CN116189794A - Rammed earth water salt content measurement method - Google Patents

Abstract

The invention discloses a method for measuring the water salt content of rammed earth, which comprises the following steps: initializing a population, compacting the population scale in a water and salt content prediction model, establishing a generalized regression network to predict the water and salt content of the compacted soil, calculating the root mean square error between the actual value and the predicted value of the water content or the salt content as the fitness of particles, updating the state of the particles, executing crossover and mutation operations, judging whether convergence conditions are met, and predicting the water and salt content of the compacted soil. The invention utilizes the generalized regression neural network optimized by the improved GA-PSO algorithm, can indirectly obtain the water content and the salt content of the rammed soil by measuring the capacitance value, the conductivity, the soil temperature and the soil density parameters of the rammed soil, avoids the defects of field sampling and time and labor waste in laboratory detection in laboratory measurement, greatly improves the modeling speed compared with the traditional GA algorithm, avoids the problem of sinking into a local optimal solution compared with the traditional PSO algorithm, and is a method with better comprehensive performance.

Description

A method for measuring water-salt content of rammed earth

Technical Field

The invention belongs to the technical field of soil measurement, and in particular is a method for quickly measuring the moisture content and salt content of rammed soil.

Background Art

Earthen sites are the remains of earthen buildings with high historical and cultural value. As most earthen sites are located in natural open-air environments, they are easily damaged by various human and natural factors. In view of the common cracks, erosion, collapse and other diseases of earthen sites, physical reinforcement methods are usually used to reinforce and repair earthen sites. Among them, the top reinforcement using rammed earth is a common and effective method that can effectively stabilize the original structure. The building materials used are rammed earth similar to the original site made by ramming technology, so they have good integrity and meet the principle of maintaining the status quo and not destroying the appearance in the protection and reinforcement of earthen sites. In order to monitor the reinforcement effect of the rammed earth top, it is necessary to measure its moisture content and salt content.

At present, the measurement methods for the moisture content and salt content of rammed earth are still mainly based on traditional laboratory measurements (drying method and immersion liquid measurement method). The drying method is a measurement method that determines the moisture content based on the change in mass of the soil sample before and after drying. The principle of the immersion liquid measurement method is to prepare the soil sample into an extract and calculate the soil salt content by calculating the conductivity. Neither of these two methods can achieve fast and real-time on-site measurement of parameters. Since rammed earth is a kind of soil, many of its characteristics are similar to those of agricultural soil, but there are also many differences. For example, rammed earth needs to add additives such as quicklime to ensure strength and stiffness, the density of rammed earth usually meets certain standards, and the measurement of rammed earth buildings requires certain continuity and real-time performance. When using the dielectric property measurement method to measure the moisture content and salt content of rammed earth, it is necessary to fully consider parameters such as capacitance, conductivity, soil temperature, and soil density to achieve accurate prediction of moisture content and salt content. Therefore, it is necessary to establish a multivariate regression algorithm for the above parameters for moisture content and salt content.

Summary of the invention

The object of the present invention is to provide a multivariate regression algorithm for measuring the moisture content and salt content of rammed earth, so as to achieve rapid and accurate measurement and prediction of the moisture content and salt content of rammed earth.

The technical solution to achieve the purpose of the present invention is:

A method for measuring water-salt content of rammed earth comprises the following steps:

Step 1: Initialize the population and set the population size in the GRNN model optimized by GA-PSO;

Step 2: Establish a generalized regression network model, taking the capacitance, conductivity, soil temperature, and soil density of the rammed earth as input parameters, and the predicted moisture content and salt content of the rammed earth as output parameters;

Step 3: Calculate the fitness of each particle:

The root mean square error between the actual value and the predicted value of the rammed soil moisture content or salt content is taken as the fitness of the particle;

Step 4: Update the state of the particle:

Update the velocity and displacement of each particle in the particle swarm algorithm;

Step 5: Perform crossover and mutation operations:

Introduce crossover operation to the update of particle position and velocity, and introduce mutation operation to the update of particle position and velocity;

Step 6: Determine whether the convergence condition is met. If the optimized generalized regression network model meets the accuracy requirement, a generalized regression network is established to predict the moisture content and salt content of the rammed earth. If the accuracy does not meet the requirement, return to step 2.

