Deformation monitoring data prediction method based on multi-scale features, medium and equipment

Abstract

The invention relates to the technical field of deformation monitoring, in particular to a deformation monitoring data prediction method based on multi-scale features. Comprising the following steps: S1, acquiring original deformation monitoring data of a certain research area, performing average pooling, and introducing a residual iterative calculation method in the pooling process to perform multi-scale feature extraction to obtain a multi-scale feature matrix F; s2, designing according to the multi-scale feature matrix F to obtain a prediction neural network model; s3, training the prediction neural network model by adopting a batch gradient descent method to obtain a trained prediction model; and S4, inputting the features of the first P time points in the multi-scale feature matrix F in the S1 into the trained prediction model, and outputting a prediction result of the deformation monitoring data. According to the method, features are extracted through average pooling under different time windows, residual iterative calculation and a multi-layer neural network structure are introduced to capture changes of deformation data on different time scales, and a good effect is shown in deformation monitoring data prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of deformation monitoring, and in particular to a deformation monitoring data prediction method based on multi-scale features. Background Art

[0002] Deformation monitoring is an important monitoring indicator in the fields of engineering safety monitoring and geological disaster monitoring, and plays an important role in early warning and forecasting of risks and disasters. Predicting the future trend of deformation monitoring data is one of the important application directions of data science and artificial intelligence. However, deformation monitoring data usually has complex periodicity and variation characteristics, which brings great challenges to prediction.

[0003] Commonly used deformation monitoring methods include simple moving average, weighted moving average, exponential smoothing, weighted exponential smoothing and other methods. These methods are based on the average or smoothing of historical data in the previous period for prediction. However, deformation monitoring data is affected by complex factors such as different geological structures, weather, and external disturbances. It is difficult to determine how long the historical data should be used for prediction, resulting in poor prediction results of the above algorithms, or algorithms that perform well on one data set are difficult to migrate to other data sets, which has a negative impact on actual engineering applications. Summary of the invention

[0004] The present invention aims to provide a deformation monitoring data prediction method based on multi-scale features that can better capture the change pattern of deformation data at different time scales. The specific technical solution is as follows:

[0005] The present invention provides a deformation monitoring data prediction method based on multi-scale features, comprising the following steps:

[0006] S1: Obtain the original deformation monitoring data of a certain study area, perform average pooling on the original deformation monitoring data, and introduce the residual iterative calculation method in the pooling process to extract multi-scale features to obtain the multi-scale feature matrix F;

[0007] S2: The prediction neural network model is designed based on the multi-scale feature matrix F;

[0008] S3: Divide the multi-scale feature matrix F into training data and verification data, use the training data to train the prediction neural network model using the batch gradient descent method, and obtain the trained prediction model;

[0009] S4: Take the features of the first P time points in the verification data and input them into the trained prediction model for prediction, and output the prediction result of the deformation monitoring data, that is, the prediction vector of the deformation monitoring data.

[0010] Optionally, the S1 includes:

[0011] S1.1. Obtain the original deformation monitoring data of a certain study area, and define N time windows of different lengths based on the original deformation monitoring data. The lengths of the time windows are in the order of w1, w2, … w i …,w N , where: i is the number of the time window, 1≤i≤N;

[0012] S1.2. Average pooling is performed on the original deformation monitoring data, and the residual iterative calculation method is introduced in the pooling process to extract multi-scale features of the original deformation monitoring data to obtain a multi-scale feature matrix F, specifically:

[0013] S1.2.1. Perform average pooling of the original deformation monitoring data centered at the current time point to obtain the feature F1(t) of the time window w1, and then use the original deformation monitoring data to subtract the feature F1(t) to obtain the residual R1(t);

[0014] S1.2.2, for time windows w2,…w i …,w N Perform feature extraction to obtain a multi-scale feature matrix F, which is:

[0015] For the residual R i-1 (t) Perform average pooling centered at the current time point to obtain the time window w i Features F i (t), and then use the residual R i-1 (t) minus characteristic F i (t) Get the residual R i (t);

[0016] After completing the feature extraction of all time windows, a multi-scale feature matrix F with a shape of T×N is obtained. The specific expression is as follows:

[0017]

[0018] Where: T is the length of the original deformation monitoring data.

