and chemical properties of the original soil and can accurately evaluate the applicability of the original soil.

[0097] The loss function of the five-layer neural network curing agent constraint system:

[0098] The choice of is based on the following considerations: the cross entropy loss function has good performance for classification problems; the logarithmic function can increase the impact of the error on the weight update; the introduction of the negative sign makes the optimization goal minimize the loss function. This function can speed up the convergence of the model and improve the classification accuracy.

[0099] Optimization model of curing agent component ratio:

[0100] The construction process of is as follows: First, determine the optimization target, including performance indicators such as strength, workability, and durability; then, use the experimental method to obtain the target value y i ; Finally, establish the prediction model f i (x). Weight coefficient w i Determined by the analytic hierarchy process, it reflects the importance of different performance indicators. The model can achieve multi-objective optimization of curing agent ratio.

[0101] Engineering Performance Matrix:

[0102] The construction is based on the following principle: the matrix element S ij Determined by standard test methods; different curing periods reflect the law of performance development; different performance indicators reflect the applicability of the project. The matrix realizes the systematic expression of performance data.

[0103] Comprehensive performance satisfaction calculation formula:

[0104] The derivation process of is as follows: First, establish the evaluation factor set and the evaluation level set; then, determine the membership degree γ by fuzzy statistics method. j ; Finally, the weight coefficient β is determined by using the hierarchical analysis method i . This formula realizes the comprehensive evaluation of qualitative and quantitative indicators.

[0105] Reactivity index calculation formula:

[0106] I r=λ1H+λ2E+λ3M+λ4O is constructed as follows: various indicators are measured by experimental methods; weight coefficients λ1, λ2, λ3, λ4 are determined by the Delphi method, including: expert scoring; statistical analysis; consistency test. This formula can quantitatively characterize the activity level of the curing reaction.

[0107] The innovations of these equations are: adopting a multi-level evaluation system to achieve a comprehensive evaluation of the performance of solidified soil; introducing an adaptive adjustment mechanism to improve the stability and adaptability of the model; considering multiple influencing factors to improve the accuracy of the evaluation results; establishing a correlation between performance indicators to reflect the complexity of the solidification process. Compared with existing technologies, this method has better universality and reliability.

[0108] Specifically, the principle of the present invention is: the core technical principle of the present invention is to use the powerful feature extraction and nonlinear mapping capabilities of multi-layer neural networks to establish a correlation model between original soil characteristics, curing agent ratio and curing effect. The introduction of Gaussian kernel function enables the model to process high-dimensional nonlinear data and extract deep feature information through multi-layer convolution operations. The use of radial basis function ensures the continuity and smoothness of feature mapping and improves the generalization ability of the model. The application of batch normalization reduces internal covariate offset and accelerates the convergence speed of the model.

[0109] This method adopts a hierarchical evaluation system and uses multiple neural network models to deal with problems at different levels. The original soil evaluation system focuses on the physical and chemical properties of the soil, the curing agent constraint system is responsible for ratio optimization, the performance evaluation system realizes comprehensive performance evaluation, and the reaction mechanism system predicts the chemical reaction process. This hierarchical processing method is in line with the problem decomposition idea in engineering practice and can effectively reduce the complexity of the problem.

[0110] By introducing an adaptive adjustment mechanism, this method can automatically adjust the model parameters according to the changes in characteristic parameters, improving the adaptability and stability of the model. The combination of multi-objective optimization and genetic algorithm ensures the global optimality of the ratio scheme. The application of fuzzy comprehensive evaluation method realizes the unified evaluation of qualitative and quantitative indicators. The organic combination of these technical principles constructs a complete theoretical system for curing agent ratio design.

[0111] A specific embodiment 1 of the present invention is provided below, and the specific implementation method of each step in this embodiment 1 is described in detail as follows.

