Prediction method for unfrozen water content of frozen soil based on neural network
Through the neural network prediction model, combined with principal component analysis and genetic algorithm to optimize the BP neural network, the high cost and pollution problems of measurement of unfrozen water content in frozen soil are solved, and high-precision, low-cost, non-destructive in-situ measurement and real-time are achieved. monitor.
Patent Information
- Application Number
- CN202410890405.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-04
- Publication Date
- 2024-12-03
AI Technical Summary
The existing methods for measuring unfrozen water content in frozen soils have high costs, inability to achieve in-situ measurements, and may contaminate soil.
Using a neural network-based method, the BP neural network is optimized by collecting amplitude ratio data and temperature data of the frozen soil, combining principal component analysis and genetic algorithms to construct a prediction model for unfrozen water content in the frozen soil, and realize non-destructive in-situ measurement and real-time monitor.
It achieves high-precision and low-cost large-scale deployment, avoids soil pollution, and can monitor the unfrozen water content in real time.
Smart Images

Figure CN119066405A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of measuring the unfrozen water content of frozen soil, and in particular to a method for predicting the unfrozen water content of frozen soil based on a neural network. Background Art
[0002] The ice and water content in frozen soil will change with the change of seasonal temperature. The unfrozen water content in frozen soil is related to the overall strength of the soil and the ground's support for buildings. In the process of railway and highway construction, management and maintenance, the strength of frozen soil and the unfrozen water content of frozen soil are indispensable considerations.
[0003] The existing methods for measuring the unmoved water content of frozen soil mainly include time domain reflectometry (TDR), frequency domain reflectometry (FDR), neutron method, nuclear magnetic resonance method, etc. The time domain reflectometry is based on measuring the speed of electromagnetic waves propagating in the soil to measure the dielectric constant of the soil; the frequency domain reflectometry uses the principle of electromagnetic pulses to measure the dielectric constant of the soil according to the frequency of electromagnetic waves propagating in the soil. However, TDR and FDR are expensive and the cost is too high. They cannot be deployed on a large scale at present. The neutron method has radiation, which can cause soil contamination and affect human health. The nuclear magnetic resonance method equipment is expensive and difficult to measure in situ. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a method for predicting the unfrozen water content of frozen soil based on a neural network, which does not require destruction of the soil structure, can achieve in-situ measurement, and can realize real-time monitoring without causing pollution to the soil, thereby solving the problem of difficulty in in-situ measurement and pollution in the prior art.
[0005] The present invention is achieved in this way.
[0006] A method for predicting unfrozen soil water content based on a neural network, the method comprising:
[0007] Collect the amplitude difference between the sine signal of different frequencies after passing through frozen soil and the amplitude of the original sine signal, i.e., the frozen soil amplitude ratio data;
[0008] The principal component analysis method is used to reduce the dimension of the amplitude ratio data, extract the principal components of the amplitude ratio data, take the first principal component containing the most information in the amplitude ratio data, and calculate the projection value s1 of each sample on the first principal component; use the temperature sensor to collect the frozen soil temperature data T and determine the soil type st;
[0009] A prediction neural network model of the projection value s1, frozen soil temperature T, soil type st and unfrozen soil water content is established through a neural network, and the unfrozen soil water content is predicted using the prediction neural network model.