[0019] Optionally, the specific formula for average pooling is as follows:

[0020]

[0021] Where: x t Deformation monitoring data value at the tth time point; The time window w is used i For the pooling result at time point t, k i represents the time window w i There are k on the left and right sides of the internal time point t i Monitoring data, τ is the time interval for deformation monitoring data collection; j is the time window w i The time point number of the monitoring data within;

[0022] In S1.2, the specific formula for introducing the residual iterative calculation method to extract multi-scale features from the original deformation monitoring data is as follows:

[0023] For time window w1:

[0024]

[0025] in: is the pooling result of time window w1 at time point t;

[0026] For time windows w2,…w i …,w N :

[0027]

[0028] Among them: F i (t) is the time window w i The initial features at time point t; R i-1 (t) is the time window w i-1 The residual at time point t; To use the time window w i Pooling result at time point t.

[0029] Optionally, the S2 includes:

[0030] S2.1. Design the convolutional layer of the prediction neural network model; specifically:

[0031] S2.1.1, let the input multi-scale feature matrix be Among them: The multi-scale feature matrix is The data shape is m×N, that is, the N-dimensional multi-scale features of the first m consecutive time points;

[0032] S2.1.2, based on one-dimensional convolutional neural network, extract the multi-scale feature matrix: The features of the convolutional neural network are flattened into the convolution output vector z. The specific formula is as follows:

[0033]

[0034] Where: ol is the convolution kernel weight, b o is the bias term, σ is the ReLU activation function, and h ois the output feature of the convolution layer, the result of the oth channel at time point t, C is the width of the convolution kernel, 1≤C≤m-1;

[0035] S2.2. Design the first fully connected layer and the second fully connected layer of the prediction neural network model; specifically:

[0036] S2.2.1. The first fully connected layer maps the convolution output vector to M fc1 neurons, and nonlinearity is introduced through the ReLU activation function. The specific formula is as follows:

[0037] z′=σ(ω fc1 z+b fc1 );

[0038] Where: z′ is the output result of the first fully connected layer; ω fc1 and b fc1 They are M in the first fully connected layer. fc1 The weight matrix and bias vector of each neuron, σ is the ReLU activation function;

[0039] S2.2.2, the second fully connected layer will M fc2 The output of each neuron is mapped to the final predicted output. The specific formula is as follows:

[0040]

[0041] Where: fc2 and b fc2 They are M in the second fully connected layer. fc2 The weight matrix and bias vector of each neuron, It is the final prediction output result, that is, the predicted deformation monitoring data value at the next Q time points.

[0042] Optionally, the S3 includes:

[0043] S3.1. Construct the loss function Loss based on the mean square error and the predicted value difference loss;

[0044] S3.2. Use batch gradient descent method to train the prediction neural network model until Loss≤∈ stops training, save the model parameters to obtain the trained prediction model, where: ∈ is the set threshold.

[0045] Optionally, in S3.1, the specific formula of the loss function Loss is as follows:

[0046] Loss=Loss MSE +λ·Loss diff ;

[0047]

[0048]

[0049] Among them: Loss MSE is the mean square error; Loss diff is the data increment error; y true,n is the nth true value of the real data; y true,1 is the first true value of the real data; y pred,n is the nth predicted value of the trained prediction model prediction data, y pred,1 is the first predicted value of the data predicted by the trained prediction model; λ is the weight coefficient.

[0050] Optionally, in S3.2, the batch gradient descent method is used to train the prediction neural network model, specifically: the original deformation monitoring data is divided into multiple batches, forward propagation, loss calculation, back propagation and parameter update are performed, and the loss function value is gradually reduced through multiple rounds of iterative training, wherein: the Adam optimizer is used for updating when the parameters are updated.

[0051] Optionally, the S4 specifically includes:

[0052] S4.1. Take the features of the first P time points of the validation data as the input of the trained prediction model:

[0053] S4.2. Based on the trained prediction model, the deformation monitoring data values ​​at the next Q time points are calculated through forward propagation, and the prediction vector of the deformation monitoring data with a length of Q is output. The specific formula is as follows:

[0054]

[0055] Among them: Model represents the trained prediction model;

[0056] After S4.2, the method further includes: inputting the verification data into the trained prediction model for prediction to obtain a corresponding prediction vector, and comparing the vector with the true value to verify the trained prediction model.