[0112] The specific implementation method of step S01 is to conduct stratified sampling and determine the basic property indicators of the beach aquaculture sludge area. This step aims to obtain the basic characteristic parameters of the original soil and provide data support for the subsequent curing agent ratio design. The original soil sample collection adopts the stratified sampling method, and the sampling depth is divided into three layers: 0 cm to 20 cm in the surface layer, 20 cm to 60 cm in the middle layer, and 60 cm to 100 cm in the bottom layer. The sampling point spacing is set to 100 meters to 200 meters to ensure that the sample is representative. The sampling weight of each layer at each sampling point is 2 kg to 5 kg, and the sample is reduced by four points. The moisture content was determined by drying method at a drying temperature of 105 degrees Celsius; the organic matter content was determined by potassium dichromate volumetric method, and the organic matter content was obtained by measuring the organic carbon content in the sample; the particle size distribution was determined by a laser particle size analyzer with a measurement range of 0.1 microns to 1000 microns; the pH was determined by the potentiometric method with a soil-water ratio of 1 to 2.5; the cation exchange capacity was determined by the ammonium acetate exchange method, and the exchangeable cation content was determined by a flame photometer; the salt content was determined by the conductivity method with a soil-water ratio of 1 to 5. The formula for calculating the moisture content of the original soil is: In the formula, w is the moisture content, m1 is the mass of wet soil, m2 is the mass of the dried soil sample, and m3 is the mass of the container. The organic matter content calculation formula is: OM = 0.58 × C org ; In the formula, OM is the organic matter content, C org is the organic carbon content, and 0.58 is the empirical coefficient.

[0113] The specific implementation method of step S02 is to pre-treat the original soil sample and determine the ion content and organic matter characteristics. This step is intended to obtain the chemical composition characteristics of the original soil and provide basic data for establishing an original soil evaluation system. The original soil sample is dried in a constant temperature drying oven at 105 degrees Celsius for 24 hours, ground with an agate mortar, and passed through a 60-mesh standard sieve. The calcium ion content is determined by ethylenediaminetetraacetic acid titration method and calculated by the following formula: In the formula, C Ca is the calcium ion content, c1 is the concentration of EDTA standard solution, v1 is the titration volume, M is the molar mass of calcium, and m is the mass of the sample. The magnesium ion content was determined by atomic absorption spectrophotometry and calculated by the standard curve method: In the formula, C Mg is the magnesium ion content, c2 is the concentration obtained from the standard curve, v2 is the constant volume, M1 is the molar mass of magnesium, and m1 is the sample mass. The sodium ion content was determined by flame photometry, and the degree of decomposition of organic matter was determined by alkaline nitrogen method.

[0114] The specific implementation of step S03 is to establish a multi-layer neural network original soil evaluation system, which aims to quantitatively evaluate the adaptability of the original soil. The Gaussian kernel function is used to construct the initial convolution kernel, and the function expression is: In the formula, x is the input feature vector, x i is the i-th training sample, σ is the bandwidth parameter of the Gaussian kernel, and its value range is 0.1 to 1.0. The characteristic parameter calculation of the initial convolution kernel of the L-th layer is expressed as: In the formula, F L is the initial convolution kernel output of the Lth layer, w i is the weight coefficient of the i-th feature parameter, P i is the i-th characteristic parameter, and b is the bias term. The quotient of the characteristic parameter and the reference parameter K is calculated as: When R is greater than 1.5, it is reduced by 0.8 times, and when R is less than 0.5, it is enlarged by 1.2 times. The basis function of the radial basis function is expressed as: The original soil adaptability index is calculated as: I soil =α1C s +α2C i +α3C o +α4C p Where I soil is the original soil adaptability index, C s is the soil adaptability coefficient, C i is the ion activity coefficient, C o is the organic stability coefficient, C p is the physical property coefficient, α1, α2, α3, α4 are weight coefficients.

[0115] The specific implementation of step S04 is to establish a five-layer neural network curing agent constraint system, which is intended to determine the reasonable ratio range of each component of the curing agent. The loss function of the five-layer neural network curing agent constraint system is expressed as: Where L is the cross entropy loss, y i is the true label, is the predicted value. The optimization model of curing agent component ratio is expressed as: Where f(x) is the objective function, w i is the weight of the i-th performance indicator, y i is the target value, f i (x) is the predicted value of the i-th performance index. The characteristic parameters include the mass fraction of quicklime (10% to 30%), the mass fraction of slaked lime (8% to 25%), the mass fraction of cement (5% to 20%), the mass fraction of fly ash (3% to 15%) and the mass fraction of admixtures (0.5% to 5%).

