CN118037563A - Airport runway settlement prediction method and system - Google Patents

Abstract

The invention relates to the technical field of settlement prediction, in particular to an airport runway settlement prediction method and system, wherein the method comprises the following steps: step1, processing a plurality of acquired SAR images according to an SBAS-InSAR technology; step2, constructing an EnKF-BP model; step3, predicting a result through the model, and updating the state vector; step4, judging whether the iteration requirement is met, if not, returning to Step3, and if so, outputting a final prediction result. The method can be used for predicting the settlement of the airport runway.

Description

A method and system for predicting airport runway settlement

Technical Field

The present invention relates to the technical field of settlement prediction, and in particular to a method and system for predicting settlement of an airport runway.

Background technique

Traditional methods of measuring runway subsidence usually involve measuring the position changes of ground markers using tools such as levels and rangefinders. Problems with these methods include:

1) Frequent manual measurement is required, which is time-consuming and labor-intensive, and may result in human errors;

2) The accuracy is limited and high-frequency and large-scale monitoring cannot be achieved;

3) Measurements under complex terrain and environmental conditions are limited.

The development of modern technology has provided some more advanced methods for detecting airport runway subsidence, such as lidar measurement, satellite remote sensing technology, drone aerial photography, etc. The problems of these methods mainly include:

1) The cost of technical equipment is high and requires professional knowledge and skills to operate and maintain;

2) The applicability to complex weather conditions and environmental interference needs to be improved.

These methods require a lot of manpower, time and high monitoring costs, are easily affected by the environment and cannot obtain large-scale monitoring points.

At present, the research ideas for settlement prediction mainly include numerical simulation method, mathematical statistics method, various combined prediction algorithms and deep learning algorithm, specifically:

1. Numerical simulation method: used to simulate the subsidence caused by groundwater exploitation, rainfall and drainage, but the modeling is difficult and the robustness and universality of the prediction results are limited.

2. Mathematical statistics method: By analyzing the internal connections of historical data, the future subsidence trend can be predicted, but it has high requirements on data quality and is not suitable for multiple factors.

3. Combined prediction algorithm: Improve prediction accuracy by increasing dimensions or types of algorithms, but it is difficult and cannot be widely used.

4. Deep learning algorithms: can mine complex relationships, but some key issues need to be addressed to apply artificial neural networks more effectively.

Gradient dependency is an important issue when training artificial neural networks. In order to overcome the local minimum problem that may be caused by gradient dependency, improved methods such as batch gradient descent, stochastic gradient descent, and mini-batch gradient descent can be used. These methods can improve training and prediction performance, but they still need to be selected and optimized for specific situations.

Summary of the invention

The present invention provides an airport runway settlement prediction method and system, which can better predict airport runway settlement.

A method for predicting airport runway subsidence according to the present invention comprises the following steps:

Step 1: Process the acquired multiple SAR images according to SBAS-InSAR technology;

Step 2, build EnKF-BP model;

Step 3: Use the model to predict the results and update the state vector;

Step 4: Determine whether the iteration requirements are met. If not, return to Step 3. If so, output the final prediction result.

As a preferred embodiment, in Step 1, specifically:

Step 1-1, generate connection graph and interference pairs: generate time and space relationship between images;

Step 1-2, DEM data generates interference image: Use the pre-processed SAR image data to generate interference image;

Step 1-3, phase unwrapping: perform phase unwrapping on the interference image of each small area to obtain surface deformation information;

Step 1-4, select GCP points for track refinement;

Step 1-5, inversion and geocoding: perform time series analysis on the unwrapped phase data through two inversions; project the results into the real coordinate system through geocoding for easy observation.

Preferably, in the EnKF-BP model of Step 2, the structure of BP is assumed to be M ^* ; the weight ω and bias b of BP are regarded as the state X ^* of M ^* , thereby obtaining X ^* = (ω, b) ^T ;

The predicted value formula is:

In the formula, is the state prediction value of the i-th set at time k,/> is the state analysis value of the i-th set at time k-1, _qi is the model error, which obeys a Gaussian distribution with a mean of 0 and a covariance matrix of _Qk ; Y ^* is the output; H ^* represents the state observation matrix.

As a preference, Updated by the following formula:

In the formula, is the state analysis value of the ith set at time k+1; K _k+1 is the gain matrix; /> is the observation data at time k+1; _vi,k is the observation error, which obeys a Gaussian distribution with a mean of 0 and a covariance matrix of _Qk ./> is the average analysis value of all sets at time k+1; ^Pf is the prediction error variance matrix, and H is the observation operator.

The present invention provides an airport runway settlement prediction system, which adopts the above-mentioned airport runway settlement prediction method.

The present invention combines the EnKF (Ensemble Kalman filter) and the neural network BP (Back Propagation) system to accurately predict the airport runway surface settlement, and the beneficial effects are as follows:

(1) Avoiding gradient calculation: The present invention avoids the calculation of gradients through the data assimilation method, thereby avoiding problems such as gradient explosion and gradient disappearance caused by gradient training.

