Resistance tomography detection method for health condition of foundation of power transmission tower

Abstract

The invention discloses a power transmission tower foundation health condition electrical resistance tomography detection method, which is characterized in that an electrical resistance tomography system is constructed based on power transmission tower foundation health detection requirements, firstly, a simple electrical resistance tomography system is constructed, and accurate solutions of four unknown numbers are obtained through four simultaneous equations; the resistivity values are the resistivity values of the four tower leg pile foundations, and the health conditions of the tower leg pile foundations are preliminarily reflected. Furthermore, on the basis of an electrical resistance tomography system, positive problem simulation is carried out by using finite elements, and a positive problem solving database is established through millions of finite element solutions. And training a neural network by using a finite element solution result as an input parameter by using machine learning, and directly solving an inverse problem according to a result of the machine learning, so that more accurate resistivity distribution characteristics of the power transmission tower foundation are obtained, and the health condition of the power transmission tower foundation is more accurately judged.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence resistance tomography detection methods, and in particular relates to a resistance tomography detection method for the health status of a transmission tower foundation. Background Art

[0002] The western part of my country is vast and sparsely populated, and new power generation methods continue to emerge, and the power generation is also increasing; the eastern region is the main area of ​​electricity consumption, which has led to an increasingly prominent demand for the improvement of power transmission capacity. The core of all the above is power transmission, and the power transmission and transformation infrastructure project serves as the carrier platform for everything. However, the topography and geology of the western region of my country are complex and diverse, with mountains crisscrossing, peaks overlapping, canyons and basins everywhere, and most of the area is plateau, with a variety of topography and landforms such as inland arid basins, karst landforms and loess plateaus, and a variety of climate forms such as continental monsoon climate, inland arid climate and Qinghai-Tibet Plateau climate. All of these have put forward higher requirements for power transmission and transformation projects and formed new challenges. The load borne by power grid facilities will increase with the commissioning of high-voltage, long-distance and large-capacity transmission lines, and higher requirements have been put forward for the structure and form design of power grid facilities and foundation operation.

[0003] In recent years, the power grid in Northwest China has repeatedly encountered safety problems caused by the collapse of the foundation of the transmission tower, uneven settlement, landslides, and corrosion. These defects are often hidden, have a long latent time, deteriorate suddenly, and the accident hazards are particularly huge, which has caused great harm to the safety, reliability and stability of the power grid. Traditional monitoring facilities often monitor the infrastructure of power transmission and transformation lines. However, when the infrastructure is in poor condition, it often means that it has reached a point where it is difficult to remedy. Therefore, it is urgent to develop more systematic monitoring facilities and more accurate methods to achieve systematic monitoring of the foundation range involved in the infrastructure and foundation, which will help to conduct in-depth research and analysis on the health status of power grid facilities and foundations under various working conditions, and thus comprehensively improve the safe operation level of high-voltage transmission lines.

[0004] Compared with monitoring the health of the foundation of power facilities, an effective method is to monitor the resistivity characteristics of the foundation around the concrete foundation to reflect its stress and health. The resistivity tomography method is not only applicable to concrete foundations, but also to monitoring the deformation of the earth. It can monitor the resistivity of power transmission and transformation facilities and their foundations in real time, so as to accurately predict the uneven settlement of the foundation of power facilities, the stress state of the foundation concrete, the safety status of the foundation, etc., and then realize the monitoring and early warning of cracks, settlement, landslides, etc. in the ground around the foundation of power transmission facilities. Summary of the invention

[0005] The present invention aims to solve the problem of low accuracy in the health detection of the foundation of a transmission tower in the background technology, especially the technical problem that the resistance tomography result is not accurate and the detection accuracy is greatly reduced due to the inverse problem that the resistance tomography method needs to solve. The present invention aims to provide a resistance tomography detection method for the health status of the foundation of a transmission tower, which greatly improves the calculation accuracy and the detection accuracy.

