Double-row pile side slope earthquake displacement prediction and landslide sliding surface position determination calculation method based on GA-BP neural network

Abstract

The invention discloses a double-row pile side slope earthquake displacement prediction and landslide sliding surface position determination calculation method based on a GA-BP neural network, particularly relates to the field of pile body displacement monitoring, and aims to optimize the BP neural network through a GA genetic algorithm so as to improve the accuracy of double-row pile side slope earthquake displacement prediction and landslide sliding surface position determination. And constructing a double-row pile slope earthquake displacement prediction model by using a deep learning means. According to the method, the double-row anti-slide pile serves as a research object, data of a plurality of different acceleration sensors at a pile body are collected, and basic data information is obtained; performing acceleration baseline adjustment on the acquired data by using an SPECTR response spectrum calculation tool, and putting the processed data into a GA-BP neural network model for training; and related parameters such as training times and variation probability in the model are determined through continuous debugging, so that a GA-BP neural network double-row pile side slope seismic displacement time sequence prediction model is established, and double-row pile side slope seismic displacement prediction and landslide sliding surface position determination are realized.

Description

Technical Field

[0001] The invention belongs to the field of construction technology, and in particular relates to the field of pile displacement monitoring, and a calculation method for predicting earthquake displacement of double-row pile slopes and determining the position of a landslide sliding surface based on a GA-BP neural network. Background Art

[0002] The prediction of earthquake displacement of double-row pile slopes is an effective means to study the damage of piles and soil. Through the prediction of soil displacement, two pieces of information can be obtained: first, where the pile foundation will produce a sliding surface under the influence of an earthquake; second, where the pile foundation is the weakest under the influence of an earthquake. Therefore, the prediction of earthquake displacement of double-row pile slopes can understand the degree of damage to the pile body under the action of an earthquake, which is of great significance for the post-earthquake repair and reinforcement of the pile body.

[0003] The traditional BP neural network is a widely used and efficient artificial neural network, but it has a slow convergence speed and is prone to falling into local minima during training. Therefore, if only the BP neural network is used, the workload for weight and threshold retrieval is huge. The algorithm theorem of the GA genetic algorithm draws on the natural selection mechanism of the biological world. Adding it to the BP neural network can theoretically ensure that the screening of weights and thresholds is an optimization process, fundamentally determining the optimal threshold and weight, which can not only reduce the number of training times but also improve the accuracy of training results. Therefore, this patent proposes a GA-BP algorithm for predicting the displacement of double-row piles.

[0004] (1) Frequent geological disasters: Earthquake is a natural disaster caused by changes in the earth's internal structure and crustal movement. Research on earthquake prediction is conducive to the faster and safer implementation of earthquake prevention and control plans.

[0005] (2) Advances in monitoring technology: With the application of sensor technology, computer technology, network communication technology, etc. in the field of earthquake monitoring, earthquake monitoring and early warning are gradually developing in the direction of automation and intelligence.

[0006] (3) Artificial neural network model: BP neural network is a multi-layer feedforward network trained by error back propagation (abbreviated as error back propagation). Its algorithm is called BP algorithm. This algorithm is related to learning and storing a large number of input-output pattern mappings, without revealing the mathematical equations describing this mapping relationship in advance. Its basic idea is the gradient descent method, which uses gradient search technology to continuously adjust the weights and thresholds of the network through back propagation, in order to minimize the error between the actual output value of the network and the expected output value.

[0007] (4) Data model limitations: The quality of the data is likely to determine the evaluation performance and accuracy of the model. The factors involved include accuracy, completeness, consistency, and interpretability. Generally speaking, directly acquired data information contains a large number of missing values, noise, and outliers, which is not conducive to the training of algorithm models.

[0008] (5) Prevention and control decision-making needs: Earthquakes are natural phenomena in which the earth's crust rapidly releases energy and generates seismic waves. Earthquakes usually involve large movements of soil, so their potential destructive power is relatively large, and their impact on the surrounding environment and human activities is more significant. The double-row pile slope earthquake displacement prediction model can quickly locate the location and extent of damage and help determine the control plan.