[0116] The specific implementation of step S05 is to determine the initial range of the curing agent ratio based on the neural network curing agent constraint system. This step aims to establish a curing agent ratio optimization model. The original soil adaptability index is used as the network input layer, and the characteristic parameters are used as the hidden layer mapping parameters. The ratio optimization adopts the stochastic gradient descent method, and the update formula of the learning rate η is: Where η t is the learning rate of the tth iteration, η0 is the initial learning rate, and its value is 0.01. The weight update formula is: In the formula, w t is the weight of the tth iteration, is the gradient of the objective function with respect to the weight.

[0117] The specific implementation of step S06 is to conduct a curing agent ratio test, which aims to obtain the performance indicators of the test samples under different ratio conditions. The Latin hypercube sampling method is used to design the test plan for the characteristic parameters, and the sampling point calculation formula is: n = k × m 1 / 4 ; Where n is the number of sampling points, k is the empirical coefficient, ranging from 5 to 10, and m is the number of characteristic parameters. The moisture content gradient of the test samples is from 15% to 40%, with an interval of 5%. The slump measurement value calculation formula is: H = H0-H1; where H is the slump, H0 is the initial height of the sample, and H1 is the final height of the sample. The initial setting time and final setting time are determined by the Vicat apparatus, and the organic stability index calculation formula is: In the formula, S o is the organic matter stability index, C1 is the organic matter content after solidification, and C0 is the organic matter content before solidification.

[0118] The specific implementation of step S07 is to establish a multi-layer neural network performance evaluation system, which is intended to conduct a comprehensive performance evaluation of the test samples. The optimization objective function of the local linear embedding algorithm is: In the formula, x i is the i-th sample point, w ij is the weight coefficient, and k is the number of neighboring points. The formula for calculating the feature vector of the performance initial convolution kernel is: Where V L is the performance vector, β i is the weight coefficient, F i They are fluidity index, coagulation time index, chemical stability index and organic matter decomposition index. The weight calculation formula of the attention mechanism is: In the formula, a i is the attention weight of the i-th feature, e i is the characteristic energy value. The calculation formula for the comprehensive performance score of the test sample is: In the formula, S is the comprehensive performance score, γi is the weight of the i-th performance indicator, P i is the normalized performance index value.

[0119] The specific implementation of step S08 is to establish a multi-layer neural network reaction mechanism system, which is intended to evaluate the activity of the curing reaction. The eigenvalue equation of the principal component analysis method is: |A-λI|=0; where A is the covariance matrix, λ is the eigenvalue, and I is the unit matrix. The calculation formula for the initial convolution kernel of the reaction is: In the formula, R L is the reaction vector, δ i is the weight coefficient, I i They are hydration reaction index, ion exchange index, mineral formation index and organic matter transformation index. The update formula of the gated cycle unit is: In the formula, h t is the hidden state at the current moment, z t To update the gate, is the candidate hidden state. The calculation formula of the reaction activity index is: r =λ1H+λ2E+λ3M+λ4O; where, i r is the reaction activity index, H is the hydration reaction index, E is the ion exchange index, M is the mineral formation index, O is the organic matter transformation index, λ1, λ2, λ3, λ4 are weight coefficients.

[0120] The specific implementation of steps S09 and S10 is to prepare standard test blocks and perform engineering performance tests. The engineering performance matrix is ​​expressed as: In the formula, S ij is the jth performance index in the i-th curing period. The strength growth coefficient calculation formula is: In the formula, k s is the strength growth coefficient, f t is the compressive strength at age t, f 28 is the 28-day compressive strength. The permeability coefficient calculation formula is: In the formula, k p is the permeability coefficient, Q is the amount of permeated water, L is the sample height, H is the water head height, A is the sample cross-sectional area, h is the water pressure, and T is the permeation time. The durability coefficient calculation formula is: In the formula, k d is the durability factor, f n is the strength after cycling, and f0 is the initial strength.

[0121] The specific implementation method of steps S11 to S15 is to establish a multi-objective optimization function and solve the optimal ratio scheme through a genetic algorithm. The membership function of the fuzzy comprehensive evaluation is: In the formula, μ(x) is the membership function, x1 is the upper limit, x0 is the lower limit, and k is the adjustment coefficient. The comprehensive performance satisfaction calculation formula is: In the formula, D is the performance satisfaction, β i is the weight of the i-th evaluation factor, r ij is the fuzzy relation matrix element, γ j is the membership degree of the jth evaluation level. The fitness function of the genetic algorithm is: In the formula, F is the fitness value, and f(x) is the objective function value. The crossover operator uses arithmetic crossover, x i+1 =αx i +(1-α)y i ; In the formula, x i+1 is a new individual, x i ,y i is the parent individual, α is the crossover coefficient. The mutation operator uses non-uniform mutation, x new =x+r(b u -x) or x new =xr(xb l );where x new is the value after mutation, b u , b l are the upper and lower bounds respectively, and r is a random number between 0 and 1.