[0006] In order to solve the technical problem, the technical solution of the present invention is:

[0007] A method for detecting the health status of a transmission tower foundation by electrical resistance tomography, the method comprising:

[0008] Electrodes are buried at different depths around the transmission tower pile foundation to build a multi-layer health preliminary judgment system and electrical resistance tomography system, with each layer of electrodes evenly distributed along the circumference of the transmission tower;

[0009] The current is measured by multiple electrodes of the health preliminary judgment system, and the equation is established to solve the regional resistivity to obtain the preliminary judgment result of the pile foundation health;

[0010] Conducting current measurement through multiple electrodes of the electrical resistance tomography system to construct a current matrix raw data set;

[0011] Finite element methods are used to simulate scenarios with different locations, different areas, and implanted materials with different resistivity values, calculate the current response between each electrode pair under known voltage, and build a massive database of direct problems;

[0012] Based on the positive problem database, a neural network model is constructed and trained, and the regularization optimization model is used to obtain the optimized neural network model;

[0013] The original data set is input into the trained neural network model, and the predicted underground resistivity distribution is output.

[0014] Furthermore, the method also includes: judging the health status of the transmission tower foundation by analyzing the predicted bottom resistivity distribution at different depths.

[0015] Furthermore, a multi-layer health preliminary judgment system and electrical resistance tomography system are constructed, including:

[0016] Eight electrodes of the health preliminary judgment system and eight, sixteen or twenty-four electrodes of the electrical resistance tomography system are buried at the depth of each layer, and the electrodes are spaced and distributed on a circumference.

[0017] Furthermore, the 8 electrodes of the preliminary health judgment system are used to make a preliminary judgment on the basic health conditions of the four legs and feet:

[0018] af=A+C

[0019] be=B+D

[0020] hc=A+B

[0021] gd=C+D

[0022] Solve the above four equations simultaneously, A, B, C, D are four unknowns, and the four current parameters af, be, hc, gd can be obtained through testing; thus, the resistivity of the four areas A, B, C, D are obtained, which correspond to the four pile foundations respectively, and a preliminary judgment can be made on the health status of the four pile foundations.

[0023] Furthermore, the current measurement and current matrix construction include:

[0024] Eight electrodes were used for electrical resistance tomography. A specific voltage was applied between the a' and b' electrode pairs, and then the current between any two electrodes was measured, with a total of 28 sets of current values ​​recorded. Similarly, the voltage was repeatedly applied and the current was measured between other electrode pairs, and the 28 sets of current values ​​for each measurement were recorded in turn. In this way, an 8×28 current matrix was eventually obtained, which contained the current response data between all electrodes.

[0025] Furthermore, the positive problem database includes current matrices and corresponding resistivity data under different resistivity distributions.

[0026] Furthermore, the input of the neural network model is a current matrix, and the output is a resistivity distribution; the neural network model adopts one of a multi-layer perceptron MLP, a convolutional neural network CNN, and a deep neural network DNN, and the regularization method adopts L2 regularization or Dropout regularization.

[0027] Compared with the prior art, the advantages of the present invention are:

[0028] Multi-level health detection and accurate prediction: The present invention can realize multi-level health detection of the transmission tower foundation by burying multiple electrodes at different depths, combining the health preliminary judgment system and the resistance tomography system. The preliminary judgment system uses simple electrode measurement data to quickly obtain the health status of the pile foundation, while the resistance tomography system provides a more accurate health assessment through fine resistivity distribution calculation. This multi-level detection method can effectively improve the accuracy and reliability of detection.

[0029] Finite element simulation and large-scale data support: Through finite element direct problem simulation, a large amount of current response data under different resistivity distribution conditions is generated, and a huge direct problem database is constructed. This direct problem database based on real physical simulation provides a large amount of high-quality data for neural network training, ensuring that the model can learn the complex nonlinear relationship between current response and resistivity distribution, and improving the accuracy and stability of prediction.

[0030] Inverse problem solving based on machine learning: Using neural networks (such as MLP) to solve inverse problems can efficiently handle complex nonlinear relationships. By taking the current matrix as input and combining it with large-scale training data, the neural network can quickly infer the resistivity distribution of the underground medium. This process transforms the traditional complex inverse problem solving process into a machine learning prediction task, greatly simplifying the calculation process and significantly improving the solution efficiency.