[0057] The present invention also provides a readable storage medium having computer program instructions stored thereon, and when the computer program instructions are executed by a processor, the deformation monitoring data prediction method based on multi-scale features as described above is implemented.

[0058] The present invention also provides an electronic device, comprising: at least one processor, at least one memory and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, the deformation monitoring data prediction method based on multi-scale features as described above is implemented.

[0059] The present invention utilizes average pooling under different time windows to extract features to better capture the change patterns of deformation data at different time scales. The extracted multi-scale features reduce the length of the data, retain the main trends, and remove high-frequency noise. On this basis, prediction is performed through convolutional neural networks and fully connected layers, and by introducing residual iterative calculation, multi-layer neural network structure, optimized loss function and optimization method, the prediction method of the present invention shows better effect in deformation monitoring data prediction.

[0060] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0062] Figure 1 is a schematic diagram of prediction results of Example 1 of the deformation monitoring data prediction method based on multi-scale features in an embodiment of the present invention;

[0063] Figure 2 is a schematic diagram of prediction results of Example 2 of the deformation monitoring data prediction method based on multi-scale features in an embodiment of the present invention;

[0064] Figure 3 is a schematic diagram of prediction results of Example 3 of the deformation monitoring data prediction method based on multi-scale features in an embodiment of the present invention;

[0065] Figure 4 is a schematic diagram of prediction results of Example 4 of the deformation monitoring data prediction method based on multi-scale features in an embodiment of the present invention;

[0066] Figure 5 is a schematic diagram of prediction results of Example 5 of the deformation monitoring data prediction method based on multi-scale features in an embodiment of the present invention;

[0067] Figure 6 is a schematic diagram of prediction results of Example 6 of the deformation monitoring data prediction method based on multi-scale features in an embodiment of the present invention;

[0068] Figure 7is a schematic diagram of prediction results of Example 7 of the deformation monitoring data prediction method based on multi-scale features in an embodiment of the present invention;

[0069] Figure 8 4 is a schematic diagram of prediction results of Example 8 of the deformation monitoring data prediction method based on multi-scale features in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0071] In one embodiment, a deformation monitoring data prediction method based on multi-scale features includes the following steps:

[0072] 1. Data preprocessing and multi-scale feature extraction

[0073] S1: Obtain the original deformation monitoring data of a certain study area, perform average pooling on the original deformation monitoring data, and introduce the residual iterative calculation method in the pooling process to extract multi-scale features to obtain the multi-scale feature matrix F;

[0074] First, multi-scale feature extraction is performed on the original deformation monitoring data. Specifically, the average pooling operation of the deformation monitoring data with different time window lengths can capture the long-term and short-term change characteristics of the deformation monitoring data. The entire feature extraction process is divided into the following steps:

[0075] S1.1. Obtain the original deformation monitoring data of a certain study area, and define N time windows of different lengths based on the original deformation monitoring data. The lengths of the time windows are in the order of w1, w2, … w i …,w N , where: i is the number of the time window, 1≤i≤N;

[0076] Determine time windows of different lengths, such as 12 hours, 6 hours, 2 hours, etc. Each window length corresponds to a different scale and is used to extract corresponding features. In actual operation, the time window length determined can be determined according to the data situation. To facilitate subsequent calculations, define N different time window lengths w1, w2, ..., w N To extract multi-scale features. w1 to w N The time windows are getting shorter and shorter, w1 is the longest time window, w N It is the shortest time window.

[0077] In this implementation, the data structure of the deformation monitoring data is (id, timestamp, value), where id is the serial number of the deformation monitoring data, timestamp is the timestamp of obtaining the monitoring data, and value is the value of the deformation monitoring data, which can be a single deformation amount or a deformation rate.

