GB-RAR deformation monitoring and early warning method based on hybrid algorithm optimized prediction model

Abstract

The application discloses a GB-RAR deformation monitoring and early warning method based on a hybrid algorithm optimized prediction model, and relates to the technical field of deformation monitoring and early warning. The method comprises the following steps: acquiring GB-RAR deformation time series data of a research object by using a ground-based real aperture radar system; constructing a neural network deformation prediction model with multidimensional environmental factor time series data as input and deformation time series data as prediction output, and training the model by using a constructed historical sample data set; performing difference processing on the deformation prediction time series data and the pretreated GB-RAR deformation time series data to obtain a prediction residual sequence and a standard deviation sequence arranged in time sequence; taking the median of the standard deviation sequence as a stability benchmark and calculating an alarm threshold corresponding to a current sliding window; and if a continuous number of sliding windows in the prediction residual sequence are all marked as structural abnormalities, determining that the continuous number of sliding windows are structural early warning events. The application realizes early warning, accurate alarm and reliable alarm of a complex research object, and improves the operation safety early warning and pre-control capability.

Description

Technical Field

[0001] This invention relates to the field of deformation monitoring and early warning of research objects, and in particular to a GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model. Background Technology

[0002] Deformation monitoring in many public places is a crucial aspect of public safety. These public places typically involve complex and multifunctional buildings (such as residential or commercial buildings, high-rise buildings, bridges, etc.) and roads. In building deformation monitoring projects, continuous deformation at the millimeter or even sub-millimeter level is often considered an early sign of collapse risk, thus requiring high accuracy, rapid response, and early warning. Deformation monitoring of buildings is a current technological challenge. Ground-based true aperture radar (GB-RAR) technology, with its high precision, high spatiotemporal resolution, and all-weather operation, is well-suited for building deformation monitoring and provides a high-precision data foundation for in-depth analysis of building dynamic behavior. Building deformation is influenced by both environmental factors (including meteorological data, load data, and construction activities; meteorological data includes wind force or direction, rainfall, temperature, humidity, sunlight, atmospheric disturbance, etc.; load data includes the gravity load from the upper floors to the lower floors and the heavy objects piled up; construction activities include construction activities around the building, such as subway or tunnel shield construction, etc.) and the building's own structural performance. Buildings may experience short-term abrupt deformation due to environmental factors, while deformation due to the building's own structural performance is generally affected by the service life of the building's steel reinforcement and structure. The impact of the building's own structural performance on building deformation is cumulative over a long period of time and is the most important aspect of building deformation monitoring and early warning. Ground-based true aperture radar (GB-RAR) technology can monitor building deformation. However, the raw GB-RAR observation data obtained is strain data under the combined conditions of current environmental factors and the building's own structural performance. Inevitably, deformation changes will occur due to short-term environmental factors. If such deformation changes are used as early warning, it is clear that the deformation changes caused by the building's own structural performance are not being acquired. Existing technology cannot provide early warning of deformation changes caused by the building's own structural performance, and cannot effectively monitor and warn of deformation changes caused by structural performance problems over time. Therefore, it is impossible to conduct effective monitoring and early warning. Summary of the Invention

[0003] The purpose of this invention is to provide a GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model. This method utilizes the deformation prediction time series data output by the neural network deformation prediction model and the GB-RAR deformation time series data for joint processing to achieve accurate perception of anomalies in the studied structure. Furthermore, it employs intelligent early warning with adaptive dynamic thresholds, enabling effective judgment and early warning of complex research object structures. This achieves early warning, accurate alarm, and reliable alarm, improving the early warning and control capabilities for operational safety, thereby ensuring the long-term safe operation and maintenance of critical infrastructure.

[0004] The objective of this invention is achieved through the following technical solution: A GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model, the method comprising: S1. Deploy a ground-based real aperture radar system at the research object location, and use the ground-based real aperture radar system to monitor and acquire the GB-RAR deformation time series data of the research object. Perform data preprocessing on the GB-RAR deformation time series data, including outlier removal and filling, and filtering.

[0005] S2. Construct a neural network deformation prediction model that takes multi-dimensional environmental factor time series data as input and deformation time series data as prediction output, and train the model using the constructed historical sample dataset. The neural network deformation prediction model outputs deformation prediction time series data.