(2) Continuous learning capability: The present invention uses an integrated Kalman filter (EnKF) for parameter optimization, which can continuously update parameters when there are new observations, thus achieving continuous learning capability. This enables the present invention to adapt to new data in a timely manner and make corrections based on observations, thereby improving accuracy and adaptability.

(3) Improved accuracy: Compared with the traditional gradient-optimized BP model, the EnKF-optimized BP model performs better in terms of accuracy. By capturing the changes in observed values and appropriately modifying the model, the present invention can more accurately predict the target value and improve the prediction accuracy.

(4) InSAR has the advantages of high precision, all-weather capability, large-area coverage, high temporal and spatial resolution, long-term monitoring capability, and non-contact, making it an important tool for monitoring and analyzing surface deformation.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for predicting airport runway subsidence in an embodiment;

FIG2 is a schematic diagram of the BP structure and back propagation in the embodiment.

Detailed ways

In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments are only for explaining the present invention and are not intended to limit it.

Example

As shown in FIG1 , this embodiment provides a method for predicting airport runway subsidence (integrating SBAS-InSAR technology and EnKF-BP), which includes the following steps:

Step 1: Process the acquired multiple SAR images according to SBAS-InSAR technology.

Specifically:

Step 1-1, Generate connection graph and interference pairs: Generate the time and space relationship between images.

Step 1-2, generate interference image from DEM data: use the preprocessed SAR image data to generate interference image; the interference image reflects the surface changes at different time points after removing the terrain phase.

Step 1-3, phase unwrapping: perform phase unwrapping on the interference image of each small area to obtain surface deformation information.

Step 1-4: Select GCP points for orbit refinement. This improves the accuracy of satellite orbit calculation and reduces surface deformation errors.

Step 1-5, inversion and geocoding: perform time series analysis on the unwrapped phase data through two inversions to extract the changing trend and rate of surface deformation; project the results into the real coordinate system through geocoding for easy observation.

Step 2: Build the EnKF-BP model, including determining the network structure, determining the number of neurons, setting hyperparameters, initializing parameters (initializing the state vector set), etc.

Step 3: Use the model to predict the results and update the state vector.

Step 4: Determine whether the iteration requirements are met. If not, return to Step 3. If so, output the final prediction result.

InSAR (Interferometric Synthetic Aperture Radar)

InSAR (Interferometric Synthetic Aperture Radar) is a method that uses radar imaging technology to monitor surface deformation. Its process generally includes the following steps:

Data collection: The radar image data of the earth's surface is obtained through the synthetic aperture radar system carried by satellites or aircraft. These data include radar interferometric images at different time points, recording the reflection characteristics of the earth's surface at different time periods.

Interferometric image registration: radar image data from the same area at different time points are registered to ensure their consistency in space and time. This step usually requires precise registration and correction to eliminate image distortion caused by different imaging conditions and the rotation of the earth.

Phase unwrapping: Next, the registered radar interferometer image is subjected to phase unwrapping. This step calculates the deformation information of the surface by comparing the phase differences of radar signals at different time points, and then obtains the deformation image of the surface.

Deformation analysis: The deformation image obtained by phase analysis can be used to quantitatively analyze and monitor the deformation of the surface. By analyzing the changing characteristics of the deformation image, earthquakes, geological activities, surface subsidence and other phenomena can be identified, thus providing important data support for geological disaster monitoring, resource exploration and environmental monitoring.

Ensemble Kalman Filter (EnKF)

The basic steps of EnKF are divided into two steps:

Assume there are N sets, k = 0, and initialize each set at time,

(1) Prediction

In the prediction step, the update of the state vector, i.e., the state analysis value, is determined by (2.1), where Gaussian noise is added to the update. The prediction value is calculated based on the state vector and the observation operator.

In the formula, is the state analysis value of the i-th set at the moment;/> is the predicted value of the state at time k+1; M _k,k+1 is the state change relationship from time k to time k+1, which is generally a nonlinear model operator; _qi is the model error, which obeys a Gaussian distribution with a mean of 0 and a covariance matrix of _Qk ; _Yi is the predicted value of the i-th state set; H is the model observation operator, which is the mapping relationship between the state vector and the predicted value.

(2) Update

In the formula, is the state analysis value of the i-th set at time k+1; K _k+1 is the gain matrix; /> is the observation data at time k+1; _vi,k is the observation error, which obeys a Gaussian distribution with a mean of 0 and a covariance matrix of _Qk ./> is the average analysis value of all sets at time k+1; ^Pf is the prediction error variance matrix, and H is the observation operator.

BP Neural Network

BP usually uses activation functions to increase the nonlinearity of the network. The activation function can be sigmoid, ReLU, tanh, etc. Most of the parameter update methods used in back propagation are gradient optimization calculation methods. Its feedforward propagation is as follows:

Where _xi , _hj, and _yk represent the node values in the input, hidden, and output layers. N, M, and m are the number of neurons in the input, hidden, and output layers; _ωji is the weight connecting the input _xi and the jth neuron in the hidden layer; _bj and _b′j represent the bias in the hidden layer; _ωkj is the weight connecting the jth neuron ( _hj ) in the hidden layer and the output _yk ; _f1 and _f2 are the activation functions in the hidden and output layers.