[0078] S1.2. Average pooling is performed on the original deformation monitoring data, and the residual iterative calculation method is introduced in the pooling process to extract multi-scale features of the original deformation monitoring data to obtain a multi-scale feature matrix F, in which each row corresponds to a different scale feature at a time point, specifically:

[0079] In order to ensure that the multi-scale features of each time point reflect the data of a period of time around it, the present invention averages the data in the time window with the current time point as the center during the pooling process, which avoids the phenomenon of bias towards a certain moment. Assume that the original deformation monitoring data sequence is X = [x1, x2, ..., x T ], where x t is the deformation value at the tth time point, and T is the total number of time points. In the average pooling process, residual iterative calculation is introduced. First, the longest time window is used to average pool the data, and the obtained feature data is subtracted from the original data to obtain the residual; then the residual is pooled using a shorter time window, and features are continuously extracted until the pooling operation of all windows is completed. Specifically:

[0080] For each time window w i , first use the longest window w1 for pooling to get the initial feature F1:

[0081]

[0082] Then calculate the residual R1: R1(t) = x t -F1(t)

[0083] Pool the residual to the next layer, using a shorter window w2:

[0084]

[0085] Similarly, the residuals are pooled sequentially until the shortest time window w N , and finally obtain the multi-scale feature matrix F

[0086] The detailed steps are as follows:

[0087] S1.2.1. Perform average pooling of the original deformation monitoring data centered at the current time point to obtain the feature F1(t) of the time window w1, and then use the original deformation monitoring data to subtract the feature F1(t) to obtain the residual R1(t);

[0088] S1.2.2, for time windows w2,…w i …,w N Perform feature extraction to obtain a multi-scale feature matrix F, which is:

[0089] For the residual R i-1 (t) Perform average pooling centered at the current time point to obtain the time window w i Features F i (t), and then use the residual R i-1 (t) minus characteristic F i (t) Get the residual R i (t);

[0090] After completing the feature extraction of all time windows, a multi-scale feature matrix F with a shape of T×N is obtained. The specific expression is as follows:

[0091]

[0092] Where: T is the length of the original deformation monitoring data.

[0093] The specific formula for average pooling is as follows:

[0094]

[0095] Where: x t Deformation monitoring data value at the tth time point; The time window w is used i For the pooling result at time point t, k i represents the time window w i There are k on the left and right sides of the internal time point t i Monitoring data, τ is the time interval for deformation monitoring data collection; j is the time window w i The time point number of the monitoring data within.

[0096] In S1.2, the specific formula for introducing the residual iterative calculation method to extract multi-scale features from the original deformation monitoring data is as follows:

[0097] For time window w1:

[0098]

[0099] in: is the pooling result of time window w1 at time point t;

[0100] For time windows w2,…w i …,w N :

[0101]

[0102] Among them: F i (t) is the time window w i The initial features at time point t; R i-1 (t) is the time window w i-1 The residual at time point t; is the time window w i Pooling result at time point t.

[0103] 2. Prediction Network Model Design

[0104] S2: The prediction neural network model is designed based on the multi-scale feature matrix F;

[0105] After feature extraction is completed, a neural network model is designed to predict future time series data based on the multi-scale features of the previous time points.

[0106] S2 includes:

[0107] First, a one-dimensional convolutional neural network (1D CNN) is used to extract features from the input multi-scale feature matrix. Through convolution kernels of different sizes, the convolution layer can capture the local patterns of multi-scale features while retaining multi-scale information.

[0108] S2.1. Design the convolutional layer of the prediction neural network model; specifically:

[0109] S2.1.1, let the input multi-scale feature matrix be Among them: The multi-scale feature matrix is The data shape is m×N, that is, the N-dimensional multi-scale features of the first m consecutive time points, where each row represents the N-dimensional multi-scale features of a time point;

[0110] S2.1.2, based on one-dimensional convolutional neural network, extract the multi-scale feature matrix: The features of the convolutional neural network are flattened into the convolution output vector z. The specific formula is as follows:

[0111]

[0112] Where: ol is the convolution kernel weight, b o is the bias term, σ is the ReLU activation function, and h o is the output feature of the convolution layer, the result of the oth channel at time point t, C is the width of the convolution kernel, 1≤C≤m-1;

[0113] After the convolution operation, the output is flattened into a vector form so that it can be input into the fully connected layer.

[0114] S2.2. Design the first fully connected layer and the second fully connected layer of the prediction neural network model; specifically:

[0115] The flattened convolutional layer outputs a vector, which is input into a two-layer fully connected network. The fully connected layer maps it to a higher-dimensional space through a linear transformation.