[0122] Specifically, the specific implementation of steps S11 to S15 is as follows.

[0123] The specific implementation method of step S11 is to construct a multi-objective optimization function, which aims to establish a curing agent ratio optimization model. The analytic hierarchy process is used to determine the weight coefficients of the comprehensive performance score, the reactivity index, and the engineering performance matrix. First, a judgment matrix is ​​constructed, and the 1 to 9 scaling method is used to compare each index in pairs; then the maximum eigenvalue and the corresponding eigenvector of the judgment matrix are calculated; finally, the eigenvector is normalized to obtain the weight coefficient. The weight coefficient of the comprehensive performance score is 0.4, the weight coefficient of the reactivity index is 0.3, and the weight coefficient of the engineering performance matrix is ​​0.3. The organic matter content correction coefficient is negatively correlated with the organic matter content. When the organic matter content is greater than 5%, the correction coefficient decreases by 0.05 for every 1 percentage point increase.

[0124] The specific implementation method of step S12 is to use a genetic algorithm to solve the optimal ratio scheme, which is intended to obtain the optimal ratio of the curing agent. The genetic algorithm is used to iteratively solve the multi-objective optimization function, the population size is set to 100, the crossover probability is 0.8, the mutation probability is 0.1, and the number of iterations is not less than 1000 times. Each generation of population establishes an inter-layer mapping relationship through an equivalent convolution kernel, calculates the fitness function value, and the selection of the initial population is randomly generated according to the ratio range determined by the experimental design scheme. The roulette selection operator is used to select excellent individuals, and the selection probability is proportional to the individual fitness. The crossover operation uses arithmetic crossover to generate new individuals under the premise of satisfying the ratio constraints. The mutation operation uses non-uniform mutation, and the variable step length gradually decreases with the increase of the number of iterations. Finally, the optimal ratio of the curing agent is obtained.

[0125] The specific implementation method of step S13 is to conduct a trial production of the production line, which is intended to verify the engineering applicability of the optimal ratio scheme. According to the optimal ratio scheme, a trial production of no less than 100 cubic meters is carried out on the production line, and samples are taken every 2 hours for inspection to measure the engineering performance indicators. During the trial production, the production process parameters are kept stable, including stirring time, stirring speed, water addition, etc. Standard test methods are used when sampling to ensure the accuracy and reliability of the data.

[0126] The specific implementation method of step S14 is to establish an engineering performance test matrix, which is intended to evaluate the comprehensive performance of the solidified soil. The engineering performance test matrix includes production stability indicators, product consistency indicators, construction adaptability indicators, organic stability indicators, etc. Production stability indicators include slump, initial setting time, and coefficient of variation of final setting time. Product consistency indicators include the discrete degree of compressive strength and impermeability. Construction adaptability indicators include workability, construction time, and pumping performance. Organic stability indicators include organic content change rate and pH stability.

[0127] The specific implementation method of step S15 is to output the final design scheme, which is intended to ensure that the curing agent ratio meets the engineering requirements. The performance satisfaction is calculated by using the fuzzy comprehensive evaluation method. First, an evaluation factor set and an evaluation grade set are established. The evaluation factors include production stability, product consistency, construction adaptability and organic matter stability. The evaluation grades are divided into four grades: excellent, good, qualified and unqualified. Then a fuzzy relationship matrix is ​​constructed to determine the membership of each indicator based on the test data. Finally, the weight vector is determined to calculate the comprehensive evaluation result. When the performance satisfaction reaches more than 0.85, the final curing agent ratio design scheme is output. When the performance satisfaction is less than 0.85, return to step S11 and re-optimize the calculation by adjusting the weight coefficient until the performance requirements are met. The calculation of performance satisfaction involves fuzzy mathematics theory. By establishing a fuzzy relationship matrix, the final evaluation result is obtained by weighted average method. This method can fully consider the importance of various performance indicators and achieve a comprehensive evaluation of the performance of the solidified soil.