[0010] Furthermore, the prediction neural network model is constructed by:
[0011] S1: Collect training sample data: Prepare five soil types with initial volume water content of 5%, 10%, 15%, 20%, 25% and 30%; Place the samples in a -20℃ constant temperature box, collect the amplitude ratio data of the samples at different frequencies every 60 minutes, and collect 48 times in total. Each set of data collects the amplitude ratio data of m frequency points, and records the frozen soil temperature data and the frozen soil unfrozen water content θ corresponding to the frozen soil temperature when the time domain reflectometry is used to measure the recorded frozen soil temperature w ;
[0012] S2: Perform principal component analysis on the collected amplitude ratio data and extract the principal components, including: Step 1: Standardize the matrix: For a sample, observe the amplitude ratios of m frequency points x1, x2, ... x m , the amplitude ratio matrix of n groups of frozen soil is:
[0013]
[0014] in:
[0015] Calculate mean by column and standard deviation Calculate the standardized data The amplitude ratio matrix is normalized to
[0016] Step 2: Calculate the correlation coefficient matrix:
[0017] Step 3: Calculate the eigenvector and eigenvalue: Rβ=λβ, that is, (R-λE)β=0, and obtain the eigenvalue of the correlation coefficient matrix: λ1≥λ2≥…λ n ≥0, eigenvector: β1,β2,…,β n , then the eigenvector matrix V = (β1,β2,…,β n );
[0018] Step 4: Calculate the contribution rate of each principal component:
[0019] Step 5: Calculate the projection matrix: S = XV = (s1, s2, ..., s n ), and extract the first principal component according to the principal component contribution rate to obtain vector s1;
[0020] Step 3: Take the soil amplitude ratio data s1, temperature T, and soil type st after PCA dimension reduction as input, and the unfrozen soil water content θ wThe output is imported into the genetic algorithm optimized BP neural network, and after training, a neural network model for predicting the unfrozen water content of frozen soil is obtained.
[0021] Compared with the prior art, the present invention has the following beneficial effects:
[0022] 1. The present invention obtains the amplitude ratio data, temperature data and soil type of frozen soil and inputs them into the frozen soil unfrozen water content prediction neural network model to obtain the unfrozen water content of frozen soil, thereby simplifying complex problems and achieving high measurement accuracy.
[0023] 2. The present invention has strong practicability, simple implementation, low cost, and can be deployed over a large area.
[0024] 3. The present invention does not need to destroy the soil structure, can realize in-situ measurement, and can realize real-time monitoring without causing pollution to the soil. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 A block diagram of a frozen soil temperature and amplitude ratio data acquisition system provided by an embodiment of the present invention;
[0026] Figure 2 A flowchart of a method provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0027] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0028] The present invention adopts Figure 1 The data acquisition system shown in the figure firstly generates a swept frequency sinusoidal electrical signal with a specific frequency range by controlling the signal generator of the single chip microcomputer, and at the same time controls the temperature sensor and reads the temperature of the frozen soil through the measuring probe. Furthermore, the signal generator transmits the generated swept frequency sinusoidal electrical signal to the measuring probe and the amplitude detector. Due to the different unfrozen water content in the frozen soil, the electrical signal transmitted to the frozen soil through the measuring probe will suffer different degrees of loss. The lost electrical signal is transmitted to the amplitude detector again through the measuring probe.
[0029] The amplitude detector reflects the amplitude ratio by comparing the amplitude difference between the original electrical signal and the electrical signal after loss and outputting a voltage signal.
[0030] The microcontroller reads the voltage signal output by the amplitude detector through its own ADC and performs simple data processing to obtain the amplitude ratio data.
[0031] The single-chip microcomputer sends the acquired amplitude ratio data and temperature data of different frequency points to the host computer through the serial port.
[0032] The host computer performs principal component analysis on the amplitude ratio data to obtain the projection value on the first principal component.
[0033] Finally, the projection value, temperature data, and soil type are input into the pre-built neural network model for predicting the unfrozen water content of frozen soil to obtain the unfrozen water content.
[0034] See also Figure 2 As shown, the present invention provides a method for predicting the unfrozen water content of frozen soil based on a neural network, the method comprising:
[0035] Collect the amplitude difference between the sine signal of different frequencies after passing through frozen soil and the amplitude of the original sine signal, i.e., the frozen soil amplitude ratio data;
[0036] The principal component analysis method is used to reduce the dimension of the amplitude ratio data, extract the principal components of the amplitude ratio data, take the first principal component containing the most information in the amplitude ratio data, and calculate the projection value s1 of each sample on the first principal component; use the temperature sensor to collect the frozen soil temperature data T and determine the soil type st;
[0037] A prediction neural network model of the projection value s1, frozen soil temperature T, soil type st and unfrozen soil water content is established through a neural network, and the unfrozen soil water content is predicted using the prediction neural network model.