[0006] S3. Subtract the deformation prediction time series data from the preprocessed GB-RAR deformation time series data to obtain the prediction residual sequence. Perform Fourier transform analysis on the prediction residual sequence and extract the principal frequency and principal period. The data length of the master cycle is equal to the length of the sliding window. Overlap rate is used to calculate the standard deviation of the predicted residual sequence within a sliding window, resulting in a standard deviation sequence arranged in time series. The median of the standard deviation sequence is then used as the benchmark. As a stability benchmark, the alarm threshold corresponding to the current sliding window is obtained using the following method. : , , The standard deviation of the predicted residuals for the current sliding window. , These are preset constants, This is the initial warning threshold; When the standard deviation of the predicted residuals in the predicted residual sequence is obtained from the sliding window prediction residuals Greater than the alarm threshold If the sliding window is not continuously updated, then mark it as an anomaly; if the predicted residual sequence is continuous... If all sliding windows are marked as structurally abnormal, then the continuity is determined. Each sliding window represents a structural early warning event.

[0007] To better realize this invention, the neural network deformation prediction model is constructed with a backpropagation (BP) neural network as its core. The BP neural network consists of three layers: an input layer, a hidden layer, and an output layer. The input layer is used to input time-series data of environmental factors or historical sample datasets. The hidden layer is used to extract the variable features of the input layer and pass them to the output layer. The output layer outputs the deformation prediction time-series data. The expressions for the three layers of the BP neural network are as follows:

[0008]

[0009] ,in For the output variables of the hidden layer, The activation function of the hidden layer. These are the connection weights between the input layer and the hidden layer. For the input variables of the input layer of the neural network, This is the connection threshold between the input layer and the hidden layer. For the output variables of the output layer, The activation function of the output layer. This is the connection threshold between the hidden layer and the output layer. H is the connection threshold between the hidden layer and the output layer, where H is the number of hidden layer nodes. It is the number of nodes in the input layer. It is the number of output layer nodes. It is a preset constant.

[0010] Preferably, the neural network deformation prediction model is optimized through a hybrid collaborative process of genetic algorithm (GA) and particle swarm optimization (PSO). The method includes: GA constructs a chromosome population using the model parameters of the neural network deformation prediction model as chromosome individuals and globally explores to obtain the optimal chromosome individual; the optimal chromosome individual is then used as a particle in the PSO process for local refinement and iterative processing, as shown in the following expression:

[0011] ,in Let be the velocity or displacement of particle i in the k-th iteration. Let i be the velocity or displacement of particle i in the (k-1)th iteration. Let be the velocity or displacement of particle i in the (k+1)th iteration. , , These represent inertia weight, individual learning factor, and social learning factor, respectively. Let be the individual optimal solution recorded by particle i in the (k-1)th iteration. This represents the globally optimal solution recorded by the particle swarm during the (k-1)th iteration. Let i be the position of particle i in the (k+1)th iteration. Let i be the position of particle i in the (k+1)th iteration.

[0012] Preferably, the global exploration of the genetic algorithm (GA) includes selection, crossover, and mutation operations. The selection operation uses a roulette wheel selection strategy to select chromosome individuals with low fitness. The crossover operation uses a real number crossover method to cross two chromosome individuals to generate new chromosome individuals. The mutation operation uses non-uniform mutation to mutate chromosome individuals.

[0013] Preferably, in method S1, the outlier removal and imputation method for GB-RAR deformed time series data is as follows: First, the median of the GB-RAR deformed time series data is obtained. Then, the difference between the data in the GB-RAR deformed time series data and the median is calculated, and the dispersion (MAD) is obtained according to the following expression: , To obtain GB-RAR variant timing data The median of To obtain GB-RAR variant timing data and The median of the difference This is a preset constant; Data in GB-RAR deformation time series data The data is calculated using the following formula. The corresponding determination coefficient C: If the data If the corresponding decision coefficient C is greater than the decision coefficient threshold, then the data is removed. Linear interpolation was used to fill in the removed data.

[0014] Preferably, in method S1, the filtering process uses a Savitzky-Golay filter to filter the GB-RAR deformed time series data after outlier removal and imputation according to the following formula: Where t is the position index of the current data point to be smoothed in the global sequence, h is the half-width of one side of the filtering window, and J is the index of the data points before and after the filtering window with position index t as the center point. Here is the filter function. This represents the data before and after the center point centered at position index t within the filter window.