[0128] In order to better understand and implement the present invention, the following is an example 2 of a specific application scenario of the present invention: The technical team used the curing agent mix ratio design method of the present invention to conduct research, and the specific implementation process is as follows: Sampling points are arranged in the research area at a sampling interval of 150 meters, and each sampling point collects original soil samples at the surface layer of 0 cm to 20 cm, the middle layer of 20 cm to 60 cm, and the bottom layer of 60 cm to 100 cm, respectively, and 3 kg of samples are collected from each layer. After measurement, the basic physical indicators of the original soil are shown in Table 1 below:

[0129] Table 1 Test results of basic physical indicators of original soil

[0130] Test items surface layer Middle level Bottom Layer average value Moisture content / % 42.5 38.2 35.6 38.8 Organic matter content / % 9.8 8.5 7.2 8.5 Clay content / % 18.5 16.8 15.2 16.8 Powder content / % 65.2 68.5 70.2 68.0 Sand content / % 16.3 14.7 14.6 15.2 pH 7.6 7.8 8.0 7.8 Cation exchange capacity / (cmol / kg) 28.5 26.2 24.8 26.5 Salt content / % 0.92 0.88 0.82 0.87

[0131] The original soil sample was dried at 105 degrees Celsius for 24 hours, ground through a 60-mesh standard sieve, and the ion content and organic matter characteristics were measured. The results are shown in Table 2 below:

[0132] Table 2 Test results of chemical properties of raw soil

[0133] Test items surface layer Middle level Bottom Layer average value Calcium ion content / % 0.45 0.42 0.38 0.42 Magnesium ion content / % 0.42 0.38 0.35 0.38 Sodium ion content / % 0.68 0.65 0.62 0.65 Alkaline nitrogen content / (mg / kg) 135 122 118 125

[0134] The original soil evaluation system with multi-layer neural network was used for evaluation, and the original soil adaptability index was calculated to be 0.68. The specific indicators are shown in Table 3 below:

[0135] Table 3 Evaluation results of original soil adaptability

[0136] Evaluation indicators Numeric Soil adaptability coefficient 0.72 Ion activity coefficient 0.65 Organic stability factor 0.62 Physical property coefficient 0.75 Adaptability Index 0.68

[0137] Based on the five-layer neural network curing agent constraint system, combined with the original soil adaptability index, the initial range of curing agent ratio is determined: quicklime 0% to 3%, slaked lime 0% to 5%, cement 0% to 3%, fly ash 0% to 5%, admixture 0.3% to 0.8%. The admixture is composed of the following components: water reducer (polycarboxylic acid series) accounts for 0.2%, early strength agent (calcium chloride) accounts for 0.1%, swelling agent (aluminum sulfate) accounts for 0.1%, and organic stabilizer (biological enzyme) accounts for 0.1%. The selected material indicators are: quicklime activity is not less than 85%, cement strength grade is 42.5, and fly ash fineness is not more than 45 microns, accounting for not less than 85%.

[0138] The Latin hypercube sampling method was used to design 30 test ratios. After preliminary screening tests, the following 6 ratios were selected for in-depth research:

[0139] Mixing ratio 1: 1.5% quicklime + 2.5% cement + 2.5% fly ash + 0.5% admixture;

[0140] Mix ratio 2: 2.5% quicklime + 2.0% slaked lime + 2.0% fly ash + 0.5% admixture;

[0141] Mixing ratio 3: 1.2% quicklime + 4.2% slaked lime + 2.2% fly ash + 0.5% admixture;

[0142] Mixing ratio 4: 2.2% quicklime + 1.2% slaked lime + 4.2% fly ash + 0.5% admixture;

[0143] Mixing ratio 5: 1.8% quicklime + 1.8% slaked lime + 3.0% fly ash + 0.5% admixture;

[0144] Mix ratio 6: 2.0% quicklime + 3.0% slaked lime + 1.8% fly ash + 0.5% admixture.

