Sequential InSAR nuclear power plant ground surface settlement prediction method based on SSA-LSTM
By using a time-series InSAR method based on SSA-LSTM to predict land subsidence in nuclear power plants, the problem of insufficient coverage and real-time performance of traditional monitoring methods in nuclear power plants has been solved. This method achieves high-precision subsidence prediction and monitoring, providing reliable technical support for the safe operation of nuclear power plants.
Patent Information
- Application Number
- CN202511668545.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-14
- Publication Date
- 2026-02-06
AI Technical Summary
Traditional manual inspections and single-point monitoring methods in nuclear power plants have limited coverage, insufficient real-time performance, and difficulty in data integration, making it difficult to meet the needs of comprehensive and high-precision safety supervision. They are also prone to failure in extreme weather or geological disaster situations.
A time-series InSAR method for predicting land subsidence in nuclear power plants based on SSA-LSTM was adopted. By acquiring multi-temporal time-series InSAR monitoring data of land subsidence, PS-InSAR technology was used for preprocessing, and the hyperparameters of LSTM were optimized by combining the sparrow search algorithm. The time-series InSAR subsidence prediction model was constructed and trained, and the model performance was evaluated.
This improved the accuracy and stability of the LSTM model in predicting land subsidence in nuclear power plants, reduced noise interference, ensured the reliability of input data, and provided accurate and reliable time-series prediction support, meeting the needs of land subsidence monitoring and early warning in nuclear power plant areas.
Smart Images
Figure CN121479520A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of nuclear power plant ground surface settlement prediction, in particular to a time series InSAR nuclear power plant ground surface settlement prediction method based on SSA-LSTM. BACKGROUND
[0002] As a large country of nuclear power development, the number of nuclear power plants and supporting facilities continues to grow. As of now, there are 102 nuclear power units in operation, under construction and approved for construction in China, with a total installed capacity of 113 million kilowatts, and the overall scale of nuclear power has ranked first in the world for the first time. At present, there are 28 nuclear power units under construction in China, with a total installed capacity of 33.65 million kilowatts, and the installed capacity of the units under construction has maintained the first place in the world for 18 consecutive years; the number of commercial nuclear power units in China reaches 58, with a total installed capacity of 60.96 million kilowatts. In 2024, the cumulative power generation of nuclear power units in operation in China will reach 444.7 billion kilowatt-hours, accounting for 4.72% of the total power generation in China, ranking second in the world, which is equivalent to reducing the burning of 127 million tons of standard coal and reducing carbon dioxide emissions by 334 million tons. Not only is the scale of nuclear power continuing to expand, but the self-innovation capability of China's nuclear power technology has also significantly enhanced. Nuclear power plants, as the core base for clean energy production, are also high-risk special facilities, and their safe operation is directly related to the safety of millions of residents and the long-term stability of the ecological environment.
[0003] However, there are significant differences in the geographical location, geological and hydrological conditions, reactor types, operation time and enterprise management level of nuclear power plants in China. Some early-built nuclear power plants face problems such as equipment aging and technology upgrading, while newly-built units need to cope with the regulatory challenges brought by new reactor technologies. Traditional manual inspection and single-point monitoring methods have limitations such as limited coverage, insufficient real-time performance, and difficulty in data integration, making it difficult to fully grasp the safety status of the complex system of nuclear power plants. Especially in the case of extreme weather, geological disasters and other emergencies, traditional monitoring methods are prone to failure and cannot meet the needs of all-around and high-precision safety supervision. SUMMARY
[0004] To solve the above technical problems, the technical scheme adopted by the present application is as follows: According to a first aspect of the present application, a time series InSAR nuclear power plant ground surface settlement prediction method based on SSA-LSTM is provided, which comprises the following steps: S100, acquiring multi-temporal ground surface settlement time series InSAR monitoring data of a nuclear power plant area, and preprocessing the multi-temporal ground surface settlement time series InSAR monitoring data; S200, extracting the preprocessed time series InSAR monitoring data, normalizing it after dividing it into a training set and a test set; S300, using sparrow search algorithm to optimize the hyperparameters of long short-term memory network LSTM, constructing and training a time series InSAR settlement prediction model based on LSTM; S400, performing anti-normalization and integration on the predicted settlement data output by the time series InSAR settlement prediction model, finally outputting the time series InSAR settlement prediction result, and completing the model performance evaluation.