[0015] Preferably, the initial warning threshold The expression is as follows: , It represents the maximum value in the standard deviation sequence.

[0016] Preferably, the ground-based true aperture radar system monitors the deformation of the research object, including north-south and east-west displacements. The deformation displacement data expression in the GB-RAR deformation time series data is as follows: ,in This represents the deformation and displacement data at the current moment. This is the current north-south displacement data. This represents the north-south displacement data at the initial moment. This is the east-west displacement data at the current moment. This represents the east-west displacement data at the initial moment.

[0017] Preferably, the roulette wheel selection strategy selects chromosomes with low fitness. The expression is as follows: ,in Let be the fitness value of chromosome individual i. For preset coefficients, Let be the number of chromosomes in the population. The expression for crossover of two chromosomes using the real number crossover method is as follows:

[0018] ,in The gene formed after crossing over at position j in chromosome individual K. The gene formed after crossing over at the j-th gene locus in chromosome individual i. The preset coefficients are used; the expression for the non-uniform mutation operation is as follows: ,in , They are respectively The maximum and minimum values, This indicates that the direction of variation is towards the maximum value. This indicates that the direction of variation is towards the minimum value. This represents the current iteration number. The maximum number of iterations, The parameters are used to adjust the variation in radians.

[0019] Preferably, the neural network deformation prediction model is trained using historical sample datasets, and the root mean square error (RMSE) and coefficient of determination (R²) are used to constrain the model prediction.

[0020] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) This invention utilizes the deformation prediction time series data predicted by the neural network deformation prediction model and the GB-RAR deformation time series data to achieve accurate perception of the abnormality of the research structure. It also adopts intelligent early warning with adaptive dynamic threshold to realize effective judgment and early warning of the complex research object structure, realize early warning, accurate alarm and reliable alarm, improve the early warning and control capability of operation safety, and thus realize the long-term safe operation and maintenance of major infrastructure.

[0021] (2) This invention constructs a neural network deformation prediction model with multi-dimensional environmental factor time series data as input and deformation time series data as prediction output. The model is trained using a constructed historical sample dataset. The neural network deformation prediction model is optimized by a hybrid algorithm of genetic algorithm (GA) and particle swarm optimization (PSO). The hybrid algorithm uses the global search capability of genetic algorithm (GA) to explore the solution space extensively and identify the spatial range in which a superior solution is more likely to be obtained. Then, the particle swarm optimization algorithm (PSO), which has a higher local search efficiency, is used to develop the area near the superior solution in detail. This can effectively realize the collaborative work of "exploration" and "development", which greatly improves the modeling capability of nonlinear time series. The neural network deformation prediction model is more outstanding in terms of prediction accuracy, stability and generalization performance, making the deformation time series data with multi-dimensional environmental factor time series data as input more accurate and reliable in prediction.

[0022] (3) This invention uses the residual sequence after high-precision prediction as the observation window, introduces the Fast Fourier Transform (FFT) algorithm into the early warning analysis, adopts the Fast Fourier Transform to adaptively determine the analysis scale according to the fluctuation of the fault data sequence, and adjusts the threshold of the early warning according to the real-time data fluctuation under actual conditions, ensuring that the early warning has a certain degree of flexibility. It also removes occasional interference factors through continuous triggering, transforming the early warning judgment from simple conditional logical reasoning into an intelligent judgment process that can understand the behavior pattern of the building and clearly distinguish between normal fluctuations and dangerous signs. It significantly reduces the false alarm and missed alarm rate at the source level and realizes the essential leap from delayed alarm to advanced early warning. Attached Figure Description

[0023] Figure 1 This is a flowchart of the GB-RAR deformation monitoring and early warning method of the present invention; Figure 2 This is a schematic diagram illustrating the principle of the collaborative optimization of the neural network deformation prediction model by the genetic algorithm (GA) and particle swarm optimization (PSO) in the embodiment. Figure 3 The following is a schematic diagram illustrating the north-south and east-west displacements of a research object before and after filtering, as an example in the embodiments. Figure 4The following is an example of the effect of a research object after optimization by a hybrid algorithm of genetic algorithm (GA) and particle swarm optimization (PSO) in the embodiment; Figure 5 This is a statistical effect diagram of the deformation prediction time series data output by the neural network deformation prediction model in the embodiment; Figure 6 The following diagram illustrates a case study where a structural early warning event serves as an alarm point for a research object. Detailed Implementation