[0038] The method for constructing the neural network model for predicting the unfrozen water content of frozen soil is as follows:
[0039] Step 1: Training sample data collection, prepare five soil types with initial volume water content of 5%, 10%, 15%, 20%, 25% and 30%, a total of 30 groups of samples. Place them in a -20℃ constant temperature box, collect the amplitude ratio data of the frozen soil at different frequencies every 60 minutes, a total of 48 times, each group of data has m frequency points of amplitude ratio data, and record the frozen soil temperature data and the frozen soil unfrozen water content θ at that time measured by TDR. w .
[0040] Step 2: Perform principal component analysis on the collected amplitude ratio data to extract its principal components, including: Step 1: Standardize the matrix: For a sample, observe the amplitude ratios of m frequency points x1, x2, ... x m , the amplitude ratio matrix of n groups of frozen soil is:
[0041]
[0042] in:
[0043] Calculate mean by column and standard deviation Calculate the standardized data The original sample matrix is normalized to
[0044] The second step is to calculate the correlation coefficient matrix:
[0045] The third step is to calculate the eigenvector and eigenvalue: Rβ=λβ, that is, (R-λE)β=0, from which the eigenvalue of the correlation coefficient matrix can be obtained: λ1≥λ2≥…λ n ≥0, with eigenvectors: β1,β2,…,β n , then the eigenvector matrix V = (β1,β2,…,β n );
[0046] The fourth step is to calculate the contribution rate of each principal component:
[0047] The fifth step is to calculate the projection matrix: S = XV = (s1, s2, ..., s n ), and extract the first principal component according to the principal component contribution rate to obtain vector s1;
[0048] Step 3: Take the soil amplitude ratio data s1, temperature T, and soil type st after PCA dimension reduction as input, and the unfrozen soil water content θ w As the output, it is imported into the genetic algorithm optimized BP neural network. Since the model has 3 inputs and 1 output, the BP neural network structure is set to 3-7-1, that is, the input layer has 3 nodes, the hidden layer has 7 nodes, and the output layer has 1 node. There are 3×7+7=28 weights, 7+1=8 thresholds, so the genetic algorithm individual coding length is 28+8=36. After training, the neural network model for predicting the unfrozen water content of frozen soil is obtained.
[0049] The above genetic algorithm optimizes the BP neural network to achieve the following process:
[0050] 1. Randomly initialized population: The individual encoding method is real number encoding. Each individual is a real number string, which consists of four parts: the connection weight between the input layer and the hidden layer, the hidden layer threshold, the connection weight between the hidden layer and the output layer, and the output layer threshold. The individual contains all the weights and thresholds of the neural network. When the network structure is known, a neural network with a certain structure, weights, and thresholds can be constructed.
[0051] 2. Calculate the fitness value of the population and find the best individual: Get the initial weight and threshold of the BP neural network based on the individual, train the BP neural network with the training data and predict the system output, and take the absolute value of the error between the predicted output and the expected output as the individual fitness value F. The calculation formula is:
[0052]
[0053] Where n is the number of network output nodes; y i is the expected output of the ith node of the BP neural network; o i is the predicted output of the i-th node; k is the coefficient.
[0054] 3. Selection operation: The genetic algorithm selection operation consists of a variety of methods such as roulette method and tournament method. The present invention selects the roulette method, that is, a selection strategy based on fitness ratio. The selection probability p of each individual i is i The calculation formula is:
[0055]
[0056] In the formula, F i is the fitness value of individual i; k is the coefficient; N is the number of individuals in the population.