[0005] Further, in step S200, the time series InSAR monitoring data is divided into a training set and a test set according to a ratio of 4:1.
[0006] Further, in step S300, the LSTM hyperparameters optimized by the sparrow search algorithm include the initial learning rate, the regularization parameter, and the number of hidden layer nodes of the LSTM structure of the LSTM network.
[0007] Further, in step S300, the specific process of optimizing the LSTM hyperparameters by the sparrow search algorithm includes: S310, determining the number of sparrow populations, the maximum number of iterations, the number of optimization parameters, and the upper and lower bounds of parameter values, and initializing the value of the SSA optimization algorithm; S320, calculating the fitness value of each sparrow according to the number of sparrow populations and updating the position, using the position iteration formula of the producer, beggar and sparrow with danger consciousness to update the position of the optimal individual in the iteration sparrow population and the global optimal position, and saving the position found by iteration; S330, determining whether the maximum number of iterations is reached, if yes, then exiting the loop to obtain the optimal network parameters, otherwise continuing the above updating and iteration steps until the iteration ends, and outputting the optimal network parameters.
[0008] Further, in step S100, the PS-InSAR technology is used for preprocessing the time series InSAR monitoring data of multi-temporal ground surface settlement, and the processing process includes the steps of main image selection, registration, interference, inversion, and geographic coding.
[0009] Further, in the processing process of the PS-InSAR technology, the permanent scatterer points are selected by calculating the amplitude deviation value, and the amplitude deviation value is the ratio of the amplitude standard deviation to the mean value. When the amplitude deviation value is less than or equal to the set threshold value, the pixel point is selected as the permanent scatterer point.
[0010] Further, in step S400, the evaluation indexes used for model performance evaluation include the root mean square error, the mean absolute error, and the absolute average percentage error.
[0011] Further, in step S300, the time series InSAR settlement prediction model based on LSTM is constructed using an adaptive moment estimation optimizer.
[0012] Further, the initial learning rate is in the range of [1e-3, 1e-2], the regularization coefficient is in the range of [1e-4, 1e-1], and the number of hidden layer nodes is in the range of [10, 30].
[0013] Further, the sparrow population number is 20, the population dimension is 3, and the maximum iteration number is 8.
[0014] The present application has at least the following beneficial effects: The SSA-LSTM-based time series InSAR nuclear power plant ground subsidence prediction method provided by the present application optimizes the hyperparameters of the long short-term memory network (LSTM) by using the sparrow search algorithm (SSA), effectively improving the accuracy and stability of the LSTM model in the prediction of InSAR time series data of nuclear power plant ground subsidence, solving the problem of insufficient prediction accuracy caused by the difficulty in determining the hyperparameters of the traditional LSTM model; at the same time, the original InSAR data is preprocessed by combining the PS-InSAR technology, the noise interference is reduced by selecting permanent scatterer points, and the reliability of the input model data is ensured; in addition, by reasonably dividing the training set and the test set, using scientific normalization and denormalization processing, and using multi-dimensional model performance evaluation indexes, the effectiveness and credibility of the prediction results are further guaranteed, which can provide accurate and reliable time series prediction support for the ground subsidence monitoring and early warning of the nuclear power plant area, and has high practical value and application prospect. BRIEF DESCRIPTION OF DRAWINGS
[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0016] Figure 1 The flowchart of the SSA-LSTM-based time series InSAR nuclear power plant ground subsidence prediction method provided by the present application is shown in the figure. Figure 2 The PS-InSAR technology route schematic diagram provided by the present application is shown in the figure. Figure 3 The overall technical route schematic diagram of the SSA-LSTM-based time series InSAR nuclear power plant ground subsidence prediction method provided by the present application is shown in the figure. Figure 4 The prediction result comparison schematic diagram provided by the present application is shown in the figure. Figure 5 The prediction error schematic diagram provided by the present application is shown in the figure. DETAILED DESCRIPTION
[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.
[0018] It should be noted that, based on the present disclosure, a person of ordinary skill in the art should appreciate that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, an apparatus can be implemented and / or a method can be practiced using any number of the aspects set forth herein. In addition, such an apparatus can be implemented and / or such a method can be practiced using other structure and / or functionality in addition to or other than one or more of the aspects set forth herein.