Figure 2 is a schematic diagram of the BP structure and back propagation, where y is the actual output of the network and Y is the expected output of the network. is the error between the actual output and the expected output. is greater than the given error requirement, then/> The weights and biases are updated by backpropagating along the neural network.

EnKF-BP network construction

Assume that the structure of BP for a particular problem (number of network layers, number of nodes in each layer, and connection status between nodes) has been determined, denoted by M ^* . The weight ω and bias b in BP are regarded as the state X ^* of M ^* , thus obtaining X ^* = (ω, b) ^T.

The predicted value formula is:

In the formula, is the state prediction value of the i-th set at time k,/> is the state analysis value of the i-th set at time k-1, _qi is the model error, which obeys a Gaussian distribution with a mean of 0 and a covariance matrix of _Qk ; Y ^* is the output, which is the mapping relationship between the state vector and the predicted value; H ^* represents the state observation matrix.

The parameter update in back propagation is replaced by EnKF, which is started when new observations are available and adjusts the parameters of the model to integrate the new observations to make new predictions. The combination of BP and EnKF has several hyperparameters that need to be determined based on prior information of the actual situation.

This embodiment provides an airport runway settlement prediction system, which adopts the above-mentioned airport runway settlement prediction method.

This embodiment combines SBAS-InSAR technology with EnKF-BP to predict airport runway subsidence, which has the following advantages:

High-precision monitoring: The use of SBAS-InSAR technology can achieve non-contact, large-scale and continuous monitoring, providing sub-centimeter-level settlement monitoring accuracy. This can accurately monitor the settlement of airport runways and detect potential problems in a timely manner.

Timely prediction: By introducing the EnKF-BP model, the timeliness and gradient optimization issues of the traditional BP neural network model are solved. The EnKF (ensemble Kalman filter) optimization algorithm can better adapt to environmental changes, provide more accurate prediction results, and speed up the prediction without retraining the model when the amount of data is large.

By accurately predicting the settlement of airport runways, this embodiment can help airport management departments to detect potential settlement problems early, speed up data processing, improve prediction efficiency, and provide timely feedback and decision support for airport management departments. It can reduce maintenance costs, arrange maintenance plans reasonably, and avoid expensive repair costs caused by unexpected problems. This helps reduce possible safety risks and delays in airport operations.

The above is a schematic description of the present invention and its implementation methods, which is not restrictive. The drawings show only one implementation method of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by it and does not deviate from the purpose of the invention, he or she can design a structure and an embodiment similar to the technical solution without creativity, which should fall within the protection scope of the present invention.

Claims (5)

1. A method for predicting airport runway subsidence, characterized in that it comprises the following steps: Step 1: Process the acquired multiple SAR images according to SBAS-InSAR technology; Step 2, build EnKF-BP model; Step 3: Use the model to predict the results and update the state vector; Step 4: Determine whether the iteration requirements are met. If not, return to Step 3. If so, output the final prediction result. 2. The method for predicting airport runway subsidence according to claim 1, characterized in that: in Step 1, specifically: Step 1-1, generate connection graph and interference pairs: generate time and space relationship between images; Step 1-2, DEM data generates interference image: Use the pre-processed SAR image data to generate interference image; Step 1-3, phase unwrapping: perform phase unwrapping on the interference image of each small area to obtain surface deformation information; Step 1-4, select GCP points for track refinement; Step 1-5, inversion and geocoding: perform time series analysis on the unwrapped phase data through two inversions; project the results into the real coordinate system through geocoding for easy observation. 3. The method for predicting airport runway subsidence according to claim 2, characterized in that: in the EnKF-BP model of Step 2, the structure of BP is assumed to be M ^* ; the weight ω and the bias b of BP are regarded as the state X ^* of M ^* , thereby obtaining X ^* = (ω, b) ^T ; The predicted value formula is: In the formula, is the state prediction value of the i-th set at time k,/> is the state analysis value of the i-th set at time k-1, _qi is the model error, which obeys a Gaussian distribution with a mean of 0 and a covariance matrix of _Qk ; Y ^* is the output; H ^* represents the state observation matrix. 4. The method for predicting airport runway subsidence according to claim 3, characterized in that: Updated by the following formula: In the formula, is the state analysis value of the ith set at time k+1; K _k+1 is the gain matrix; /> is the observation data at time k+1; v _{i, k} is the observation error, which obeys a Gaussian distribution with a mean of 0 and a covariance matrix of Q _k , /> is the average analysis value of all sets at time k+1; ^Pf is the prediction error variance matrix, and H is the observation operator. 5. An airport runway settlement prediction system, characterized in that it adopts an airport runway settlement prediction method as described in any one of claims 1-4.