[0009] In summary, the study of double-row pile slope seismic displacement prediction and landslide sliding surface position determination through sensor technology and the application of GA-BP neural network algorithm is of great significance to reducing casualties and property losses caused by earthquakes. Summary of the invention

[0010] The present invention provides a double-row pile slope earthquake displacement prediction and landslide sliding surface position determination calculation method based on GA-BP neural network, the purpose of which is to optimize the BP neural network through GA genetic algorithm to improve the accuracy of double-row pile displacement prediction and sliding surface position.

[0011] To this end, the present invention adopts the following technical solutions:

[0012] The method for predicting the earthquake displacement of double-row pile slope and determining the position of landslide sliding surface based on GA-BP neural network includes the following steps:

[0013] Step 1: According to the experimental requirements, carry out the double-row anti-sliding pile test on the shaking table to obtain the acceleration data required for the experiment.

[0014] Step 2: Data preprocessing: first adjust the acceleration baseline of the data obtained from step 1, then use bandpass filtering to process the noise of the baseline processed data, and finally integrate the filtered acceleration data to obtain the velocity and displacement data.

[0015] Step 3: Constructing a GA-BP neural network double-row pile slope seismic displacement prediction model

[0016] (1) BP Neural Network

[0017] The BP neural network consists of three parts, including the input layer, output layer and hidden layer; the input layer neurons are a m , the hidden layer neurons are b u , the output layer neuron is c n ; Assume ω mnis the connection weight between the mth neuron in the input layer and the uth neuron in the hidden layer, v un is the connection weight between the u-th neuron in the hidden layer and the n-th neuron in the output layer, then the neurons in the hidden layer and the neurons in the output layer are obtained in turn as shown in equations (1) and (2):

[0018]

[0019] Among them, the activation function selects the Sigmoid function to map the variable to [0,1], as shown in formula (3):

[0020]

[0021] Determine the hidden layer of the neural network, use a loop to traverse the hidden layer nodes and training errors within the range; the number of hidden layers is calculated using the formula Or h = 2n + m; (h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is an adjustable constant from 1 to 10;)

[0022] (2) GA

[0023] Genetic algorithms include selection operations, crossover operations, and mutation operations. The specific operation methods are as follows:

[0024] Selection operation: Call the selection function (SelectionFcn), after calculating the fitness of each individual in the population, use the selection process to determine which individual in the population will be used for reproduction and produce the next generation, select individuals with high values, and pass their genetic material to the next generation;

[0025] Crossover operation: Call the crossover function (CrossoverFcn). In order to create a pair of new individuals, some chromosomes of the parent samples selected from the current generation are exchanged (crossed) to create two new chromosomes representing the offspring;

[0026] Mutation operation: calling the mutation function (MutationFcn). The purpose of the mutation operation is to periodically update the population randomly, introduce new patterns into the chromosome, and search in unknown areas of the solution space. Mutation is achieved by randomly changing one or more chromosome values.

[0027] (3)GA-BP neural network

[0028] Use GA genetic algorithm to optimize the internal weights and thresholds of BP neural network to find the global optimum;

[0029] Step 4: Output the results, obtain the GA-BP training prediction results in step 3, and calculate the average displacement prediction value of each monitoring point.

[0030] Step 5: Select evaluation metrics

[0031] The mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R 2 ) is the evaluation index;

[0032] MSE is the average of the squared errors between the predicted value and the true value. When the predicted value is exactly the same as the true value, MSE is equal to 0, which is a perfect model. The larger the error, the larger the MSE value.

[0033] RMSE is the root mean square value of MSE;

[0034] The MAE range is [0, +∞). The smaller the MAE value, the higher the prediction accuracy. The larger the error, the larger the MAE value.

[0035] The MAPE range is [0, +∞), a MAPE of 0% indicates a perfect model, and a MAPE greater than 100% indicates a poor model;

[0036] R 2 The value range is [0,1], R 2 The closer the value is to 1, the better the data fit is;

[0037] The specific calculation formula is as follows:

[0038]

[0039]

[0040] Where n is the number of samples; y i is the true value; is the predicted value.