[0019] Reference will be made to the drawings Figure 1 A SSA-LSTM-based time-series InSAR nuclear power plant ground surface subsidence prediction method will be introduced below with reference to the flowchart of the SSA-LSTM-based time-series InSAR nuclear power plant ground surface subsidence prediction method shown in the drawing.
[0020] The SSA-LSTM-based time-series InSAR nuclear power plant ground surface subsidence prediction method can include the following steps: S100, acquiring multi-temporal ground surface subsidence time-series InSAR monitoring data of a nuclear power plant area, and preprocessing the multi-temporal ground surface subsidence time-series InSAR monitoring data.
[0021] Further, the preprocessing of the multi-temporal ground surface subsidence time-series InSAR monitoring data adopts PS-InSAR technology, and the processing process includes the steps of main image selection, registration, interference, inversion, and geographic coding.
[0022] In this embodiment, the multi-temporal ground surface subsidence time-series InSAR monitoring data of the nuclear power plant area is acquired by synthetic aperture radar (SAR), and the original data is preprocessed by using PS-InSAR technology.
[0023] The preprocessing process includes main image selection (selecting an image in the middle of the time series and having good connectivity with other images as the main image), registration (unifying the pixel point position information of all images), interference (processing phase information, generating a differential interferogram in combination with an external DEM, and selecting a permanent scatterer point), inversion (estimating a deformation rate and eliminating atmospheric phase influence), and geographic coding (converting the PS point from a SAR coordinate system to a geographic coordinate system), and finally obtaining the purified ground surface subsidence monitoring data.
[0024] Beneficial effects: PS-InSAR technology can effectively separate the reference ellipsoid, atmospheric, noise and other interference phase, extract the real ground deformation information, reduce the impact of noise on subsequent prediction, provide high-quality input data for the model, and lay the foundation for prediction accuracy.
[0025] PS-InSAR technology processing method: The PS-InSAR technology route is shown in Figure 2 , and the processing process mainly includes main image selection, registration, interference, inversion, and geographic coding. The main image selection is to select a best image as the main image according to the time and space characteristics of the imported data set. The main image is generally in the middle position of the long time sequence and needs to be connected with the remaining slave images. The registration is to uniformly process the position information of all main and auxiliary images to obtain consistent pixel point information. The interference is to process the phase in the SAR image data. After removing the flat and terrain phase, the external DEM data is used to generate a differential interferogram, and the permanent scatterer point is selected by the amplitude dispersion index. The inversion is to identify the permanent scatterer point, estimate the deformation rate using a linear model, and further solve the average deformation rate after eliminating the atmospheric phase. The geographic coding is to convert the PS point from the SAR coordinate system to the geographic coordinate system to obtain the line-of-sight deformation result.
[0026] After long time sequence screening, the stable points with high reflectivity in the image are selected as PS points. According to the phase information of the selected points, the phases of the reference ellipsoid, atmosphere, noise and other factors affecting the coherence of SAR image are separated to obtain the phase information reflecting the obvious terrain change. If the image has high signal-to-noise ratio, the phase dispersion value can be approximately represented by the amplitude dispersion value. The amplitude dispersion value is calculated by the ratio of the standard deviation to the mean value of the amplitude. The value determines the stability of the PS point, and the points with high amplitude dispersion value are selected as effective data in the experiment. The amplitude dispersion value calculation formula for identifying PS points is: ; In the formula: is the ratio of the standard deviation to the mean value of the amplitude of each pixel (i, j) in the image; Th is the set threshold value. If ≤Th, it is considered that the pixel point meets the requirements, and the pixel point is selected as a permanent scatterer point.
[0027] In the SAR image data processing process, the phase of the PS point can be represented as: ; In the formula: def is the deformation phase; topo is the reference DEM induced terrain phase; flatis the phase of atmospheric delay; atm is the phase of atmospheric delay; noise is the phase of noise.
[0028] ; In the formula: Lambda 、 Theta 、 R respectively, radar wavelength, incidence angle, slant range; B ⊥ is the vertical baseline; Δ h err is the DEM error.