[0031] Regularization technology improves the generalization ability of the model: L2 regularization is introduced in the neural network training process to prevent overfitting by limiting the weight size of the model. The application of regularization technology not only improves the generalization ability of the model, but also ensures that the model can still maintain high accuracy and stability when facing different test data.

[0032] Comprehensive foundation health assessment: Through in-depth analysis of the electrical resistance tomography results, a detailed understanding of the foundation health at different depth levels can be obtained. The solution can effectively identify abnormal changes in underground resistivity. For example, high resistivity may indicate dry soil or cracks, while low resistivity may indicate underground structural damage or moisture. This comprehensive foundation health assessment helps to detect potential problems in the transmission tower foundation in a timely manner and avoid serious structural damage.

[0033] Improve detection efficiency and automation: Compared with traditional foundation health detection methods, the solution based on electrical resistance tomography combined with machine learning has greatly improved the efficiency and automation of detection. The inverse problem solving can provide high-precision underground resistivity distribution in a short time through rapid prediction of neural networks, reducing the difficulty of manual analysis and complex calculations in traditional methods, and improving the overall detection speed and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 , transmission tower foundation structure diagram and electrode distribution diagram;

[0035] Figure 2 , an electrode distribution diagram using 8 electrodes in the preliminary health judgment system;

[0036] Figure 3 , Electrode distribution diagram using 8 electrodes in electrical resistance tomography. DETAILED DESCRIPTION

[0037] The specific implementation mode of the present invention is described below in conjunction with embodiments:

[0038] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0039] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0040] Embodiment 1:

[0041] like Figure 1 and 2 As shown, electrodes are buried at different depths underground around the transmission tower to construct several layers of electrical resistance tomography systems.

[0042] 8+8 electrodes, 8+16 electrodes, or 8+24 electrodes are buried at each depth, of which 8 electrodes are used to make a preliminary judgment (actual measurement) on the health status of the foundation at the four pile foundations of the tower base, and the other 8, 16 or 24 electrodes are used for electrical resistance tomography (simulation to obtain massive data) to more accurately calculate the resistivity distribution of the earth around the foundation of the transmission tower. Each layer is evenly distributed along the circumference.

[0043] The buried depth of each electrode layer includes 1m~6m per layer, 1m~7m per layer, 1m~8m per layer, etc.

[0044] 8 electrodes are used to make a preliminary assessment of the health of the foundations of the four legs;

[0045] af=A+C

[0046] be=B+D

[0047] hc=A+B

[0048] gd=C+D

[0049] Solve the above four equations simultaneously, A, B, C, D are four unknowns, and the four parameters af, be, hc, gd can be obtained through testing. Thus, the resistivity of the four regions A, B, C, and D is obtained, corresponding to the four pile foundations respectively. Therefore, a preliminary judgment can be made on the health status of the four pile foundations.

[0050] like Figure 3 As shown, another 8, 16 or 24 electrodes are used for electrical resistance tomography to make a more accurate prediction of the health of the foundation. The inverse problem of electrical resistance tomography is solved using the finite element big data machine learning method. The specific steps are as follows (taking 8-electrode electrical resistance tomography as an example):

[0051] Current measurement and matrix construction:

[0052] A specific voltage is applied between the a'b' electrode pair, and then the current between any two electrodes is measured, and a total of 28 sets of current values ​​are recorded. Similarly, the voltage is repeatedly applied and the current is measured between other electrode pairs such as b'c', and the 28 sets of current values ​​measured each time are recorded in sequence. In this way, an 8×28-order current matrix is ​​finally obtained, which contains the current response data between all electrodes and provides the original data for the subsequent solution of the inverse problem.

[0053] Finite element simulation of the direct problem:

[0054] In order to generate enough data to train the machine learning model, the current response under different resistivity distribution conditions is first simulated by the finite element method. Specifically, the finite element direct problem solution simulates the scenario of implanting materials with different resistivity values ​​at different locations and different areas in the region, and calculates the current response between each electrode pair under a known voltage. These direct problem simulation data constitute a database of millions of direct problems.