[0145] After the evaluation of the multi-layer neural network performance evaluation system and the reaction mechanism system, combined with the engineering performance test results, the above 6 groups of ratio schemes were optimized, and finally 4 groups of optimal ratio schemes were determined: Scheme 1 is 1% quicklime + 2% cement + 2% fly ash + 0.5% admixture; Scheme 2 is 2% quicklime + 2% slaked lime + 2% fly ash + 0.5% admixture; Scheme 3 is 1% quicklime + 4% slaked lime + 2% fly ash + 0.5% admixture; Scheme 4 is 2% quicklime + 1% slaked lime + 4% fly ash + 0.5% admixture. These 4 groups of ratio schemes optimize the amount of curing agent while ensuring engineering performance.

[0146] Next, the multi-layer neural network performance evaluation system was used to evaluate the optimal solution. The results are shown in Table 4 below:

[0147] Table 4 Performance evaluation results of the preferred solution

[0148] Evaluation indicators Solution 1 Solution 2 Solution 3 Solution 4 Liquidity Indicators 92 88 90 85 Setting time index 88 90 85 92 Chemical stability index 90 92 88 85 Organic matter decomposition index 85 82 80 88 Comprehensive score 89 88 86 87

[0149] The curing reaction activity was evaluated using a multi-layer neural network reaction mechanism system, and the results are shown in Table 5 below:

[0150] Table 5 Reactivity evaluation results of the preferred scheme

[0151] Reaction indicators Solution 1 Solution 2 Solution 3 Solution 4 Hydration reaction index 0.88 0.92 0.85 0.90 Ion exchange index 0.85 0.88 0.90 0.82 Mineral generation index 0.90 0.85 0.82 0.88 Organic matter conversion index 0.82 0.80 0.78 0.85 Reactivity Index 0.86 0.86 0.84 0.86

[0152] The mixture was prepared according to the preferred ratio. Six 70 mm × 70 mm × 70 mm standard test blocks were prepared for each ratio. The mixture was cured in a standard curing room at a temperature of 20 ± 2 degrees Celsius and a relative humidity of more than 95%. The engineering performance was measured after 3 days, 7 days, and 28 days. The results are shown in Table 6 below:

[0153] Table 6 Engineering performance test results of the preferred solution

[0154]

[0155] The four preferred solutions were tested on a production line of 150 cubic meters respectively, and samples were taken every 2 hours to test the engineering performance indicators. The test results are shown in Table 7 below:

[0156] Table 7 Performance satisfaction evaluation results of the preferred solution

[0157]

[0158]

[0159] After pilot testing, the optimal mix ratio was obtained: the first one, 1% quicklime + 2% cement + 2% fly ash + 0.5% admixture; the second one, 2% quicklime + 2% slaked lime + 2% fly ash + 0.5% admixture; the third one, 1% quicklime + 4% slaked lime + 2% fly ash + 0.5% admixture; the fourth one, 2% quicklime + 1% slaked lime + 4% fly ash + 0.5% admixture.

[0160] The traditional curing agent ratio design method mainly relies on experience to determine the initial ratio range, and performs orthogonal test optimization through a single performance index. It is difficult to accurately evaluate the influence of the original soil characteristics on the curing effect, and it is also impossible to achieve accurate optimization of the curing agent ratio. The present invention adopts a multi-layer neural network evaluation system, and through a systematic evaluation of the original soil characteristics, establishes a correlation model between the original soil characteristics, the curing agent ratio and the curing effect, and achieves accurate optimization of the ratio scheme. From the test results, the comprehensive satisfaction of Scheme 1 reaches 0.89, which is significantly better than the traditional method. This scheme not only meets the strength requirements, but also has good durability, especially in terms of organic stability and construction adaptability. The optimal ratio scheme determined by the method of the present invention achieves the optimization of the curing agent dosage while ensuring the engineering performance, and can be promoted and applied in this project.

[0161] It should be noted that the variables involved in the present invention are explained in detail as shown in Table 8 below.