[0029] def The deformation phase can be divided into nonlinear and linear deformation phases: ; In the formula: linear is the linear deformation phase; non-linear is the nonlinear phase; D non-linear is the nonlinear deformation; Δv is the linear deformation rate; T is the time interval relative to the main image. The PS points are differenced according to the Delaunay triangular grid edges, and the obtained phase is represented as: ; ; In the formula: diff is the final unwrapped phase value. Through the phase value, the deformation increment and the elevation increment are solved, so as to obtain the deformation rate field of the region of interest.
[0030] S200, extract the preprocessed time series InSAR monitoring data, divide it into a training set and a test set, and then perform normalization processing.
[0031] Further, in the step S200, the time series InSAR monitoring data is divided into a training set and a test set according to a ratio of 4:1.
[0032] From the preprocessed time series InSAR monitoring data, extract the effective time series sequence, divide the data according to a ratio of 4:1 (80% for the training set and 20% for the test set); adopt normalization processing (such as mapping the data to the interval [0, 1] or [-1, 1]) to eliminate the influence of different orders of magnitude of data, so that the model training is more stable.
[0033] The division ratio of 4:1 ensures that the training set has sufficient data to fit the data law, and the test set can effectively verify the generalization ability of the model; the normalization processing avoids the model weight bias caused by the difference in data magnitude, and improves the training efficiency and stability.
[0034] S300, using sparrow search algorithm to optimize the hyperparameters of long short-term memory network LSTM, constructing and training a time series InSAR settlement prediction model based on LSTM.
[0035] Further, in the step S300, the LSTM hyperparameters optimized by the sparrow search algorithm include the initial learning rate of the LSTM network, the regularization parameter, and the number of hidden layer nodes of the LSTM structure.
[0036] Further, in the step S300, the specific process of optimizing the LSTM hyperparameters by the sparrow search algorithm includes: S310, determining the number of sparrow populations, the maximum number of iterations, the number of optimization parameters, and the upper and lower bounds of parameter values, and initializing the value of the SSA optimization algorithm.
[0037] S320, calculating the fitness value of each sparrow according to the number of sparrow populations and updating the position, using the position iteration formula of the producer, beggar and sparrow with danger consciousness to update the position of the optimal individual in the iteration sparrow population and the global optimal position, and saving the position found by iteration.
[0038] S330, determining whether the maximum number of iterations is reached, if yes, then exiting the loop to obtain the optimal network parameters, otherwise continuing the above updating and iteration steps until the iteration ends, and outputting the optimal network parameters.
[0039] Further, the initial learning rate is in the range of [1e-3, 1e-2], the regularization coefficient is in the range of [1e-4, 1e-1], and the number of hidden layer nodes is in the range of [10, 30].
[0040] Further, the number of sparrow populations is 20, the population dimension is 3, and the maximum number of iterations is 8.
[0041] In this embodiment, the sparrow search algorithm is used to optimize the LSTM hyperparameters; the specific optimization method is as follows: Assume that the initial position of the sparrow population searching for food is represented by matrix X, F X represents the fitness value of its population: ; ; In the formula, n is the population number, d is the variable dimension, F X Each row value in F
[0042] The producer, as the leader of the population searching for food, has a better fitness value, which guides the beggars to search for food and is responsible for the migration of the beggars. The position of the producer is iterated as follows: ; where t represents the current iteration number, X t i，j represents the value of the jth dimension of the ith sparrow at the tth iteration, j = 1, 2, …, d; iter max is a constant of the maximum iteration number, a is a random number between 0 and 1, Q is a normally distributed random number, L represents a 1 x d matrix, the elements of which are all 1, and R2 represents an alarm value, and ST represents a safety threshold.
[0043] When R2 < ST, it means that the producer can search for food in a larger range without predators around, and when R2 ≥ ST, it means that the producer has found danger and needs to guide the population to move to a safe area.
[0044] The beggars in the population have low fitness values, so they need to constantly pay attention to the dynamics of the producer. If the producer finds better food, they will quickly compete for the food. If they fail in the competition, the beggars will continue to pay attention to the dynamics of the producer. Therefore, the position of the beggars is iterated as follows: ; where X p represents the latest food position of the producer, X worst represents the current worst position, A represents a 1 x d matrix, the elements of which are randomly 1 or -1; A + = A T (AA T ) -1 . If i > n / 2, it means that the ith beggar in the population is in an unfavorable fitness value.