[0041] Step 6: Result Analysis

[0042] The evaluation index selected in step 5 is used to determine whether the prediction results of the GA-BP neural network double-row pile slope seismic displacement model are accurate. If the training error meets the evaluation index requirements, then the deformation of the pile body under the seismic load is analyzed based on the average displacement prediction value calculated in step 4, and then the weak points of the pile body and the location of the sliding surface are determined; if the training error does not meet the requirements, return to step 3 for retraining, and repeat this cycle until the error standard is met.

[0043] The beneficial effects of the present invention are:

[0044] 1. Prediction method of seismic displacement of double-row pile slope using GA-BP algorithm under earthquake action

[0045] The displacement of double-row anti-sliding piles under seismic loads cannot be directly observed, so it is necessary to use other directly available data to predict their displacement changes. BP neural network has powerful nonlinear mapping capabilities and flexibility, among which weights and thresholds are two very important parameters. The weights determine the intensity of information transmission and processing between different neurons, while the threshold determines whether the neuron is activated and the degree of activation. Therefore, the two parameters can directly affect the performance and output of the training results. The GA genetic algorithm is used to optimize the BP neural network (optimize thresholds and weights), which can select the optimal weights and thresholds, thereby improving the accuracy of the BP neural network in predicting the displacement of double-row anti-sliding piles.

[0046] 2. Predict the positive and negative mutations of the seismic displacement values ​​of the double-row pile slope based on the GA-BP algorithm to determine the position of the landslide sliding surface

[0047] After using the GA-BP algorithm to predict the seismic displacement value of the double-row pile slope, the mean displacement of each monitoring point in the 10″~50″ time period is calculated and its movement is analyzed. When an earthquake occurs, dynamic loads are generated, causing the anti-slide piles and the surrounding soil to interact, which can cause a sudden change in the stress distribution of the soil. The positive and negative mutations in the displacement values ​​of the anti-slide piles can just reflect the stress changes. The positive and negative mutations in the displacement values ​​usually lead to the occurrence of sliding surface phenomena. Therefore, the approximate location interval of the sliding surface formation can be determined based on the changes in the displacement values, so as to timely discover and deal with the displacement changes of the anti-slide piles and ensure the stability of the anti-slide piles. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is the algorithm flow chart of the present invention;

[0049] Figure 2 It is a topological structure diagram of the BP neural network of the present invention;

[0050] Figure 3 is the image of Sigmoid function;

[0051] Figure 4 It is a flowchart of the genetic algorithm of the present invention;

[0052] Figure 5 It is a step diagram of optimizing BP neural network by genetic algorithm of the present invention;

[0053] Figure 6 is a time history diagram before and after acceleration data processing according to an embodiment of the present invention;

[0054] Figure 7 is a comparison diagram of prediction errors of an embodiment of the present invention;

[0055] Figure 8 is a comparison chart of prediction results of an embodiment of the present invention;

[0056] Fig. 9is a comparison chart of regression results of an embodiment of the present invention;

[0057] Fig.10 is a displacement prediction diagram of an embodiment of the present invention;

[0058] Fig.11 This is a diagram showing the deformation of the anti-sliding No. 3 pile according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments:

[0060] Step 1: The shaking table experiment was set to 0.4g and the seismic wave was loaded from the X direction. The seismic wave loaded in this experiment was the EL seismic wave; after the EL seismic wave test was loaded, four points on the river side of pile 3 were selected as monitoring points and their acceleration data were obtained.

[0061] Step 2: Use the SPECTR response spectrum calculation tool to perform linear baseline correction on the acceleration time history using the acceleration data obtained in step 1 to avoid baseline drift of the acceleration time history caused by low-frequency noise, which affects the subsequent integral solution of velocity and displacement time history. Then use the origin software to perform a 3-110Hz FFT bandpass filter operation on the experimental data to remove some noise in the experimental data. After removing the noise, perform an integral operation on the acceleration time history to obtain the velocity and displacement time history. Figure 6 It can be seen that the acceleration 10 seconds before and 50 seconds after has almost no change, so the acceleration data of 10″~50″ is selected for subsequent model training.