[0024] The present invention will be further described in detail below with reference to embodiments: Example like Figure 1 As shown, a GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model is presented, the method comprising: S1. A ground-based real aperture radar system is deployed at the research object (in this embodiment, a super high-rise building is used as an example). The ground-based real aperture radar system is used to monitor and acquire the GB-RAR deformation time series data of the research object. The GB-RAR deformation time series data is preprocessed, including outlier removal and imputation, and filtering. In some embodiments, the outlier removal and imputation method for GB-RAR deformation time series data is as follows: First, the median of the GB-RAR deformation time series data is obtained (by utilizing the median's insensitivity to extreme values, outliers that deviate from the main data distribution can be effectively identified and removed, thereby avoiding these gross errors from having a misleading impact on subsequent model training and early warning judgment). The difference between the data in the GB-RAR deformation time series data and the median is calculated, and the dispersion (MAD) is obtained according to the following expression: , To obtain GB-RAR variant timing data The median of To obtain GB-RAR variant timing data and The median of the difference This is a preset constant (in this embodiment). The value is 1.4826.

[0025] Data in GB-RAR deformation time series data The data is calculated using the following formula. The corresponding determination coefficient C: If the data If the corresponding decision coefficient C is greater than the decision coefficient threshold (in this embodiment, the decision coefficient threshold is 2.5; the specific value of the decision coefficient threshold is determined based on the actual research object), then the data is removed. Linear interpolation was used to fill in the removed data (ensuring the continuity of the time series, significantly improving the basic quality of the monitoring data, and laying a reliable data foundation for high-precision prediction).

[0026] In some embodiments, the preprocessing of the filtering in this invention employs a Savitzky-Golay filter to filter the outlier-removed and imputed GB-RAR deformation time-series data (the outlier-removed and imputed GB-RAR deformation time-series data may still contain high-frequency random fluctuations caused by factors such as environmental thermal noise and equipment background noise. To smooth out these noises while preserving the signal characteristics representing the true deformation trend of the building to the greatest extent, this invention uses filtering processing) according to the following formula: , where t is the position index of the current data point to be smoothed in the global sequence, h is the half-width of one side of the filter window (i.e., the general width range of the filter window), and J is the index of the data points before and after the filter window with the position index t as the center point, with a value range of [−h, h]; Here is the filter function. This represents the data before and after the center point t within the filtering window. This embodiment uses a Savitzky-Golay filter for filtering, which effectively smooths noise while better preserving the signal's local extrema and width characteristics, avoiding signal distortion that may be caused by traditional low-pass filtering.

[0027] This embodiment takes a super high-rise building in the Greenland Center in Wuhan as an example. The ground-based true aperture radar system monitors the deformation of the research object, including north-south displacement and east-west displacement. The deformation displacement data expression in the GB-RAR deformation time series data is as follows: ,in This represents the deformation and displacement data at the current moment. This is the current north-south displacement data. This represents the north-south displacement data at the initial moment. This is the east-west displacement data at the current moment. The data represents the east-west displacement at the initial time. Outliers in the deformation displacement data of the GB-RAR deformation time series data were removed and imputed. Then, a Savitzky-Golay filter was used to filter the GB-RAR deformation time series data after outlier removal and imputed. In a case study of a super high-rise building in the Greenland Center in Wuhan, the deformation monitoring of the research object included north-south and east-west displacements. A comparison was made before and after SG filtering of the north-south and east-west displacements using the Savitzky-Golay filter. Figure 3As shown, analysis of ground-based radar monitoring data without SG filtering revealed significant fluctuations in the original displacement data, accompanied by numerous sharp spikes. After MAD combined with Savitzky-Golay polynomial filtering noise reduction technology, the smoothness of the displacement curve was significantly improved, and the spike phenomenon was greatly reduced. Specifically, around 500 minutes into the monitoring period, the original data produced huge outliers deviating from the normal sequence. The algorithm proposed in this application effectively removed and interpolated these outliers, improving the accuracy and rationality of the monitoring data and providing a necessary prerequisite for subsequent prediction modeling.