[0045] Some sparrows in the population are on the periphery of the group. If they find danger, they will quickly move to a safe area. The remaining sparrows in the middle of the group will move randomly to find a safe position. The position of the sparrow with a sense of danger in the population is iterated as follows: ; where X best represents the current best fitness position, β represents a step control parameter, which is a random number obeying a normal distribution with a mean of 0 and a variance of 1, K represents a random number in the interval [-1, 1], which represents the moving direction and coefficient of the step of the sparrow in the population, f i represents the fitness value of the ith sparrow, f g represents the global best fitness value, and f pε represents a constant approaching 0 to avoid zero division error.
[0046] f i >f g representing that the sparrow is in the periphery of the group and needs to move to a safe location after discovering danger, i =f g representing that the sparrow in the middle of the population is aware of the danger and needs to find a sparrow in a safe location.
[0047] The SSA algorithm uses the danger awareness behavior of the sparrow population, and the above formula iteratively updates the latest position of all sparrows in the population. According to the position information of the sparrow population, the optimal individual position in the population and the global optimal fitness value are obtained, so as to determine the optimal weight and threshold of the LSTM model.
[0048] After the above preprocessing, the data is used for experiments, and the first 80% of the data is selected as the training set, and the remaining 20% of the data is selected as the test set. The optimizer of the SSA-LSTM model selects the adaptive moment estimation (Adam) optimizer, sets the number of sparrow populations to 20, the population dimension to 3, the initial learning rate to [1e-3, 1e-2], the regularization coefficient to [1e-4, 1e-1], the number of hidden layer nodes to [10, 30], the maximum number of iterations to 8, and the training number to 1000 times.
[0049] The SSA algorithm can efficiently search for the global optimal hyperparameter, solve the problem of insufficient prediction accuracy and overfitting / underfitting caused by the difficulty in determining the hyperparameter of a single LSTM model, improve the model training efficiency and fitting ability for time series data, and make the prediction more accurate.
[0050] S400, the predicted settlement data output by the time series InSAR settlement prediction model is de-normalized and integrated, and finally the time series InSAR settlement prediction result is output, and the model performance evaluation is completed.
[0051] SSA-LSTM algorithm specific implementation: (1) Taking the initial learning rate of the LSTM network, the regularization parameter, and the number of hidden layer nodes of the LSTM structure as the optimization object, determining the number of sparrow populations, the maximum number of iterations, the number of optimization parameters, and the upper and lower bounds of parameter values (initial learning rate, regularization coefficient, and number of hidden layer nodes), and initializing the value of the SSA optimization algorithm.
[0052] (2) Calculate the fitness value of each sparrow according to the number of sparrow populations and update the position, and use the above formula to update the optimal individual position and global optimal position in the iterative sparrow population, and save the position found by iteration.
[0053] (3) judging whether the maximum iteration number is reached, if yes, exiting the loop to obtain the optimal network parameter, otherwise, continuing the above step S200 until the iteration ends and outputting the optimal network parameter.
[0054] The InSAR data is input into the trained model by using the optimal network parameter output by the optimization algorithm, and the output of the test set is performed.
[0055] The evaluation indexes used in the evaluation of the model performance include root mean squared error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE).
[0056] The normalized prediction data output by the model is de-normalized (restored to the real ground subsidence order of magnitude), and the time series prediction result is integrated; the root mean squared error (RMSE), the mean absolute error (MAE) and the mean absolute percentage error (MAPE) are used to evaluate the model performance and verify the reliability of the prediction result.
[0057] The prediction accuracy of the model is evaluated by the RMSE, the actual prediction error is evaluated by the MAE, and the model accuracy is evaluated by the MAPE. The evaluation value of the SSA-LSTM relative to the actual data is smaller, indicating that the prediction effect is better and the prediction accuracy is accurate.
[0058] The de-normalization makes the prediction result fit the actual physical meaning, which is convenient for engineering application; the multi-dimensional evaluation index can comprehensively reflect the prediction accuracy of the model, and ensures the result to be reliable, thereby providing a reliable basis for the ground subsidence early warning of the nuclear power plant. The overall technical route is as shown in Figure 3 .
[0059] According to the above parameter setting, the deformation data of a certain nuclear power plant is analyzed, and the prediction result of the SSA-LSTM network model is as shown in Figure 4 .
[0060] The prediction results of the SSA-LSTM model and the comparative model in the embodiment of the present application are compared as shown in Figure 5 , and it can be known from Figure 5 that the prediction result of the present application is closer to the true value, so the present application has higher prediction accuracy.