[0062] Step 3: The input layer neurons of the BP neural network are the acceleration and velocity obtained in step 2, and the output layer neurons are the soil displacement prediction results we want to obtain. Since the number of output layer neurons is 1, the BP neural network model in this paper can be called a single-output BP neural network model. According to the formula h=2n+m, the number of BP hidden layer neurons is 5. If the predicted value of the training part is highly consistent with the measured value, while the predicted value of the test part is significantly different from the measured value, or the predicted value of the test part is highly consistent with the measured value, while the predicted value of the training part is significantly different from the measured value, then they are not good prediction models. Therefore, the accuracy of the entire prediction model needs to consider the errors of the test set and the training set. Other BP related parameters are shown in Table 1.

[0063] Table 1 BP neural network related parameters

[0064]

[0065] For the GA genetic algorithm, three fitness functions are called: selection roulette, crossover two point and Gaussian mutation. Other related parameters are shown in Table 2.

[0066] Table 2 Genetic algorithm parameter settings

[0067]

[0068] According to the above parameters, the training results are as follows Figure 7 and 8 shown.

[0069] Step 4: Get the training results in step 3 and calculate the average displacement prediction results of each monitoring point on the river side within 40 seconds.

[0070] Step 5: Mean absolute error (MAE) is the average value of absolute errors, reflecting the actual situation of the prediction error; mean square error (MSE) is the ratio of the sum of squares of the deviations between the prediction and the actual value to the total number of samples; root mean square error (RMSE) is the square root of the ratio of the sum of squares of the deviations between the prediction and the actual value to the total number of samples; mean absolute percentage error (MAPE) is the ratio of the absolute value of the deviation between the prediction and the actual value to the actual value, taking the average value. The specific evaluation index results are shown in the following table.

[0071] Table 3 Comparison of evaluation index results

[0072]

[0073] Absolute coefficient (R 2 ) results are as follows Fig. 9 shown.

[0074] From the above results, we can see that the GA-BP algorithm not only has higher prediction accuracy than the BP algorithm, but also has a smaller training error than the BP algorithm.

[0075] Step 6: Based on the results of step 4, draw a double-row pile slope earthquake displacement prediction diagram, such as Fig.10 shown.

[0076] from Fig.10 It can be seen that the displacement deformation at point 4 of the pile body is the largest, and the displacement value at point 2 becomes negative. Fig.11The deformation analysis of the anti-slide pile under earthquake action shows that under the dynamic load of earthquake, the anti-slide pile may be subjected to thrust or tension from the soil, causing its displacement to change. If the anti-slide pile is subjected to pressure from the soil, and this pressure is greater than the compressive strength of the pile body, the anti-slide pile will move toward the landslide body, that is, the displacement value changes from zero to a positive value; if the anti-slide pile is subjected to tension from the soil, and this tension is greater than the tensile strength of the pile body, the pile body may break or produce reverse deformation, resulting in a negative displacement. The change in soil stress can be reflected in the change in the positive and negative signs of the displacement value. Combined with the construction direction of the anti-slide pile from top to bottom, it can be seen that there will be a sudden stress change between monitoring points 2 and 3, which will cause a sliding surface. Monitoring point 4 is the weakest part of the pile body.