Step 7: Use the optimized generalized regression network to predict the moisture content and salt content of rammed earth.

Compared with the prior art, the present invention has the following significant advantages:

(1) The generalized regression neural network model can be used to inversely predict the moisture content and salt content of rammed soil. By measuring the capacitance, conductivity, soil temperature, and soil density parameters of the rammed soil, the moisture content and salt content of the rammed soil can be indirectly obtained, avoiding the disadvantages of the laboratory measurement method that requires on-site sampling and time-consuming and laborious laboratory testing, and the moisture content and salt content can be obtained instantly on-site.

(2) The generalized regression neural network optimized by the improved GA-PSO algorithm improves the accuracy of the regression model. Compared with the single GA algorithm, it greatly improves the modeling speed. Compared with the single PSO algorithm, it avoids the problem of falling into the local optimal solution. It is a method with better comprehensive performance.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is the flow chart of the GRNN algorithm optimized by GA-PSO.

Figure 2 is a diagram of the GRNN network structure.

FIG3 is a schematic diagram of particle-to-particle update in the PSO algorithm.

Figure 4 is a comparison chart of moisture content measurement data and predicted data.

Figure 5 is a comparison chart of the measured data and the predicted data of salt content.

DETAILED DESCRIPTION

The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

A method for measuring the water-salt content of rammed earth in this embodiment includes the following steps:

Step 1: Initialize the population and set the population size in the algorithm model.

Particle swarm optimization (PSO) is an optimization algorithm inspired by the predation behavior of bird flocks. It has a fast convergence speed and its basic idea is to find the optimal solution through collaboration and information sharing between groups and individuals. Particle swarm optimization abstracts each feasible solution into a particle point and extends it to N-dimensional space, so that each individual's position vector _Di = ( _d1 , _d2 , ..., _dN ) and flight speed vector _Vi = ( _v1 , _v2 , ..., _vN ) in N-dimensional space are obtained. Each particle has a fitness, whose value is determined by the objective function. Particles find positions in space according to conditions and adjust their position vectors and flight speed vectors. In this process, each particle will generate an individual historical optimal solution p_best and obtain the global optimal solution g_best in the population through communication. According to p_best and g_best, they continue to adjust their position vectors and speed vectors to finally reach the optimal solution.

Step 2: Establish a generalized regression network model to predict the water-salt content of rammed earth

The generalized regression neural network (GRNN) is a radial basis neural network based on mathematical statistics. It is usually used for regression approximation of nonlinear functions. It has a fast training speed, much smaller than the BP neural network, and has higher accuracy under the condition of small samples. As shown in Figure 2, the GRNN network consists of a four-layer structure, including an input layer, a pattern layer, a summation layer, and an output layer. The number of neurons in the input layer is the same as the vector dimension in the training sample, and the input vector X is directly output to the pattern layer. In the prediction of the water and salt content of rammed soil, the input layer parameters are set to the capacitance, conductivity, soil temperature, and soil density of the rammed soil, and the output layer is the water content and salt content of the rammed soil.

The theoretical basis of GRNN is nonlinear regression analysis. Assuming x and y are two random variables, _x0 is the observed value of x, and the joint probability density of x and y is f(x, y), then:

为输入为x₀条件下y的预测输出，f(x₀，y)为x₀与y的联合概率密度。in

is the predicted output of y under the condition that the input is _x0 , and f( _x0 , y) is the joint probability density of _x0 and y.

表示样本数据集，其中n为样本总数，x_i，y_i为第i个样本的观察值。对样本数据集应用Parzen非参数估计，可以得到估算概率密度函数：use

Represents a sample data set, where n is the total number of samples, and x _i , y _i are the observed values of the i-th sample. Applying Parzen nonparametric estimation to the sample data set, we can get the estimated probability density function:

Where p is the dimension of the random variable x, and σ is the width coefficient (standard deviation) of the Gaussian function, also known as the smoothing factor.