[0061] The present application has the following beneficial effects compared with the prior art: The application fuses the technical advantages of sparrow search algorithm (SSA) and long short-term memory network (LSTM), optimizes the network parameters of LSTM through SSA, and constructs an SSA-LSTM prediction model. The model can efficiently search for the optimal or approximate optimal parameter combination in a complex parameter space, can shorten the optimization time and improve the solution quality, and thus effectively improves the training efficiency and prediction accuracy of the model. The model can be widely applied to the settlement prediction of typical geological disaster scenes such as the nuclear power plant area, urban surface and landslide body, provides technical support for the early warning of geological disasters, and thus guarantees the safety of people's life and property.
[0062] In addition, although the various steps of the methods in the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that the steps must be performed in this particular order, or that all of the steps shown must be performed to achieve the desired results. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, one step can be divided into multiple steps, etc.
[0063] Although some specific embodiments of the present application have been described in detail by examples, those skilled in the art should understand that the above examples are only for illustration, not for limiting the scope of the present application. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the present application.
Claims
1. A time-series InSAR method for predicting surface subsidence in nuclear power plants based on SSA-LSTM, characterized in that, The method includes the following steps: S100 acquires multi-temporal time-series InSAR monitoring data of land subsidence in the nuclear power plant area and preprocesses the multi-temporal time-series InSAR monitoring data of land subsidence. S200: Extract the preprocessed time-series InSAR monitoring data, divide it into training and test sets, and then perform normalization processing. S300 uses the Sparrow Search algorithm to optimize the hyperparameters of the Long Short-Term Memory (LSTM) network, and constructs and trains a time-series InSAR subsidence prediction model based on LSTM. S400 performs inverse normalization and integration on the predicted settlement data output by the time-series InSAR settlement prediction model, and finally outputs the time-series InSAR settlement prediction results and completes the model performance evaluation.
2. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 1, characterized in that, In step S200, the time-series InSAR monitoring data is divided into a training set and a test set in a 4:1 ratio.
3. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 1, characterized in that, In step S300, the LSTM hyperparameters optimized using the sparrow search algorithm include the initial learning rate of the LSTM network, the regularization parameter, and the number of hidden layer nodes in the LSTM structure.
4. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 1, characterized in that, In step S300, the specific process of optimizing the LSTM hyperparameters using the sparrow search algorithm includes: S310, determine the sparrow population size, maximum number of iterations, number of optimization parameters, upper and lower bounds of parameter values, and initialize the SSA optimization algorithm values; S320: Calculate the fitness value of each sparrow based on the number of sparrows in the population and update its position. Use the position iteration formula of producers, beggars and sparrows with danger awareness to update the optimal individual position and the global optimal position in the sparrow population, and save the position found in the iteration. S330: Determine if the maximum number of iterations has been reached. If so, exit the loop and obtain the optimal network parameters. Otherwise, continue the above update and iteration steps until the iteration ends and output the optimal network parameters.
5. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 1, characterized in that, In step S100, the preprocessing of multi-temporal surface subsidence time series InSAR monitoring data is performed using PS-InSAR technology. The processing steps include: master image selection, registration, interferometry, inversion, and geocoding.
6. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 5, characterized in that, In the PS-InSAR technology processing, permanent scatterer points are selected by calculating the amplitude deviation value. The amplitude deviation value is the ratio of the amplitude standard deviation to the mean. When the amplitude deviation value is less than or equal to the set threshold, the pixel point is selected as the permanent scatterer point.
7. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 1, characterized in that, In step S400, the evaluation metrics used for model performance evaluation include root mean square error, mean absolute error, and absolute mean percentage error.
8. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 1, characterized in that, In step S300, the constructed LSTM-based time-series InSAR settlement prediction model employs an adaptive moment estimator optimizer.
9. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 3, characterized in that, The initial learning rate is in the range of [1e-3, 1e-2], the regularization coefficient is in the range of [1e-4, 1e-1], and the number of hidden layer nodes is in the range of [10, 30].
10. The method for predicting surface subsidence of nuclear power plants based on SSA-LSTM according to claim 4, characterized in that, The sparrow population has a size of 20, a population dimension of 3, and a maximum number of iterations of 8.