[0057] 4. Crossover operation: Since individuals are coded with real numbers, real number crossover is used. The kth chromosome a k and a l The crossover operation method at position j is as follows:
[0058]
[0059] Where b is a random number between [0,1].
[0060] 5. Mutation operation: select the i-th gene a ij Perform mutation, the mutation operation is as follows
[0061]
[0062] Where r is a random number between [0,1].
[0063] 6. Determine whether the evolution is finished. If not, return to step 2.
[0064] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for predicting the unfrozen water content of frozen soil based on a neural network, characterized in that: The method includes: Collect the amplitude difference between the sine signal of different frequencies after passing through frozen soil and the amplitude of the original sine signal, i.e., the frozen soil amplitude ratio data; The principal component analysis method is used to reduce the dimension of the amplitude ratio data, extract the principal components of the amplitude ratio data, take the first principal component containing the most information in the amplitude ratio data, and calculate the projection value s1 of each sample on the first principal component; use the temperature sensor to collect the frozen soil temperature data T and determine the soil type st; A prediction neural network model of the projection value s1, frozen soil temperature T, soil type st and unfrozen soil water content is established through a neural network, and the unfrozen soil water content is predicted using the prediction neural network model.
2. The method for predicting the unfrozen water content of frozen soil based on a neural network according to claim 1, characterized in that: The construction method of the prediction neural network model is: S1: Collect training sample data: Prepare five soil types with initial volume water content of 5%, 10%, 15%, 20%, 25% and 30%; Place the samples in a -20℃ constant temperature box, collect the amplitude ratio data of the samples at different frequencies every 60 minutes, and collect 48 times in total. Each set of data collects the amplitude ratio data of m frequency points, and records the frozen soil temperature data and the frozen soil unfrozen water content θ corresponding to the frozen soil temperature when the time domain reflectometry is used to measure the recorded frozen soil temperature w ; S2: Perform principal component analysis on the collected amplitude ratio data to extract the principal components.
3. The method for predicting the unfrozen water content of frozen soil based on a neural network according to claim 2, characterized in that: The collected amplitude ratio data is subjected to principal component analysis, and the steps of extracting the principal component include: The first step is to standardize the matrix: for a sample, observe the amplitude ratio of m frequency points x1, x2, ... x m , the amplitude ratio matrix of n groups of frozen soil is: in: Calculate mean by column and standard deviation Calculate the standardized data The amplitude ratio matrix is normalized to Step 2: Calculate the correlation coefficient matrix: Step 3: Calculate the eigenvector and eigenvalue: Rβ=λβ, that is, (R-λE)β=0, and obtain the eigenvalue of the correlation coefficient matrix: λ1≥λ2≥…λ n ≥0, eigenvector: β1,β2,…,β n , then the eigenvector matrix V = (β1,β2,…,β n ); Step 4: Calculate the contribution rate of each principal component: Step 5: Calculate the projection matrix: S = XV = (s1, s2, ..., sn), and extract the first principal component according to the principal component contribution rate to obtain vector s1; Step 3: Take the soil amplitude ratio data s1, temperature T, and soil type st after PCA dimension reduction as input, and the unfrozen soil water content θ w The output is imported into the genetic algorithm optimized BP neural network, and after training, a neural network model for predicting the unfrozen water content of frozen soil is obtained.
Citation Information
Patent Citations
Calculation method of unfrozen water content in frozen soil based on clay colloidal chemical electric double layer theory
CN109142444A
Algae bloom prediction method based on principal component analysis and BP neural network
CN110046645A
Capacitive soil moisture monitoring system and method
CN111693580A
Method for detecting content of unfrozen water in frozen soil
CN113009107A
Method for predicting surface quality of workpiece based on optimized generalized regression neural network
CN114139772A
Cited By
Multi-parameter integrated detection method and device for frozen soil, terminal and storage medium
CN119959517A