Claims

1. A method for predicting the earthquake displacement of double-row pile slopes and determining the position of landslide sliding surface based on GA-BP neural network, characterized in that: The following steps are involved: Step 1: According to the experimental requirements, carry out the double-row anti-sliding pile test on the shaking table to obtain the acceleration data required for the experiment; Step 2: Data preprocessing: first adjust the acceleration baseline of the data obtained in step 1, then filter the baseline processed data with a bandpass filter, and finally integrate the filtered acceleration data to obtain velocity and displacement data; Step 3: Construct GA-BP neural network displacement prediction model (1) BP Neural Network The BP neural network consists of three parts, including the input layer, output layer and hidden layer; the input layer neurons are a m , the hidden layer neurons are b u , the output layer neuron is c n ; Assume ω mn is the connection weight between the mth neuron in the input layer and the uth neuron in the hidden layer, v un is the connection weight between the u-th neuron in the hidden layer and the n-th neuron in the output layer, then the neurons in the hidden layer and the neurons in the output layer are obtained in turn as shown in equations (1) and (2): Among them, the activation function selects the Sigmoid function to map the variable to [0,1], as shown in formula (3): Determine the hidden layer of the neural network, use a loop to traverse the hidden layer nodes and training errors within the range; the number of hidden layers is calculated using the formula Or h = 2n + m; (h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is an adjustable constant from 1 to 10;) (2) GA Genetic algorithms include selection operations, crossover operations, and mutation operations. The specific operation methods are as follows: Selection operation: Call the selection function (SelectionFcn), after calculating the fitness of each individual in the population, use the selection process to determine which individual in the population will be used for reproduction and produce the next generation, select individuals with high values, and pass their genetic material to the next generation; Crossover operation: Call the crossover function (CrossoverFcn). In order to create a pair of new individuals, some chromosomes of the parent samples selected from the current generation are exchanged (crossed) to create two new chromosomes representing the offspring; Mutation operation: Call the mutation function (MutationFcn). The purpose of the mutation operation is to periodically update the population randomly, introduce new patterns into the chromosome, and search in unknown areas of the solution space. Mutation is achieved by randomly changing one or more chromosome values. (3)GA-BP neural network Use GA genetic algorithm to optimize the internal weights and thresholds of BP neural network to find the global optimum; Step 4: Output the results, obtain the GA-BP training prediction results in step 3, and sort and calculate the average displacement prediction value of each monitoring point; Step 5: Select evaluation metrics The mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R 2 ) is the evaluation index; MSE is the average of the squared errors between the predicted value and the true value. When the predicted value is exactly the same as the true value, MSE is equal to 0, which is a perfect model. The larger the error, the larger the MSE value. RMSE is the root mean square value of MSE; The MAE range is [0, +∞). The smaller the MAE value, the higher the prediction accuracy. The larger the error, the larger the MAE value. The MAPE range is [0, +∞), a MAPE of 0% indicates a perfect model, and a MAPE greater than 100% indicates a poor model; R 2 The value range is [0,1], R 2 The closer the value is to 1, the better the data fit is; The specific calculation formula is as follows: Where n is the number of samples; y i is the true value; is the predicted value; Step 6: Analyze the results The evaluation index selected in step 5 is used to determine whether the results of the GA-BP neural network displacement prediction model are accurate. If the training error meets the evaluation index requirements, then the deformation of the pile body under the seismic load is analyzed based on the average displacement prediction value calculated in step 4, and then the weak points of the pile body and the location of the sliding surface are determined; if the training error does not meet the requirements, return to step 3 for retraining, and repeat this cycle until the error standard is met.

2. The method for predicting the earthquake displacement of double-row pile slopes and determining the position of landslide sliding surface based on GA-BP neural network according to claim 1 is characterized in that: In step 3, the specific steps of GA genetic algorithm optimizing BP neural network are as follows: 1) Obtain the initialization weights and thresholds of the BP neural network; 2) Use the training set data for forward propagation and error calculation; 3) The weights and thresholds of the BP neural network are used as chromosomes to construct the initial population of the GA genetic algorithm; 4) Iteratively optimize the population using selection, crossover, and mutation operations of genetic algorithms; 5) Evaluate the performance of each individual according to the fitness function, and select individuals with higher fitness values ​​for the next step of reproduction; 6) Generate new individuals through crossover and mutation operations and update the population; 7) Use the updated weights and thresholds to train the BP neural network, repeating steps 2) and 3) until the stopping condition of the minimum error of the training target is reached; 8) Obtain the optimized BP neural network model.

Images