[0055] Establishment of the problem database:

[0056] By solving finite element direct problems millions of times, a huge direct problem database was established. This database records the current matrix and corresponding resistivity information under different resistivity distributions. This database is the core input for machine learning model training, ensuring that the neural network can learn the complex relationship between current response and resistivity distribution.

[0057] Machine Learning Model Training for Neural Networks:

[0058] The input of the neural network model is the current matrix, and the output is the resistivity distribution. Commonly used neural network architectures include multi-layer perceptron (MLP), convolutional neural network (CNN), deep neural network (DNN), etc.

[0059] MLP captures the basic relationship between the current matrix and resistivity through hidden layers and is suitable for simple structures.

[0060] CNN extracts local features of the matrix through convolution kernels, which is suitable for processing two-dimensional data and can capture local resistivity changes in current response.

[0061] DNN is suitable for complex nonlinear scenarios and can capture the deep features of large-scale positive problem data.

[0062] During training, the model can use the following regularization techniques to improve generalization:

[0063] L2 regularization (weight decay): By limiting the model parameters, overfitting can be avoided and the performance of the model on unknown data can be improved.

[0064] Dropout regularization: Randomly ignore some neurons during the training process so that the model uses a different combination of neurons each time it is trained, improving the robustness and generalization ability of the model.

[0065] The generalization ability of the model is not only improved through regularization techniques, but also uses data augmentation and cross-validation to ensure its stable performance under different resistivity distribution scenarios.

[0066] Inverse problem solving:

[0067] When the measured current matrix is ​​input into the trained neural network, the network will directly predict the resistivity distribution of the underground medium through the mapping relationship it has learned. This process transforms the solution of the inverse problem into an image recognition task. The current matrix is ​​used as the input "image", and the network recognizes the features in the image to invert the underground resistivity distribution.

[0068] Judgment of foundation health condition:

[0069] The health of the transmission tower foundation can be determined by analyzing the resistivity distribution of the formation at different depths (such as 1 meter, 2 meters, etc.). Changes in resistivity can reflect the physical state of the underground medium. For example, high resistivity may indicate dry soil, while low resistivity may indicate the presence of moisture or cracks in the underground structure. Ultimately, this method can provide more accurate predictions and help to detect potential problems in the foundation in a timely manner.

[0070] Embodiment 2:

[0071] Model selection and construction

[0072] For the mapping problem of current matrix to resistivity distribution, the following common neural network architectures can be used:

[0073] (1) Multilayer Perceptron (MLP)

[0074] MLP is a fully connected neural network suitable for processing tabular data or vector inputs, such as current matrices. MLP is characterized by nonlinear mapping through hidden layers and is suitable for simpler data patterns.

[0075] Input layer: The 8×28 current matrix is ​​flattened into a one-dimensional vector (224 input nodes).

[0076] Hidden layer: Generally, 2 to 3 hidden layers are selected, with 100-200 neurons in each layer. The activation function is ReLU (Rectified Linear Unit).

[0077] Output layer: The output is a resistivity distribution map. The shape of this output can be determined according to the spatial resolution you want to predict.

[0078] from tensorflow.keras.models import Sequential

[0079] from tensorflow.keras.layers import Dense

[0080] #Build MLP model

[0081] model=Sequential()

[0082] model.add(Dense(128,input_dim=224,activation='relu'))

[0083] model.add(Dense(128,activation='relu'))

[0084] model.add(Dense(128,activation='relu'))

[0085] model.add(Dense(1,activation='linear'))#Assume the output is a single-value resistivity

[0086] (2) Convolutional Neural Network (CNN)

[0087] CNN extracts local features through convolution kernels and is suitable for processing two-dimensional current matrix data. For resistance tomography problems, CNN can well capture the local resistivity changes in the current matrix.

[0088] Input layer: The 8×28 current matrix is ​​regarded as a two-dimensional input.

[0089] Convolutional layer: Multiple convolutional layers can be stacked, using 3×3 or 5×5 convolution kernels and ReLU activation functions.

[0090] Pooling layer: Use Max Pooling to reduce parameters and extract key features.