[0116] S2.2.1. The first fully connected layer maps the convolution output vector to M fc1 neurons, and nonlinearity is introduced through the ReLU activation function. The specific formula is as follows:

[0117] z′=σ(ω fc1 z+b fc1 );

[0118] Where: z' is the output result of the first fully connected layer; ω fc1 and b fc1 They are M in the first fully connected layer. fc1 The weight matrix and bias vector of each neuron, σ is the ReLU activation function;

[0119] S2.2.2, the second fully connected layer will M fc2 The output of each neuron is mapped to the final predicted output. The specific formula is as follows:

[0120]

[0121] Where: fc2 and b fc2 They are M in the second fully connected layer. fc2 The weight matrix and bias vector of each neuron, It is the final prediction output result, that is, the predicted deformation monitoring data value at the next Q time points.

[0122] 3. Loss Function and Optimization

[0123] By minimizing the loss function Loss, we can update the parameters of the neural network to improve the accuracy of the prediction. The details are as follows:

[0124] S3: Divide the multi-scale feature matrix F into training data and verification data, use the training data to train the prediction neural network model using the batch gradient descent method, and obtain the trained prediction model;

[0125] S3 includes:

[0126] S3.1. Construct the loss function Loss based on the mean square error and the predicted value difference loss;

[0127] In S3.1, the specific formula of the loss function Loss is as follows:

[0128] Loss=Loss MSE +λ·Loss diff ;

[0129]

[0130]

[0131] Among them: Loss MSE is the mean square error; Loss diff is the data increment error; y true,n is the nth true value of the real data; y true,1 is the first true value of the real data; y pred,n is the nth predicted value of the trained prediction model prediction data, y pred,1 is the first predicted value of the data predicted by the trained prediction model; λ is the weight coefficient.

[0132] S3.2. Use batch gradient descent method to train the prediction neural network model until Loss≤∈ stops training, save the model parameters to obtain the trained prediction model, where: ∈ is the set threshold.

[0133] In S3.2, the batch gradient descent method is used to train the prediction neural network model. Specifically, the original deformation monitoring data is divided into multiple batches, and forward propagation, loss calculation, back propagation and parameter update are performed. Through multiple rounds of iterative training, the loss function value is gradually reduced, wherein: the Adam optimizer is used for parameter update.

[0134] S4: Take the features of the first P time points in the verification data and input them into the trained prediction model for prediction, and output the prediction result of the deformation monitoring data, that is, the prediction vector of the deformation monitoring data.

[0135] 4. Prediction and Verification

[0136] After the training is completed, the trained model is recorded as Model, which can be used to predict deformation monitoring data. The specific steps are as follows:

[0137] S4 specifically includes:

[0138] S4.1. Take the features of the first P time points of the validation data as the input of the trained prediction model:

[0139] S4.2. Based on the trained prediction model, the deformation monitoring data values ​​at the next Q time points are calculated through forward propagation, and the prediction vector of the deformation monitoring data with a length of Q is output. The specific formula is as follows:

[0140]

[0141] Among them: Model represents the trained prediction model;

[0142] After S4.2, the method further includes: inputting the verification data into the trained prediction model to obtain a corresponding prediction vector, and comparing it with the true value to verify the trained prediction model. Through visual comparison, the degree of consistency between the model prediction result and the actual data can be intuitively observed.

[0143] In this embodiment, P=10, Q=5, and the features of 10 time points are respectively taken from the verification data and input into the trained prediction model to predict the prediction data of the next 5 time points. Figures 1 to 8 There are eight examples in this embodiment, and the 10 time points selected in the eight examples and the 5 predicted time points are different from each other. Figure 1 This is the deformation monitoring data result at the predicted time point t∈[100,104] in the test set of Example 1; Figure 2 This is the deformation monitoring data result of the predicted time point t∈[150,154] in the test set of Example 2; Figure 3 The deformation monitoring data result at the predicted time point t∈[200,204] in the test set of Example 3; Figure 4 This is the deformation monitoring data result of the predicted time point t∈[500,504] in the test set of Example 4; Figure 5 This is the deformation monitoring data result of the predicted time point t∈[555,559] in the test set of Example 5; Figure 6 The deformation monitoring data result of the predicted time point t∈[777,781] in the test set of Example 6; Figure 7 The deformation monitoring data result of the predicted time point t∈[1111,1115] in the test set of Example 7; Figure 8 This is the deformation monitoring data result of the predicted time point t∈[5000,5004] in the test set of Example 8. Figures 1 to 8 It can be seen that the trend of the predicted data is consistent with the trend of the real data, and the deviation between the predicted data and the real data is very small, which meets the requirements of data prediction and subsequent early warning analysis.