[0028] S2. Construct a neural network deformation prediction model that takes multi-dimensional environmental factor time-series data as input and deformation time-series data (GB-RAR deformation time-series data processed by outlier removal and imputation and filtering) as the prediction output. Train the model using a constructed historical sample dataset. The neural network deformation prediction model outputs deformation prediction time-series data. Environmental factor time-series data consists of time-series data composed of environmental factors, including meteorological data, load data, and construction activities. Meteorological data includes wind force or direction, rainfall, temperature, humidity, light intensity, atmospheric disturbances, etc. Load data includes the gravity load from the upper to the lower floors of the building and the heavy objects piled up, etc. Construction activities include construction activities around the building, such as subway or tunnel boring machine construction. The historical sample dataset is a combination of historically correlated historical environmental factor time-series data and historical deformation time-series data (the historical sample dataset includes historical environmental factor time-series data and historical deformation time-series data, which are aligned according to time sequence). Preferably, the neural network deformation prediction model is constructed with a backpropagation (BP) neural network as its core. The BP neural network consists of three layers: an input layer, a hidden layer, and an output layer. The input layer is used to input time-series data of environmental factors or historical sample datasets. The hidden layer is used to extract the variable features of the input layer and pass them to the output layer. The output layer outputs the deformation prediction time-series data. The expressions for the three layers of the BP neural network are as follows:

[0029]

[0030] ,in For the output variables of the hidden layer, The activation function of the hidden layer. These are the connection weights between the input layer and the hidden layer. For the input variables of the input layer of the neural network, This is the connection threshold between the input layer and the hidden layer. For the output variables of the output layer, The activation function of the output layer. This is the connection threshold between the hidden layer and the output layer. H is the connection threshold between the hidden layer and the output layer, where H is the number of hidden layer nodes. It is the number of nodes in the input layer. It is the number of output layer nodes. It is a preset constant.

[0031] In some embodiments, such as Figure 2 As shown, the neural network deformation prediction model is optimized using a hybrid approach combining Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). This hybrid approach effectively addresses the problems of BP neural networks, such as susceptibility to local minima, slow convergence, and sensitivity to initial weights. The methods include: Figure 2 As shown, the Genetic Algorithm (GA) constructs a chromosome population using the model parameters (including weight and threshold parameters) of the neural network deformation prediction model and globally explores to obtain the optimal chromosome individual. The global exploration of the GA algorithm includes selection, crossover, and mutation operations. The selection operation uses a roulette wheel selection strategy to choose chromosome individuals with low fitness. The probability of selecting chromosome individuals with low fitness using the roulette wheel selection strategy is [not specified]. The expression is as follows: ,in Let be the fitness value of chromosome individual i. For preset coefficients, Let be the number of chromosomes in the population. The crossover operation uses the real-number crossover method to generate a new chromosome from two individual chromosomes. The expression for the real-number crossover method for two individual chromosomes is as follows:

[0032] ,in The gene formed after crossing over at position j in chromosome individual K. The gene formed after crossing over at the j-th gene locus in chromosome individual i. These are preset coefficients. The mutation operation uses non-uniform mutation to process individual chromosome variations. The expression for the non-uniform mutation operation is as follows: ,in , They are respectively The maximum and minimum values, This indicates that the direction of variation is towards the maximum value. This indicates that the direction of variation is towards the minimum value. This represents the current iteration number. The maximum number of iterations, The parameters are used to adjust the degree of mutation. The global exploration of the Genetic Algorithm (GA) through selection, crossover, and mutation operations can extensively search for potential advantageous regions in the solution space, providing a high-quality initial solution group for subsequent optimization.

[0033] Particle Swarm Optimization (PSO) is used in the local refinement stage of neural network deformation prediction models, such as... Figure 2 As shown, the optimal chromosome individuals obtained in the global exploration phase of the genetic algorithm (GA) are used as particles in the particle swarm optimization (PSO) process (each particle represents a chromosome individual, i.e., represents the model network parameters) for local fine-tuning iteration. The expression is as follows:

[0034] ,in Let be the velocity or displacement of particle i in the k-th iteration. Let i be the velocity or displacement of particle i in the (k-1)th iteration. Let be the velocity or displacement of particle i in the (k+1)th iteration. , , These represent inertia weight, individual learning factor, and social learning factor, respectively. Let be the individual optimal solution recorded by particle i in the (k-1)th iteration. This represents the globally optimal solution recorded by the particle swarm during the (k-1)th iteration. Let i be the position of particle i in the (k+1)th iteration. Let represent the position of particle i in the (k+1)th iteration. This invention employs a hybrid algorithm combining Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), fully leveraging the strong global exploration capability of GA and the efficient local exploration capability of PSO. This effectively solves the problems of BP neural networks easily getting trapped in local minima, slow convergence speed, and sensitivity to initial weights in neural network deformation prediction models. The neural network deformation prediction model is trained using historical sample datasets, and the root mean square error (RMSE) and coefficient of determination (R²) are used for model prediction constraints. The expression for RMSE is as follows: The coefficient of determination R² is expressed as follows: n is the number of samples. Let i be the true value of the i-th sample. Let be the predicted value for the i-th sample. The root mean square error (RMSE) is the average value of the entire sample. A smaller RMSE indicates higher prediction accuracy; the coefficient of determination... The closer the value is to 1, the more variation the model explains, meaning the better the predictive performance of the neural network deformation prediction model. In the case study of a super high-rise building in the Greenland Center in Wuhan, for example... Figure 4As shown, a dataset from N monitoring periods was used for predictive analysis, with 70% used for training and 30% for testing. To minimize the network error of the neural network deformation prediction model, the maximum number of iterations of the BP neural network was set to 500, the number of particles in the Particle Swarm Optimization (PSO) algorithm was set to 30, and the inertia weight was initially set from 0.9 and then linearly decreased to 0.4 to facilitate global search in the early stage and fine-grained local search in the later stage. The individual and social learning factors were set to 2, the maximum velocity was set to 1, the minimum velocity to -1, the crossover probability to 0.8, and the mutation probability to 0.1. Two optimal individuals were retained and directly entered into the next generation for iterative operation. Through the training and iteration of the neural network deformation prediction model, the root mean square error (RMSE) and the coefficient of determination (R²) were used to evaluate and constrain the model's prediction accuracy. In the case of a super high-rise building in the Greenland Center in Wuhan, the prediction accuracy evaluation of the neural network deformation prediction model is shown in Table 1.

[0035] By comparing the evaluation accuracy of the models, it can be fully demonstrated that the model is relatively suitable for the engineering projects of this application and exhibits extremely high accuracy. This not only verifies the model's prediction accuracy but also provides a guarantee for the subsequent selection of warning thresholds. The trained neural network deformation prediction model outputs deformation prediction time-series data, see [link to relevant documentation]. Figure 5 As shown, the neural network deformation prediction model takes multi-dimensional environmental factor time series data as input and predicts deformation prediction time series data as output.

[0036] S3. Subtract the deformation prediction time series data (i.e., the deformation prediction time series data obtained in step S2) from the preprocessed GB-RAR deformation time series data to obtain the prediction residual sequence. For the predicted residual sequence By performing Fourier transform analysis and extracting the principal frequency and principal period, this invention predicts the residual sequence. The sequence is segmented based on characteristic patterns (preferably periodic patterns of fluctuation periods), and then Fourier transform analysis is performed on each segment. For example, Fourier transform analysis is performed on the prediction residual time series of a certain segment in the prediction residual sequence to extract the dominant energy frequency. The corresponding principal period is This invention is based on The data length of the master cycle is equal to the length of the sliding window. The overlap rate is used to calculate the standard deviation of the predicted residual sequence within a sliding window. (Used to quantify the degree of fluctuation in prediction error during this period) The calculation yields a standard deviation series arranged in time sequence. In the case of a super high-rise building in the Greenland Center in Wuhan, The value is set to 2 (so that the length of the sliding window covers two complete fluctuation cycles). The value is 0.5 (i.e., 50%).

[0037] In each sliding window, the median of the series is taken as the standard deviation. As a stability benchmark, the alarm threshold corresponding to the current sliding window is obtained using the following method. : , , The standard deviation of the predicted residuals for the current sliding window. , These are preset constants, This is the initial warning threshold. Preferably, the initial warning threshold... The expression is as follows: , The maximum value in the standard deviation sequence is used in the case of a super high-rise building in the Greenland Center in Wuhan. The value is set to 3 (based on the sigma principle, since 3 times the standard deviation can cover approximately 99.7% of the fluctuating data, in order to reduce the false alarm rate and enhance the identification of outliers exceeding the normal range, this embodiment sets the initial warning threshold to 3). Set to 3 times the maximum standard deviation, that is: ).