[0091] Fully connected layer: The output of the convolutional layer is flattened and connected to the fully connected layer, and finally the resistivity distribution is output.

[0092] from tensorflow.keras.models import Sequential

[0093] from tensorflow.keras.layers import Conv2D,MaxPooling2D,Flatten,Dense

[0094] #Build CNN model

[0095] model=Sequential()

[0096] model.add(Conv2D(32,(3,3),activation='relu',input_shape=(8,28,1)))

[0097] model.add(MaxPooling2D((2,2)))

[0098] model.add(Flatten())

[0099] model.add(Dense(128,activation='relu'))

[0100] model.add(Dense(1,activation='linear'))#Assume the output is a single-value resistivity

[0101] L2 regularization adds a penalty term of the square of the weight to the loss function to avoid excessive weights and improve the generalization ability of the model.

[0102] from tensorflow.keras.regularizers import l2

[0103] #Add L2 regularization to the hidden layer

[0104] model.add(Dense(128,activation='relu',kernel_regularizer=l2(0.001)))

[0105] Among them, the parameter in l2(0.001) is the regularization strength. The larger the value, the greater the penalty.

[0106] Dropout is an effective method to prevent overfitting. It randomly ignores a part of neurons in each training so that the model cannot rely on certain specific features, thereby improving the robustness of the model.

[0107] from tensorflow.keras.layers import Dropout

[0108] #Add Dropout to the hidden layer

[0109] model.add(Dense(128,activation='relu'))

[0110] model.add(Dropout(0.5))#50% of neurons will be randomly ignored

[0111] Before starting model training, data preprocessing is a very important step, including:

[0112] Normalization: Normalize the input current matrix data to a smaller range (such as 0 to 1) to increase training speed and reduce numerical instability problems.

[0113] from sklearn.preprocessing import MinMaxScaler

[0114] scaler = MinMaxScaler()

[0115] X_train_scaled = scaler.fit_transform(X_train) # Normalize the current matrix

[0116] Data enhancement: If the amount of data is not large enough, data enhancement techniques can be used, such as adding noise, rotating the current matrix, etc., to simulate more actual situations and improve the robustness of the model.

[0117] Model training:

[0118] Loss function: The commonly used loss function is mean square error (MSE), which is suitable for regression problems.

[0119] Optimizer: You can use the Adam optimizer because it performs well in most scenarios.

[0120] model.compile(optimizer='adam',loss='mean_squared_error')

[0121] #Model training

[0122] history=model.fit(X_train_scaled,y_train,epochs=100,batch_size=32,validation_split=0.2,callbacks=[early_stopping])

[0123] Model evaluation: During the training process, the model performance is monitored through the validation set to avoid overfitting. Cross-validation: k-fold cross-validation can be used to further evaluate the stability of the model.

[0124] Model deployment: Once the model is trained, the actual performance of the model can be evaluated using the test set. If the test results are good, the model can be deployed in actual applications, that is, the actual measured current matrix is ​​input into the trained model to predict the underground resistivity distribution.

[0125] Embodiment 3:

[0126] Inverse problem solving is particularly important in electrical resistance tomography because it infers the resistivity distribution of the subsurface by analyzing the current matrix. This process usually involves complex nonlinear mapping relationships, and neural networks (such as MLP) are an effective tool to solve this nonlinear inverse problem.

[0127] Next, we will describe in detail how to use MLP and L2 regularization to solve the inverse problem in electrical resistance tomography in combination with the steps of inverse problem solving. The process includes data generation, establishment of the positive problem database, model training, application of regularization, and solution and evaluation of the inverse problem.

[0128] 1. Definition of Inverse Problem

[0129] The inverse problem here refers to the inference of the output (the resistivity distribution of the ground) given the input (the measured current matrix). Specifically, the inverse problem solution aims to invert the resistivity distribution of the ground through the measured current values ​​between the electrodes (the current matrix).

[0130] Key Points:

[0131] Input: Measured current matrix (e.g., 8×28 matrix, flattened into a 224-dimensional vector).