[0162] Table 8 Variable explanation table

[0163]

[0164]

[0165] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for designing the proportion of a curing agent for a nomadic fluidized solidified soil production line, characterized in that: The following steps are involved: Original soil samples are collected, and the moisture content, organic matter content, particle size distribution, pH, cation exchange capacity, and salt content of the original soil samples are measured; a multi-layer neural network original soil evaluation system is established, and an initial convolution kernel is constructed through a Gaussian kernel function, and the initial convolution kernel contains multiple characteristic parameters, including soil quality indicators, chemical indicators, and physical indicators; a five-layer neural network curing agent constraint system is established, and the first layer of initial convolution kernel is constructed through the least squares method, and the characteristic parameters of the first layer of initial convolution kernel include the mass fraction of quicklime, the mass fraction of slaked lime, the mass fraction of cement, the mass fraction of fly ash, and the mass fraction of admixtures; a multi-layer neural network performance evaluation system is established, and a local linear embedding algorithm is used to construct the L-th layer performance initial convolution kernel and the L-th layer performance equivalent convolution kernel; based on the multi-layer neural network performance evaluation system, the optimal ratio of the curing agent is determined, and verified through trial production on the production line.

2. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 1 is characterized in that: The first initial parameter of the Lth layer initial convolution kernel in the multi-layer neural network original soil evaluation system is the reference parameter K, and the reference parameter K is a non-zero number. The quotient of the M feature parameters and the reference parameter K is calculated by the radial basis function algorithm to obtain the Lth layer equivalent convolution kernel. When the quotient is greater than the first threshold, the M feature parameters are reduced by m times. When the quotient is less than the second threshold, the M feature parameters are expanded by n times, where L represents the number of layers, L∈[3,8].

3. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 1 is characterized in that: In the five-layer neural network curing agent constraint system, the mass fraction of the admixture includes the mass fraction of the water reducer, the mass fraction of the early strength agent, the mass fraction of the swelling reducer, and the mass fraction of the organic stabilizer. The characteristic parameters of the first layer initial convolution kernel are processed by the regularization method, and the parameter quotient is calculated based on the radial basis function algorithm to obtain the first layer equivalent convolution kernel.

4. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 1 is characterized in that: The curing agent component is mixed with the standardized soil sample according to a ratio, and the Latin hypercube sampling method is used to design test schemes for the characteristic parameters based on the five-layer neural network curing agent constraint system. A plurality of groups of test samples are prepared according to a moisture content gradient, and the moisture content gradient interval is 5%. The slump, initial setting time, final setting time, pH, and organic stability index of the test samples are measured.

5. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 1 is characterized in that: The L-th layer performance initial convolution kernel includes a fluidity index, a coagulation time index, a chemical stability index, and an organic matter decomposition index. The L-th layer performance equivalent convolution kernel is obtained by calculating the quotient of the L-th layer performance initial convolution kernel and the benchmark parameter K through a radial basis function algorithm.

6. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 1 is characterized in that: A multi-layer neural network reaction mechanism system is established, and the principal component analysis method is used to construct the initial reaction convolution kernel, which includes hydration reaction indicators, ion exchange indicators, mineral generation indicators, and organic matter conversion indicators. The quotient of the initial reaction convolution kernel and the benchmark parameter K is calculated by the radial basis function algorithm, and the reaction equivalent convolution kernel is determined in combination with the first threshold and the second threshold.

7. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 1 is characterized in that: The test samples were made into standard cubic test blocks of 70 mm × 70 mm × 70 mm, and cured in a standard curing room at a temperature of 20±2 degrees Celsius and a relative humidity of more than 95% for 3 days, 7 days, and 28 days, and the unconfined compressive strength, flexural strength, impermeability, frost resistance, and salt resistance of the standard cubic test blocks were measured.

8. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 1 is characterized in that: According to the optimal ratio of the curing agent, a trial production of no less than 100 cubic meters is carried out on the production line, and samples are taken every 2 hours to obtain engineering performance test data and establish an engineering performance test matrix. The engineering performance test matrix includes production stability indicators, product consistency indicators, construction adaptability indicators, and organic matter stability indicators.

9. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 8, characterized in that: The performance satisfaction is calculated by fuzzy comprehensive evaluation method. When the performance satisfaction reaches above 0.85, the final curing agent ratio design scheme is output. When the performance satisfaction is lower than 0.85, the weight coefficient is adjusted and the optimization calculation is performed again.

10. The method for designing the curing agent mix ratio of the nomadic fluidized solidified soil production line according to claim 1, characterized in that: A genetic algorithm is used to iteratively solve the multi-objective optimization function. Each generation of the population establishes an inter-layer mapping relationship through the equivalent convolution kernel. The number of iterations is not less than 1000 times, and the optimal ratio of the curing agent is calculated.

Images