带入式(1)，得：Will

Substituting into formula (1), we get:

The number of neurons in the pattern layer is the same as the number of training samples, that is, each neuron has a corresponding training sample. In the pattern layer, the Euclid distance between the test sample and all training samples is calculated. The transfer function of the pattern layer is:

Where X is the network input vector, _Xi is the training sample corresponding to the i-th neuron; n is the number of training samples, which is also the number of neurons in the pattern layer.

Two types of neurons are used in the summation layer for summation. The first calculation directly sums the output of the pattern layer. The formula is:

The second calculation is to perform a weighted summation on the output of the pattern layer. The connection weight between the rth neuron in the pattern layer and the tth molecule in the summation layer is multiplied by the neuron in the rth pattern layer and summed. The formula is:

The transfer functions of these two types of neurons are:

Where S _D is the transfer function when the weight is 1, S _Nt is the transfer function of the t-th neuron, _wrt is the connection weight between the r-th neuron in the pattern layer and the t-th neuron in the summation layer, and s is the number of weighted summation neurons in the pattern layer.

The output of the output layer neuron is:

Where q is the dimension of the output vector. In the prediction algorithm of water-salt content of rammed earth, q is 2.

In this regression model, the input layer parameters are the capacitance, conductivity, temperature and rammed earth density measured by the measurement module, and the output layer is the moisture content and salt content. Since the weight factor of each observation Yi is the exponent of the square of the Euclid distance between the corresponding sample _Xi and X. Therefore, the accuracy of the predicted value depends largely on the value of the smooth factor σ. When σ is very large, -(XX _r ) ^T (XX _r )/2σ ² approaches 0, so the predicted value is approximately equal to the average value of the dependent variable of the training sample; when σ approaches 0, exp[-(XX _r ) ^T (XX _r )/2σ ² ] approaches 0, and the distance between Y and the training sample is very close. Only when the predicted point is in the training sample, the predicted value will have a good accuracy, and when the predicted point is not in the training sample, the prediction effect will be very poor, that is, the network generalization ability is poor. Therefore, selecting a suitable smooth factor is very important for the generalized regression neural network, which is related to the accuracy and generalization ability of the model. In order to select the most suitable smoothing factor, the traditional empirical method has low accuracy and is time-consuming, so the improved GA-PSO algorithm is used to optimize the smoothing factor.

Step 3: Calculate the fitness of each particle

When predicting the moisture content or salt content of rammed soil, the root mean square error (RMSE) between the actual value of the moisture content or salt content and the predicted value is used as the fitness of the particle, and the fitness of each particle is calculated. The root mean square error is calculated as follows:

为第i个样本的预测值，y_i为第i个样本的实际值。in

is the predicted value of the ith sample, and _yi is the actual value of the ith sample.

Step 4: Update the state of the particle

The speed and displacement of each particle in the particle swarm algorithm are calculated according to the following two formulas:

分别是第i个粒子在第k轮和第k+1轮迭代的速度矢量；

是分别是第i个粒子在第k轮和第k+1轮迭代的位置矢量；

是第k轮的个体最优解，

是第k轮的群体最优解；ω是惯性因子；r₁、r₂是介于(0，1)之间的随机数；c₁、c₂是学习因子。Where i = 1, 2, ... M, M is the total number of particles in the particle group;

are the velocity vectors of the i-th particle in the k-th and k+1-th iterations respectively;

are the position vectors of the i-th particle in the k-th and k+1-th iterations respectively;

is the individual optimal solution in round k,

is the optimal solution of the group in the kth round; ω is the inertia factor; r ₁ and r ₂ are random numbers between (0, 1); c ₁ and c ₂ are learning factors.

Step 5: Perform crossover and mutation operations

The crossover operation and mutation operation in the improved GA-PSO algorithm are similar to those in the genetic algorithm.

Step 5.1: Perform crossover operation

The update of particle position and velocity introduces cross operation, and the particle position and velocity update formula is:

是分别是第j个粒子在第k轮和第k+1轮迭代的位置矢量；

分别是第j个粒子在第k轮和第k+1轮迭代的速度矢量；α₁、α₂为区间(0，1)的随机数。in,

are the position vectors of the jth particle in the kth and k+1th iterations respectively;

are the velocity vectors of the jth particle in the kth and k+1th iterations respectively; α ₁ and α ₂ are random numbers in the interval (0, 1).