[0132] Output: Resistivity distribution at different locations in the subsurface.

[0133] 2. Finite element simulation and database establishment of the positive problem

[0134] In order to let the neural network learn the mapping relationship between the current matrix and the resistivity distribution, it is necessary to first establish a positive problem database. The positive problem is to calculate the current response between each electrode when the underground resistivity distribution is known.

[0135] (1) Solving the problem

[0136] The finite element method (FEM) is used to simulate the direct problem. The simulation steps include:

[0137] Construct a subsurface model assuming different resistivities at different locations.

[0138] The finite element method is used to calculate the current response between electrodes under a known resistivity distribution.

[0139] Generate a large number of current matrices under different resistivity distributions and build a positive problem database.

[0140] This process generates millions of current matrices and their corresponding resistivity information under different resistivity distributions.

[0141] (2) Construction of the positive problem database

[0142] Generate a large number of data samples through finite element simulation and store these samples as a positive problem database:

[0143] Input of the database: current matrix under different resistivity distributions.

[0144] Output of the database: resistivity distribution.

[0145] These data will be used as training sets for training neural network models.

[0146] 3. Construction and training of neural network model (MLP+L2 regularization)

[0147] Once the positive question database is built, the next step is to train a neural network model based on this data. Here, a multi-layer perceptron (MLP) model is used, using L2 regularization to prevent overfitting.

[0148] (1) MLP model structure

[0149] Input layer: The input is a flattened current matrix (8×28 flattened into a 224-dimensional vector).

[0150] Hidden layer: Set 2 to 3 hidden layers, each layer contains 128 to 256 neurons. The activation function uses ReLU.

[0151] Output layer: The output is the subsurface resistivity distribution. The shape of the output depends on your model (it can be a scalar or a vector of resistivity distribution).

[0152] (2) L2 Regularization

[0153] To prevent overfitting, L2 regularization (also known as weight decay) is used to limit the weight values ​​of the model to a smaller range. L2 regularization adds a penalty term of the sum of squared weights to the loss function, so that the model does not rely on excessively large weight values ​​during optimization, improving the generalization ability of the model.

[0154] (3) Loss Function and Optimizer

[0155] The inverse problem is a regression problem, so the commonly used loss function is the mean square error (MSE), which measures the difference between the model output and the true resistivity distribution. The optimizer can use Adam, which is stable and converges quickly in practice.

[0156] (4) Data division and training

[0157] The data in the positive problem database is divided into a training set and a validation set. The training set is used to train the model, and the validation set is used to monitor whether the model is overfitting during the training process.

[0158] 4. Inverse problem solving process

[0159] After the model training is completed, the step of solving the inverse problem is to input the actual measured current matrix into the trained MLP model, and the model will output the predicted value of the underground resistivity. This process regards the measured current matrix as the input "image", and the underground resistivity distribution is the "feature" identified by the neural network.

[0160] (1) Input current matrix

[0161] When the new current matrix is ​​obtained (measured by electrodes around the transmission towers), it is normalized and then used as the input of the MLP model.

[0162] (2) Output underground resistivity distribution

[0163] The output of the model will be a prediction of the subsurface resistivity. This prediction can represent the resistivity at a certain location or the resistivity distribution throughout the subsurface.

[0164] 5. Evaluation of Inverse Problem Solving

[0165] In order to ensure the accuracy of the model, the solution of the inverse problem needs to be evaluated. The performance of the model can be evaluated using the following methods:

[0166] Mean Square Error (MSE): The mean square error between the resistivity distribution output by the model and the true resistivity distribution.

[0167] Cross-validation: K-fold cross-validation is used to ensure that the model performs stably on different datasets.

[0168] Actual scenario verification: Use the actual measured current matrix for testing to evaluate the generalization ability of the model.

[0169] 6. Result interpretation and foundation health assessment

[0170] Based on the resistivity distribution output by the model, the health of the transmission tower foundation can be further analyzed:

[0171] High resistivity: May indicate dry or voided subsoil.

[0172] Low resistivity: May indicate underground moisture or structural damage.