[0038] When the standard deviation of the predicted residuals in the predicted residual sequence is obtained from the sliding window prediction residuals Greater than the alarm threshold If so, the sliding window is marked as an anomaly; further, if the predicted residual sequence is continuous... If all sliding windows are marked as structurally abnormal, then the continuity is determined. Each sliding window represents a structural early warning event. The deformation prediction time series data output by the neural network deformation prediction model is the prediction data under the condition that the structure under study has not experienced any abnormalities (mainly caused by environmental factors in the time series data). If the deformation prediction time series data is subtracted from the preprocessed GB-RAR deformation time series data, the standard deviation of the prediction residual sequence is obtained. Greater than the alarm threshold This indicates a structural anomaly; if it is continuous... If all sliding windows show structural anomalies, then the research object has experienced a structural anomaly, and this is determined to be a structural warning event. This invention utilizes the deformation prediction time-series data predicted by a neural network deformation prediction model, combined with GB-RAR deformation time-series data, to achieve accurate perception of structural anomalies. An adaptive dynamic threshold (i.e., alarm threshold) is employed. Intelligent early warning systems enable effective analysis and early warning of complex research objects (such as buildings), achieving early warning, accurate alarms, and reliable alarms, thus improving operational safety early warning and control capabilities and ensuring the long-term safe operation and maintenance of major infrastructure. For example... Figure 6 As shown in the example diagram, this embodiment illustrates a research object with structural early warning events as alarm points. The alarm point first appears at the 924th epoch of the index, and subsequent alarms are issued continuously due to multiple exceeding of limits. After removing duplicate data in overlapping windows, the model alarms at the 2343-minute mark of the deformation dataset, at which point the deformation value is 8.61 mm. In the progressive test setting, progressive deformation starts at the 2146-minute mark. This indicates that a danger signal can be issued approximately 3.28 hours after the building shows a dangerous trend, and the deformation value at this time is approximately 0.963% of the deformation limit, achieving the effect of early warning and providing sufficient time for building maintenance and personnel evacuation, thus reducing the probability of safety accidents.

[0039] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model, characterized in that: The methods include: S1. Deploy a ground-based real aperture radar system at the research object location, use the ground-based real aperture radar system to monitor and acquire GB-RAR deformation time series data of the research object, and perform data preprocessing on the GB-RAR deformation time series data, including outlier removal and filling, and filtering. S2. Construct a neural network deformation prediction model with multi-dimensional environmental factor time series data as input and deformation time series data as prediction output. Train the model using the constructed historical sample dataset. The neural network deformation prediction model outputs deformation prediction time series data. S3. Subtract the deformation prediction time series data from the preprocessed GB-RAR deformation time series data to obtain the prediction residual sequence. Perform Fourier transform analysis on the prediction residual sequence and extract the principal frequency and principal period. The data length of the master cycle is equal to the length of the sliding window. Overlap rate is used to calculate the standard deviation of the predicted residual sequence within a sliding window, resulting in a standard deviation sequence arranged in time series. The median of the standard deviation sequence is then used as the benchmark. As a stability benchmark, the alarm threshold corresponding to the current sliding window is obtained using the following method. : , , The standard deviation of the predicted residuals for the current sliding window. , These are preset constants, This is the initial warning threshold; When the standard deviation of the predicted residuals in the predicted residual sequence is obtained from the sliding window prediction residuals Greater than the alarm threshold If the sliding window is not continuously updated, then mark it as an anomaly; if the predicted residual sequence is continuous... If all sliding windows are marked as structurally abnormal, then the continuity is determined. Each sliding window represents a structural early warning event.

2. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 1, characterized in that: The neural network deformation prediction model is built around a backpropagation (BP) neural network, which consists of three layers: an input layer, a hidden layer, and an output layer. The input layer is used to input time-series data of environmental factors or historical sample datasets. The hidden layer is used to extract the variable features of the input layer and pass them to the output layer. The output layer outputs the deformation prediction time-series data. The expressions for the three layers of the BP neural network are as follows: ； ； ,in For the output variables of the hidden layer, The activation function of the hidden layer. These are the connection weights between the input layer and the hidden layer. For the input variables of the input layer of the neural network, This is the connection threshold between the input layer and the hidden layer. For the output variables of the output layer, The activation function of the output layer. This is the connection threshold between the hidden layer and the output layer. H is the connection threshold between the hidden layer and the output layer, where H is the number of hidden layer nodes. It is the number of nodes in the input layer. It is the number of output layer nodes. It is a preset constant.

3. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 1 or 2, characterized in that: The neural network deformation prediction model is optimized through a hybrid collaborative process using genetic algorithm (GA) and particle swarm optimization (PSO). The method includes: GA constructs a chromosome population using the model parameters of the neural network deformation prediction model as chromosome individuals and globally explores to obtain the optimal chromosome individual; the optimal chromosome individual is then used as a particle in the PSO process for local refinement and iterative optimization, as shown in the following expression: ； ,in Let be the velocity or displacement of particle i in the k-th iteration. Let i be the velocity or displacement of particle i in the (k-1)th iteration. Let be the velocity or displacement of particle i in the (k+1)th iteration. , , These represent inertia weight, individual learning factor, and social learning factor, respectively. Let be the individual optimal solution recorded by particle i in the (k-1)th iteration. This represents the globally optimal solution recorded by the particle swarm during the (k-1)th iteration. Let i be the position of particle i in the (k+1)th iteration. Let i be the position of particle i in the (k+1)th iteration.

4. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 3, characterized in that: The global exploration of the genetic algorithm GA includes selection, crossover, and mutation operations. The selection operation uses a roulette wheel selection strategy to select chromosome individuals with low fitness. The crossover operation uses a real number crossover method to cross two chromosome individuals to generate new chromosome individuals. The mutation operation uses non-uniform mutation to mutate chromosome individuals.

5. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 1, characterized in that: In method S1, the outlier removal and imputation method for GB-RAR deformed time series data is as follows: First, the median of the GB-RAR deformed time series data is obtained. Then, the difference between the median and the data in the GB-RAR deformed time series data is calculated, and the dispersion (MAD) is obtained according to the following expression: , To obtain GB-RAR variant timing data The median of To obtain GB-RAR variant timing data and The median of the difference This is a preset constant; Data in GB-RAR deformation time series data The data is calculated using the following formula. The corresponding determination coefficient C: If the data If the corresponding decision coefficient C is greater than the decision coefficient threshold, then the data is removed. Linear interpolation was used to fill in the removed data.

6. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 1 or 5, characterized in that: In method S1, the Savitzky-Golay filter is used to filter the GB-RAR deformed time series data after outlier removal and imputation according to the following formula: Where t is the position index of the current data point to be smoothed in the global sequence, h is the half-width of one side of the filtering window, and J is the index of the data points before and after the filtering window with position index t as the center point. Here is the filter function. This represents the data before and after the center point centered at position index t within the filter window.

7. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 1, characterized in that: Initial warning threshold The expression is as follows: , It represents the maximum value in the standard deviation sequence.

8. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 1, characterized in that: The ground-based true aperture radar system monitors the deformation of the research object, including north-south and east-west displacements. The deformation displacement data expression in the GB-RAR deformation time series data is as follows: ,in This represents the deformation and displacement data at the current moment. This is the current north-south displacement data. This represents the north-south displacement data at the initial moment. This is the east-west displacement data at the current moment. This represents the east-west displacement data at the initial moment.

9. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 4, characterized in that: The roulette wheel selection strategy selects chromosomes with low fitness. The expression is as follows: ,in Let be the fitness value of chromosome individual i. For preset coefficients, Let be the number of chromosomes in the population. The expression for crossover of two chromosomes using the real number crossover method is as follows: ； ,in The gene formed after crossing over at position j in chromosome individual K. The gene formed after crossing over at the j-th gene locus in chromosome individual i. The preset coefficients are used; the expression for the non-uniform mutation operation is as follows: ,in , They are respectively The maximum and minimum values, This indicates that the direction of variation is towards the maximum value. This indicates that the direction of variation is towards the minimum value. This represents the current iteration number. The maximum number of iterations, The parameters are used to adjust the variation in radians.

10. The GB-RAR deformation monitoring and early warning method based on a hybrid algorithm-optimized prediction model according to claim 1, characterized in that: The neural network deformation prediction model is trained using historical sample datasets, and the root mean square error (RMSE) and coefficient of determination (R²) are used to constrain the model prediction.

Images