Step 5.2: Perform mutation operation:

The mutation operation mutates individuals according to a certain probability to achieve the purpose of optimizing individuals and enriching the group. This is also an important reason why the algorithm is not easy to fall into the local optimal solution. The mutation operation is performed as follows:

Where P _m and P′ _m are the individuals before and after mutation, P _max and P _min are the upper and lower limits of the solution space, β is the mutation probability, k is the current iteration number, and mg is the total iteration number.

Step 6: Determine whether the convergence conditions are met

If the optimized GRNN model meets the accuracy requirements, stop; otherwise, return to step 2 and adjust the parameters to continue training.

Step 7: Determine the final GRNN model and make predictions

The optimized GRNN model can predict the moisture content and salt content of rammed earth.

Example

Data collection was performed on the homemade rammed earth samples to measure the capacitance, conductivity, soil temperature, and soil density of the rammed earth. The moisture content and salt content were measured using the laboratory method. The data collected in the experiment were predicted using the patented method, and the predicted results were compared with the measurement results obtained by the laboratory method. The results obtained are shown in Figures 4 and 5, where Figure 4 is a comparison between the moisture content measurement data and the predicted data, and Figure 5 is a comparison between the salt content measurement data and the predicted data. The experiment shows that the predicted value obtained by this method has a good correlation with the measured value, and the accuracy is high.

Claims (5)

1. A method for measuring the water-salt content of rammed earth, characterized in that it comprises the following steps: Step 1: Initialize the population and set the population size in the GRNN model optimized by GA-PSO; Step 2: Establish a generalized regression network model, taking the capacitance, conductivity, soil temperature, and soil density of the rammed earth as input parameters, and the predicted moisture content and salt content of the rammed earth as output parameters; Step 3: Calculate the fitness of each particle: The root mean square error between the actual value and the predicted value of the rammed soil moisture content or salt content is taken as the fitness of the particle; Step 4: Update the state of the particle: Update the velocity and displacement of each particle in the particle swarm algorithm; Step 5: Perform crossover and mutation operations: Introduce crossover operation to the update of particle position and velocity, and introduce mutation operation to the update of particle position and velocity; Step 6: Determine whether the convergence condition is met. If the optimized generalized regression network model meets the accuracy requirement, a generalized regression network is established to predict the moisture content and salt content of the rammed earth. If the accuracy does not meet the requirement, return to step 2. Step 7: Use the optimized generalized regression network to predict the moisture content and salt content of rammed earth. 2. The method for measuring the water-salt content of rammed earth according to claim 1 is characterized in that the crossover and mutation operations are performed: specifically comprising: The update formula for performing the crossover operation is:

in,

is the position vector of the jth particle at the kth iteration;

is the velocity vector of the jth particle in the kth iteration; α ₁ and α ₂ are random numbers in the interval (0,1); The update formula for performing mutation operation is:

Where P _m and P′ _m are the individuals before and after mutation, P _max and P _min are the upper and lower limits of the solution space, β is the mutation probability, and mg is the total number of iterations. 3. The method for measuring water and salt content of rammed earth according to claim 1, characterized in that establishing a generalized regression network model comprises: Set the smoothing factor of the GRNN network, input the rammed earth capacitance, conductivity, soil temperature, and soil density, and set the output to be the rammed earth moisture content and salt content; calculate the input layer, pattern layer, summation layer, and output layer. 4. The method for measuring the water and salt content of rammed earth according to claim 1 is characterized in that the fitness of each particle is calculated, and the root mean square error RMSE between the actual value and the predicted value of the water content or salt content is calculated as follows:

in

is the predicted value of the ith sample, _yi is the actual value of the ith sample, and n is the total number of samples. 5. The method for measuring water-salt content of rammed earth according to claim 1, characterized in that the state of the particles is updated: The following two formulas are performed:

Where i = 1, 2, ... M, M is the total number of particles in the particle group;

is the velocity vector of the i-th particle in the k-th iteration;

is the position vector of the i-th particle at the k-th iteration;

is the individual optimal solution in round k,

is the optimal solution of the group in the kth round; ω is the inertia factor; r ₁ and r ₂ are random numbers between (0,1); c ₁ and c ₂ are learning factors.

Images