[0173] Through the resistivity distribution at different depths, a detailed prediction of the foundation's health status can be obtained, and potential problems can be discovered in a timely manner.

[0174] The entire inverse problem solving process is achieved through the following steps:

[0175] Forward problem simulation and database construction: Generate the correspondence between massive current matrix and resistivity distribution.

[0176] MLP model training: L2 regularization is used to prevent overfitting, and the model is trained to predict the resistivity distribution from the current matrix.

[0177] Inverse problem solving: The actual measured current matrix is ​​input into the model to predict the underground resistivity.

[0178] Result evaluation: Evaluate model performance through validation sets, cross-validation, etc., and apply it to actual scenarios.

[0179] This approach enables efficient and accurate solution of the inverse problem in electrical resistance tomography and ultimately the health of the transmission tower foundation to be assessed.

[0180] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

[0181] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A method for detecting the health status of a transmission tower foundation by electrical resistance tomography, characterized in that: The method comprises: Electrodes are buried at different depths around the transmission tower pile foundation to build a multi-layer health preliminary judgment system and electrical resistance tomography system, with each layer of electrodes evenly distributed along the circumference of the transmission tower; The current is measured by multiple electrodes of the health preliminary judgment system, and the equation is established to solve the regional resistivity to obtain the preliminary judgment result of the pile foundation health; Conducting current measurement through multiple electrodes of the electrical resistance tomography system to construct a current matrix raw data set; Finite element methods are used to simulate scenarios with different locations, different areas, and implanted materials with different resistivity values, calculate the current response between each electrode pair under known voltage, and build a massive database of direct problems; Based on the positive problem database, a neural network model is constructed and trained, and the regularization optimization model is used to obtain the optimized neural network model; The original data set is input into the trained neural network model, and the predicted underground resistivity distribution is output.

2. The method for detecting the health status of a transmission tower foundation by electrical resistance tomography according to claim 1, characterized in that: The method also includes: determining the health of the transmission tower foundation by analyzing the predicted bottom resistivity distribution at different depths.

3. The method for detecting the health status of a transmission tower foundation by electrical resistance tomography according to claim 1, characterized in that: Construct a multi-layer health preliminary judgment system and electrical resistance tomography system, including: Eight electrodes of the health preliminary judgment system and eight, sixteen or twenty-four electrodes of the electrical resistance tomography system are buried at the depth of each layer, and the electrodes are spaced and distributed on a circumference.

4. The method for detecting the health status of a transmission tower foundation by electrical resistance tomography according to claim 1, characterized in that: The 8 electrodes of the preliminary health judgment system are used to make a preliminary judgment on the health status of the four foot and leg foundations: af=A+C be=B+D hc=A+B gd=C+D Solve the above four equations simultaneously, A, B, C, D are four unknowns, and the four current parameters af, be, hc, gd can be obtained through testing; thus, the resistivity of the four areas A, B, C, D are obtained, which correspond to the four pile foundations respectively, and a preliminary judgment can be made on the health status of the four pile foundations.

5. The method for detecting the health status of a transmission tower foundation by electrical resistance tomography according to claim 1, characterized in that: The current measurement and current matrix construction include: Eight electrodes were used for electrical resistance tomography. A specific voltage was applied between the a' and b' electrode pairs, and then the current between any two electrodes was measured, with a total of 28 sets of current values ​​recorded. Similarly, the voltage was repeatedly applied and the current was measured between other electrode pairs, and the 28 sets of current values ​​for each measurement were recorded in turn. In this way, an 8×28-order current matrix was eventually obtained, which included the current response data between all electrodes.

6. The method for detecting the health status of a transmission tower foundation by electrical resistance tomography according to claim 1, characterized in that: The positive problem database includes current matrices and corresponding resistivity data under different resistivity distributions.

7. The method for detecting the health status of a transmission tower foundation by electrical resistance tomography according to claim 1, characterized in that: The input of the neural network model is a current matrix, and the output is a resistivity distribution; the neural network model adopts one of a multi-layer perceptron MLP, a convolutional neural network CNN, and a deep neural network DNN, and the regularization method adopts L2 regularization or Dropout regularization.

Images