[0144] The deformation monitoring data prediction method based on multi-scale features in this embodiment utilizes average pooling under different time windows to extract features to better capture the change patterns of deformation data at different time scales. The extracted multi-scale features reduce the length of the data, retain the main trends, and remove high-frequency noise. On this basis, prediction is performed through convolutional neural networks and fully connected layers, and by introducing residual iterative calculation, multi-layer neural network structure, and optimized loss function and optimization method, the prediction method of the present invention shows better results in deformation monitoring data prediction.

[0145] This embodiment provides a readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the deformation monitoring data prediction method based on multi-scale features as described above is implemented.

[0146] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0147] This embodiment also includes an electronic device, including: at least one processor, at least one memory, and computer program instructions stored in the memory. When the computer program instructions are executed by the processor, the deformation monitoring data prediction method based on multi-scale features as described above is performed.

[0148] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the electronic device.

[0149] The electronic device may be a computing device such as a mobile phone, a desktop computer, a notebook, a PDA, a cloud server, etc. The electronic device may include, but is not limited to, a processor and a memory. For example, the electronic device may also include an input / output device, a network access device, a bus, etc.

[0150] The processor may be a central processing unit (CPU), a GPU or other general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the electronic device, and uses various interfaces and lines to connect various parts of the entire electronic device.

[0151] The memory can be used to store the computer program and / or module, and the processor implements the computer program by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0152] Wherein, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0153] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A deformation monitoring data prediction method based on multi-scale features, characterized in that: The steps include: S1: Obtain the original deformation monitoring data of a certain study area, perform average pooling on the original deformation monitoring data, and introduce the residual iterative calculation method in the pooling process to extract multi-scale features to obtain the multi-scale feature matrix F; S2: The prediction neural network model is designed based on the multi-scale feature matrix F; S3: Divide the multi-scale feature matrix F into training data and verification data, use the training data to train the prediction neural network model using the batch gradient descent method, and obtain the trained prediction model; S4: Take the features of the first P time points in the verification data and input them into the trained prediction model for prediction, and output the prediction result of the deformation monitoring data, that is, the prediction vector of the deformation monitoring data.

2. The deformation monitoring data prediction method based on multi-scale features according to claim 1 is characterized in that: The S1 includes: S1.

1. Obtain the original deformation monitoring data of a certain study area, and define N time windows of different lengths based on the original deformation monitoring data. The lengths of the time windows are in the order of w1, w2, … w i …,w N , where: i is the number of the time window, 1≤i≤N; S1.

2. Average pooling is performed on the original deformation monitoring data, and the residual iterative calculation method is introduced in the pooling process to extract multi-scale features of the original deformation monitoring data to obtain a multi-scale feature matrix F, specifically: S1.2.

1. Perform average pooling of the original deformation monitoring data centered at the current time point to obtain the feature F1(t) of the time window w1, and then use the original deformation monitoring data to subtract the feature F1(t) to obtain the residual R1(t); S1.2.2, for time windows w2,…w i …,w N Perform feature extraction to obtain a multi-scale feature matrix F, which is: For the residual R i-1 (t) Perform average pooling centered at the current time point to obtain the time window w i Features F i (t), and then use the residual R i-1 (t) minus characteristic F i (t) Get the residual R i (t); After completing the feature extraction of all time windows, a multi-scale feature matrix F with a shape of T×N is obtained. The specific expression is as follows: Where: T is the length of the original deformation monitoring data.

3. The deformation monitoring data prediction method based on multi-scale features according to claim 2 is characterized in that: The specific formula for average pooling is as follows: Where: x t Deformation monitoring data value at the tth time point; The time window w is used i For the pooling result at time point t, k i represents the time window w i There are k on the left and right sides of the internal time point t i Monitoring data, τ is the time interval for deformation monitoring data collection; j is the time window w i The time point number of the monitoring data within; In S1.2, the specific formula for introducing the residual iterative calculation method to extract multi-scale features from the original deformation monitoring data is as follows: For time window w1: in: is the pooling result of time window w1 at time point t; For time windows w2,…w i …,w N : Among them: F i (t) is the time window w i The initial features at time point t; R i-1 (t) is the time window w i-1 The residual at time point t; To use the time window w i Pooling result at time point t.

4. The deformation monitoring data prediction method based on multi-scale features according to claim 3 is characterized in that: The S2 includes: S2.

1. Design the convolutional layer of the prediction neural network model; specifically: S2.1.1, let the input multi-scale feature matrix be Among them: The multi-scale feature matrix is The data shape is m×N, that is, N-dimensional multi-scale features of m consecutive time points; S2.1.2, based on one-dimensional convolutional neural network, extract the multi-scale feature matrix: The features of the convolutional neural network are flattened into the convolution output vector z. The specific formula is as follows: Where: ol is the convolution kernel weight, b o is the bias term, σ is the ReLU activation function, and h o is the output feature of the convolution layer, the result of the oth channel at time point t, C is the width of the convolution kernel, 1≤C≤m-1; S2.

2. Design the first fully connected layer and the second fully connected layer of the prediction neural network model; specifically: S2.2.

1. The first fully connected layer maps the convolution output vector to M fc1 neurons, and nonlinearity is introduced through the ReLU activation function. The specific formula is as follows: z′=σ(ω fc1 with+with fc1 ); Where: z′ is the output result of the first fully connected layer; ω fc1 and b fc1 They are M in the first fully connected layer. fc1 The weight matrix and bias vector of each neuron, σ is the ReLU activation function; S2.2.2, the second fully connected layer will M fc2 The output of each neuron is mapped to the final predicted output. The specific formula is as follows: Where: fc2 and b fc2 They are M in the second fully connected layer. fc2 The weight matrix and bias vector of each neuron, It is the final prediction output result, that is, the predicted deformation monitoring data value at the next Q time points.

5. The deformation monitoring data prediction method based on multi-scale features according to claim 4 is characterized in that: The S3 includes: S3.

1. Construct the loss function Loss based on the mean square error and the predicted value difference loss; S3.

2. Use batch gradient descent method to train the prediction neural network model until Loss≤∈ stops training, save the model parameters to obtain the trained prediction model, where: ∈ is the set threshold.

6. The deformation monitoring data prediction method based on multi-scale features according to claim 5 is characterized in that: In S3.1, the specific formula of the loss function Loss is as follows: Loss=Loss MSE +λ·Loss diff ; Loss diff (and pred,Q -and pred,1 -(and true,Q -and true,1 )) 2 ; Among them: Loss MSE is the mean square error; Loss diff is the data increment error; y true,n is the nth true value of the real data; y true,1 is the first true value of the real data; y pred,n is the nth predicted value of the predicted data by the trained prediction model, y pred,1 Predict the first predicted value of the data for the trained prediction model; λ is the weight coefficient.

7. The deformation monitoring data prediction method based on multi-scale features according to claim 5 is characterized in that: In S3.2, the batch gradient descent method is used to train the prediction neural network model, specifically: the original deformation monitoring data is divided into multiple batches, forward propagation, loss calculation, back propagation and parameter update are performed, and the loss function value is gradually reduced through multiple rounds of iterative training, wherein: the Adam optimizer is used for updating when the parameters are updated.

8. The deformation monitoring data prediction method based on multi-scale features according to claim 7 is characterized in that: The S4 specifically includes: S4.

1. Take the features of the first P time points of the validation data as the input of the trained prediction model: S4.

2. Based on the trained prediction model, the deformation monitoring data values ​​at the next Q time points are calculated through forward propagation, and the prediction vector of the deformation monitoring data with a length of Q is output. The specific formula is as follows: Among them: Model represents the trained prediction model; After S4.2, the method further includes: inputting the verification data into the trained prediction model for prediction to obtain a corresponding prediction vector, and comparing the vector with the true value to verify the trained prediction model.

9. A readable storage medium, characterized in that: Computer program instructions are stored thereon, and when the computer program instructions are executed by a processor, a deformation monitoring data prediction method based on multi-scale features as described in any one of claims 1 to 8 is implemented.

10. An electronic device, characterized in that: include: At least one processor, at least one memory, and computer program instructions stored in the memory, when the computer program instructions are executed by the processor, the deformation monitoring data prediction method based on multi-scale features as described in any one of claims 1 to 8.

Images